d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
7eb884645c
freebsd-skq/sys/dev/e1000/if_igb.c

6387 lines
176 KiB
C
Raw Normal View History

This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/******************************************************************************
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
Copyright (c) 2001-2013, Intel Corporation
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
1. Redistributions of source code must retain the above copyright notice,
this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
3. Neither the name of the Intel Corporation nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "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 COPYRIGHT OWNER OR CONTRIBUTORS 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.
******************************************************************************/
/*$FreeBSD$*/
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
Re-add opt_inet.h, as we did in r193862 and lost yet again. Approved by: re (kib)
2009-07-14 19:32:36 +00:00
#include "opt_inet.h"
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
#include "opt_inet6.h"
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
#include "opt_rss.h"
Change the E1000 driver option header handling to match the ixgbe driver. As it was, when building them as a module INET and INET6 are not defined. In these drivers it does not cause a panic, however it does result in different behavior in the ioctl routine when you are using a module vs static, and I think the behavior should be the same. MFC after: 3 days
2013-07-12 22:36:26 +00:00
#ifdef HAVE_KERNEL_OPTION_HEADERS
#include "opt_device_polling.h"
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#include "opt_altq.h"
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
#endif
#include <sys/param.h>
#include <sys/systm.h>
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
#include <sys/buf_ring.h>
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
#include <sys/bus.h>
#include <sys/endian.h>
#include <sys/kernel.h>
#include <sys/kthread.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/module.h>
#include <sys/rman.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
#include <sys/eventhandler.h>
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
#include <sys/pcpu.h>
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
#include <sys/smp.h>
#include <machine/smp.h>
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
#include <machine/bus.h>
#include <machine/resource.h>
#include <net/bpf.h>
#include <net/ethernet.h>
#include <net/if.h>
The r48589 promised to remove implicit inclusion of if_var.h soon. Prepare to this event, adding if_var.h to files that do need it. Also, include all includes that now are included due to implicit pollution via if_var.h Sponsored by: Netflix Sponsored by: Nginx, Inc.
2013-10-26 17:58:36 +00:00
#include <net/if_var.h>
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
#include <net/if_arp.h>
#include <net/if_dl.h>
#include <net/if_media.h>
#include <net/if_types.h>
#include <net/if_vlan_var.h>
#include <netinet/in_systm.h>
#include <netinet/in.h>
#include <netinet/if_ether.h>
#include <netinet/ip.h>
#include <netinet/ip6.h>
#include <netinet/tcp.h>
Include netinet/tcp_lro.h, unbreak the build
2008-07-30 23:54:23 +00:00
#include <netinet/tcp_lro.h>
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
#include <netinet/udp.h>
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
#ifdef RSS
#include <netinet/in_rss.h>
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
#include <machine/in_cksum.h>
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
#include <dev/led/led.h>
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
#include <dev/pci/pcivar.h>
#include <dev/pci/pcireg.h>
#include "e1000_api.h"
#include "e1000_82575.h"
#include "if_igb.h"
/*********************************************************************
* Set this to one to display debug statistics
*********************************************************************/
int igb_display_debug_stats = 0;
/*********************************************************************
* Driver version:
*********************************************************************/
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
char igb_driver_version[] = "version - 2.4.0";
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
* PCI Device ID Table
*
* Used by probe to select devices to load on
* Last field stores an index into e1000_strings
* Last entry must be all 0s
*
* { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index }
*********************************************************************/
static igb_vendor_info_t igb_vendor_info_array[] =
{
{ 0x8086, E1000_DEV_ID_82575EB_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82575EB_FIBER_SERDES,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82575GB_QUAD_COPPER,
PCI_ANY_ID, PCI_ANY_ID, 0},
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
{ 0x8086, E1000_DEV_ID_82576, PCI_ANY_ID, PCI_ANY_ID, 0},
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
{ 0x8086, E1000_DEV_ID_82576_NS, PCI_ANY_ID, PCI_ANY_ID, 0},
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
{ 0x8086, E1000_DEV_ID_82576_NS_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0},
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
{ 0x8086, E1000_DEV_ID_82576_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82576_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0},
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
{ 0x8086, E1000_DEV_ID_82576_SERDES_QUAD,
PCI_ANY_ID, PCI_ANY_ID, 0},
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
{ 0x8086, E1000_DEV_ID_82576_QUAD_COPPER,
PCI_ANY_ID, PCI_ANY_ID, 0},
Update code from Intel: - Sync shared code with Intel internal - New client chipset support added - em driver - fixes to 82574, limit queues to 1 but use MSIX - em driver - large changes in TX checksum offload and tso code, thanks to yongari. - some small changes for watchdog issues. - igb driver - local timer watchdog code was missing locking this and a couple other watchdog related fixes. - bug in rx discard found by Andrew Boyer, check for null pointer MFC: a week
2010-09-28 00:13:15 +00:00
{ 0x8086, E1000_DEV_ID_82576_QUAD_COPPER_ET2,
PCI_ANY_ID, PCI_ANY_ID, 0},
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
{ 0x8086, E1000_DEV_ID_82576_VF, PCI_ANY_ID, PCI_ANY_ID, 0},
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
{ 0x8086, E1000_DEV_ID_82580_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82580_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82580_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82580_SGMII, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82580_COPPER_DUAL,
PCI_ANY_ID, PCI_ANY_ID, 0},
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
{ 0x8086, E1000_DEV_ID_82580_QUAD_FIBER,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_DH89XXCC_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_DH89XXCC_SGMII, PCI_ANY_ID, PCI_ANY_ID, 0},
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
{ 0x8086, E1000_DEV_ID_DH89XXCC_SFP, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_DH89XXCC_BACKPLANE,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I350_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I350_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I350_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I350_SGMII, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I350_VF, PCI_ANY_ID, PCI_ANY_ID, 0},
Sync with Intel internal source: shared code update and small changes in core required Add support for new i210/i211 devices Improve queue calculation based on mac type MFC after:5 days
2012-07-05 20:26:57 +00:00
{ 0x8086, E1000_DEV_ID_I210_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I210_COPPER_IT, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I210_COPPER_OEM1,
PCI_ANY_ID, PCI_ANY_ID, 0},
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
{ 0x8086, E1000_DEV_ID_I210_COPPER_FLASHLESS,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I210_SERDES_FLASHLESS,
PCI_ANY_ID, PCI_ANY_ID, 0},
Sync with Intel internal source: shared code update and small changes in core required Add support for new i210/i211 devices Improve queue calculation based on mac type MFC after:5 days
2012-07-05 20:26:57 +00:00
{ 0x8086, E1000_DEV_ID_I210_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I210_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I210_SGMII, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I211_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
{ 0x8086, E1000_DEV_ID_I354_BACKPLANE_1GBPS,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_I354_SGMII, PCI_ANY_ID, PCI_ANY_ID, 0},
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* required last entry */
{ 0, 0, 0, 0, 0}
};
/*********************************************************************
* Table of branding strings for all supported NICs.
*********************************************************************/
static char *igb_strings[] = {
"Intel(R) PRO/1000 Network Connection"
};
/*********************************************************************
* Function prototypes
*********************************************************************/
static int igb_probe(device_t);
static int igb_attach(device_t);
static int igb_detach(device_t);
static int igb_shutdown(device_t);
static int igb_suspend(device_t);
static int igb_resume(device_t);
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
static int igb_mq_start(struct ifnet *, struct mbuf *);
drbr_enqueue() awlays consumes mbuf, no matter did it fail or not. The mbuf pointer is no longer valid, so can't be reused after. Fix igb_mq_start() where mbuf pointer was used after drbr_enqueue(). This eventually leads us to all invocations of igb_mq_start_locked() called with third argument as NULL. This allows us to simplify this function. Submitted by: Karim Fodil-Lemelin <fodillemlinkarim gmail.com> Reviewed by: jfv
2012-11-26 20:03:57 +00:00
static int igb_mq_start_locked(struct ifnet *, struct tx_ring *);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
static void igb_qflush(struct ifnet *);
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
static void igb_deferred_mq_start(void *, int);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#else
static void igb_start(struct ifnet *);
static void igb_start_locked(struct tx_ring *, struct ifnet *ifp);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static int igb_ioctl(struct ifnet *, u_long, caddr_t);
- Provide igb_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-19 11:49:41 +00:00
static uint64_t igb_get_counter(if_t, ift_counter);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static void igb_init(void *);
static void igb_init_locked(struct adapter *);
static void igb_stop(void *);
static void igb_media_status(struct ifnet *, struct ifmediareq *);
static int igb_media_change(struct ifnet *);
static void igb_identify_hardware(struct adapter *);
static int igb_allocate_pci_resources(struct adapter *);
static int igb_allocate_msix(struct adapter *);
static int igb_allocate_legacy(struct adapter *);
static int igb_setup_msix(struct adapter *);
static void igb_free_pci_resources(struct adapter *);
static void igb_local_timer(void *);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
static void igb_reset(struct adapter *);
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
static int igb_setup_interface(device_t, struct adapter *);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static int igb_allocate_queues(struct adapter *);
static void igb_configure_queues(struct adapter *);
static int igb_allocate_transmit_buffers(struct tx_ring *);
static void igb_setup_transmit_structures(struct adapter *);
static void igb_setup_transmit_ring(struct tx_ring *);
static void igb_initialize_transmit_units(struct adapter *);
static void igb_free_transmit_structures(struct adapter *);
static void igb_free_transmit_buffers(struct tx_ring *);
static int igb_allocate_receive_buffers(struct rx_ring *);
static int igb_setup_receive_structures(struct adapter *);
static int igb_setup_receive_ring(struct rx_ring *);
static void igb_initialize_receive_units(struct adapter *);
static void igb_free_receive_structures(struct adapter *);
static void igb_free_receive_buffers(struct rx_ring *);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
static void igb_free_receive_ring(struct rx_ring *);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static void igb_enable_intr(struct adapter *);
static void igb_disable_intr(struct adapter *);
static void igb_update_stats_counters(struct adapter *);
static bool igb_txeof(struct tx_ring *);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
static __inline void igb_rx_discard(struct rx_ring *, int);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
static __inline void igb_rx_input(struct rx_ring *,
struct ifnet *, struct mbuf *, u32);
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
static bool igb_rxeof(struct igb_queue *, int, int *);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
static void igb_rx_checksum(u32, struct mbuf *, u32);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
static int igb_tx_ctx_setup(struct tx_ring *,
struct mbuf *, u32 *, u32 *);
static int igb_tso_setup(struct tx_ring *,
struct mbuf *, u32 *, u32 *);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static void igb_set_promisc(struct adapter *);
static void igb_disable_promisc(struct adapter *);
static void igb_set_multi(struct adapter *);
static void igb_update_link_status(struct adapter *);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
static void igb_refresh_mbufs(struct rx_ring *, int);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
static void igb_register_vlan(void *, struct ifnet *, u16);
static void igb_unregister_vlan(void *, struct ifnet *, u16);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
static void igb_setup_vlan_hw_support(struct adapter *);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static int igb_xmit(struct tx_ring *, struct mbuf **);
static int igb_dma_malloc(struct adapter *, bus_size_t,
struct igb_dma_alloc *, int);
static void igb_dma_free(struct adapter *, struct igb_dma_alloc *);
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
static int igb_sysctl_nvm_info(SYSCTL_HANDLER_ARGS);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static void igb_print_nvm_info(struct adapter *);
static int igb_is_valid_ether_addr(u8 *);
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
static void igb_add_hw_stats(struct adapter *);
static void igb_vf_init_stats(struct adapter *);
static void igb_update_vf_stats_counters(struct adapter *);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Management and WOL Support */
static void igb_init_manageability(struct adapter *);
static void igb_release_manageability(struct adapter *);
static void igb_get_hw_control(struct adapter *);
static void igb_release_hw_control(struct adapter *);
static void igb_enable_wakeup(device_t);
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
static void igb_led_func(void *, int);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static int igb_irq_fast(void *);
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
static void igb_msix_que(void *);
static void igb_msix_link(void *);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
static void igb_handle_que(void *context, int pending);
Add a link tasklet so updates can be sleepable.
2010-01-27 20:12:04 +00:00
static void igb_handle_link(void *context, int pending);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
static void igb_handle_link_locked(struct adapter *);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
static void igb_set_sysctl_value(struct adapter *, const char *,
const char *, int *, int);
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
static int igb_set_flowcntl(SYSCTL_HANDLER_ARGS);
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
static int igb_sysctl_dmac(SYSCTL_HANDLER_ARGS);
Change the interface to the Energy Efficient Ethernet (EEE) setting in the igb and em driver. This was necessitated by a shared code change that I was given late in the game, a data type changed from bool to int, in the last update I dealt with it by a cast, but it was pointed out (thanks jhb) that there was a potential problem with this. John suggested this safer approach, and it is fine with me... MFC after:2 days (to catch the 9.1 update)
2012-07-07 20:21:05 +00:00
static int igb_sysctl_eee(SYSCTL_HANDLER_ARGS);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#ifdef DEVICE_POLLING
static poll_handler_t igb_poll;
#endif /* POLLING */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
* FreeBSD Device Interface Entry Points
*********************************************************************/
static device_method_t igb_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, igb_probe),
DEVMETHOD(device_attach, igb_attach),
DEVMETHOD(device_detach, igb_detach),
DEVMETHOD(device_shutdown, igb_shutdown),
DEVMETHOD(device_suspend, igb_suspend),
DEVMETHOD(device_resume, igb_resume),
Use DEVMETHOD_END macro defined in sys/bus.h instead of {0, 0} sentinel on device_method_t arrays Reviewed by: cognet Approved by: cognet
2013-01-30 18:01:20 +00:00
DEVMETHOD_END
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
};
static driver_t igb_driver = {
"igb", igb_methods, sizeof(struct adapter),
};
static devclass_t igb_devclass;
DRIVER_MODULE(igb, pci, igb_driver, igb_devclass, 0, 0);
MODULE_DEPEND(igb, pci, 1, 1, 1);
MODULE_DEPEND(igb, ether, 1, 1, 1);
/*********************************************************************
* Tunable default values.
*********************************************************************/
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
static SYSCTL_NODE(_hw, OID_AUTO, igb, CTLFLAG_RD, 0, "IGB driver parameters");
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
/* Descriptor defaults */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static int igb_rxd = IGB_DEFAULT_RXD;
static int igb_txd = IGB_DEFAULT_TXD;
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
SYSCTL_INT(_hw_igb, OID_AUTO, rxd, CTLFLAG_RDTUN, &igb_rxd, 0,
"Number of receive descriptors per queue");
SYSCTL_INT(_hw_igb, OID_AUTO, txd, CTLFLAG_RDTUN, &igb_txd, 0,
"Number of transmit descriptors per queue");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
/*
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
** AIM: Adaptive Interrupt Moderation
** which means that the interrupt rate
** is varied over time based on the
** traffic for that interrupt vector
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
*/
static int igb_enable_aim = TRUE;
Pull in r267961 and r267973 again. Fix for issues reported will follow.
2014-06-28 03:56:17 +00:00
SYSCTL_INT(_hw_igb, OID_AUTO, enable_aim, CTLFLAG_RWTUN, &igb_enable_aim, 0,
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
"Enable adaptive interrupt moderation");
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/*
* MSIX should be the default for best performance,
* but this allows it to be forced off for testing.
*/
static int igb_enable_msix = 1;
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
SYSCTL_INT(_hw_igb, OID_AUTO, enable_msix, CTLFLAG_RDTUN, &igb_enable_msix, 0,
"Enable MSI-X interrupts");
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/*
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
** Tuneable Interrupt rate
*/
static int igb_max_interrupt_rate = 8000;
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
SYSCTL_INT(_hw_igb, OID_AUTO, max_interrupt_rate, CTLFLAG_RDTUN,
&igb_max_interrupt_rate, 0, "Maximum interrupts per second");
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
#ifndef IGB_LEGACY_TX
This fixes a out-of-order problem with several of the newer drivers. The basic problem was that the driver was pulling the mbuf off the drbr ring and then when sending with xmit(), encounting a full transmit ring. Thus the lower layer xmit() function would return an error, and the drivers would then append the data back on to the ring. For TCP this is a horrible scenario sure to bring on a fast-retransmit. The fix is to use drbr_peek() to pull the data pointer but not remove it from the ring. If it fails then we either call the new drbr_putback or drbr_advance method. Advance moves it forward (we do this sometimes when the xmit() function frees the mbuf). When we succeed we always call advance. The putback will always copy the mbuf back to the top of the ring. Note that the putback *cannot* be used with a drbr_dequeue() only with drbr_peek(). We most of the time, in putback, would not need to copy it back since most likey the mbuf is still the same, but sometimes xmit() functions will change the mbuf via a pullup or other call. So the optimial case for the single consumer is to always copy it back. If we ever do a multiple_consumer (for lagg?) we will need a test and atomic in the put back possibly a seperate putback_mc() in the ring buf. Reviewed by: jhb@freebsd.org, jlv@freebsd.org
2013-02-07 15:20:54 +00:00
/*
** Tuneable number of buffers in the buf-ring (drbr_xxx)
*/
static int igb_buf_ring_size = IGB_BR_SIZE;
SYSCTL_INT(_hw_igb, OID_AUTO, buf_ring_size, CTLFLAG_RDTUN,
&igb_buf_ring_size, 0, "Size of the bufring");
#endif
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
/*
** Header split causes the packet header to
** be dma'd to a seperate mbuf from the payload.
** this can have memory alignment benefits. But
** another plus is that small packets often fit
** into the header and thus use no cluster. Its
** a very workload dependent type feature.
*/
Consistently use types in e1000 driver code: - Two struct members eee_disable are used in a function that expects an int *, so declare them int, not bool. - igb_tx_ctx_setup() returns a boolean value, so declare it bool, not int. - igb_header_split is passed to TUNABLE_INT, so delcare it int, not bool. - igb_tso_setup() returns a bool, so declare it bool, not boolean_t. - Do not re-define bool/true/false if the symbols already exist. MFC after: 2 weeks Sponsored by: Isilon Systems, LLC
2011-12-12 18:27:34 +00:00
static int igb_header_split = FALSE;
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
SYSCTL_INT(_hw_igb, OID_AUTO, header_split, CTLFLAG_RDTUN, &igb_header_split, 0,
"Enable receive mbuf header split");
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
This delta has a few important items: PR 122839 is fixed in both em and in igb Second, the issue on building modules since the static kernel build changes is now resolved. I was not able to get the fancier directory hierarchy working, but this works, both em and igb build as modules now. Third, there is now support in em for two new NICs, Hartwell (or 82574) is a low cost PCIE dual port adapter that has MSIX, for this release it uses 3 vectors only, RX, TX, and LINK. In the next release I will add a second TX and RX queue. Also, there is support here for ICH10, the followon to ICH9. Both of these are early releases, general availability will follow soon. Fourth: On Hartwell and ICH10 we now have IEEE 1588 PTP support, I have implemented this in a provisional way so that early adopters may try and comment on the functionality. The IOCTL structure may change. This feature is off by default, you need to edit the Makefile and add the EM_TIMESYNC define to get the code. Enjoy all!!
2008-04-25 21:19:41 +00:00
/*
Correct the multicast handling in the E1000 drivers as was done in ixgbe, thanks to Mike Karels for this fix. When exiting promiscuous mode MPE bit was being unconditionally cleared, this should not be done if we are in MAX multicast groups.
2013-04-03 23:39:54 +00:00
** This will autoconfigure based on the
** number of CPUs and max supported
** MSIX messages if left at 0.
This delta has a few important items: PR 122839 is fixed in both em and in igb Second, the issue on building modules since the static kernel build changes is now resolved. I was not able to get the fancier directory hierarchy working, but this works, both em and igb build as modules now. Third, there is now support in em for two new NICs, Hartwell (or 82574) is a low cost PCIE dual port adapter that has MSIX, for this release it uses 3 vectors only, RX, TX, and LINK. In the next release I will add a second TX and RX queue. Also, there is support here for ICH10, the followon to ICH9. Both of these are early releases, general availability will follow soon. Fourth: On Hartwell and ICH10 we now have IEEE 1588 PTP support, I have implemented this in a provisional way so that early adopters may try and comment on the functionality. The IOCTL structure may change. This feature is off by default, you need to edit the Makefile and add the EM_TIMESYNC define to get the code. Enjoy all!!
2008-04-25 21:19:41 +00:00
*/
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
static int igb_num_queues = 0;
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
SYSCTL_INT(_hw_igb, OID_AUTO, num_queues, CTLFLAG_RDTUN, &igb_num_queues, 0,
"Number of queues to configure, 0 indicates autoconfigure");
Modify the binding of queues to attach to as many CPUs as possible when using more than one igb(4) adapter. This means that queues will not be bound to the same CPUs if there are more CPUs availble. This is only applicable to a system that has multiple interfaces. Obtained from: Yahoo! Inc. MFC after: 3 days
2012-05-10 00:00:28 +00:00
/*
** Global variable to store last used CPU when binding queues
** to CPUs in igb_allocate_msix. Starts at CPU_FIRST and increments when a
** queue is bound to a cpu.
*/
static int igb_last_bind_cpu = -1;
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
/* How many packets rxeof tries to clean at a time */
static int igb_rx_process_limit = 100;
SYSCTL_INT(_hw_igb, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
&igb_rx_process_limit, 0,
"Maximum number of received packets to process at a time, -1 means unlimited");
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP /* see ixgbe.c for details */
#include <dev/netmap/if_igb_netmap.h>
#endif /* DEV_NETMAP */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
* Device identification routine
*
* igb_probe determines if the driver should be loaded on
* adapter based on PCI vendor/device id of the adapter.
*
* return BUS_PROBE_DEFAULT on success, positive on failure
*********************************************************************/
static int
igb_probe(device_t dev)
{
char adapter_name[60];
uint16_t pci_vendor_id = 0;
uint16_t pci_device_id = 0;
uint16_t pci_subvendor_id = 0;
uint16_t pci_subdevice_id = 0;
igb_vendor_info_t *ent;
INIT_DEBUGOUT("igb_probe: begin");
pci_vendor_id = pci_get_vendor(dev);
if (pci_vendor_id != IGB_VENDOR_ID)
return (ENXIO);
pci_device_id = pci_get_device(dev);
pci_subvendor_id = pci_get_subvendor(dev);
pci_subdevice_id = pci_get_subdevice(dev);
ent = igb_vendor_info_array;
while (ent->vendor_id != 0) {
if ((pci_vendor_id == ent->vendor_id) &&
(pci_device_id == ent->device_id) &&
((pci_subvendor_id == ent->subvendor_id) ||
(ent->subvendor_id == PCI_ANY_ID)) &&
((pci_subdevice_id == ent->subdevice_id) ||
(ent->subdevice_id == PCI_ANY_ID))) {
sprintf(adapter_name, "%s %s",
igb_strings[ent->index],
igb_driver_version);
device_set_desc_copy(dev, adapter_name);
return (BUS_PROBE_DEFAULT);
}
ent++;
}
return (ENXIO);
}
/*********************************************************************
* Device initialization routine
*
* The attach entry point is called when the driver is being loaded.
* This routine identifies the type of hardware, allocates all resources
* and initializes the hardware.
*
* return 0 on success, positive on failure
*********************************************************************/
static int
igb_attach(device_t dev)
{
struct adapter *adapter;
int error = 0;
u16 eeprom_data;
INIT_DEBUGOUT("igb_attach: begin");
This isn't functionally identical. In some cases a hint to disable unit 0 would in fact disable all units. This reverts r241856 Approved by: cperciva (implicit)
2012-10-22 13:06:09 +00:00
if (resource_disabled("igb", device_get_unit(dev))) {
device_printf(dev, "Disabled by device hint\n");
return (ENXIO);
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
adapter = device_get_softc(dev);
adapter->dev = adapter->osdep.dev = dev;
IGB_CORE_LOCK_INIT(adapter, device_get_nameunit(dev));
/* SYSCTL stuff */
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
OID_AUTO, "nvm", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
igb_sysctl_nvm_info, "I", "NVM Information");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
igb_set_sysctl_value(adapter, "enable_aim",
"Interrupt Moderation", &adapter->enable_aim,
igb_enable_aim);
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
OID_AUTO, "fc", CTLTYPE_INT|CTLFLAG_RW,
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
adapter, 0, igb_set_flowcntl, "I", "Flow Control");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
callout_init_mtx(&adapter->timer, &adapter->core_mtx, 0);
/* Determine hardware and mac info */
igb_identify_hardware(adapter);
/* Setup PCI resources */
if (igb_allocate_pci_resources(adapter)) {
device_printf(dev, "Allocation of PCI resources failed\n");
error = ENXIO;
goto err_pci;
}
/* Do Shared Code initialization */
if (e1000_setup_init_funcs(&adapter->hw, TRUE)) {
device_printf(dev, "Setup of Shared code failed\n");
error = ENXIO;
goto err_pci;
}
e1000_get_bus_info(&adapter->hw);
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
/* Sysctl for limiting the amount of work done in the taskqueue */
igb_set_sysctl_value(adapter, "rx_processing_limit",
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
"max number of rx packets to process",
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
&adapter->rx_process_limit, igb_rx_process_limit);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
* Validate number of transmit and receive descriptors. It
* must not exceed hardware maximum, and must be multiple
* of E1000_DBA_ALIGN.
*/
if (((igb_txd * sizeof(struct e1000_tx_desc)) % IGB_DBA_ALIGN) != 0 ||
(igb_txd > IGB_MAX_TXD) || (igb_txd < IGB_MIN_TXD)) {
device_printf(dev, "Using %d TX descriptors instead of %d!\n",
IGB_DEFAULT_TXD, igb_txd);
adapter->num_tx_desc = IGB_DEFAULT_TXD;
} else
adapter->num_tx_desc = igb_txd;
if (((igb_rxd * sizeof(struct e1000_rx_desc)) % IGB_DBA_ALIGN) != 0 ||
(igb_rxd > IGB_MAX_RXD) || (igb_rxd < IGB_MIN_RXD)) {
device_printf(dev, "Using %d RX descriptors instead of %d!\n",
IGB_DEFAULT_RXD, igb_rxd);
adapter->num_rx_desc = IGB_DEFAULT_RXD;
} else
adapter->num_rx_desc = igb_rxd;
adapter->hw.mac.autoneg = DO_AUTO_NEG;
adapter->hw.phy.autoneg_wait_to_complete = FALSE;
adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
/* Copper options */
if (adapter->hw.phy.media_type == e1000_media_type_copper) {
adapter->hw.phy.mdix = AUTO_ALL_MODES;
adapter->hw.phy.disable_polarity_correction = FALSE;
adapter->hw.phy.ms_type = IGB_MASTER_SLAVE;
}
/*
* Set the frame limits assuming
* standard ethernet sized frames.
*/
adapter->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHERNET_FCS_SIZE;
/*
** Allocate and Setup Queues
*/
if (igb_allocate_queues(adapter)) {
error = ENOMEM;
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
goto err_pci;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
/* Allocate the appropriate stats memory */
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->vf_ifp) {
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
adapter->stats =
(struct e1000_vf_stats *)malloc(sizeof \
(struct e1000_vf_stats), M_DEVBUF, M_NOWAIT | M_ZERO);
igb_vf_init_stats(adapter);
} else
adapter->stats =
(struct e1000_hw_stats *)malloc(sizeof \
(struct e1000_hw_stats), M_DEVBUF, M_NOWAIT | M_ZERO);
Make sure not to access unallocated stats memory. Reviewed by: jfv
2010-08-27 23:50:13 +00:00
if (adapter->stats == NULL) {
device_printf(dev, "Can not allocate stats memory\n");
error = ENOMEM;
goto err_late;
}
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
Do not allocate multicast array memory in multicast filter configuration function. For failed memory allocations, em(4)/lem(4) called panic(9) which is not acceptable on production box. igb(4)/ixgb(4)/ix(4) allocated the required memory in stack which consumed 768 bytes of stack memory which looks too big. To address these issues, allocate multicast array memory in device attach time and make multicast configuration success under any conditions. This change also removes the excessive use of memory in stack. Reviewed by: jfv
2010-08-28 00:34:22 +00:00
/* Allocate multicast array memory. */
adapter->mta = malloc(sizeof(u8) * ETH_ADDR_LEN *
MAX_NUM_MULTICAST_ADDRESSES, M_DEVBUF, M_NOWAIT);
if (adapter->mta == NULL) {
device_printf(dev, "Can not allocate multicast setup array\n");
error = ENOMEM;
goto err_late;
}
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
/* Some adapter-specific advanced features */
if (adapter->hw.mac.type >= e1000_i350) {
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
OID_AUTO, "dmac", CTLTYPE_INT|CTLFLAG_RW,
adapter, 0, igb_sysctl_dmac, "I", "DMA Coalesce");
Change the interface to the Energy Efficient Ethernet (EEE) setting in the igb and em driver. This was necessitated by a shared code change that I was given late in the game, a data type changed from bool to int, in the last update I dealt with it by a cast, but it was pointed out (thanks jhb) that there was a potential problem with this. John suggested this safer approach, and it is fine with me... MFC after:2 days (to catch the 9.1 update)
2012-07-07 20:21:05 +00:00
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
OID_AUTO, "eee_disabled", CTLTYPE_INT|CTLFLAG_RW,
adapter, 0, igb_sysctl_eee, "I",
"Disable Energy Efficient Ethernet");
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (adapter->hw.phy.media_type == e1000_media_type_copper) {
if (adapter->hw.mac.type == e1000_i354)
e1000_set_eee_i354(&adapter->hw);
else
e1000_set_eee_i350(&adapter->hw);
}
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
}
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/*
** Start from a known state, this is
** important in reading the nvm and
** mac from that.
*/
e1000_reset_hw(&adapter->hw);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Make sure we have a good EEPROM before we read from it */
Sync with Intel internal source: shared code update and small changes in core required Add support for new i210/i211 devices Improve queue calculation based on mac type MFC after:5 days
2012-07-05 20:26:57 +00:00
if (((adapter->hw.mac.type != e1000_i210) &&
(adapter->hw.mac.type != e1000_i211)) &&
(e1000_validate_nvm_checksum(&adapter->hw) < 0)) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
** Some PCI-E parts fail the first check due to
** the link being in sleep state, call it again,
** if it fails a second time its a real issue.
*/
if (e1000_validate_nvm_checksum(&adapter->hw) < 0) {
device_printf(dev,
"The EEPROM Checksum Is Not Valid\n");
error = EIO;
goto err_late;
}
}
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/*
** Copy the permanent MAC address out of the EEPROM
*/
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (e1000_read_mac_addr(&adapter->hw) < 0) {
device_printf(dev, "EEPROM read error while reading MAC"
" address\n");
error = EIO;
goto err_late;
}
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/* Check its sanity */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (!igb_is_valid_ether_addr(adapter->hw.mac.addr)) {
device_printf(dev, "Invalid MAC address\n");
error = EIO;
goto err_late;
}
/* Setup OS specific network interface */
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
if (igb_setup_interface(dev, adapter) != 0)
goto err_late;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* Now get a good starting state */
igb_reset(adapter);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Initialize statistics */
igb_update_stats_counters(adapter);
adapter->hw.mac.get_link_status = 1;
igb_update_link_status(adapter);
/* Indicate SOL/IDER usage */
if (e1000_check_reset_block(&adapter->hw))
device_printf(dev,
"PHY reset is blocked due to SOL/IDER session.\n");
/* Determine if we have to control management hardware */
adapter->has_manage = e1000_enable_mng_pass_thru(&adapter->hw);
/*
* Setup Wake-on-Lan
*/
/* APME bit in EEPROM is mapped to WUC.APME */
eeprom_data = E1000_READ_REG(&adapter->hw, E1000_WUC) & E1000_WUC_APME;
if (eeprom_data)
adapter->wol = E1000_WUFC_MAG;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
/* Register for VLAN events */
adapter->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
Improvement on the last change, this gives a precise way to tell the one and only interface that a vlan event is for. Thanks to John Baldwin for the patch. Approved by: re
2009-07-24 21:35:52 +00:00
igb_register_vlan, adapter, EVENTHANDLER_PRI_FIRST);
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
adapter->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
Improvement on the last change, this gives a precise way to tell the one and only interface that a vlan event is for. Thanks to John Baldwin for the patch. Approved by: re
2009-07-24 21:35:52 +00:00
igb_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST);
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
igb_add_hw_stats(adapter);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Tell the stack that the interface is not active */
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
adapter->ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
adapter->ifp->if_drv_flags |= IFF_DRV_OACTIVE;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
adapter->led_dev = led_create(igb_led_func, adapter,
device_get_nameunit(dev));
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
/*
** Configure Interrupts
*/
if ((adapter->msix > 1) && (igb_enable_msix))
error = igb_allocate_msix(adapter);
else /* MSI or Legacy */
error = igb_allocate_legacy(adapter);
if (error)
goto err_late;
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
igb_netmap_attach(adapter);
#endif /* DEV_NETMAP */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
INIT_DEBUGOUT("igb_attach: end");
return (0);
err_late:
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
igb_detach(dev);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
igb_free_transmit_structures(adapter);
igb_free_receive_structures(adapter);
igb_release_hw_control(adapter);
err_pci:
igb_free_pci_resources(adapter);
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
if (adapter->ifp != NULL)
if_free(adapter->ifp);
Do not allocate multicast array memory in multicast filter configuration function. For failed memory allocations, em(4)/lem(4) called panic(9) which is not acceptable on production box. igb(4)/ixgb(4)/ix(4) allocated the required memory in stack which consumed 768 bytes of stack memory which looks too big. To address these issues, allocate multicast array memory in device attach time and make multicast configuration success under any conditions. This change also removes the excessive use of memory in stack. Reviewed by: jfv
2010-08-28 00:34:22 +00:00
free(adapter->mta, M_DEVBUF);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IGB_CORE_LOCK_DESTROY(adapter);
return (error);
}
/*********************************************************************
* Device removal routine
*
* The detach entry point is called when the driver is being removed.
* This routine stops the adapter and deallocates all the resources
* that were allocated for driver operation.
*
* return 0 on success, positive on failure
*********************************************************************/
static int
igb_detach(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
struct ifnet *ifp = adapter->ifp;
INIT_DEBUGOUT("igb_detach: begin");
/* Make sure VLANS are not using driver */
if (adapter->ifp->if_vlantrunk != NULL) {
device_printf(dev,"Vlan in use, detach first\n");
return (EBUSY);
}
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
ether_ifdetach(adapter->ifp);
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
if (adapter->led_dev != NULL)
led_destroy(adapter->led_dev);
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#ifdef DEVICE_POLLING
if (ifp->if_capenable & IFCAP_POLLING)
ether_poll_deregister(ifp);
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IGB_CORE_LOCK(adapter);
adapter->in_detach = 1;
igb_stop(adapter);
IGB_CORE_UNLOCK(adapter);
e1000_phy_hw_reset(&adapter->hw);
/* Give control back to firmware */
igb_release_manageability(adapter);
igb_release_hw_control(adapter);
if (adapter->wol) {
E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN);
E1000_WRITE_REG(&adapter->hw, E1000_WUFC, adapter->wol);
igb_enable_wakeup(dev);
}
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
/* Unregister VLAN events */
if (adapter->vlan_attach != NULL)
EVENTHANDLER_DEREGISTER(vlan_config, adapter->vlan_attach);
if (adapter->vlan_detach != NULL)
EVENTHANDLER_DEREGISTER(vlan_unconfig, adapter->vlan_detach);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
callout_drain(&adapter->timer);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
netmap_detach(adapter->ifp);
#endif /* DEV_NETMAP */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
igb_free_pci_resources(adapter);
bus_generic_detach(dev);
if_free(ifp);
igb_free_transmit_structures(adapter);
igb_free_receive_structures(adapter);
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
if (adapter->mta != NULL)
free(adapter->mta, M_DEVBUF);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IGB_CORE_LOCK_DESTROY(adapter);
return (0);
}
/*********************************************************************
*
* Shutdown entry point
*
**********************************************************************/
static int
igb_shutdown(device_t dev)
{
return igb_suspend(dev);
}
/*
* Suspend/resume device methods.
*/
static int
igb_suspend(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
IGB_CORE_LOCK(adapter);
igb_stop(adapter);
igb_release_manageability(adapter);
igb_release_hw_control(adapter);
if (adapter->wol) {
E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN);
E1000_WRITE_REG(&adapter->hw, E1000_WUFC, adapter->wol);
igb_enable_wakeup(dev);
}
IGB_CORE_UNLOCK(adapter);
return bus_generic_suspend(dev);
}
static int
igb_resume(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
struct tx_ring *txr = adapter->tx_rings;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
struct ifnet *ifp = adapter->ifp;
IGB_CORE_LOCK(adapter);
igb_init_locked(adapter);
igb_init_manageability(adapter);
if ((ifp->if_flags & IFF_UP) &&
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
(ifp->if_drv_flags & IFF_DRV_RUNNING) && adapter->link_active) {
for (int i = 0; i < adapter->num_queues; i++, txr++) {
IGB_TX_LOCK(txr);
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
/* Process the stack queue only if not depleted */
if (((txr->queue_status & IGB_QUEUE_DEPLETED) == 0) &&
!drbr_empty(ifp, txr->br))
drbr_enqueue() awlays consumes mbuf, no matter did it fail or not. The mbuf pointer is no longer valid, so can't be reused after. Fix igb_mq_start() where mbuf pointer was used after drbr_enqueue(). This eventually leads us to all invocations of igb_mq_start_locked() called with third argument as NULL. This allows us to simplify this function. Submitted by: Karim Fodil-Lemelin <fodillemlinkarim gmail.com> Reviewed by: jfv
2012-11-26 20:03:57 +00:00
igb_mq_start_locked(ifp, txr);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#else
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
igb_start_locked(txr, ifp);
#endif
IGB_TX_UNLOCK(txr);
}
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IGB_CORE_UNLOCK(adapter);
return bus_generic_resume(dev);
}
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifdef IGB_LEGACY_TX
Commit a portion of 233708 I missed earlier and don't include the definition of igb_start() and igb_start_locked() (nor set if_start in the ifnet) when igb(4) uses if_transmit.
2012-06-01 15:52:41 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
* Transmit entry point
*
* igb_start is called by the stack to initiate a transmit.
* The driver will remain in this routine as long as there are
* packets to transmit and transmit resources are available.
* In case resources are not available stack is notified and
* the packet is requeued.
**********************************************************************/
static void
igb_start_locked(struct tx_ring *txr, struct ifnet *ifp)
{
struct adapter *adapter = ifp->if_softc;
struct mbuf *m_head;
IGB_TX_LOCK_ASSERT(txr);
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) !=
IFF_DRV_RUNNING)
return;
if (!adapter->link_active)
return;
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
/* Call cleanup if number of TX descriptors low */
if (txr->tx_avail <= IGB_TX_CLEANUP_THRESHOLD)
igb_txeof(txr);
Add a missing fragment in the tx msix handler to invoke another if all work is not done. Sync the igb driver with changes suggested by yongari and made in em, these made sense to be in both drivers.
2010-04-14 20:55:33 +00:00
while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
if (txr->tx_avail <= IGB_MAX_SCATTER) {
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
txr->queue_status |= IGB_QUEUE_DEPLETED;
Add a missing fragment in the tx msix handler to invoke another if all work is not done. Sync the igb driver with changes suggested by yongari and made in em, these made sense to be in both drivers.
2010-04-14 20:55:33 +00:00
break;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
if (m_head == NULL)
break;
/*
* Encapsulation can modify our pointer, and or make it
* NULL on failure. In that event, we can't requeue.
*/
if (igb_xmit(txr, &m_head)) {
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
if (m_head != NULL)
IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
if (txr->tx_avail <= IGB_MAX_SCATTER)
txr->queue_status |= IGB_QUEUE_DEPLETED;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
break;
}
/* Send a copy of the frame to the BPF listener */
ETHER_BPF_MTAP(ifp, m_head);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* Set watchdog on */
Add a missing fragment in the tx msix handler to invoke another if all work is not done. Sync the igb driver with changes suggested by yongari and made in em, these made sense to be in both drivers.
2010-04-14 20:55:33 +00:00
txr->watchdog_time = ticks;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
txr->queue_status |= IGB_QUEUE_WORKING;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/*
* Legacy TX driver routine, called from the
* stack, always uses tx[0], and spins for it.
* Should not be used with multiqueue tx
*/
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static void
igb_start(struct ifnet *ifp)
{
struct adapter *adapter = ifp->if_softc;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
struct tx_ring *txr = adapter->tx_rings;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
IGB_TX_LOCK(txr);
igb_start_locked(txr, ifp);
IGB_TX_UNLOCK(txr);
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
return;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#else /* ~IGB_LEGACY_TX */
Sync with Intel internal source: shared code update and small changes in core required Add support for new i210/i211 devices Improve queue calculation based on mac type MFC after:5 days
2012-07-05 20:26:57 +00:00
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/*
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
** Multiqueue Transmit Entry:
** quick turnaround to the stack
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
**
*/
static int
igb_mq_start(struct ifnet *ifp, struct mbuf *m)
{
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
struct adapter *adapter = ifp->if_softc;
struct igb_queue *que;
struct tx_ring *txr;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
int i, err = 0;
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
#ifdef RSS
uint32_t bucket_id;
#endif
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/* Which queue to use */
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
/*
* When doing RSS, map it to the same outbound queue
* as the incoming flow would be mapped to.
*
* If everything is setup correctly, it should be the
* same bucket that the current CPU we're on is.
*/
if ((m->m_flags & M_FLOWID) != 0) {
#ifdef RSS
if (rss_hash2bucket(m->m_pkthdr.flowid,
M_HASHTYPE_GET(m), &bucket_id) == 0) {
/* XXX TODO: spit out something if bucket_id > num_queues? */
i = bucket_id % adapter->num_queues;
} else {
#endif
i = m->m_pkthdr.flowid % adapter->num_queues;
#ifdef RSS
}
#endif
} else {
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
i = curcpu % adapter->num_queues;
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
txr = &adapter->tx_rings[i];
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
que = &adapter->queues[i];
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
err = drbr_enqueue(ifp, txr->br, m);
Alter the mq_start routine to do a TRYLOCK and call to the locked routine rather than just queueing. The former code was an attempt at getting UDP performance up, but there have been customer reports of problems with it, so the ixgbe approach seems the best solution for now.
2013-08-13 00:25:39 +00:00
if (err)
return (err);
if (IGB_TX_TRYLOCK(txr)) {
make sure if_transmit returns 0 if the mbuf is enqueued. ixgbe/ixv.c still needs a similar fix but it takes a little more restructuring of the code. MFC after: 3 days
2014-06-06 20:49:56 +00:00
igb_mq_start_locked(ifp, txr);
Alter the mq_start routine to do a TRYLOCK and call to the locked routine rather than just queueing. The former code was an attempt at getting UDP performance up, but there have been customer reports of problems with it, so the ixgbe approach seems the best solution for now.
2013-08-13 00:25:39 +00:00
IGB_TX_UNLOCK(txr);
} else
taskqueue_enqueue(que->tq, &txr->txq_task);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
make sure if_transmit returns 0 if the mbuf is enqueued. ixgbe/ixv.c still needs a similar fix but it takes a little more restructuring of the code. MFC after: 3 days
2014-06-06 20:49:56 +00:00
return (0);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
}
static int
drbr_enqueue() awlays consumes mbuf, no matter did it fail or not. The mbuf pointer is no longer valid, so can't be reused after. Fix igb_mq_start() where mbuf pointer was used after drbr_enqueue(). This eventually leads us to all invocations of igb_mq_start_locked() called with third argument as NULL. This allows us to simplify this function. Submitted by: Karim Fodil-Lemelin <fodillemlinkarim gmail.com> Reviewed by: jfv
2012-11-26 20:03:57 +00:00
igb_mq_start_locked(struct ifnet *ifp, struct tx_ring *txr)
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
{
struct adapter *adapter = txr->adapter;
struct mbuf *next;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
int err = 0, enq = 0;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
IGB_TX_LOCK_ASSERT(txr);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
if (((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) ||
drbr_enqueue() awlays consumes mbuf, no matter did it fail or not. The mbuf pointer is no longer valid, so can't be reused after. Fix igb_mq_start() where mbuf pointer was used after drbr_enqueue(). This eventually leads us to all invocations of igb_mq_start_locked() called with third argument as NULL. This allows us to simplify this function. Submitted by: Karim Fodil-Lemelin <fodillemlinkarim gmail.com> Reviewed by: jfv
2012-11-26 20:03:57 +00:00
adapter->link_active == 0)
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
return (ENETDOWN);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
Add a missing fragment in the tx msix handler to invoke another if all work is not done. Sync the igb driver with changes suggested by yongari and made in em, these made sense to be in both drivers.
2010-04-14 20:55:33 +00:00
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/* Process the queue */
This fixes a out-of-order problem with several of the newer drivers. The basic problem was that the driver was pulling the mbuf off the drbr ring and then when sending with xmit(), encounting a full transmit ring. Thus the lower layer xmit() function would return an error, and the drivers would then append the data back on to the ring. For TCP this is a horrible scenario sure to bring on a fast-retransmit. The fix is to use drbr_peek() to pull the data pointer but not remove it from the ring. If it fails then we either call the new drbr_putback or drbr_advance method. Advance moves it forward (we do this sometimes when the xmit() function frees the mbuf). When we succeed we always call advance. The putback will always copy the mbuf back to the top of the ring. Note that the putback *cannot* be used with a drbr_dequeue() only with drbr_peek(). We most of the time, in putback, would not need to copy it back since most likey the mbuf is still the same, but sometimes xmit() functions will change the mbuf via a pullup or other call. So the optimial case for the single consumer is to always copy it back. If we ever do a multiple_consumer (for lagg?) we will need a test and atomic in the put back possibly a seperate putback_mc() in the ring buf. Reviewed by: jhb@freebsd.org, jlv@freebsd.org
2013-02-07 15:20:54 +00:00
while ((next = drbr_peek(ifp, txr->br)) != NULL) {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if ((err = igb_xmit(txr, &next)) != 0) {
This fixes a out-of-order problem with several of the newer drivers. The basic problem was that the driver was pulling the mbuf off the drbr ring and then when sending with xmit(), encounting a full transmit ring. Thus the lower layer xmit() function would return an error, and the drivers would then append the data back on to the ring. For TCP this is a horrible scenario sure to bring on a fast-retransmit. The fix is to use drbr_peek() to pull the data pointer but not remove it from the ring. If it fails then we either call the new drbr_putback or drbr_advance method. Advance moves it forward (we do this sometimes when the xmit() function frees the mbuf). When we succeed we always call advance. The putback will always copy the mbuf back to the top of the ring. Note that the putback *cannot* be used with a drbr_dequeue() only with drbr_peek(). We most of the time, in putback, would not need to copy it back since most likey the mbuf is still the same, but sometimes xmit() functions will change the mbuf via a pullup or other call. So the optimial case for the single consumer is to always copy it back. If we ever do a multiple_consumer (for lagg?) we will need a test and atomic in the put back possibly a seperate putback_mc() in the ring buf. Reviewed by: jhb@freebsd.org, jlv@freebsd.org
2013-02-07 15:20:54 +00:00
if (next == NULL) {
/* It was freed, move forward */
drbr_advance(ifp, txr->br);
} else {
/*
* Still have one left, it may not be
* the same since the transmit function
* may have changed it.
*/
drbr_putback(ifp, txr->br, next);
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
break;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
}
This fixes a out-of-order problem with several of the newer drivers. The basic problem was that the driver was pulling the mbuf off the drbr ring and then when sending with xmit(), encounting a full transmit ring. Thus the lower layer xmit() function would return an error, and the drivers would then append the data back on to the ring. For TCP this is a horrible scenario sure to bring on a fast-retransmit. The fix is to use drbr_peek() to pull the data pointer but not remove it from the ring. If it fails then we either call the new drbr_putback or drbr_advance method. Advance moves it forward (we do this sometimes when the xmit() function frees the mbuf). When we succeed we always call advance. The putback will always copy the mbuf back to the top of the ring. Note that the putback *cannot* be used with a drbr_dequeue() only with drbr_peek(). We most of the time, in putback, would not need to copy it back since most likey the mbuf is still the same, but sometimes xmit() functions will change the mbuf via a pullup or other call. So the optimial case for the single consumer is to always copy it back. If we ever do a multiple_consumer (for lagg?) we will need a test and atomic in the put back possibly a seperate putback_mc() in the ring buf. Reviewed by: jhb@freebsd.org, jlv@freebsd.org
2013-02-07 15:20:54 +00:00
drbr_advance(ifp, txr->br);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
enq++;
- Provide igb_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-19 11:49:41 +00:00
if_inc_counter(ifp, IFCOUNTER_OBYTES, next->m_pkthdr.len);
The drbr(9) API appeared to be so unclear, that most drivers in tree used it incorrectly, which lead to inaccurate overrated if_obytes accounting. The drbr(9) used to update ifnet stats on drbr_enqueue(), which is not accurate since enqueuing doesn't imply successful processing by driver. Dequeuing neither mean that. Most drivers also called drbr_stats_update() which did accounting again, leading to doubled if_obytes statistics. And in case of severe transmitting, when a packet could be several times enqueued and dequeued it could have been accounted several times. o Thus, make drbr(9) API thinner. Now drbr(9) merely chooses between ALTQ queueing or buf_ring(9) queueing. - It doesn't touch the buf_ring stats any more. - It doesn't touch ifnet stats anymore. - drbr_stats_update() no longer exists. o buf_ring(9) handles its stats itself: - It handles br_drops itself. - br_prod_bytes stats are dropped. Rationale: no one ever reads them but update of a common counter on every packet negatively affects performance due to excessive cache invalidation. - buf_ring_enqueue_bytes() reduced to buf_ring_enqueue(), since we no longer account bytes. o Drivers handle their stats theirselves: if_obytes, if_omcasts. o mlx4(4), igb(4), em(4), vxge(4), oce(4) and ixv(4) no longer use drbr_stats_update(), and update ifnet stats theirselves. o bxe(4) was the most correct driver, it didn't call drbr_stats_update(), thus it was the only driver accurate under moderate load. Now it also maintains stats itself. o ixgbe(4) had already taken stats from hardware, so just - drop software stats updating. - take multicast packet count from hardware as well. o mxge(4) just no longer needs NO_SLOW_STATS define. o cxgb(4), cxgbe(4) need no change, since they obtain stats from hardware. Reviewed by: jfv, gnn
2012-09-28 18:28:27 +00:00
if (next->m_flags & M_MCAST)
- Provide igb_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-19 11:49:41 +00:00
if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
ETHER_BPF_MTAP(ifp, next);
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
break;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
}
if (enq > 0) {
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/* Set the watchdog */
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
txr->queue_status |= IGB_QUEUE_WORKING;
Add a missing fragment in the tx msix handler to invoke another if all work is not done. Sync the igb driver with changes suggested by yongari and made in em, these made sense to be in both drivers.
2010-04-14 20:55:33 +00:00
txr->watchdog_time = ticks;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
}
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
if (txr->tx_avail <= IGB_TX_CLEANUP_THRESHOLD)
igb_txeof(txr);
if (txr->tx_avail <= IGB_MAX_SCATTER)
txr->queue_status |= IGB_QUEUE_DEPLETED;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
return (err);
}
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
/*
* Called from a taskqueue to drain queued transmit packets.
*/
static void
igb_deferred_mq_start(void *arg, int pending)
{
struct tx_ring *txr = arg;
struct adapter *adapter = txr->adapter;
struct ifnet *ifp = adapter->ifp;
IGB_TX_LOCK(txr);
if (!drbr_empty(ifp, txr->br))
drbr_enqueue() awlays consumes mbuf, no matter did it fail or not. The mbuf pointer is no longer valid, so can't be reused after. Fix igb_mq_start() where mbuf pointer was used after drbr_enqueue(). This eventually leads us to all invocations of igb_mq_start_locked() called with third argument as NULL. This allows us to simplify this function. Submitted by: Karim Fodil-Lemelin <fodillemlinkarim gmail.com> Reviewed by: jfv
2012-11-26 20:03:57 +00:00
igb_mq_start_locked(ifp, txr);
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
IGB_TX_UNLOCK(txr);
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/*
** Flush all ring buffers
*/
static void
igb_qflush(struct ifnet *ifp)
{
struct adapter *adapter = ifp->if_softc;
struct tx_ring *txr = adapter->tx_rings;
struct mbuf *m;
for (int i = 0; i < adapter->num_queues; i++, txr++) {
IGB_TX_LOCK(txr);
while ((m = buf_ring_dequeue_sc(txr->br)) != NULL)
m_freem(m);
IGB_TX_UNLOCK(txr);
}
if_qflush(ifp);
}
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#endif /* ~IGB_LEGACY_TX */
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
* Ioctl entry point
*
* igb_ioctl is called when the user wants to configure the
* interface.
*
* return 0 on success, positive on failure
**********************************************************************/
static int
igb_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
{
struct adapter *adapter = ifp->if_softc;
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
struct ifreq *ifr = (struct ifreq *)data;
#if defined(INET) || defined(INET6)
struct ifaddr *ifa = (struct ifaddr *)data;
Hide AF_INET specific ioctl handling under #ifdef INET. MFC after: 2 months
2008-11-06 11:11:25 +00:00
#endif
Last change still had an issue, one more time...
2011-12-11 18:46:14 +00:00
bool avoid_reset = FALSE;
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
int error = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (adapter->in_detach)
return (error);
switch (command) {
case SIOCSIFADDR:
Hide AF_INET specific ioctl handling under #ifdef INET. MFC after: 2 months
2008-11-06 11:11:25 +00:00
#ifdef INET
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
if (ifa->ifa_addr->sa_family == AF_INET)
avoid_reset = TRUE;
#endif
Correct LINT build issues in the ioctl code.
2011-12-11 09:37:25 +00:00
#ifdef INET6
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
if (ifa->ifa_addr->sa_family == AF_INET6)
avoid_reset = TRUE;
#endif
/*
** Calling init results in link renegotiation,
** so we avoid doing it when possible.
*/
if (avoid_reset) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
ifp->if_flags |= IFF_UP;
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
igb_init(adapter);
Last change still had an issue, one more time...
2011-12-11 18:46:14 +00:00
#ifdef INET
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (!(ifp->if_flags & IFF_NOARP))
arp_ifinit(ifp, ifa);
Last change still had an issue, one more time...
2011-12-11 18:46:14 +00:00
#endif
A fix to make the LINT-NOINET build happy, if this works out the ixgbe driver should be changed as well.
2011-07-07 00:46:50 +00:00
} else
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
error = ether_ioctl(ifp, command, data);
break;
case SIOCSIFMTU:
{
int max_frame_size;
IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFMTU (Set Interface MTU)");
IGB_CORE_LOCK(adapter);
max_frame_size = 9234;
if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
ETHER_CRC_LEN) {
IGB_CORE_UNLOCK(adapter);
error = EINVAL;
break;
}
ifp->if_mtu = ifr->ifr_mtu;
adapter->max_frame_size =
ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
igb_init_locked(adapter);
IGB_CORE_UNLOCK(adapter);
break;
}
case SIOCSIFFLAGS:
IOCTL_DEBUGOUT("ioctl rcv'd:\
SIOCSIFFLAGS (Set Interface Flags)");
IGB_CORE_LOCK(adapter);
if (ifp->if_flags & IFF_UP) {
if ((ifp->if_drv_flags & IFF_DRV_RUNNING)) {
if ((ifp->if_flags ^ adapter->if_flags) &
This delta has a few important items: PR 122839 is fixed in both em and in igb Second, the issue on building modules since the static kernel build changes is now resolved. I was not able to get the fancier directory hierarchy working, but this works, both em and igb build as modules now. Third, there is now support in em for two new NICs, Hartwell (or 82574) is a low cost PCIE dual port adapter that has MSIX, for this release it uses 3 vectors only, RX, TX, and LINK. In the next release I will add a second TX and RX queue. Also, there is support here for ICH10, the followon to ICH9. Both of these are early releases, general availability will follow soon. Fourth: On Hartwell and ICH10 we now have IEEE 1588 PTP support, I have implemented this in a provisional way so that early adopters may try and comment on the functionality. The IOCTL structure may change. This feature is off by default, you need to edit the Makefile and add the EM_TIMESYNC define to get the code. Enjoy all!!
2008-04-25 21:19:41 +00:00
(IFF_PROMISC | IFF_ALLMULTI)) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
igb_disable_promisc(adapter);
igb_set_promisc(adapter);
}
} else
igb_init_locked(adapter);
} else
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
igb_stop(adapter);
adapter->if_flags = ifp->if_flags;
IGB_CORE_UNLOCK(adapter);
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
IOCTL_DEBUGOUT("ioctl rcv'd: SIOC(ADD|DEL)MULTI");
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
IGB_CORE_LOCK(adapter);
igb_disable_intr(adapter);
igb_set_multi(adapter);
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#ifdef DEVICE_POLLING
if (!(ifp->if_capenable & IFCAP_POLLING))
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
igb_enable_intr(adapter);
IGB_CORE_UNLOCK(adapter);
}
break;
case SIOCSIFMEDIA:
/* Check SOL/IDER usage */
IGB_CORE_LOCK(adapter);
if (e1000_check_reset_block(&adapter->hw)) {
IGB_CORE_UNLOCK(adapter);
device_printf(adapter->dev, "Media change is"
" blocked due to SOL/IDER session.\n");
break;
}
IGB_CORE_UNLOCK(adapter);
case SIOCGIFMEDIA:
IOCTL_DEBUGOUT("ioctl rcv'd: \
SIOCxIFMEDIA (Get/Set Interface Media)");
error = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
break;
case SIOCSIFCAP:
{
int mask, reinit;
IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFCAP (Set Capabilities)");
reinit = 0;
mask = ifr->ifr_reqcap ^ ifp->if_capenable;
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#ifdef DEVICE_POLLING
if (mask & IFCAP_POLLING) {
if (ifr->ifr_reqcap & IFCAP_POLLING) {
error = ether_poll_register(igb_poll, ifp);
if (error)
return (error);
IGB_CORE_LOCK(adapter);
igb_disable_intr(adapter);
ifp->if_capenable |= IFCAP_POLLING;
IGB_CORE_UNLOCK(adapter);
} else {
error = ether_poll_deregister(ifp);
/* Enable interrupt even in error case */
IGB_CORE_LOCK(adapter);
igb_enable_intr(adapter);
ifp->if_capenable &= ~IFCAP_POLLING;
IGB_CORE_UNLOCK(adapter);
}
}
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (mask & IFCAP_HWCSUM) {
ifp->if_capenable ^= IFCAP_HWCSUM;
reinit = 1;
}
if (mask & IFCAP_TSO4) {
ifp->if_capenable ^= IFCAP_TSO4;
reinit = 1;
}
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (mask & IFCAP_TSO6) {
ifp->if_capenable ^= IFCAP_TSO6;
reinit = 1;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (mask & IFCAP_VLAN_HWTAGGING) {
ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
reinit = 1;
}
Important fix got clobbered in the em driver, keeping VLAN HWFILTER from being used by default, this breaks stacked pseudo devices, and as it turns out, also breaks virtual machines that happen to use VLANS (didn't know that before :). Put the fix back into the em driver, and for good measure add the same code to the igb driver where it should have been anyway.
2010-04-08 00:50:43 +00:00
if (mask & IFCAP_VLAN_HWFILTER) {
ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
reinit = 1;
}
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
if (mask & IFCAP_VLAN_HWTSO) {
ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
reinit = 1;
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (mask & IFCAP_LRO) {
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
ifp->if_capenable ^= IFCAP_LRO;
reinit = 1;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING))
igb_init(adapter);
VLAN_CAPABILITIES(ifp);
break;
}
default:
error = ether_ioctl(ifp, command, data);
break;
}
return (error);
}
/*********************************************************************
* Init entry point
*
* This routine is used in two ways. It is used by the stack as
* init entry point in network interface structure. It is also used
* by the driver as a hw/sw initialization routine to get to a
* consistent state.
*
* return 0 on success, positive on failure
**********************************************************************/
static void
igb_init_locked(struct adapter *adapter)
{
struct ifnet *ifp = adapter->ifp;
device_t dev = adapter->dev;
INIT_DEBUGOUT("igb_init: begin");
IGB_CORE_LOCK_ASSERT(adapter);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
igb_disable_intr(adapter);
callout_stop(&adapter->timer);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Get the latest mac address, User can use a LAA */
bcopy(IF_LLADDR(adapter->ifp), adapter->hw.mac.addr,
ETHER_ADDR_LEN);
/* Put the address into the Receive Address Array */
e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
igb_reset(adapter);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
igb_update_link_status(adapter);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_VET, ETHERTYPE_VLAN);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Set hardware offload abilities */
ifp->if_hwassist = 0;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
if (ifp->if_capenable & IFCAP_TXCSUM) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP);
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
#if __FreeBSD_version >= 800000
if (adapter->hw.mac.type == e1000_82576)
ifp->if_hwassist |= CSUM_SCTP;
#endif
}
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (ifp->if_capenable & IFCAP_TSO)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
ifp->if_hwassist |= CSUM_TSO;
/* Configure for OS presence */
igb_init_manageability(adapter);
/* Prepare transmit descriptors and buffers */
igb_setup_transmit_structures(adapter);
igb_initialize_transmit_units(adapter);
/* Setup Multicast table */
igb_set_multi(adapter);
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/*
** Figure out the desired mbuf pool
** for doing jumbo/packetsplit
*/
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
if (adapter->max_frame_size <= 2048)
adapter->rx_mbuf_sz = MCLBYTES;
else if (adapter->max_frame_size <= 4096)
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
adapter->rx_mbuf_sz = MJUMPAGESIZE;
else
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
adapter->rx_mbuf_sz = MJUM9BYTES;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Prepare receive descriptors and buffers */
if (igb_setup_receive_structures(adapter)) {
device_printf(dev, "Could not setup receive structures\n");
return;
}
igb_initialize_receive_units(adapter);
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
/* Enable VLAN support */
if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING)
igb_setup_vlan_hw_support(adapter);
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Don't lose promiscuous settings */
igb_set_promisc(adapter);
ifp->if_drv_flags |= IFF_DRV_RUNNING;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
callout_reset(&adapter->timer, hz, igb_local_timer, adapter);
e1000_clear_hw_cntrs_base_generic(&adapter->hw);
if (adapter->msix > 1) /* Set up queue routing */
igb_configure_queues(adapter);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* this clears any pending interrupts */
E1000_READ_REG(&adapter->hw, E1000_ICR);
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#ifdef DEVICE_POLLING
/*
* Only enable interrupts if we are not polling, make sure
* they are off otherwise.
*/
if (ifp->if_capenable & IFCAP_POLLING)
igb_disable_intr(adapter);
else
#endif /* DEVICE_POLLING */
{
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
igb_enable_intr(adapter);
E1000_WRITE_REG(&adapter->hw, E1000_ICS, E1000_ICS_LSC);
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
/* Set Energy Efficient Ethernet */
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (adapter->hw.phy.media_type == e1000_media_type_copper) {
if (adapter->hw.mac.type == e1000_i354)
e1000_set_eee_i354(&adapter->hw);
else
e1000_set_eee_i350(&adapter->hw);
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
static void
igb_init(void *arg)
{
struct adapter *adapter = arg;
IGB_CORE_LOCK(adapter);
igb_init_locked(adapter);
IGB_CORE_UNLOCK(adapter);
}
static void
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
igb_handle_que(void *context, int pending)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct igb_queue *que = context;
struct adapter *adapter = que->adapter;
struct tx_ring *txr = que->txr;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
struct ifnet *ifp = adapter->ifp;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Add a missing fragment in the tx msix handler to invoke another if all work is not done. Sync the igb driver with changes suggested by yongari and made in em, these made sense to be in both drivers.
2010-04-14 20:55:33 +00:00
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
bool more;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
more = igb_rxeof(que, adapter->rx_process_limit, NULL);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
Add a missing fragment in the tx msix handler to invoke another if all work is not done. Sync the igb driver with changes suggested by yongari and made in em, these made sense to be in both drivers.
2010-04-14 20:55:33 +00:00
IGB_TX_LOCK(txr);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
igb_txeof(txr);
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
/* Process the stack queue only if not depleted */
if (((txr->queue_status & IGB_QUEUE_DEPLETED) == 0) &&
!drbr_empty(ifp, txr->br))
drbr_enqueue() awlays consumes mbuf, no matter did it fail or not. The mbuf pointer is no longer valid, so can't be reused after. Fix igb_mq_start() where mbuf pointer was used after drbr_enqueue(). This eventually leads us to all invocations of igb_mq_start_locked() called with third argument as NULL. This allows us to simplify this function. Submitted by: Karim Fodil-Lemelin <fodillemlinkarim gmail.com> Reviewed by: jfv
2012-11-26 20:03:57 +00:00
igb_mq_start_locked(ifp, txr);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
#else
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
igb_start_locked(txr, ifp);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
#endif
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
IGB_TX_UNLOCK(txr);
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
/* Do we need another? */
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
if (more) {
Add a missing fragment in the tx msix handler to invoke another if all work is not done. Sync the igb driver with changes suggested by yongari and made in em, these made sense to be in both drivers.
2010-04-14 20:55:33 +00:00
taskqueue_enqueue(que->tq, &que->que_task);
return;
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
}
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#ifdef DEVICE_POLLING
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
if (ifp->if_capenable & IFCAP_POLLING)
return;
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#endif
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
/* Reenable this interrupt */
if (que->eims)
E1000_WRITE_REG(&adapter->hw, E1000_EIMS, que->eims);
else
igb_enable_intr(adapter);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Add a link tasklet so updates can be sleepable.
2010-01-27 20:12:04 +00:00
/* Deal with link in a sleepable context */
static void
igb_handle_link(void *context, int pending)
{
struct adapter *adapter = context;
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
IGB_CORE_LOCK(adapter);
igb_handle_link_locked(adapter);
IGB_CORE_UNLOCK(adapter);
}
static void
igb_handle_link_locked(struct adapter *adapter)
{
struct tx_ring *txr = adapter->tx_rings;
struct ifnet *ifp = adapter->ifp;
IGB_CORE_LOCK_ASSERT(adapter);
Add a link tasklet so updates can be sleepable.
2010-01-27 20:12:04 +00:00
adapter->hw.mac.get_link_status = 1;
igb_update_link_status(adapter);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) && adapter->link_active) {
for (int i = 0; i < adapter->num_queues; i++, txr++) {
IGB_TX_LOCK(txr);
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
/* Process the stack queue only if not depleted */
if (((txr->queue_status & IGB_QUEUE_DEPLETED) == 0) &&
!drbr_empty(ifp, txr->br))
drbr_enqueue() awlays consumes mbuf, no matter did it fail or not. The mbuf pointer is no longer valid, so can't be reused after. Fix igb_mq_start() where mbuf pointer was used after drbr_enqueue(). This eventually leads us to all invocations of igb_mq_start_locked() called with third argument as NULL. This allows us to simplify this function. Submitted by: Karim Fodil-Lemelin <fodillemlinkarim gmail.com> Reviewed by: jfv
2012-11-26 20:03:57 +00:00
igb_mq_start_locked(ifp, txr);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#else
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
igb_start_locked(txr, ifp);
#endif
IGB_TX_UNLOCK(txr);
}
}
Add a link tasklet so updates can be sleepable.
2010-01-27 20:12:04 +00:00
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
* MSI/Legacy Deferred
* Interrupt Service routine
*
*********************************************************************/
static int
igb_irq_fast(void *arg)
{
Change to have legacy interrupts use the same handler had a flaw, thanks to John Baldwin for finding it. Change which queue legacy tasks are enqueued on. MFC: soonest
2010-06-15 21:11:51 +00:00
struct adapter *adapter = arg;
struct igb_queue *que = adapter->queues;
u32 reg_icr;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
/* Hot eject? */
if (reg_icr == 0xffffffff)
return FILTER_STRAY;
/* Definitely not our interrupt. */
if (reg_icr == 0x0)
return FILTER_STRAY;
if ((reg_icr & E1000_ICR_INT_ASSERTED) == 0)
return FILTER_STRAY;
/*
* Mask interrupts until the taskqueue is finished running. This is
* cheap, just assume that it is needed. This also works around the
* MSI message reordering errata on certain systems.
*/
igb_disable_intr(adapter);
Change to have legacy interrupts use the same handler had a flaw, thanks to John Baldwin for finding it. Change which queue legacy tasks are enqueued on. MFC: soonest
2010-06-15 21:11:51 +00:00
taskqueue_enqueue(que->tq, &que->que_task);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Link status change */
Add a link tasklet so updates can be sleepable.
2010-01-27 20:12:04 +00:00
if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))
Change to have legacy interrupts use the same handler had a flaw, thanks to John Baldwin for finding it. Change which queue legacy tasks are enqueued on. MFC: soonest
2010-06-15 21:11:51 +00:00
taskqueue_enqueue(que->tq, &adapter->link_task);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (reg_icr & E1000_ICR_RXO)
adapter->rx_overruns++;
return FILTER_HANDLED;
}
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#ifdef DEVICE_POLLING
#if __FreeBSD_version >= 800000
#define POLL_RETURN_COUNT(a) (a)
static int
#else
#define POLL_RETURN_COUNT(a)
static void
#endif
igb_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
{
The POLL code was missed in the queue conversion, change the argument type to igb_rxeof() to the correct type. Note, any users of POLLING must be sure and set the number of queues to 1 for things to work correctly.
2010-03-31 23:24:42 +00:00
struct adapter *adapter = ifp->if_softc;
Make the polling interface in igb able to handle multiqueue, and correct the rxdone handling. Update the polling man page to include igb as well. Thanks to Mark Johnston for these changes.
2012-08-06 22:43:49 +00:00
struct igb_queue *que;
struct tx_ring *txr;
The POLL code was missed in the queue conversion, change the argument type to igb_rxeof() to the correct type. Note, any users of POLLING must be sure and set the number of queues to 1 for things to work correctly.
2010-03-31 23:24:42 +00:00
u32 reg_icr, rx_done = 0;
u32 loop = IGB_MAX_LOOP;
bool more;
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
IGB_CORE_LOCK(adapter);
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
IGB_CORE_UNLOCK(adapter);
return POLL_RETURN_COUNT(rx_done);
}
if (cmd == POLL_AND_CHECK_STATUS) {
reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
/* Link status change */
if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
igb_handle_link_locked(adapter);
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
if (reg_icr & E1000_ICR_RXO)
adapter->rx_overruns++;
}
IGB_CORE_UNLOCK(adapter);
Make the polling interface in igb able to handle multiqueue, and correct the rxdone handling. Update the polling man page to include igb as well. Thanks to Mark Johnston for these changes.
2012-08-06 22:43:49 +00:00
for (int i = 0; i < adapter->num_queues; i++) {
que = &adapter->queues[i];
txr = que->txr;
igb_rxeof(que, count, &rx_done);
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
Make the polling interface in igb able to handle multiqueue, and correct the rxdone handling. Update the polling man page to include igb as well. Thanks to Mark Johnston for these changes.
2012-08-06 22:43:49 +00:00
IGB_TX_LOCK(txr);
do {
more = igb_txeof(txr);
} while (loop-- && more);
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Make the polling interface in igb able to handle multiqueue, and correct the rxdone handling. Update the polling man page to include igb as well. Thanks to Mark Johnston for these changes.
2012-08-06 22:43:49 +00:00
if (!drbr_empty(ifp, txr->br))
drbr_enqueue() awlays consumes mbuf, no matter did it fail or not. The mbuf pointer is no longer valid, so can't be reused after. Fix igb_mq_start() where mbuf pointer was used after drbr_enqueue(). This eventually leads us to all invocations of igb_mq_start_locked() called with third argument as NULL. This allows us to simplify this function. Submitted by: Karim Fodil-Lemelin <fodillemlinkarim gmail.com> Reviewed by: jfv
2012-11-26 20:03:57 +00:00
igb_mq_start_locked(ifp, txr);
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#else
Make the polling interface in igb able to handle multiqueue, and correct the rxdone handling. Update the polling man page to include igb as well. Thanks to Mark Johnston for these changes.
2012-08-06 22:43:49 +00:00
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
igb_start_locked(txr, ifp);
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#endif
Make the polling interface in igb able to handle multiqueue, and correct the rxdone handling. Update the polling man page to include igb as well. Thanks to Mark Johnston for these changes.
2012-08-06 22:43:49 +00:00
IGB_TX_UNLOCK(txr);
}
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
return POLL_RETURN_COUNT(rx_done);
}
#endif /* DEVICE_POLLING */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
* MSIX Que Interrupt Service routine
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*
**********************************************************************/
static void
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
igb_msix_que(void *arg)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct igb_queue *que = arg;
struct adapter *adapter = que->adapter;
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
struct ifnet *ifp = adapter->ifp;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct tx_ring *txr = que->txr;
struct rx_ring *rxr = que->rxr;
u32 newitr = 0;
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
bool more_rx;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
When the igb driver is static there are cases when early interrupts occur, resulting in a panic in refresh_mbufs, to prevent this add a check in the interrupt handler for DRV_RUNNING. MFC after: 1 day (critical for 9.2)
2013-08-06 18:00:53 +00:00
/* Ignore spurious interrupts */
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
return;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_EIMC, que->eims);
++que->irqs;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
IGB_TX_LOCK(txr);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
igb_txeof(txr);
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
/* Process the stack queue only if not depleted */
if (((txr->queue_status & IGB_QUEUE_DEPLETED) == 0) &&
!drbr_empty(ifp, txr->br))
drbr_enqueue() awlays consumes mbuf, no matter did it fail or not. The mbuf pointer is no longer valid, so can't be reused after. Fix igb_mq_start() where mbuf pointer was used after drbr_enqueue(). This eventually leads us to all invocations of igb_mq_start_locked() called with third argument as NULL. This allows us to simplify this function. Submitted by: Karim Fodil-Lemelin <fodillemlinkarim gmail.com> Reviewed by: jfv
2012-11-26 20:03:57 +00:00
igb_mq_start_locked(ifp, txr);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#else
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
igb_start_locked(txr, ifp);
#endif
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
IGB_TX_UNLOCK(txr);
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
more_rx = igb_rxeof(que, adapter->rx_process_limit, NULL);
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
Restore the sysctl changes from 223676 and 227309 lost in the previous import: - Add read-only sysctls for all of the tunables supported by the igb and em drivers. - Make the per-instance 'enable_aim' sysctl truly per-instance by having it change a per-instance variable (which is used to control AIM) rather than having all of the per-instance sysctls operate on a single global variable. While here, restore the previously existing hw.igb.rx_processing_limit tunable as it is very useful to be able to set a default tunable that applies to all adapters in the system.
2011-12-21 20:10:11 +00:00
if (adapter->enable_aim == FALSE)
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
goto no_calc;
/*
** Do Adaptive Interrupt Moderation:
** - Write out last calculated setting
** - Calculate based on average size over
** the last interval.
*/
if (que->eitr_setting)
E1000_WRITE_REG(&adapter->hw,
E1000_EITR(que->msix), que->eitr_setting);
que->eitr_setting = 0;
/* Idle, do nothing */
if ((txr->bytes == 0) && (rxr->bytes == 0))
goto no_calc;
/* Used half Default if sub-gig */
if (adapter->link_speed != 1000)
newitr = IGB_DEFAULT_ITR / 2;
else {
if ((txr->bytes) && (txr->packets))
newitr = txr->bytes/txr->packets;
if ((rxr->bytes) && (rxr->packets))
newitr = max(newitr,
(rxr->bytes / rxr->packets));
newitr += 24; /* account for hardware frame, crc */
/* set an upper boundary */
newitr = min(newitr, 3000);
/* Be nice to the mid range */
if ((newitr > 300) && (newitr < 1200))
newitr = (newitr / 3);
else
newitr = (newitr / 2);
}
newitr &= 0x7FFC; /* Mask invalid bits */
if (adapter->hw.mac.type == e1000_82575)
newitr |= newitr << 16;
else
Thanks to Michael Tuexen for catching this, bit set that keeps the clock from being reset when writing to EITR was incorrect, also there is a shared code #define for it anyway.
2010-04-09 21:16:45 +00:00
newitr |= E1000_EITR_CNT_IGNR;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* save for next interrupt */
que->eitr_setting = newitr;
/* Reset state */
txr->bytes = 0;
txr->packets = 0;
rxr->bytes = 0;
rxr->packets = 0;
no_calc:
/* Schedule a clean task if needed*/
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
if (more_rx)
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
taskqueue_enqueue(que->tq, &que->que_task);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
else
/* Reenable this interrupt */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_EIMS, que->eims);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
return;
}
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/*********************************************************************
*
* MSIX Link Interrupt Service routine
*
**********************************************************************/
static void
igb_msix_link(void *arg)
{
struct adapter *adapter = arg;
u32 icr;
++adapter->link_irq;
icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
if (!(icr & E1000_ICR_LSC))
goto spurious;
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
igb_handle_link(adapter, 0);
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
spurious:
/* Rearm */
E1000_WRITE_REG(&adapter->hw, E1000_IMS, E1000_IMS_LSC);
E1000_WRITE_REG(&adapter->hw, E1000_EIMS, adapter->link_mask);
return;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
* Media Ioctl callback
*
* This routine is called whenever the user queries the status of
* the interface using ifconfig.
*
**********************************************************************/
static void
igb_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
{
struct adapter *adapter = ifp->if_softc;
INIT_DEBUGOUT("igb_media_status: begin");
IGB_CORE_LOCK(adapter);
igb_update_link_status(adapter);
ifmr->ifm_status = IFM_AVALID;
ifmr->ifm_active = IFM_ETHER;
if (!adapter->link_active) {
IGB_CORE_UNLOCK(adapter);
return;
}
ifmr->ifm_status |= IFM_ACTIVE;
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
switch (adapter->link_speed) {
case 10:
ifmr->ifm_active |= IFM_10_T;
break;
case 100:
/*
** Support for 100Mb SFP - these are Fiber
** but the media type appears as serdes
*/
if (adapter->hw.phy.media_type ==
e1000_media_type_internal_serdes)
ifmr->ifm_active |= IFM_100_FX;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
else
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
ifmr->ifm_active |= IFM_100_TX;
break;
case 1000:
ifmr->ifm_active |= IFM_1000_T;
break;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
case 2500:
ifmr->ifm_active |= IFM_2500_SX;
break;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
if (adapter->link_duplex == FULL_DUPLEX)
ifmr->ifm_active |= IFM_FDX;
else
ifmr->ifm_active |= IFM_HDX;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IGB_CORE_UNLOCK(adapter);
}
/*********************************************************************
*
* Media Ioctl callback
*
* This routine is called when the user changes speed/duplex using
* media/mediopt option with ifconfig.
*
**********************************************************************/
static int
igb_media_change(struct ifnet *ifp)
{
struct adapter *adapter = ifp->if_softc;
struct ifmedia *ifm = &adapter->media;
INIT_DEBUGOUT("igb_media_change: begin");
if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
return (EINVAL);
IGB_CORE_LOCK(adapter);
switch (IFM_SUBTYPE(ifm->ifm_media)) {
case IFM_AUTO:
adapter->hw.mac.autoneg = DO_AUTO_NEG;
adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
break;
case IFM_1000_LX:
case IFM_1000_SX:
case IFM_1000_T:
adapter->hw.mac.autoneg = DO_AUTO_NEG;
adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
break;
case IFM_100_TX:
adapter->hw.mac.autoneg = FALSE;
adapter->hw.phy.autoneg_advertised = 0;
if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
else
adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
break;
case IFM_10_T:
adapter->hw.mac.autoneg = FALSE;
adapter->hw.phy.autoneg_advertised = 0;
if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
else
adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
break;
default:
device_printf(adapter->dev, "Unsupported media type\n");
}
igb_init_locked(adapter);
IGB_CORE_UNLOCK(adapter);
return (0);
}
/*********************************************************************
*
* This routine maps the mbufs to Advanced TX descriptors.
*
**********************************************************************/
static int
igb_xmit(struct tx_ring *txr, struct mbuf **m_headp)
{
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
struct adapter *adapter = txr->adapter;
u32 olinfo_status = 0, cmd_type_len;
int i, j, error, nsegs;
int first;
bool remap = TRUE;
struct mbuf *m_head;
bus_dma_segment_t segs[IGB_MAX_SCATTER];
bus_dmamap_t map;
struct igb_tx_buf *txbuf;
union e1000_adv_tx_desc *txd = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
m_head = *m_headp;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* Basic descriptor defines */
cmd_type_len = (E1000_ADVTXD_DTYP_DATA |
E1000_ADVTXD_DCMD_IFCS | E1000_ADVTXD_DCMD_DEXT);
if (m_head->m_flags & M_VLANTAG)
cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
/*
* Important to capture the first descriptor
* used because it will contain the index of
* the one we tell the hardware to report back
*/
first = txr->next_avail_desc;
txbuf = &txr->tx_buffers[first];
map = txbuf->map;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
/*
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
* Map the packet for DMA.
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*/
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
retry:
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
error = bus_dmamap_load_mbuf_sg(txr->txtag, map,
*m_headp, segs, &nsegs, BUS_DMA_NOWAIT);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (__predict_false(error)) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
struct mbuf *m;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
switch (error) {
case EFBIG:
/* Try it again? - one try */
if (remap == TRUE) {
remap = FALSE;
m = m_defrag(*m_headp, M_NOWAIT);
if (m == NULL) {
adapter->mbuf_defrag_failed++;
m_freem(*m_headp);
*m_headp = NULL;
return (ENOBUFS);
}
*m_headp = m;
goto retry;
} else
return (error);
case ENOMEM:
txr->no_tx_dma_setup++;
return (error);
default:
txr->no_tx_dma_setup++;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
m_freem(*m_headp);
*m_headp = NULL;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
return (error);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
}
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* Make certain there are enough descriptors */
if (nsegs > txr->tx_avail - 2) {
txr->no_desc_avail++;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
bus_dmamap_unload(txr->txtag, map);
return (ENOBUFS);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
m_head = *m_headp;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/*
** Set up the appropriate offload context
** this will consume the first descriptor
*/
error = igb_tx_ctx_setup(txr, m_head, &cmd_type_len, &olinfo_status);
if (__predict_false(error)) {
m_freem(*m_headp);
*m_headp = NULL;
return (error);
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* 82575 needs the queue index added */
if (adapter->hw.mac.type == e1000_82575)
olinfo_status |= txr->me << 4;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
i = txr->next_avail_desc;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
for (j = 0; j < nsegs; j++) {
bus_size_t seglen;
bus_addr_t segaddr;
txbuf = &txr->tx_buffers[i];
txd = &txr->tx_base[i];
seglen = segs[j].ds_len;
segaddr = htole64(segs[j].ds_addr);
txd->read.buffer_addr = segaddr;
txd->read.cmd_type_len = htole32(E1000_TXD_CMD_IFCS |
cmd_type_len | seglen);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
txd->read.olinfo_status = htole32(olinfo_status);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (++i == txr->num_desc)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
i = 0;
}
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
txd->read.cmd_type_len |=
htole32(E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
txr->tx_avail -= nsegs;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
txr->next_avail_desc = i;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
txbuf->m_head = m_head;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
/*
** Here we swap the map so the last descriptor,
** which gets the completion interrupt has the
** real map, and the first descriptor gets the
** unused map from this descriptor.
*/
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
txr->tx_buffers[first].map = txbuf->map;
txbuf->map = map;
bus_dmamap_sync(txr->txtag, map, BUS_DMASYNC_PREWRITE);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* Set the EOP descriptor that will be marked done */
txbuf = &txr->tx_buffers[first];
txbuf->eop = txd;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
* Advance the Transmit Descriptor Tail (Tdt), this tells the
* hardware that this frame is available to transmit.
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*/
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
++txr->total_packets;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_TDT(txr->me), i);
return (0);
}
static void
igb_set_promisc(struct adapter *adapter)
{
struct ifnet *ifp = adapter->ifp;
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
struct e1000_hw *hw = &adapter->hw;
u32 reg;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->vf_ifp) {
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
e1000_promisc_set_vf(hw, e1000_promisc_enabled);
return;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
reg = E1000_READ_REG(hw, E1000_RCTL);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (ifp->if_flags & IFF_PROMISC) {
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
reg |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
E1000_WRITE_REG(hw, E1000_RCTL, reg);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
} else if (ifp->if_flags & IFF_ALLMULTI) {
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
reg |= E1000_RCTL_MPE;
reg &= ~E1000_RCTL_UPE;
E1000_WRITE_REG(hw, E1000_RCTL, reg);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
}
static void
igb_disable_promisc(struct adapter *adapter)
{
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
struct e1000_hw *hw = &adapter->hw;
Correct the multicast handling in the E1000 drivers as was done in ixgbe, thanks to Mike Karels for this fix. When exiting promiscuous mode MPE bit was being unconditionally cleared, this should not be done if we are in MAX multicast groups.
2013-04-03 23:39:54 +00:00
struct ifnet *ifp = adapter->ifp;
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
u32 reg;
Correct the multicast handling in the E1000 drivers as was done in ixgbe, thanks to Mike Karels for this fix. When exiting promiscuous mode MPE bit was being unconditionally cleared, this should not be done if we are in MAX multicast groups.
2013-04-03 23:39:54 +00:00
int mcnt = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->vf_ifp) {
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
e1000_promisc_set_vf(hw, e1000_promisc_disabled);
return;
}
reg = E1000_READ_REG(hw, E1000_RCTL);
reg &= (~E1000_RCTL_UPE);
Correct the multicast handling in the E1000 drivers as was done in ixgbe, thanks to Mike Karels for this fix. When exiting promiscuous mode MPE bit was being unconditionally cleared, this should not be done if we are in MAX multicast groups.
2013-04-03 23:39:54 +00:00
if (ifp->if_flags & IFF_ALLMULTI)
mcnt = MAX_NUM_MULTICAST_ADDRESSES;
else {
struct ifmultiaddr *ifma;
#if __FreeBSD_version < 800000
IF_ADDR_LOCK(ifp);
#else
if_maddr_rlock(ifp);
#endif
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
break;
mcnt++;
}
#if __FreeBSD_version < 800000
IF_ADDR_UNLOCK(ifp);
#else
if_maddr_runlock(ifp);
#endif
}
/* Don't disable if in MAX groups */
if (mcnt < MAX_NUM_MULTICAST_ADDRESSES)
reg &= (~E1000_RCTL_MPE);
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
E1000_WRITE_REG(hw, E1000_RCTL, reg);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*********************************************************************
* Multicast Update
*
* This routine is called whenever multicast address list is updated.
*
**********************************************************************/
static void
igb_set_multi(struct adapter *adapter)
{
struct ifnet *ifp = adapter->ifp;
struct ifmultiaddr *ifma;
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
u32 reg_rctl = 0;
Do not allocate multicast array memory in multicast filter configuration function. For failed memory allocations, em(4)/lem(4) called panic(9) which is not acceptable on production box. igb(4)/ixgb(4)/ix(4) allocated the required memory in stack which consumed 768 bytes of stack memory which looks too big. To address these issues, allocate multicast array memory in device attach time and make multicast configuration success under any conditions. This change also removes the excessive use of memory in stack. Reviewed by: jfv
2010-08-28 00:34:22 +00:00
u8 *mta;
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
int mcnt = 0;
IOCTL_DEBUGOUT("igb_set_multi: begin");
Do not allocate multicast array memory in multicast filter configuration function. For failed memory allocations, em(4)/lem(4) called panic(9) which is not acceptable on production box. igb(4)/ixgb(4)/ix(4) allocated the required memory in stack which consumed 768 bytes of stack memory which looks too big. To address these issues, allocate multicast array memory in device attach time and make multicast configuration success under any conditions. This change also removes the excessive use of memory in stack. Reviewed by: jfv
2010-08-28 00:34:22 +00:00
mta = adapter->mta;
bzero(mta, sizeof(uint8_t) * ETH_ADDR_LEN *
MAX_NUM_MULTICAST_ADDRESSES);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
#if __FreeBSD_version < 800000
IF_ADDR_LOCK(ifp);
#else
Use if_maddr_rlock()/if_maddr_runlock() rather than IF_ADDR_LOCK()/ IF_ADDR_UNLOCK() across network device drivers when accessing the per-interface multicast address list, if_multiaddrs. This will allow us to change the locking strategy without affecting our driver programming interface or binary interface. For two wireless drivers, remove unnecessary locking, since they don't actually access the multicast address list. Approved by: re (kib) MFC after: 6 weeks
2009-06-26 11:45:06 +00:00
if_maddr_rlock(ifp);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
break;
bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
&mta[mcnt * ETH_ADDR_LEN], ETH_ADDR_LEN);
mcnt++;
}
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
#if __FreeBSD_version < 800000
IF_ADDR_UNLOCK(ifp);
#else
Use if_maddr_rlock()/if_maddr_runlock() rather than IF_ADDR_LOCK()/ IF_ADDR_UNLOCK() across network device drivers when accessing the per-interface multicast address list, if_multiaddrs. This will allow us to change the locking strategy without affecting our driver programming interface or binary interface. For two wireless drivers, remove unnecessary locking, since they don't actually access the multicast address list. Approved by: re (kib) MFC after: 6 weeks
2009-06-26 11:45:06 +00:00
if_maddr_runlock(ifp);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
#endif
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
reg_rctl |= E1000_RCTL_MPE;
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
} else
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
e1000_update_mc_addr_list(&adapter->hw, mta, mcnt);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*********************************************************************
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
* Timer routine:
* This routine checks for link status,
* updates statistics, and does the watchdog.
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*
**********************************************************************/
static void
igb_local_timer(void *arg)
{
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
struct adapter *adapter = arg;
device_t dev = adapter->dev;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
struct ifnet *ifp = adapter->ifp;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
struct tx_ring *txr = adapter->tx_rings;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
struct igb_queue *que = adapter->queues;
int hung = 0, busy = 0;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IGB_CORE_LOCK_ASSERT(adapter);
igb_update_link_status(adapter);
igb_update_stats_counters(adapter);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/*
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
** Check the TX queues status
** - central locked handling of OACTIVE
** - watchdog only if all queues show hung
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
*/
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
for (int i = 0; i < adapter->num_queues; i++, que++, txr++) {
if ((txr->queue_status & IGB_QUEUE_HUNG) &&
(adapter->pause_frames == 0))
++hung;
if (txr->queue_status & IGB_QUEUE_DEPLETED)
++busy;
if ((txr->queue_status & IGB_QUEUE_IDLE) == 0)
taskqueue_enqueue(que->tq, &que->que_task);
}
if (hung == adapter->num_queues)
goto timeout;
if (busy == adapter->num_queues)
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
else if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) &&
(busy < adapter->num_queues))
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
adapter->pause_frames = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
callout_reset(&adapter->timer, hz, igb_local_timer, adapter);
Somehow the RX ring depletion fix got partially removed, replace the missing pieces.
2011-02-11 19:49:07 +00:00
#ifndef DEVICE_POLLING
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
/* Schedule all queue interrupts - deadlock protection */
Somehow the RX ring depletion fix got partially removed, replace the missing pieces.
2011-02-11 19:49:07 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_EICS, adapter->que_mask);
#endif
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
return;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
timeout:
device_printf(adapter->dev, "Watchdog timeout -- resetting\n");
device_printf(dev,"Queue(%d) tdh = %d, hw tdt = %d\n", txr->me,
E1000_READ_REG(&adapter->hw, E1000_TDH(txr->me)),
E1000_READ_REG(&adapter->hw, E1000_TDT(txr->me)));
device_printf(dev,"TX(%d) desc avail = %d,"
"Next TX to Clean = %d\n",
txr->me, txr->tx_avail, txr->next_to_clean);
adapter->ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
adapter->watchdog_events++;
igb_init_locked(adapter);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
static void
igb_update_link_status(struct adapter *adapter)
{
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
struct e1000_hw *hw = &adapter->hw;
struct e1000_fc_info *fc = &hw->fc;
struct ifnet *ifp = adapter->ifp;
device_t dev = adapter->dev;
struct tx_ring *txr = adapter->tx_rings;
u32 link_check, thstat, ctrl;
char *flowctl = NULL;
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
link_check = thstat = ctrl = 0;
This update primarily addresses the ability to have both the em and the igb driver static in the kernel. But it also reflects some other bug fixes in my development stream at Intel. PR 122373 is also fixed in this code.
2008-04-02 22:00:36 +00:00
/* Get the cached link value or read for real */
switch (hw->phy.media_type) {
case e1000_media_type_copper:
if (hw->mac.get_link_status) {
/* Do the work to read phy */
e1000_check_for_link(hw);
link_check = !hw->mac.get_link_status;
} else
link_check = TRUE;
break;
case e1000_media_type_fiber:
e1000_check_for_link(hw);
link_check = (E1000_READ_REG(hw, E1000_STATUS) &
E1000_STATUS_LU);
break;
case e1000_media_type_internal_serdes:
e1000_check_for_link(hw);
link_check = adapter->hw.mac.serdes_has_link;
break;
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
/* VF device is type_unknown */
This update primarily addresses the ability to have both the em and the igb driver static in the kernel. But it also reflects some other bug fixes in my development stream at Intel. PR 122373 is also fixed in this code.
2008-04-02 22:00:36 +00:00
case e1000_media_type_unknown:
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
e1000_check_for_link(hw);
link_check = !hw->mac.get_link_status;
/* Fall thru */
default:
This update primarily addresses the ability to have both the em and the igb driver static in the kernel. But it also reflects some other bug fixes in my development stream at Intel. PR 122373 is also fixed in this code.
2008-04-02 22:00:36 +00:00
break;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
/* Check for thermal downshift or shutdown */
if (hw->mac.type == e1000_i350) {
thstat = E1000_READ_REG(hw, E1000_THSTAT);
ctrl = E1000_READ_REG(hw, E1000_CTRL_EXT);
}
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
/* Get the flow control for display */
switch (fc->current_mode) {
case e1000_fc_rx_pause:
flowctl = "RX";
break;
case e1000_fc_tx_pause:
flowctl = "TX";
break;
case e1000_fc_full:
flowctl = "Full";
break;
case e1000_fc_none:
default:
flowctl = "None";
break;
}
This update primarily addresses the ability to have both the em and the igb driver static in the kernel. But it also reflects some other bug fixes in my development stream at Intel. PR 122373 is also fixed in this code.
2008-04-02 22:00:36 +00:00
/* Now we check if a transition has happened */
if (link_check && (adapter->link_active == 0)) {
e1000_get_speed_and_duplex(&adapter->hw,
&adapter->link_speed, &adapter->link_duplex);
if (bootverbose)
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
device_printf(dev, "Link is up %d Mbps %s,"
" Flow Control: %s\n",
This update primarily addresses the ability to have both the em and the igb driver static in the kernel. But it also reflects some other bug fixes in my development stream at Intel. PR 122373 is also fixed in this code.
2008-04-02 22:00:36 +00:00
adapter->link_speed,
((adapter->link_duplex == FULL_DUPLEX) ?
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
"Full Duplex" : "Half Duplex"), flowctl);
This update primarily addresses the ability to have both the em and the igb driver static in the kernel. But it also reflects some other bug fixes in my development stream at Intel. PR 122373 is also fixed in this code.
2008-04-02 22:00:36 +00:00
adapter->link_active = 1;
ifp->if_baudrate = adapter->link_speed * 1000000;
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
(thstat & E1000_THSTAT_LINK_THROTTLE))
device_printf(dev, "Link: thermal downshift\n");
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* Delay Link Up for Phy update */
if (((hw->mac.type == e1000_i210) ||
(hw->mac.type == e1000_i211)) &&
(hw->phy.id == I210_I_PHY_ID))
msec_delay(I210_LINK_DELAY);
/* Reset if the media type changed. */
if (hw->dev_spec._82575.media_changed) {
hw->dev_spec._82575.media_changed = false;
adapter->flags |= IGB_MEDIA_RESET;
igb_reset(adapter);
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* This can sleep */
This update primarily addresses the ability to have both the em and the igb driver static in the kernel. But it also reflects some other bug fixes in my development stream at Intel. PR 122373 is also fixed in this code.
2008-04-02 22:00:36 +00:00
if_link_state_change(ifp, LINK_STATE_UP);
} else if (!link_check && (adapter->link_active == 1)) {
ifp->if_baudrate = adapter->link_speed = 0;
adapter->link_duplex = 0;
if (bootverbose)
device_printf(dev, "Link is Down\n");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) &&
(thstat & E1000_THSTAT_PWR_DOWN))
device_printf(dev, "Link: thermal shutdown\n");
This update primarily addresses the ability to have both the em and the igb driver static in the kernel. But it also reflects some other bug fixes in my development stream at Intel. PR 122373 is also fixed in this code.
2008-04-02 22:00:36 +00:00
adapter->link_active = 0;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* This can sleep */
This update primarily addresses the ability to have both the em and the igb driver static in the kernel. But it also reflects some other bug fixes in my development stream at Intel. PR 122373 is also fixed in this code.
2008-04-02 22:00:36 +00:00
if_link_state_change(ifp, LINK_STATE_DOWN);
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
/* Reset queue state */
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++)
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
txr->queue_status = IGB_QUEUE_IDLE;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
}
/*********************************************************************
*
* This routine disables all traffic on the adapter by issuing a
* global reset on the MAC and deallocates TX/RX buffers.
*
**********************************************************************/
static void
igb_stop(void *arg)
{
struct adapter *adapter = arg;
struct ifnet *ifp = adapter->ifp;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct tx_ring *txr = adapter->tx_rings;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IGB_CORE_LOCK_ASSERT(adapter);
INIT_DEBUGOUT("igb_stop: begin");
igb_disable_intr(adapter);
callout_stop(&adapter->timer);
/* Tell the stack that the interface is no longer active */
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
/* Disarm watchdog timer. */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++) {
IGB_TX_LOCK(txr);
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
txr->queue_status = IGB_QUEUE_IDLE;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
IGB_TX_UNLOCK(txr);
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
e1000_reset_hw(&adapter->hw);
E1000_WRITE_REG(&adapter->hw, E1000_WUC, 0);
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
e1000_led_off(&adapter->hw);
e1000_cleanup_led(&adapter->hw);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*********************************************************************
*
* Determine hardware revision.
*
**********************************************************************/
static void
igb_identify_hardware(struct adapter *adapter)
{
device_t dev = adapter->dev;
/* Make sure our PCI config space has the necessary stuff set */
Update PCI drivers to no longer look at the MEMIO-enabled bit in the PCI command register. The lazy BAR allocation code in FreeBSD sometimes disables this bit when it detects a range conflict, and will re-enable it on demand when a driver allocates the BAR. Thus, the bit is no longer a reliable indication of capability, and should not be checked. This results in the elimination of a lot of code from drivers, and also gives the opportunity to simplify a lot of drivers to use a helper API to set the busmaster enable bit. This changes fixes some recent reports of disk controllers and their associated drives/enclosures disappearing during boot. Submitted by: jhb Reviewed by: jfv, marius, achadd, achim MFC after: 1 day
2013-08-12 23:30:01 +00:00
pci_enable_busmaster(dev);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
adapter->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
/* Save off the information about this board */
adapter->hw.vendor_id = pci_get_vendor(dev);
adapter->hw.device_id = pci_get_device(dev);
adapter->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1);
adapter->hw.subsystem_vendor_id =
pci_read_config(dev, PCIR_SUBVEND_0, 2);
adapter->hw.subsystem_device_id =
pci_read_config(dev, PCIR_SUBDEV_0, 2);
Put the early setting of the MAC type back, its removal resulted in broken code in MSIX setup.
2010-08-06 20:55:49 +00:00
/* Set MAC type early for PCI setup */
e1000_set_mac_type(&adapter->hw);
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
/* Are we a VF device? */
if ((adapter->hw.mac.type == e1000_vfadapt) ||
(adapter->hw.mac.type == e1000_vfadapt_i350))
adapter->vf_ifp = 1;
else
adapter->vf_ifp = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
static int
igb_allocate_pci_resources(struct adapter *adapter)
{
device_t dev = adapter->dev;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
int rid;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
rid = PCIR_BAR(0);
adapter->pci_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
&rid, RF_ACTIVE);
if (adapter->pci_mem == NULL) {
device_printf(dev, "Unable to allocate bus resource: memory\n");
return (ENXIO);
}
adapter->osdep.mem_bus_space_tag =
rman_get_bustag(adapter->pci_mem);
adapter->osdep.mem_bus_space_handle =
rman_get_bushandle(adapter->pci_mem);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
adapter->hw.hw_addr = (u8 *)&adapter->osdep.mem_bus_space_handle;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
adapter->num_queues = 1; /* Defaults for Legacy or MSI */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* This will setup either MSI/X or MSI */
adapter->msix = igb_setup_msix(adapter);
adapter->hw.back = &adapter->osdep;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
return (0);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*********************************************************************
*
* Setup the Legacy or MSI Interrupt handler
*
**********************************************************************/
static int
igb_allocate_legacy(struct adapter *adapter)
{
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
device_t dev = adapter->dev;
struct igb_queue *que = adapter->queues;
Fix compilation with IGB_LEGACY_TX defined. PR: 185909 Submitted by: Aurelien Rougemont <beorn binaries.fr>
2014-01-25 20:39:23 +00:00
#ifndef IGB_LEGACY_TX
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
struct tx_ring *txr = adapter->tx_rings;
Fix compilation with IGB_LEGACY_TX defined. PR: 185909 Submitted by: Aurelien Rougemont <beorn binaries.fr>
2014-01-25 20:39:23 +00:00
#endif
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
int error, rid = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Turn off all interrupts */
E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/* MSI RID is 1 */
if (adapter->msix == 1)
rid = 1;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* We allocate a single interrupt resource */
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
adapter->res = bus_alloc_resource_any(dev,
SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE);
if (adapter->res == NULL) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_printf(dev, "Unable to allocate bus resource: "
"interrupt\n");
return (ENXIO);
}
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
TASK_INIT(&txr->txq_task, 0, igb_deferred_mq_start, txr);
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
* Try allocating a fast interrupt and the associated deferred
* processing contexts.
*/
Change to have legacy interrupts use the same handler had a flaw, thanks to John Baldwin for finding it. Change which queue legacy tasks are enqueued on. MFC: soonest
2010-06-15 21:11:51 +00:00
TASK_INIT(&que->que_task, 0, igb_handle_que, que);
Add a link tasklet so updates can be sleepable.
2010-01-27 20:12:04 +00:00
/* Make tasklet for deferred link handling */
TASK_INIT(&adapter->link_task, 0, igb_handle_link, adapter);
Change to have legacy interrupts use the same handler had a flaw, thanks to John Baldwin for finding it. Change which queue legacy tasks are enqueued on. MFC: soonest
2010-06-15 21:11:51 +00:00
que->tq = taskqueue_create_fast("igb_taskq", M_NOWAIT,
taskqueue_thread_enqueue, &que->tq);
taskqueue_start_threads(&que->tq, 1, PI_NET, "%s taskq",
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_get_nameunit(adapter->dev));
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if ((error = bus_setup_intr(dev, adapter->res,
INTR_TYPE_NET | INTR_MPSAFE, igb_irq_fast, NULL,
adapter, &adapter->tag)) != 0) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_printf(dev, "Failed to register fast interrupt "
"handler: %d\n", error);
Change to have legacy interrupts use the same handler had a flaw, thanks to John Baldwin for finding it. Change which queue legacy tasks are enqueued on. MFC: soonest
2010-06-15 21:11:51 +00:00
taskqueue_free(que->tq);
que->tq = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
return (error);
}
return (0);
}
/*********************************************************************
*
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
* Setup the MSIX Queue Interrupt handlers:
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*
**********************************************************************/
static int
igb_allocate_msix(struct adapter *adapter)
{
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
device_t dev = adapter->dev;
struct igb_queue *que = adapter->queues;
int error, rid, vector = 0;
Initialise these variables so gcc doesn't complain. Submitted by: luigi
2014-06-30 23:34:36 +00:00
int cpu_id = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
/* Be sure to start with all interrupts disabled */
E1000_WRITE_REG(&adapter->hw, E1000_IMC, ~0);
E1000_WRITE_FLUSH(&adapter->hw);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
#ifdef RSS
/*
* If we're doing RSS, the number of queues needs to
* match the number of RSS buckets that are configured.
*
* + If there's more queues than RSS buckets, we'll end
* up with queues that get no traffic.
*
* + If there's more RSS buckets than queues, we'll end
* up having multiple RSS buckets map to the same queue,
* so there'll be some contention.
*/
if (adapter->num_queues != rss_getnumbuckets()) {
device_printf(dev,
"%s: number of queues (%d) != number of RSS buckets (%d)"
"; performance will be impacted.\n",
__func__,
adapter->num_queues,
rss_getnumbuckets());
}
#endif
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
for (int i = 0; i < adapter->num_queues; i++, vector++, que++) {
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
rid = vector +1;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
que->res = bus_alloc_resource_any(dev,
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (que->res == NULL) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_printf(dev,
"Unable to allocate bus resource: "
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
"MSIX Queue Interrupt\n");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
return (ENXIO);
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
error = bus_setup_intr(dev, que->res,
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
INTR_TYPE_NET | INTR_MPSAFE, NULL,
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
igb_msix_que, que, &que->tag);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (error) {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
que->res = NULL;
device_printf(dev, "Failed to register Queue handler");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
return (error);
}
Put back the lost bus_describe_intr() calls.
2010-06-11 21:35:19 +00:00
#if __FreeBSD_version >= 800504
bus_describe_intr(dev, que->res, que->tag, "que %d", i);
#endif
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
que->msix = vector;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if (adapter->hw.mac.type == e1000_82575)
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
que->eims = E1000_EICR_TX_QUEUE0 << i;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
else
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
que->eims = 1 << vector;
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
#ifdef RSS
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/*
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
* The queue ID is used as the RSS layer bucket ID.
* We look up the queue ID -> RSS CPU ID and select
* that.
*/
cpu_id = rss_getcpu(i % rss_getnumbuckets());
#else
/*
* Bind the msix vector, and thus the
* rings to the corresponding cpu.
*
* This just happens to match the default RSS round-robin
* bucket -> queue -> CPU allocation.
*/
Modify the binding of queues to attach to as many CPUs as possible when using more than one igb(4) adapter. This means that queues will not be bound to the same CPUs if there are more CPUs availble. This is only applicable to a system that has multiple interfaces. Obtained from: Yahoo! Inc. MFC after: 3 days
2012-05-10 00:00:28 +00:00
if (adapter->num_queues > 1) {
if (igb_last_bind_cpu < 0)
igb_last_bind_cpu = CPU_FIRST();
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
cpu_id = igb_last_bind_cpu;
}
#endif
if (adapter->num_queues > 1) {
bus_bind_intr(dev, que->res, cpu_id);
#ifdef RSS
device_printf(dev,
"Bound queue %d to RSS bucket %d\n",
i, cpu_id);
#else
Modify the binding of queues to attach to as many CPUs as possible when using more than one igb(4) adapter. This means that queues will not be bound to the same CPUs if there are more CPUs availble. This is only applicable to a system that has multiple interfaces. Obtained from: Yahoo! Inc. MFC after: 3 days
2012-05-10 00:00:28 +00:00
device_printf(dev,
"Bound queue %d to cpu %d\n",
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
i, cpu_id);
#endif
Modify the binding of queues to attach to as many CPUs as possible when using more than one igb(4) adapter. This means that queues will not be bound to the same CPUs if there are more CPUs availble. This is only applicable to a system that has multiple interfaces. Obtained from: Yahoo! Inc. MFC after: 3 days
2012-05-10 00:00:28 +00:00
}
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
TASK_INIT(&que->txr->txq_task, 0, igb_deferred_mq_start,
que->txr);
#endif
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* Make tasklet for deferred handling */
TASK_INIT(&que->que_task, 0, igb_handle_que, que);
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
que->tq = taskqueue_create("igb_que", M_NOWAIT,
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
taskqueue_thread_enqueue, &que->tq);
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
if (adapter->num_queues > 1) {
/*
* Only pin the taskqueue thread to a CPU if
* RSS is in use.
*
* This again just happens to match the default RSS
* round-robin bucket -> queue -> CPU allocation.
*/
#ifdef RSS
taskqueue_start_threads_pinned(&que->tq, 1, PI_NET,
cpu_id,
"%s que (bucket %d)",
device_get_nameunit(adapter->dev),
cpu_id);
#else
taskqueue_start_threads(&que->tq, 1, PI_NET,
"%s que (qid %d)",
device_get_nameunit(adapter->dev),
cpu_id);
#endif
} else {
taskqueue_start_threads(&que->tq, 1, PI_NET, "%s que",
device_get_nameunit(adapter->dev));
}
/* Finally update the last bound CPU id */
if (adapter->num_queues > 1)
igb_last_bind_cpu = CPU_NEXT(igb_last_bind_cpu);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/* And Link */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rid = vector + 1;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
adapter->res = bus_alloc_resource_any(dev,
SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE);
if (adapter->res == NULL) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_printf(dev,
"Unable to allocate bus resource: "
"MSIX Link Interrupt\n");
return (ENXIO);
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if ((error = bus_setup_intr(dev, adapter->res,
INTR_TYPE_NET | INTR_MPSAFE, NULL,
igb_msix_link, adapter, &adapter->tag)) != 0) {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
device_printf(dev, "Failed to register Link handler");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
return (error);
}
Put back the lost bus_describe_intr() calls.
2010-06-11 21:35:19 +00:00
#if __FreeBSD_version >= 800504
bus_describe_intr(dev, adapter->res, adapter->tag, "link");
#endif
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
adapter->linkvec = vector;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
return (0);
}
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static void
igb_configure_queues(struct adapter *adapter)
{
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct e1000_hw *hw = &adapter->hw;
struct igb_queue *que;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
u32 tmp, ivar = 0, newitr = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* First turn on RSS capability */
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->hw.mac.type != e1000_82575)
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
E1000_WRITE_REG(hw, E1000_GPIE,
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
E1000_GPIE_MSIX_MODE | E1000_GPIE_EIAME |
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
E1000_GPIE_PBA | E1000_GPIE_NSICR);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* Turn on MSIX */
switch (adapter->hw.mac.type) {
case e1000_82580:
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
case e1000_i350:
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
case e1000_i354:
Sync with Intel internal source: shared code update and small changes in core required Add support for new i210/i211 devices Improve queue calculation based on mac type MFC after:5 days
2012-07-05 20:26:57 +00:00
case e1000_i210:
case e1000_i211:
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
case e1000_vfadapt:
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
case e1000_vfadapt_i350:
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* RX entries */
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
for (int i = 0; i < adapter->num_queues; i++) {
u32 index = i >> 1;
ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
que = &adapter->queues[i];
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
if (i & 1) {
ivar &= 0xFF00FFFF;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
ivar |= (que->msix | E1000_IVAR_VALID) << 16;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
} else {
ivar &= 0xFFFFFF00;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
ivar |= que->msix | E1000_IVAR_VALID;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
}
E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* TX entries */
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
for (int i = 0; i < adapter->num_queues; i++) {
u32 index = i >> 1;
ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
que = &adapter->queues[i];
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
if (i & 1) {
ivar &= 0x00FFFFFF;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
ivar |= (que->msix | E1000_IVAR_VALID) << 24;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
} else {
ivar &= 0xFFFF00FF;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
ivar |= (que->msix | E1000_IVAR_VALID) << 8;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
}
E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
adapter->que_mask |= que->eims;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
}
/* And for the link interrupt */
ivar = (adapter->linkvec | E1000_IVAR_VALID) << 8;
adapter->link_mask = 1 << adapter->linkvec;
E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
break;
case e1000_82576:
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* RX entries */
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++) {
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
u32 index = i & 0x7; /* Each IVAR has two entries */
ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
que = &adapter->queues[i];
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
if (i < 8) {
ivar &= 0xFFFFFF00;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
ivar |= que->msix | E1000_IVAR_VALID;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
} else {
ivar &= 0xFF00FFFF;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
ivar |= (que->msix | E1000_IVAR_VALID) << 16;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
}
E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
adapter->que_mask |= que->eims;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* TX entries */
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++) {
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
u32 index = i & 0x7; /* Each IVAR has two entries */
ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
que = &adapter->queues[i];
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
if (i < 8) {
ivar &= 0xFFFF00FF;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
ivar |= (que->msix | E1000_IVAR_VALID) << 8;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
} else {
ivar &= 0x00FFFFFF;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
ivar |= (que->msix | E1000_IVAR_VALID) << 24;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
}
E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar);
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
adapter->que_mask |= que->eims;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
}
/* And for the link interrupt */
ivar = (adapter->linkvec | E1000_IVAR_VALID) << 8;
adapter->link_mask = 1 << adapter->linkvec;
E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
break;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
case e1000_82575:
/* enable MSI-X support*/
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
tmp = E1000_READ_REG(hw, E1000_CTRL_EXT);
tmp |= E1000_CTRL_EXT_PBA_CLR;
/* Auto-Mask interrupts upon ICR read. */
tmp |= E1000_CTRL_EXT_EIAME;
tmp |= E1000_CTRL_EXT_IRCA;
E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmp);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* Queues */
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++) {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
que = &adapter->queues[i];
tmp = E1000_EICR_RX_QUEUE0 << i;
tmp |= E1000_EICR_TX_QUEUE0 << i;
que->eims = tmp;
E1000_WRITE_REG_ARRAY(hw, E1000_MSIXBM(0),
i, que->eims);
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
adapter->que_mask |= que->eims;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/* Link */
E1000_WRITE_REG(hw, E1000_MSIXBM(adapter->linkvec),
E1000_EIMS_OTHER);
This delta has a few important items: PR 122839 is fixed in both em and in igb Second, the issue on building modules since the static kernel build changes is now resolved. I was not able to get the fancier directory hierarchy working, but this works, both em and igb build as modules now. Third, there is now support in em for two new NICs, Hartwell (or 82574) is a low cost PCIE dual port adapter that has MSIX, for this release it uses 3 vectors only, RX, TX, and LINK. In the next release I will add a second TX and RX queue. Also, there is support here for ICH10, the followon to ICH9. Both of these are early releases, general availability will follow soon. Fourth: On Hartwell and ICH10 we now have IEEE 1588 PTP support, I have implemented this in a provisional way so that early adopters may try and comment on the functionality. The IOCTL structure may change. This feature is off by default, you need to edit the Makefile and add the EM_TIMESYNC define to get the code. Enjoy all!!
2008-04-25 21:19:41 +00:00
adapter->link_mask |= E1000_EIMS_OTHER;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
default:
break;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* Set the starting interrupt rate */
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
if (igb_max_interrupt_rate > 0)
newitr = (4000000 / igb_max_interrupt_rate) & 0x7FFC;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (hw->mac.type == e1000_82575)
newitr |= newitr << 16;
else
DUH, must be tired, I missed the second instance... time for the weekend :)
2010-04-09 21:18:46 +00:00
newitr |= E1000_EITR_CNT_IGNR;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
for (int i = 0; i < adapter->num_queues; i++) {
que = &adapter->queues[i];
E1000_WRITE_REG(hw, E1000_EITR(que->msix), newitr);
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
return;
}
static void
igb_free_pci_resources(struct adapter *adapter)
{
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct igb_queue *que = adapter->queues;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
device_t dev = adapter->dev;
int rid;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/*
** There is a slight possibility of a failure mode
** in attach that will result in entering this function
** before interrupt resources have been initialized, and
** in that case we do not want to execute the loops below
** We can detect this reliably by the state of the adapter
** res pointer.
*/
if (adapter->res == NULL)
goto mem;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
* First release all the interrupt resources:
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*/
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
for (int i = 0; i < adapter->num_queues; i++, que++) {
rid = que->msix + 1;
if (que->tag != NULL) {
bus_teardown_intr(dev, que->res, que->tag);
que->tag = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (que->res != NULL)
bus_release_resource(dev,
SYS_RES_IRQ, rid, que->res);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/* Clean the Legacy or Link interrupt last */
if (adapter->linkvec) /* we are doing MSIX */
rid = adapter->linkvec + 1;
else
(adapter->msix != 0) ? (rid = 1):(rid = 0);
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
que = adapter->queues;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if (adapter->tag != NULL) {
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
taskqueue_drain(que->tq, &adapter->link_task);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
bus_teardown_intr(dev, adapter->res, adapter->tag);
adapter->tag = NULL;
}
if (adapter->res != NULL)
bus_release_resource(dev, SYS_RES_IRQ, rid, adapter->res);
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
for (int i = 0; i < adapter->num_queues; i++, que++) {
if (que->tq != NULL) {
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Reapply r223198 which was reverted in the previous vendor import. Some portions were already reapplied in r233708: - Use a dedicated task to handle deferred transmits from the if_transmit method instead of reusing the existing per-queue interrupt task. Reusing the per-queue interrupt task could result in both an interrupt thread and the taskqueue thread trying to handle received packets on a single queue resulting in out-of-order packet processing. - Call ether_ifdetach() earlier in igb_detach(). - Drain tasks and free taskqueues during igb_detach(). MFC after: 1 week
2012-04-11 21:33:45 +00:00
taskqueue_drain(que->tq, &que->txr->txq_task);
#endif
taskqueue_drain(que->tq, &que->que_task);
taskqueue_free(que->tq);
}
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
mem:
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (adapter->msix)
pci_release_msi(dev);
if (adapter->msix_mem != NULL)
bus_release_resource(dev, SYS_RES_MEMORY,
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
adapter->memrid, adapter->msix_mem);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (adapter->pci_mem != NULL)
bus_release_resource(dev, SYS_RES_MEMORY,
PCIR_BAR(0), adapter->pci_mem);
}
/*
* Setup Either MSI/X or MSI
*/
static int
igb_setup_msix(struct adapter *adapter)
{
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
device_t dev = adapter->dev;
int bar, want, queues, msgs, maxqueues;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* tuneable override */
if (igb_enable_msix == 0)
goto msi;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* First try MSI/X */
Make the various driver MSIX setup routines fallback to MSI more gracefully. This change was suggested by Marius Strobl, thank you. PR: kern/181016 MFC after: ASAP
2013-08-06 21:01:38 +00:00
msgs = pci_msix_count(dev);
if (msgs == 0)
goto msi;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/*
** Some new devices, as with ixgbe, now may
** use a different BAR, so we need to keep
** track of which is used.
*/
adapter->memrid = PCIR_BAR(IGB_MSIX_BAR);
bar = pci_read_config(dev, adapter->memrid, 4);
if (bar == 0) /* use next bar */
adapter->memrid += 4;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
adapter->msix_mem = bus_alloc_resource_any(dev,
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
SYS_RES_MEMORY, &adapter->memrid, RF_ACTIVE);
Make the various driver MSIX setup routines fallback to MSI more gracefully. This change was suggested by Marius Strobl, thank you. PR: kern/181016 MFC after: ASAP
2013-08-06 21:01:38 +00:00
if (adapter->msix_mem == NULL) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* May not be enabled */
device_printf(adapter->dev,
"Unable to map MSIX table \n");
goto msi;
}
/* Figure out a reasonable auto config value */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
queues = (mp_ncpus > (msgs-1)) ? (msgs-1) : mp_ncpus;
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
#ifdef RSS
/* If we're doing RSS, clamp at the number of RSS buckets */
if (queues > rss_getnumbuckets())
queues = rss_getnumbuckets();
#endif
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Manual override */
if (igb_num_queues != 0)
queues = igb_num_queues;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
Sync with Intel internal source: shared code update and small changes in core required Add support for new i210/i211 devices Improve queue calculation based on mac type MFC after:5 days
2012-07-05 20:26:57 +00:00
/* Sanity check based on HW */
switch (adapter->hw.mac.type) {
case e1000_82575:
maxqueues = 4;
break;
case e1000_82576:
case e1000_82580:
case e1000_i350:
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
case e1000_i354:
Sync with Intel internal source: shared code update and small changes in core required Add support for new i210/i211 devices Improve queue calculation based on mac type MFC after:5 days
2012-07-05 20:26:57 +00:00
maxqueues = 8;
break;
case e1000_i210:
maxqueues = 4;
break;
case e1000_i211:
maxqueues = 2;
break;
default: /* VF interfaces */
maxqueues = 1;
break;
}
if (queues > maxqueues)
queues = maxqueues;
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* Manual override */
if (igb_num_queues != 0)
queues = igb_num_queues;
Expose system level ixgbe sysctls. Device level sysctls are already exposed as dev.ix.<device> Fixing the case where number of queues for igb is auto-tuned and hw.igb.num_queues does not return current/updated value. Reviewed by: jfv Approved by: re (delphij) MFC after: 2 weeks
2013-10-05 19:17:56 +00:00
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/*
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
** One vector (RX/TX pair) per queue
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
** plus an additional for Link interrupt
*/
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
want = queues + 1;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (msgs >= want)
msgs = want;
else {
device_printf(adapter->dev,
"MSIX Configuration Problem, "
This delta has a few important items: PR 122839 is fixed in both em and in igb Second, the issue on building modules since the static kernel build changes is now resolved. I was not able to get the fancier directory hierarchy working, but this works, both em and igb build as modules now. Third, there is now support in em for two new NICs, Hartwell (or 82574) is a low cost PCIE dual port adapter that has MSIX, for this release it uses 3 vectors only, RX, TX, and LINK. In the next release I will add a second TX and RX queue. Also, there is support here for ICH10, the followon to ICH9. Both of these are early releases, general availability will follow soon. Fourth: On Hartwell and ICH10 we now have IEEE 1588 PTP support, I have implemented this in a provisional way so that early adopters may try and comment on the functionality. The IOCTL structure may change. This feature is off by default, you need to edit the Makefile and add the EM_TIMESYNC define to get the code. Enjoy all!!
2008-04-25 21:19:41 +00:00
"%d vectors configured, but %d queues wanted!\n",
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
msgs, want);
Make the various driver MSIX setup routines fallback to MSI more gracefully. This change was suggested by Marius Strobl, thank you. PR: kern/181016 MFC after: ASAP
2013-08-06 21:01:38 +00:00
goto msi;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Improve the MSIX setup code in the drivers, thanks to Marius for the changes. Make sure that pci_alloc_msix() does give us the vectors we need and fall back to MSI when it doesn't, also release any that were allocated when insufficient. MFC after: 3 days
2013-08-12 22:54:38 +00:00
if ((pci_alloc_msix(dev, &msgs) == 0) && (msgs == want)) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_printf(adapter->dev,
"Using MSIX interrupts with %d vectors\n", msgs);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
adapter->num_queues = queues;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
return (msgs);
}
Improve the MSIX setup code in the drivers, thanks to Marius for the changes. Make sure that pci_alloc_msix() does give us the vectors we need and fall back to MSI when it doesn't, also release any that were allocated when insufficient. MFC after: 3 days
2013-08-12 22:54:38 +00:00
/*
** If MSIX alloc failed or provided us with
** less than needed, free and fall through to MSI
*/
pci_release_msi(dev);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
msi:
Make the various driver MSIX setup routines fallback to MSI more gracefully. This change was suggested by Marius Strobl, thank you. PR: kern/181016 MFC after: ASAP
2013-08-06 21:01:38 +00:00
if (adapter->msix_mem != NULL) {
bus_release_resource(dev, SYS_RES_MEMORY,
PCIR_BAR(IGB_MSIX_BAR), adapter->msix_mem);
adapter->msix_mem = NULL;
}
msgs = 1;
if (pci_alloc_msi(dev, &msgs) == 0) {
Improve the MSIX setup code in the drivers, thanks to Marius for the changes. Make sure that pci_alloc_msix() does give us the vectors we need and fall back to MSI when it doesn't, also release any that were allocated when insufficient. MFC after: 3 days
2013-08-12 22:54:38 +00:00
device_printf(adapter->dev," Using an MSI interrupt\n");
Properly handle failures in igb_setup_msix() by returning 0 if MSI or MSI-X allocation fails. Reviewed by: jfv MFC after: 2 weeks
2012-03-01 22:13:10 +00:00
return (msgs);
}
Improve the MSIX setup code in the drivers, thanks to Marius for the changes. Make sure that pci_alloc_msix() does give us the vectors we need and fall back to MSI when it doesn't, also release any that were allocated when insufficient. MFC after: 3 days
2013-08-12 22:54:38 +00:00
device_printf(adapter->dev," Using a Legacy interrupt\n");
Properly handle failures in igb_setup_msix() by returning 0 if MSI or MSI-X allocation fails. Reviewed by: jfv MFC after: 2 weeks
2012-03-01 22:13:10 +00:00
return (0);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/*********************************************************************
*
* Initialize the DMA Coalescing feature
*
**********************************************************************/
static void
igb_init_dmac(struct adapter *adapter, u32 pba)
{
device_t dev = adapter->dev;
struct e1000_hw *hw = &adapter->hw;
u32 dmac, reg = ~E1000_DMACR_DMAC_EN;
u16 hwm;
if (hw->mac.type == e1000_i211)
return;
if (hw->mac.type > e1000_82580) {
if (adapter->dmac == 0) { /* Disabling it */
E1000_WRITE_REG(hw, E1000_DMACR, reg);
return;
} else
device_printf(dev, "DMA Coalescing enabled\n");
/* Set starting threshold */
E1000_WRITE_REG(hw, E1000_DMCTXTH, 0);
hwm = 64 * pba - adapter->max_frame_size / 16;
if (hwm < 64 * (pba - 6))
hwm = 64 * (pba - 6);
reg = E1000_READ_REG(hw, E1000_FCRTC);
reg &= ~E1000_FCRTC_RTH_COAL_MASK;
reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
& E1000_FCRTC_RTH_COAL_MASK);
E1000_WRITE_REG(hw, E1000_FCRTC, reg);
dmac = pba - adapter->max_frame_size / 512;
if (dmac < pba - 10)
dmac = pba - 10;
reg = E1000_READ_REG(hw, E1000_DMACR);
reg &= ~E1000_DMACR_DMACTHR_MASK;
reg = ((dmac << E1000_DMACR_DMACTHR_SHIFT)
& E1000_DMACR_DMACTHR_MASK);
/* transition to L0x or L1 if available..*/
reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
/* Check if status is 2.5Gb backplane connection
* before configuration of watchdog timer, which is
* in msec values in 12.8usec intervals
* watchdog timer= msec values in 32usec intervals
* for non 2.5Gb connection
*/
if (hw->mac.type == e1000_i354) {
int status = E1000_READ_REG(hw, E1000_STATUS);
if ((status & E1000_STATUS_2P5_SKU) &&
(!(status & E1000_STATUS_2P5_SKU_OVER)))
reg |= ((adapter->dmac * 5) >> 6);
else
reg |= (adapter->dmac >> 5);
} else {
reg |= (adapter->dmac >> 5);
}
E1000_WRITE_REG(hw, E1000_DMACR, reg);
#ifdef I210_OBFF_SUPPORT
/*
* Set the OBFF Rx threshold to DMA Coalescing Rx
* threshold - 2KB and enable the feature in the
* hardware for I210.
*/
if (hw->mac.type == e1000_i210) {
int obff = dmac - 2;
reg = E1000_READ_REG(hw, E1000_DOBFFCTL);
reg &= ~E1000_DOBFFCTL_OBFFTHR_MASK;
reg |= (obff & E1000_DOBFFCTL_OBFFTHR_MASK)
| E1000_DOBFFCTL_EXIT_ACT_MASK;
E1000_WRITE_REG(hw, E1000_DOBFFCTL, reg);
}
#endif
E1000_WRITE_REG(hw, E1000_DMCRTRH, 0);
/* Set the interval before transition */
reg = E1000_READ_REG(hw, E1000_DMCTLX);
if (hw->mac.type == e1000_i350)
reg |= IGB_DMCTLX_DCFLUSH_DIS;
/*
** in 2.5Gb connection, TTLX unit is 0.4 usec
** which is 0x4*2 = 0xA. But delay is still 4 usec
*/
if (hw->mac.type == e1000_i354) {
int status = E1000_READ_REG(hw, E1000_STATUS);
if ((status & E1000_STATUS_2P5_SKU) &&
(!(status & E1000_STATUS_2P5_SKU_OVER)))
reg |= 0xA;
else
reg |= 0x4;
} else {
reg |= 0x4;
}
E1000_WRITE_REG(hw, E1000_DMCTLX, reg);
/* free space in tx packet buffer to wake from DMA coal */
E1000_WRITE_REG(hw, E1000_DMCTXTH, (IGB_TXPBSIZE -
(2 * adapter->max_frame_size)) >> 6);
/* make low power state decision controlled by DMA coal */
reg = E1000_READ_REG(hw, E1000_PCIEMISC);
reg &= ~E1000_PCIEMISC_LX_DECISION;
E1000_WRITE_REG(hw, E1000_PCIEMISC, reg);
} else if (hw->mac.type == e1000_82580) {
u32 reg = E1000_READ_REG(hw, E1000_PCIEMISC);
E1000_WRITE_REG(hw, E1000_PCIEMISC,
reg & ~E1000_PCIEMISC_LX_DECISION);
E1000_WRITE_REG(hw, E1000_DMACR, 0);
}
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
* Set up an fresh starting state
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*
**********************************************************************/
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
static void
igb_reset(struct adapter *adapter)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
device_t dev = adapter->dev;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
struct e1000_hw *hw = &adapter->hw;
struct e1000_fc_info *fc = &hw->fc;
struct ifnet *ifp = adapter->ifp;
u32 pba = 0;
u16 hwm;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
INIT_DEBUGOUT("igb_reset: begin");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Let the firmware know the OS is in control */
igb_get_hw_control(adapter);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/*
* Packet Buffer Allocation (PBA)
* Writing PBA sets the receive portion of the buffer
* the remainder is used for the transmit buffer.
*/
switch (hw->mac.type) {
case e1000_82575:
pba = E1000_PBA_32K;
break;
case e1000_82576:
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
case e1000_vfadapt:
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
pba = E1000_READ_REG(hw, E1000_RXPBS);
pba &= E1000_RXPBS_SIZE_MASK_82576;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
break;
case e1000_82580:
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
case e1000_i350:
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
case e1000_i354:
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
case e1000_vfadapt_i350:
pba = E1000_READ_REG(hw, E1000_RXPBS);
pba = e1000_rxpbs_adjust_82580(pba);
break;
Sync with Intel internal source: shared code update and small changes in core required Add support for new i210/i211 devices Improve queue calculation based on mac type MFC after:5 days
2012-07-05 20:26:57 +00:00
case e1000_i210:
case e1000_i211:
pba = E1000_PBA_34K;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
default:
break;
}
/* Special needs in case of Jumbo frames */
if ((hw->mac.type == e1000_82575) && (ifp->if_mtu > ETHERMTU)) {
u32 tx_space, min_tx, min_rx;
pba = E1000_READ_REG(hw, E1000_PBA);
tx_space = pba >> 16;
pba &= 0xffff;
min_tx = (adapter->max_frame_size +
sizeof(struct e1000_tx_desc) - ETHERNET_FCS_SIZE) * 2;
min_tx = roundup2(min_tx, 1024);
min_tx >>= 10;
min_rx = adapter->max_frame_size;
min_rx = roundup2(min_rx, 1024);
min_rx >>= 10;
if (tx_space < min_tx &&
((min_tx - tx_space) < pba)) {
pba = pba - (min_tx - tx_space);
/*
* if short on rx space, rx wins
* and must trump tx adjustment
*/
if (pba < min_rx)
pba = min_rx;
}
E1000_WRITE_REG(hw, E1000_PBA, pba);
}
INIT_DEBUGOUT1("igb_init: pba=%dK",pba);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
* These parameters control the automatic generation (Tx) and
* response (Rx) to Ethernet PAUSE frames.
* - High water mark should allow for at least two frames to be
* received after sending an XOFF.
* - Low water mark works best when it is very near the high water mark.
* This allows the receiver to restart by sending XON when it has
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
* drained a bit.
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*/
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
hwm = min(((pba << 10) * 9 / 10),
((pba << 10) - 2 * adapter->max_frame_size));
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
if (hw->mac.type < e1000_82576) {
fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
fc->low_water = fc->high_water - 8;
} else {
fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
fc->low_water = fc->high_water - 16;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
fc->pause_time = IGB_FC_PAUSE_TIME;
fc->send_xon = TRUE;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
if (adapter->fc)
fc->requested_mode = adapter->fc;
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
else
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
fc->requested_mode = e1000_fc_default;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* Issue a global reset */
e1000_reset_hw(hw);
E1000_WRITE_REG(hw, E1000_WUC, 0);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* Reset for AutoMediaDetect */
if (adapter->flags & IGB_MEDIA_RESET) {
e1000_setup_init_funcs(hw, TRUE);
e1000_get_bus_info(hw);
adapter->flags &= ~IGB_MEDIA_RESET;
}
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
if (e1000_init_hw(hw) < 0)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_printf(dev, "Hardware Initialization Failed\n");
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
/* Setup DMA Coalescing */
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
igb_init_dmac(adapter, pba);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_VET, ETHERTYPE_VLAN);
e1000_get_phy_info(hw);
e1000_check_for_link(hw);
return;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*********************************************************************
*
* Setup networking device structure and register an interface.
*
**********************************************************************/
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
static int
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
igb_setup_interface(device_t dev, struct adapter *adapter)
{
struct ifnet *ifp;
INIT_DEBUGOUT("igb_setup_interface: begin");
ifp = adapter->ifp = if_alloc(IFT_ETHER);
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
if (ifp == NULL) {
device_printf(dev, "can not allocate ifnet structure\n");
return (-1);
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if_initname(ifp, device_get_name(dev), device_get_unit(dev));
ifp->if_init = igb_init;
ifp->if_softc = adapter;
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_ioctl = igb_ioctl;
- Provide igb_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-19 11:49:41 +00:00
ifp->if_get_counter = igb_get_counter;
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
ifp->if_transmit = igb_mq_start;
ifp->if_qflush = igb_qflush;
Commit a portion of 233708 I missed earlier and don't include the definition of igb_start() and igb_start_locked() (nor set if_start in the ifnet) when igb(4) uses if_transmit.
2012-06-01 15:52:41 +00:00
#else
ifp->if_start = igb_start;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
IFQ_SET_MAXLEN(&ifp->if_snd, adapter->num_tx_desc - 1);
ifp->if_snd.ifq_drv_maxlen = adapter->num_tx_desc - 1;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IFQ_SET_READY(&ifp->if_snd);
Commit a portion of 233708 I missed earlier and don't include the definition of igb_start() and igb_start_locked() (nor set if_start in the ifnet) when igb(4) uses if_transmit.
2012-06-01 15:52:41 +00:00
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
ether_ifattach(ifp, adapter->hw.mac.addr);
ifp->if_capabilities = ifp->if_capenable = 0;
Fix of a VLAN problem by jhb, the checksum capability got lost along the way. MFC: asap
2010-07-09 17:11:29 +00:00
ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
ifp->if_capabilities |= IFCAP_TSO;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
ifp->if_capabilities |= IFCAP_JUMBO_MTU;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
ifp->if_capenable = ifp->if_capabilities;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
/* Don't enable LRO by default */
ifp->if_capabilities |= IFCAP_LRO;
A few minor changes: add altq option header, add missing conditional around a buf_ring call that will break 7.3, and thanks to Fabien Thomas add POLLING support for igb and a minor related fix in the em driver.
2010-02-01 19:28:43 +00:00
#ifdef DEVICE_POLLING
ifp->if_capabilities |= IFCAP_POLLING;
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
Important fix got clobbered in the em driver, keeping VLAN HWFILTER from being used by default, this breaks stacked pseudo devices, and as it turns out, also breaks virtual machines that happen to use VLANS (didn't know that before :). Put the fix back into the em driver, and for good measure add the same code to the igb driver where it should have been anyway.
2010-04-08 00:50:43 +00:00
* Tell the upper layer(s) we
* support full VLAN capability.
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*/
Use define from if_var.h to access a field inside struct if_data, that resides in struct ifnet. Sponsored by: Nginx, Inc.
2014-08-30 19:55:54 +00:00
ifp->if_hdrlen = sizeof(struct ether_vlan_header);
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING
| IFCAP_VLAN_HWTSO
| IFCAP_VLAN_MTU;
ifp->if_capenable |= IFCAP_VLAN_HWTAGGING
| IFCAP_VLAN_HWTSO
| IFCAP_VLAN_MTU;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Important fix got clobbered in the em driver, keeping VLAN HWFILTER from being used by default, this breaks stacked pseudo devices, and as it turns out, also breaks virtual machines that happen to use VLANS (didn't know that before :). Put the fix back into the em driver, and for good measure add the same code to the igb driver where it should have been anyway.
2010-04-08 00:50:43 +00:00
/*
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
** Don't turn this on by default, if vlans are
Important fix got clobbered in the em driver, keeping VLAN HWFILTER from being used by default, this breaks stacked pseudo devices, and as it turns out, also breaks virtual machines that happen to use VLANS (didn't know that before :). Put the fix back into the em driver, and for good measure add the same code to the igb driver where it should have been anyway.
2010-04-08 00:50:43 +00:00
** created on another pseudo device (eg. lagg)
** then vlan events are not passed thru, breaking
** operation, but with HW FILTER off it works. If
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
** using vlans directly on the igb driver you can
Important fix got clobbered in the em driver, keeping VLAN HWFILTER from being used by default, this breaks stacked pseudo devices, and as it turns out, also breaks virtual machines that happen to use VLANS (didn't know that before :). Put the fix back into the em driver, and for good measure add the same code to the igb driver where it should have been anyway.
2010-04-08 00:50:43 +00:00
** enable this and get full hardware tag filtering.
*/
ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
* Specify the media types supported by this adapter and register
* callbacks to update media and link information
*/
ifmedia_init(&adapter->media, IFM_IMASK,
igb_media_change, igb_media_status);
if ((adapter->hw.phy.media_type == e1000_media_type_fiber) ||
(adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) {
ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_SX | IFM_FDX,
0, NULL);
ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_SX, 0, NULL);
} else {
ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T, 0, NULL);
ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T | IFM_FDX,
0, NULL);
ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX,
0, NULL);
ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
0, NULL);
if (adapter->hw.phy.type != e1000_phy_ife) {
ifmedia_add(&adapter->media,
IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
ifmedia_add(&adapter->media,
IFM_ETHER | IFM_1000_T, 0, NULL);
}
}
ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
return (0);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*
* Manage DMA'able memory.
*/
static void
igb_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
{
if (error)
return;
*(bus_addr_t *) arg = segs[0].ds_addr;
}
static int
igb_dma_malloc(struct adapter *adapter, bus_size_t size,
struct igb_dma_alloc *dma, int mapflags)
{
int error;
error = bus_dma_tag_create(bus_get_dma_tag(adapter->dev), /* parent */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
IGB_DBA_ALIGN, 0, /* alignment, bounds */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
size, /* maxsize */
1, /* nsegments */
size, /* maxsegsize */
0, /* flags */
NULL, /* lockfunc */
NULL, /* lockarg */
&dma->dma_tag);
if (error) {
device_printf(adapter->dev,
"%s: bus_dma_tag_create failed: %d\n",
__func__, error);
goto fail_0;
}
error = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr,
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
BUS_DMA_NOWAIT | BUS_DMA_COHERENT, &dma->dma_map);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (error) {
device_printf(adapter->dev,
"%s: bus_dmamem_alloc(%ju) failed: %d\n",
__func__, (uintmax_t)size, error);
goto fail_2;
}
dma->dma_paddr = 0;
error = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr,
size, igb_dmamap_cb, &dma->dma_paddr, mapflags | BUS_DMA_NOWAIT);
if (error || dma->dma_paddr == 0) {
device_printf(adapter->dev,
"%s: bus_dmamap_load failed: %d\n",
__func__, error);
goto fail_3;
}
return (0);
fail_3:
bus_dmamap_unload(dma->dma_tag, dma->dma_map);
fail_2:
bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
bus_dma_tag_destroy(dma->dma_tag);
fail_0:
dma->dma_tag = NULL;
return (error);
}
static void
igb_dma_free(struct adapter *adapter, struct igb_dma_alloc *dma)
{
if (dma->dma_tag == NULL)
return;
- Don't compare bus_dma map pointers for static DMA allocations against NULL to determine if bus_dmamap_unload() or bus_dmamem_free() should be called. Instead, check the associated bus and virtual addresses. - Don't clear static DMA maps to NULL. Reviewed by: jfv
2014-06-12 11:15:19 +00:00
if (dma->dma_paddr != 0) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
bus_dmamap_sync(dma->dma_tag, dma->dma_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(dma->dma_tag, dma->dma_map);
- Don't compare bus_dma map pointers for static DMA allocations against NULL to determine if bus_dmamap_unload() or bus_dmamem_free() should be called. Instead, check the associated bus and virtual addresses. - Don't clear static DMA maps to NULL. Reviewed by: jfv
2014-06-12 11:15:19 +00:00
dma->dma_paddr = 0;
}
if (dma->dma_vaddr != NULL) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
- Don't compare bus_dma map pointers for static DMA allocations against NULL to determine if bus_dmamap_unload() or bus_dmamem_free() should be called. Instead, check the associated bus and virtual addresses. - Don't clear static DMA maps to NULL. Reviewed by: jfv
2014-06-12 11:15:19 +00:00
dma->dma_vaddr = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
bus_dma_tag_destroy(dma->dma_tag);
dma->dma_tag = NULL;
}
/*********************************************************************
*
* Allocate memory for the transmit and receive rings, and then
* the descriptors associated with each, called only once at attach.
*
**********************************************************************/
static int
igb_allocate_queues(struct adapter *adapter)
{
device_t dev = adapter->dev;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct igb_queue *que = NULL;
struct tx_ring *txr = NULL;
struct rx_ring *rxr = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
int rsize, tsize, error = E1000_SUCCESS;
int txconf = 0, rxconf = 0;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* First allocate the top level queue structs */
if (!(adapter->queues =
(struct igb_queue *) malloc(sizeof(struct igb_queue) *
adapter->num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) {
device_printf(dev, "Unable to allocate queue memory\n");
error = ENOMEM;
goto fail;
}
/* Next allocate the TX ring struct memory */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (!(adapter->tx_rings =
(struct tx_ring *) malloc(sizeof(struct tx_ring) *
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
adapter->num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_printf(dev, "Unable to allocate TX ring memory\n");
error = ENOMEM;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
goto tx_fail;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* Now allocate the RX */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (!(adapter->rx_rings =
(struct rx_ring *) malloc(sizeof(struct rx_ring) *
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
adapter->num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_printf(dev, "Unable to allocate RX ring memory\n");
error = ENOMEM;
goto rx_fail;
}
tsize = roundup2(adapter->num_tx_desc *
sizeof(union e1000_adv_tx_desc), IGB_DBA_ALIGN);
/*
* Now set up the TX queues, txconf is needed to handle the
* possibility that things fail midcourse and we need to
* undo memory gracefully
*/
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++, txconf++) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Set up some basics */
txr = &adapter->tx_rings[i];
txr->adapter = adapter;
txr->me = i;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
txr->num_desc = adapter->num_tx_desc;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Initialize the TX lock */
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
snprintf(txr->mtx_name, sizeof(txr->mtx_name), "%s:tx(%d)",
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_get_nameunit(dev), txr->me);
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
mtx_init(&txr->tx_mtx, txr->mtx_name, NULL, MTX_DEF);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (igb_dma_malloc(adapter, tsize,
&txr->txdma, BUS_DMA_NOWAIT)) {
device_printf(dev,
"Unable to allocate TX Descriptor memory\n");
error = ENOMEM;
goto err_tx_desc;
}
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
txr->tx_base = (union e1000_adv_tx_desc *)txr->txdma.dma_vaddr;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
bzero((void *)txr->tx_base, tsize);
/* Now allocate transmit buffers for the ring */
if (igb_allocate_transmit_buffers(txr)) {
device_printf(dev,
"Critical Failure setting up transmit buffers\n");
error = ENOMEM;
goto err_tx_desc;
}
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/* Allocate a buf ring */
This fixes a out-of-order problem with several of the newer drivers. The basic problem was that the driver was pulling the mbuf off the drbr ring and then when sending with xmit(), encounting a full transmit ring. Thus the lower layer xmit() function would return an error, and the drivers would then append the data back on to the ring. For TCP this is a horrible scenario sure to bring on a fast-retransmit. The fix is to use drbr_peek() to pull the data pointer but not remove it from the ring. If it fails then we either call the new drbr_putback or drbr_advance method. Advance moves it forward (we do this sometimes when the xmit() function frees the mbuf). When we succeed we always call advance. The putback will always copy the mbuf back to the top of the ring. Note that the putback *cannot* be used with a drbr_dequeue() only with drbr_peek(). We most of the time, in putback, would not need to copy it back since most likey the mbuf is still the same, but sometimes xmit() functions will change the mbuf via a pullup or other call. So the optimial case for the single consumer is to always copy it back. If we ever do a multiple_consumer (for lagg?) we will need a test and atomic in the put back possibly a seperate putback_mc() in the ring buf. Reviewed by: jhb@freebsd.org, jlv@freebsd.org
2013-02-07 15:20:54 +00:00
txr->br = buf_ring_alloc(igb_buf_ring_size, M_DEVBUF,
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
M_WAITOK, &txr->tx_mtx);
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*
* Next the RX queues...
*/
rsize = roundup2(adapter->num_rx_desc *
sizeof(union e1000_adv_rx_desc), IGB_DBA_ALIGN);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++, rxconf++) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
rxr = &adapter->rx_rings[i];
rxr->adapter = adapter;
rxr->me = i;
/* Initialize the RX lock */
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
snprintf(rxr->mtx_name, sizeof(rxr->mtx_name), "%s:rx(%d)",
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
device_get_nameunit(dev), txr->me);
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
mtx_init(&rxr->rx_mtx, rxr->mtx_name, NULL, MTX_DEF);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (igb_dma_malloc(adapter, rsize,
&rxr->rxdma, BUS_DMA_NOWAIT)) {
device_printf(dev,
"Unable to allocate RxDescriptor memory\n");
error = ENOMEM;
goto err_rx_desc;
}
rxr->rx_base = (union e1000_adv_rx_desc *)rxr->rxdma.dma_vaddr;
bzero((void *)rxr->rx_base, rsize);
/* Allocate receive buffers for the ring*/
if (igb_allocate_receive_buffers(rxr)) {
device_printf(dev,
"Critical Failure setting up receive buffers\n");
error = ENOMEM;
goto err_rx_desc;
}
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/*
** Finally set up the queue holding structs
*/
for (int i = 0; i < adapter->num_queues; i++) {
que = &adapter->queues[i];
que->adapter = adapter;
que->txr = &adapter->tx_rings[i];
que->rxr = &adapter->rx_rings[i];
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
return (0);
err_rx_desc:
for (rxr = adapter->rx_rings; rxconf > 0; rxr++, rxconf--)
igb_dma_free(adapter, &rxr->rxdma);
err_tx_desc:
for (txr = adapter->tx_rings; txconf > 0; txr++, txconf--)
igb_dma_free(adapter, &txr->txdma);
free(adapter->rx_rings, M_DEVBUF);
rx_fail:
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
buf_ring_free(txr->br, M_DEVBUF);
Small changes preparing for MFC, need to conditionalize the buf_ring_free call, and lem is missing the WOL change put into em.
2010-05-14 22:18:34 +00:00
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
free(adapter->tx_rings, M_DEVBUF);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
tx_fail:
free(adapter->queues, M_DEVBUF);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
fail:
return (error);
}
/*********************************************************************
*
* Allocate memory for tx_buffer structures. The tx_buffer stores all
* the information needed to transmit a packet on the wire. This is
* called only once at attach, setup is done every reset.
*
**********************************************************************/
static int
igb_allocate_transmit_buffers(struct tx_ring *txr)
{
struct adapter *adapter = txr->adapter;
device_t dev = adapter->dev;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
struct igb_tx_buf *txbuf;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
int error, i;
/*
* Setup DMA descriptor areas.
*/
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if ((error = bus_dma_tag_create(bus_get_dma_tag(dev),
With an i386 kernel the igb driver can cause a page fault panic on initialization due to a large number of bounce pages being allocated. This is due to the dma tag requiring page alignment on mbuf mapping. This was removed some time back from the ixgbe driver and is not needed here either.
2009-11-11 19:13:40 +00:00
1, 0, /* alignment, bounds */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
IGB_TSO_SIZE, /* maxsize */
IGB_MAX_SCATTER, /* nsegments */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
PAGE_SIZE, /* maxsegsize */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
0, /* flags */
NULL, /* lockfunc */
NULL, /* lockfuncarg */
&txr->txtag))) {
device_printf(dev,"Unable to allocate TX DMA tag\n");
goto fail;
}
if (!(txr->tx_buffers =
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
(struct igb_tx_buf *) malloc(sizeof(struct igb_tx_buf) *
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
adapter->num_tx_desc, M_DEVBUF, M_NOWAIT | M_ZERO))) {
device_printf(dev, "Unable to allocate tx_buffer memory\n");
error = ENOMEM;
goto fail;
}
/* Create the descriptor buffer dma maps */
txbuf = txr->tx_buffers;
for (i = 0; i < adapter->num_tx_desc; i++, txbuf++) {
error = bus_dmamap_create(txr->txtag, 0, &txbuf->map);
if (error != 0) {
device_printf(dev, "Unable to create TX DMA map\n");
goto fail;
}
}
return 0;
fail:
/* We free all, it handles case where we are in the middle */
igb_free_transmit_structures(adapter);
return (error);
}
/*********************************************************************
*
* Initialize a transmit ring.
*
**********************************************************************/
static void
igb_setup_transmit_ring(struct tx_ring *txr)
{
struct adapter *adapter = txr->adapter;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
struct igb_tx_buf *txbuf;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
int i;
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
struct netmap_adapter *na = NA(adapter->ifp);
struct netmap_slot *slot;
#endif /* DEV_NETMAP */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/* Clear the old descriptor contents */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
IGB_TX_LOCK(txr);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
slot = netmap_reset(na, NR_TX, txr->me, 0);
#endif /* DEV_NETMAP */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
bzero((void *)txr->tx_base,
(sizeof(union e1000_adv_tx_desc)) * adapter->num_tx_desc);
/* Reset indices */
txr->next_avail_desc = 0;
txr->next_to_clean = 0;
/* Free any existing tx buffers. */
txbuf = txr->tx_buffers;
for (i = 0; i < adapter->num_tx_desc; i++, txbuf++) {
if (txbuf->m_head != NULL) {
bus_dmamap_sync(txr->txtag, txbuf->map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(txr->txtag, txbuf->map);
m_freem(txbuf->m_head);
txbuf->m_head = NULL;
}
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
if (slot) {
A bunch of netmap fixes: USERSPACE: 1. add support for devices with different number of rx and tx queues; 2. add better support for zero-copy operation, adding an extra field to the netmap ring to indicate how many buffers we have already processed but not yet released (with help from Eddie Kohler); 3. The two changes above unfortunately require an API change, so while at it add a version field and some spares to the ioctl() argument to help detect mismatches. 4. update the manual page for the two changes above; 5. update sample applications in tools/tools/netmap KERNEL: 1. simplify the internal structures moving the global wait queues to the 'struct netmap_adapter'; 2. simplify the functions that map kring<->nic ring indexes 3. normalize device-specific code, helps mainteinance; 4. start exploring the impact of micro-optimizations (prefetch etc.) in the ixgbe driver. Use 'legacy' descriptors on the tx ring and prefetch slots gives about 20% speedup at 900 MHz. Another 7-10% would come from removing the explict calls to bus_dmamap* in the core (they are effectively NOPs in this case, but it takes expensive load of the per-buffer dma maps to figure out that they are all NULL. Rx performance not investigated. I am postponing the MFC so i can import a few more improvements before merging.
2012-02-27 19:05:01 +00:00
int si = netmap_idx_n2k(&na->tx_rings[txr->me], i);
(This commit only touches code within the DEV_NETMAP blocks) Introduce some functions to map NIC ring indexes into netmap ring indexes and vice versa. This way we can implement the bound checks only in one place (and hopefully in a correct way). On passing, make the code and comments more uniform across the various drivers.
2012-02-15 23:13:29 +00:00
/* no need to set the address */
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
netmap_load_map(na, txr->txtag, txbuf->map, NMB(na, slot + si));
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
}
#endif /* DEV_NETMAP */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* clear the watch index */
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
txbuf->eop = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/* Set number of descriptors available */
txr->tx_avail = adapter->num_tx_desc;
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
IGB_TX_UNLOCK(txr);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*********************************************************************
*
* Initialize all transmit rings.
*
**********************************************************************/
static void
igb_setup_transmit_structures(struct adapter *adapter)
{
struct tx_ring *txr = adapter->tx_rings;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
igb_setup_transmit_ring(txr);
return;
}
/*********************************************************************
*
* Enable transmit unit.
*
**********************************************************************/
static void
igb_initialize_transmit_units(struct adapter *adapter)
{
struct tx_ring *txr = adapter->tx_rings;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct e1000_hw *hw = &adapter->hw;
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
u32 tctl, txdctl;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
INIT_DEBUGOUT("igb_initialize_transmit_units: begin");
tctl = txdctl = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* Setup the Tx Descriptor Rings */
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
u64 bus_addr = txr->txdma.dma_paddr;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_TDLEN(i),
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
adapter->num_tx_desc * sizeof(struct e1000_tx_desc));
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_TDBAH(i),
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
(uint32_t)(bus_addr >> 32));
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_TDBAL(i),
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
(uint32_t)bus_addr);
/* Setup the HW Tx Head and Tail descriptor pointers */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_TDT(i), 0);
E1000_WRITE_REG(hw, E1000_TDH(i), 0);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
HW_DEBUGOUT2("Base = %x, Length = %x\n",
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_READ_REG(hw, E1000_TDBAL(i)),
E1000_READ_REG(hw, E1000_TDLEN(i)));
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
txr->queue_status = IGB_QUEUE_IDLE;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
txdctl |= IGB_TX_PTHRESH;
txdctl |= IGB_TX_HTHRESH << 8;
txdctl |= IGB_TX_WTHRESH << 16;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_TXDCTL(i), txdctl);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->vf_ifp)
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
return;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
e1000_config_collision_dist(hw);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Program the Transmit Control Register */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
tctl = E1000_READ_REG(hw, E1000_TCTL);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
tctl &= ~E1000_TCTL_CT;
tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT));
/* This write will effectively turn on the transmit unit. */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_TCTL, tctl);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*********************************************************************
*
* Free all transmit rings.
*
**********************************************************************/
static void
igb_free_transmit_structures(struct adapter *adapter)
{
struct tx_ring *txr = adapter->tx_rings;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IGB_TX_LOCK(txr);
igb_free_transmit_buffers(txr);
igb_dma_free(adapter, &txr->txdma);
IGB_TX_UNLOCK(txr);
IGB_TX_LOCK_DESTROY(txr);
}
free(adapter->tx_rings, M_DEVBUF);
}
/*********************************************************************
*
* Free transmit ring related data structures.
*
**********************************************************************/
static void
igb_free_transmit_buffers(struct tx_ring *txr)
{
struct adapter *adapter = txr->adapter;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
struct igb_tx_buf *tx_buffer;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
int i;
INIT_DEBUGOUT("free_transmit_ring: begin");
if (txr->tx_buffers == NULL)
return;
tx_buffer = txr->tx_buffers;
for (i = 0; i < adapter->num_tx_desc; i++, tx_buffer++) {
if (tx_buffer->m_head != NULL) {
bus_dmamap_sync(txr->txtag, tx_buffer->map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(txr->txtag,
tx_buffer->map);
m_freem(tx_buffer->m_head);
tx_buffer->m_head = NULL;
if (tx_buffer->map != NULL) {
bus_dmamap_destroy(txr->txtag,
tx_buffer->map);
tx_buffer->map = NULL;
}
} else if (tx_buffer->map != NULL) {
bus_dmamap_unload(txr->txtag,
tx_buffer->map);
bus_dmamap_destroy(txr->txtag,
tx_buffer->map);
tx_buffer->map = NULL;
}
}
Change defines in the igb driver to allow an easier selection of the older if_start/non-multiqueue interface from the stack. This is not the default, but can be turned on in the Makefile now regardless of the OS level to allow either testing or use of ALTQ. MFC after: one week
2013-03-29 18:25:45 +00:00
#ifndef IGB_LEGACY_TX
This delta fixes two bugs: - When a vlan event occurs a check was not made that the event was actually for the interface, thus resulting in a panic. All three drivers have this vulnerability. Add a check for this condition. - Secondly, there was a duplicate buf_ring free in the em driver resulting in a panic on unload. Remove. Approved by: re
2009-07-24 16:57:49 +00:00
if (txr->br != NULL)
buf_ring_free(txr->br, M_DEVBUF);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (txr->tx_buffers != NULL) {
free(txr->tx_buffers, M_DEVBUF);
txr->tx_buffers = NULL;
}
if (txr->txtag != NULL) {
bus_dma_tag_destroy(txr->txtag);
txr->txtag = NULL;
}
return;
}
/**********************************************************************
*
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
* Setup work for hardware segmentation offload (TSO) on
* adapters using advanced tx descriptors
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*
**********************************************************************/
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
static int
igb_tso_setup(struct tx_ring *txr, struct mbuf *mp,
u32 *cmd_type_len, u32 *olinfo_status)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
struct adapter *adapter = txr->adapter;
struct e1000_adv_tx_context_desc *TXD;
u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
u32 mss_l4len_idx = 0, paylen;
u16 vtag = 0, eh_type;
int ctxd, ehdrlen, ip_hlen, tcp_hlen;
struct ether_vlan_header *eh;
#ifdef INET6
struct ip6_hdr *ip6;
#endif
#ifdef INET
struct ip *ip;
#endif
struct tcphdr *th;
/*
* Determine where frame payload starts.
* Jump over vlan headers if already present
*/
eh = mtod(mp, struct ether_vlan_header *);
if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
eh_type = eh->evl_proto;
} else {
ehdrlen = ETHER_HDR_LEN;
eh_type = eh->evl_encap_proto;
}
switch (ntohs(eh_type)) {
#ifdef INET6
case ETHERTYPE_IPV6:
ip6 = (struct ip6_hdr *)(mp->m_data + ehdrlen);
/* XXX-BZ For now we do not pretend to support ext. hdrs. */
if (ip6->ip6_nxt != IPPROTO_TCP)
return (ENXIO);
ip_hlen = sizeof(struct ip6_hdr);
ip6 = (struct ip6_hdr *)(mp->m_data + ehdrlen);
th = (struct tcphdr *)((caddr_t)ip6 + ip_hlen);
th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV6;
break;
#endif
#ifdef INET
case ETHERTYPE_IP:
ip = (struct ip *)(mp->m_data + ehdrlen);
if (ip->ip_p != IPPROTO_TCP)
return (ENXIO);
ip->ip_sum = 0;
ip_hlen = ip->ip_hl << 2;
th = (struct tcphdr *)((caddr_t)ip + ip_hlen);
th->th_sum = in_pseudo(ip->ip_src.s_addr,
ip->ip_dst.s_addr, htons(IPPROTO_TCP));
type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
/* Tell transmit desc to also do IPv4 checksum. */
*olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
break;
#endif
default:
panic("%s: CSUM_TSO but no supported IP version (0x%04x)",
__func__, ntohs(eh_type));
break;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
ctxd = txr->next_avail_desc;
TXD = (struct e1000_adv_tx_context_desc *) &txr->tx_base[ctxd];
tcp_hlen = th->th_off << 2;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* This is used in the transmit desc in encap */
paylen = mp->m_pkthdr.len - ehdrlen - ip_hlen - tcp_hlen;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* VLAN MACLEN IPLEN */
if (mp->m_flags & M_VLANTAG) {
vtag = htole16(mp->m_pkthdr.ether_vtag);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
vlan_macip_lens |= (vtag << E1000_ADVTXD_VLAN_SHIFT);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
vlan_macip_lens |= ehdrlen << E1000_ADVTXD_MACLEN_SHIFT;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
vlan_macip_lens |= ip_hlen;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
TXD->vlan_macip_lens = htole32(vlan_macip_lens);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* ADV DTYPE TUCMD */
type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* MSS L4LEN IDX */
mss_l4len_idx |= (mp->m_pkthdr.tso_segsz << E1000_ADVTXD_MSS_SHIFT);
mss_l4len_idx |= (tcp_hlen << E1000_ADVTXD_L4LEN_SHIFT);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* 82575 needs the queue index added */
if (adapter->hw.mac.type == e1000_82575)
mss_l4len_idx |= txr->me << 4;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
TXD->mss_l4len_idx = htole32(mss_l4len_idx);
TXD->seqnum_seed = htole32(0);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (++ctxd == txr->num_desc)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
ctxd = 0;
txr->tx_avail--;
txr->next_avail_desc = ctxd;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
*cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
*olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
*olinfo_status |= paylen << E1000_ADVTXD_PAYLEN_SHIFT;
++txr->tso_tx;
return (0);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
* Advanced Context Descriptor setup for VLAN, CSUM or TSO
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*
**********************************************************************/
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
static int
igb_tx_ctx_setup(struct tx_ring *txr, struct mbuf *mp,
u32 *cmd_type_len, u32 *olinfo_status)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
struct e1000_adv_tx_context_desc *TXD;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
struct adapter *adapter = txr->adapter;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
struct ether_vlan_header *eh;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
struct ip *ip;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
struct ip6_hdr *ip6;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0, mss_l4len_idx = 0;
int ehdrlen, ip_hlen = 0;
u16 etype;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
u8 ipproto = 0;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
int offload = TRUE;
int ctxd = txr->next_avail_desc;
u16 vtag = 0;
/* First check if TSO is to be used */
if (mp->m_pkthdr.csum_flags & CSUM_TSO)
return (igb_tso_setup(txr, mp, cmd_type_len, olinfo_status));
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if ((mp->m_pkthdr.csum_flags & CSUM_OFFLOAD) == 0)
offload = FALSE;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* Indicate the whole packet as payload when not doing TSO */
*olinfo_status |= mp->m_pkthdr.len << E1000_ADVTXD_PAYLEN_SHIFT;
/* Now ready a context descriptor */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
TXD = (struct e1000_adv_tx_context_desc *) &txr->tx_base[ctxd];
/*
** In advanced descriptors the vlan tag must
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
** be placed into the context descriptor. Hence
** we need to make one even if not doing offloads.
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*/
if (mp->m_flags & M_VLANTAG) {
vtag = htole16(mp->m_pkthdr.ether_vtag);
vlan_macip_lens |= (vtag << E1000_ADVTXD_VLAN_SHIFT);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
} else if (offload == FALSE) /* ... no offload to do */
return (0);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
* Determine where frame payload starts.
* Jump over vlan headers if already present,
* helpful for QinQ too.
*/
eh = mtod(mp, struct ether_vlan_header *);
if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
etype = ntohs(eh->evl_proto);
ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
} else {
etype = ntohs(eh->evl_encap_proto);
ehdrlen = ETHER_HDR_LEN;
}
/* Set the ether header length */
vlan_macip_lens |= ehdrlen << E1000_ADVTXD_MACLEN_SHIFT;
switch (etype) {
case ETHERTYPE_IP:
ip = (struct ip *)(mp->m_data + ehdrlen);
ip_hlen = ip->ip_hl << 2;
ipproto = ip->ip_p;
type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
break;
case ETHERTYPE_IPV6:
ip6 = (struct ip6_hdr *)(mp->m_data + ehdrlen);
ip_hlen = sizeof(struct ip6_hdr);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* XXX-BZ this will go badly in case of ext hdrs. */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
ipproto = ip6->ip6_nxt;
type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV6;
break;
default:
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
offload = FALSE;
break;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
vlan_macip_lens |= ip_hlen;
type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
switch (ipproto) {
case IPPROTO_TCP:
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if (mp->m_pkthdr.csum_flags & CSUM_TCP)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
break;
case IPPROTO_UDP:
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if (mp->m_pkthdr.csum_flags & CSUM_UDP)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_UDP;
break;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
#if __FreeBSD_version >= 800000
case IPPROTO_SCTP:
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if (mp->m_pkthdr.csum_flags & CSUM_SCTP)
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_SCTP;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
break;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
#endif
default:
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
offload = FALSE;
break;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (offload) /* For the TX descriptor setup */
*olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
/* 82575 needs the queue index added */
if (adapter->hw.mac.type == e1000_82575)
mss_l4len_idx = txr->me << 4;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Now copy bits into descriptor */
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
TXD->vlan_macip_lens = htole32(vlan_macip_lens);
TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
TXD->seqnum_seed = htole32(0);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
TXD->mss_l4len_idx = htole32(mss_l4len_idx);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* We've consumed the first desc, adjust counters */
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (++ctxd == txr->num_desc)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
ctxd = 0;
txr->next_avail_desc = ctxd;
--txr->tx_avail;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
return (0);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/**********************************************************************
*
* Examine each tx_buffer in the used queue. If the hardware is done
* processing the packet then free associated resources. The
* tx_buffer is put back on the free queue.
*
* TRUE return means there's work in the ring to clean, FALSE its empty.
**********************************************************************/
static bool
igb_txeof(struct tx_ring *txr)
{
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
struct adapter *adapter = txr->adapter;
struct ifnet *ifp = adapter->ifp;
u32 work, processed = 0;
u16 limit = txr->process_limit;
struct igb_tx_buf *buf;
union e1000_adv_tx_desc *txd;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
mtx_assert(&txr->tx_mtx, MA_OWNED);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
update to the latest netmap snapshot. This includes the following: - use separate memory regions for VALE ports - locking fixes - some simplifications in the NIC-specific routines - performance improvements for the VALE switch - some new features in the pkt-gen test program - documentation updates There are small API changes that require programs to be recompiled (NETMAP_API has been bumped so you will detect old binaries at runtime). In particular: - struct netmap_slot now is 16 bytes to support an extra pointer, which may save one data copy when using VALE ports or VMs; - the struct netmap_if has two extra fields; MFC after: 3 days
2013-11-01 21:21:14 +00:00
if (netmap_tx_irq(ifp, txr->me))
use netmap_rx_irq() / netmap_tx_irq() to handle interrupts in netmap mode, removing the logic from individual drivers. (note: if_lem.c not updated yet due to some other pending modifications)
2013-04-30 16:18:29 +00:00
return (FALSE);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#endif /* DEV_NETMAP */
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (txr->tx_avail == txr->num_desc) {
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
txr->queue_status = IGB_QUEUE_IDLE;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
return FALSE;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* Get work starting point */
work = txr->next_to_clean;
buf = &txr->tx_buffers[work];
txd = &txr->tx_base[work];
work -= txr->num_desc; /* The distance to ring end */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
do {
union e1000_adv_tx_desc *eop = buf->eop;
if (eop == NULL) /* No work */
break;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if ((eop->wb.status & E1000_TXD_STAT_DD) == 0)
break; /* I/O not complete */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (buf->m_head) {
txr->bytes +=
buf->m_head->m_pkthdr.len;
bus_dmamap_sync(txr->txtag,
buf->map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(txr->txtag,
buf->map);
m_freem(buf->m_head);
buf->m_head = NULL;
}
buf->eop = NULL;
++txr->tx_avail;
/* We clean the range if multi segment */
while (txd != eop) {
++txd;
++buf;
++work;
/* wrap the ring? */
if (__predict_false(!work)) {
work -= txr->num_desc;
buf = txr->tx_buffers;
txd = txr->tx_base;
}
if (buf->m_head) {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
txr->bytes +=
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
buf->m_head->m_pkthdr.len;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
bus_dmamap_sync(txr->txtag,
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
buf->map,
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(txr->txtag,
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
buf->map);
m_freem(buf->m_head);
buf->m_head = NULL;
}
++txr->tx_avail;
buf->eop = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
++txr->packets;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
++processed;
- Provide igb_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-19 11:49:41 +00:00
if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
txr->watchdog_time = ticks;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
/* Try the next packet */
++txd;
++buf;
++work;
/* reset with a wrap */
if (__predict_false(!work)) {
work -= txr->num_desc;
buf = txr->tx_buffers;
txd = txr->tx_base;
}
prefetch(txd);
} while (__predict_true(--limit));
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
work += txr->num_desc;
txr->next_to_clean = work;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
/*
** Watchdog calculation, we know there's
** work outstanding or the first return
** would have been taken, so none processed
** for too long indicates a hang.
*/
if ((!processed) && ((ticks - txr->watchdog_time) > IGB_WATCHDOG))
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
txr->queue_status |= IGB_QUEUE_HUNG;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (txr->tx_avail >= IGB_QUEUE_THRESHOLD)
txr->queue_status &= ~IGB_QUEUE_DEPLETED;
if (txr->tx_avail == txr->num_desc) {
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
txr->queue_status = IGB_QUEUE_IDLE;
return (FALSE);
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
}
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
return (TRUE);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*********************************************************************
*
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
* Refresh mbuf buffers for RX descriptor rings
* - now keeps its own state so discards due to resource
* exhaustion are unnecessary, if an mbuf cannot be obtained
* it just returns, keeping its placeholder, thus it can simply
* be recalled to try again.
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*
**********************************************************************/
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
static void
igb_refresh_mbufs(struct rx_ring *rxr, int limit)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
struct adapter *adapter = rxr->adapter;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
bus_dma_segment_t hseg[1];
bus_dma_segment_t pseg[1];
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
struct igb_rx_buf *rxbuf;
struct mbuf *mh, *mp;
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
int i, j, nsegs, error;
bool refreshed = FALSE;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
i = j = rxr->next_to_refresh;
/*
** Get one descriptor beyond
** our work mark to control
** the loop.
*/
if (++j == adapter->num_rx_desc)
j = 0;
while (j != limit) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
rxbuf = &rxr->rx_buffers[i];
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
/* No hdr mbuf used with header split off */
if (rxr->hdr_split == FALSE)
goto no_split;
if (rxbuf->m_head == NULL) {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
mh = m_gethdr(M_NOWAIT, MT_DATA);
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
if (mh == NULL)
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
goto update;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
} else
mh = rxbuf->m_head;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
mh->m_pkthdr.len = mh->m_len = MHLEN;
mh->m_len = MHLEN;
mh->m_flags |= M_PKTHDR;
/* Get the memory mapping */
error = bus_dmamap_load_mbuf_sg(rxr->htag,
rxbuf->hmap, mh, hseg, &nsegs, BUS_DMA_NOWAIT);
if (error != 0) {
printf("Refresh mbufs: hdr dmamap load"
" failure - %d\n", error);
m_free(mh);
rxbuf->m_head = NULL;
goto update;
}
rxbuf->m_head = mh;
bus_dmamap_sync(rxr->htag, rxbuf->hmap,
BUS_DMASYNC_PREREAD);
rxr->rx_base[i].read.hdr_addr =
htole64(hseg[0].ds_addr);
no_split:
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (rxbuf->m_pack == NULL) {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
mp = m_getjcl(M_NOWAIT, MT_DATA,
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
M_PKTHDR, adapter->rx_mbuf_sz);
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
if (mp == NULL)
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
goto update;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
} else
mp = rxbuf->m_pack;
mp->m_pkthdr.len = mp->m_len = adapter->rx_mbuf_sz;
/* Get the memory mapping */
error = bus_dmamap_load_mbuf_sg(rxr->ptag,
rxbuf->pmap, mp, pseg, &nsegs, BUS_DMA_NOWAIT);
if (error != 0) {
printf("Refresh mbufs: payload dmamap load"
" failure - %d\n", error);
m_free(mp);
rxbuf->m_pack = NULL;
goto update;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
}
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
rxbuf->m_pack = mp;
bus_dmamap_sync(rxr->ptag, rxbuf->pmap,
BUS_DMASYNC_PREREAD);
rxr->rx_base[i].read.pkt_addr =
htole64(pseg[0].ds_addr);
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
refreshed = TRUE; /* I feel wefreshed :) */
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
i = j; /* our next is precalculated */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
rxr->next_to_refresh = i;
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
if (++j == adapter->num_rx_desc)
j = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
update:
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
if (refreshed) /* update tail */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
E1000_WRITE_REG(&adapter->hw,
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
E1000_RDT(rxr->me), rxr->next_to_refresh);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
return;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
* Allocate memory for rx_buffer structures. Since we use one
* rx_buffer per received packet, the maximum number of rx_buffer's
* that we'll need is equal to the number of receive descriptors
* that we've allocated.
*
**********************************************************************/
static int
igb_allocate_receive_buffers(struct rx_ring *rxr)
{
struct adapter *adapter = rxr->adapter;
device_t dev = adapter->dev;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
struct igb_rx_buf *rxbuf;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
int i, bsize, error;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
bsize = sizeof(struct igb_rx_buf) * adapter->num_rx_desc;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (!(rxr->rx_buffers =
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
(struct igb_rx_buf *) malloc(bsize,
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
M_DEVBUF, M_NOWAIT | M_ZERO))) {
device_printf(dev, "Unable to allocate rx_buffer memory\n");
error = ENOMEM;
goto fail;
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if ((error = bus_dma_tag_create(bus_get_dma_tag(dev),
With an i386 kernel the igb driver can cause a page fault panic on initialization due to a large number of bounce pages being allocated. This is due to the dma tag requiring page alignment on mbuf mapping. This was removed some time back from the ixgbe driver and is not needed here either.
2009-11-11 19:13:40 +00:00
1, 0, /* alignment, bounds */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
MSIZE, /* maxsize */
1, /* nsegments */
MSIZE, /* maxsegsize */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
0, /* flags */
NULL, /* lockfunc */
NULL, /* lockfuncarg */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
&rxr->htag))) {
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
device_printf(dev, "Unable to create RX DMA tag\n");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
goto fail;
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if ((error = bus_dma_tag_create(bus_get_dma_tag(dev),
1, 0, /* alignment, bounds */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
MJUM9BYTES, /* maxsize */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
1, /* nsegments */
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
MJUM9BYTES, /* maxsegsize */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
0, /* flags */
NULL, /* lockfunc */
NULL, /* lockfuncarg */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
&rxr->ptag))) {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
device_printf(dev, "Unable to create RX payload DMA tag\n");
goto fail;
}
for (i = 0; i < adapter->num_rx_desc; i++) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
rxbuf = &rxr->rx_buffers[i];
Fix several issues with the busdma(9) KPI use in the e1000 drivers. The problems do not affect bouncing busdma in a visible way, but are critical for the dmar backend. - The bus_dmamap_create(9) is not documented to take BUS_DMA_NOWAIT flag. - Unload descriptor map after receive. - Do not reset descriptor map to NULL, bus_dmamap_load(9) requires valid map, and also this leaks the map. Reported and tested by: pho Approved by: jfv Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-11-02 09:16:11 +00:00
error = bus_dmamap_create(rxr->htag, 0, &rxbuf->hmap);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (error) {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
device_printf(dev,
"Unable to create RX head DMA maps\n");
goto fail;
}
Fix several issues with the busdma(9) KPI use in the e1000 drivers. The problems do not affect bouncing busdma in a visible way, but are critical for the dmar backend. - The bus_dmamap_create(9) is not documented to take BUS_DMA_NOWAIT flag. - Unload descriptor map after receive. - Do not reset descriptor map to NULL, bus_dmamap_load(9) requires valid map, and also this leaks the map. Reported and tested by: pho Approved by: jfv Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-11-02 09:16:11 +00:00
error = bus_dmamap_create(rxr->ptag, 0, &rxbuf->pmap);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (error) {
device_printf(dev,
"Unable to create RX packet DMA maps\n");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
goto fail;
}
}
return (0);
fail:
/* Frees all, but can handle partial completion */
igb_free_receive_structures(adapter);
return (error);
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
static void
igb_free_receive_ring(struct rx_ring *rxr)
{
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
struct adapter *adapter = rxr->adapter;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct igb_rx_buf *rxbuf;
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
for (int i = 0; i < adapter->num_rx_desc; i++) {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxbuf = &rxr->rx_buffers[i];
if (rxbuf->m_head != NULL) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
bus_dmamap_sync(rxr->htag, rxbuf->hmap,
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
BUS_DMASYNC_POSTREAD);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
bus_dmamap_unload(rxr->htag, rxbuf->hmap);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxbuf->m_head->m_flags |= M_PKTHDR;
m_freem(rxbuf->m_head);
}
if (rxbuf->m_pack != NULL) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
bus_dmamap_sync(rxr->ptag, rxbuf->pmap,
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
BUS_DMASYNC_POSTREAD);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
bus_dmamap_unload(rxr->ptag, rxbuf->pmap);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxbuf->m_pack->m_flags |= M_PKTHDR;
m_freem(rxbuf->m_pack);
}
rxbuf->m_head = NULL;
rxbuf->m_pack = NULL;
}
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
* Initialize a receive ring and its buffers.
*
**********************************************************************/
static int
igb_setup_receive_ring(struct rx_ring *rxr)
{
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
struct adapter *adapter;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
struct ifnet *ifp;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
device_t dev;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
struct igb_rx_buf *rxbuf;
bus_dma_segment_t pseg[1], hseg[1];
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
struct lro_ctrl *lro = &rxr->lro;
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
int rsize, nsegs, error = 0;
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
struct netmap_adapter *na = NA(rxr->adapter->ifp);
struct netmap_slot *slot;
#endif /* DEV_NETMAP */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
adapter = rxr->adapter;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
dev = adapter->dev;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
ifp = adapter->ifp;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
/* Clear the ring contents */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
IGB_RX_LOCK(rxr);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
slot = netmap_reset(na, NR_RX, rxr->me, 0);
#endif /* DEV_NETMAP */
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
rsize = roundup2(adapter->num_rx_desc *
sizeof(union e1000_adv_rx_desc), IGB_DBA_ALIGN);
bzero((void *)rxr->rx_base, rsize);
/*
** Free current RX buffer structures and their mbufs
*/
igb_free_receive_ring(rxr);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
/* Configure for header split? */
if (igb_header_split)
rxr->hdr_split = TRUE;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Now replenish the ring mbufs */
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
for (int j = 0; j < adapter->num_rx_desc; ++j) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
struct mbuf *mh, *mp;
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
rxbuf = &rxr->rx_buffers[j];
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
if (slot) {
reference the correct variable in a comment MFC after: 3 days
2014-05-28 06:50:16 +00:00
/* slot sj is mapped to the j-th NIC-ring entry */
A bunch of netmap fixes: USERSPACE: 1. add support for devices with different number of rx and tx queues; 2. add better support for zero-copy operation, adding an extra field to the netmap ring to indicate how many buffers we have already processed but not yet released (with help from Eddie Kohler); 3. The two changes above unfortunately require an API change, so while at it add a version field and some spares to the ioctl() argument to help detect mismatches. 4. update the manual page for the two changes above; 5. update sample applications in tools/tools/netmap KERNEL: 1. simplify the internal structures moving the global wait queues to the 'struct netmap_adapter'; 2. simplify the functions that map kring<->nic ring indexes 3. normalize device-specific code, helps mainteinance; 4. start exploring the impact of micro-optimizations (prefetch etc.) in the ixgbe driver. Use 'legacy' descriptors on the tx ring and prefetch slots gives about 20% speedup at 900 MHz. Another 7-10% would come from removing the explict calls to bus_dmamap* in the core (they are effectively NOPs in this case, but it takes expensive load of the per-buffer dma maps to figure out that they are all NULL. Rx performance not investigated. I am postponing the MFC so i can import a few more improvements before merging.
2012-02-27 19:05:01 +00:00
int sj = netmap_idx_n2k(&na->rx_rings[rxr->me], j);
small code cleanup in preparation for future modifications in the memory allocator used by netmap. No functional change, two small bug fixes: - in if_re.c add a missing bus_dmamap_sync() - in netmap.c comment out a spurious free() in an error handling block
2012-01-10 19:57:23 +00:00
uint64_t paddr;
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
void *addr;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
addr = PNMB(na, slot + sj, &paddr);
netmap_load_map(na, rxr->ptag, rxbuf->pmap, addr);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
/* Update descriptor */
small code cleanup in preparation for future modifications in the memory allocator used by netmap. No functional change, two small bug fixes: - in if_re.c add a missing bus_dmamap_sync() - in netmap.c comment out a spurious free() in an error handling block
2012-01-10 19:57:23 +00:00
rxr->rx_base[j].read.pkt_addr = htole64(paddr);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
continue;
}
#endif /* DEV_NETMAP */
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
if (rxr->hdr_split == FALSE)
goto skip_head;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* First the header */
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
rxbuf->m_head = m_gethdr(M_NOWAIT, MT_DATA);
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
if (rxbuf->m_head == NULL) {
error = ENOBUFS;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
goto fail;
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
m_adj(rxbuf->m_head, ETHER_ALIGN);
mh = rxbuf->m_head;
mh->m_len = mh->m_pkthdr.len = MHLEN;
mh->m_flags |= M_PKTHDR;
/* Get the memory mapping */
error = bus_dmamap_load_mbuf_sg(rxr->htag,
rxbuf->hmap, rxbuf->m_head, hseg,
&nsegs, BUS_DMA_NOWAIT);
if (error != 0) /* Nothing elegant to do here */
goto fail;
bus_dmamap_sync(rxr->htag,
rxbuf->hmap, BUS_DMASYNC_PREREAD);
/* Update descriptor */
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
rxr->rx_base[j].read.hdr_addr = htole64(hseg[0].ds_addr);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
skip_head:
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Now the payload cluster */
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
rxbuf->m_pack = m_getjcl(M_NOWAIT, MT_DATA,
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
M_PKTHDR, adapter->rx_mbuf_sz);
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
if (rxbuf->m_pack == NULL) {
error = ENOBUFS;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
goto fail;
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
mp = rxbuf->m_pack;
mp->m_pkthdr.len = mp->m_len = adapter->rx_mbuf_sz;
/* Get the memory mapping */
error = bus_dmamap_load_mbuf_sg(rxr->ptag,
rxbuf->pmap, mp, pseg,
&nsegs, BUS_DMA_NOWAIT);
if (error != 0)
goto fail;
bus_dmamap_sync(rxr->ptag,
rxbuf->pmap, BUS_DMASYNC_PREREAD);
/* Update descriptor */
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
rxr->rx_base[j].read.pkt_addr = htole64(pseg[0].ds_addr);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
Important fix got clobbered in the em driver, keeping VLAN HWFILTER from being used by default, this breaks stacked pseudo devices, and as it turns out, also breaks virtual machines that happen to use VLANS (didn't know that before :). Put the fix back into the em driver, and for good measure add the same code to the igb driver where it should have been anyway.
2010-04-08 00:50:43 +00:00
/* Setup our descriptor indices */
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
rxr->next_to_check = 0;
rxr->next_to_refresh = adapter->num_rx_desc - 1;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
rxr->lro_enabled = FALSE;
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
rxr->rx_split_packets = 0;
rxr->rx_bytes = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxr->fmp = NULL;
rxr->lmp = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/*
** Now set up the LRO interface, we
** also only do head split when LRO
** is enabled, since so often they
** are undesireable in similar setups.
*/
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (ifp->if_capenable & IFCAP_LRO) {
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
error = tcp_lro_init(lro);
if (error) {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
device_printf(dev, "LRO Initialization failed!\n");
goto fail;
}
This delta is primarily a fix for es2lan devices that will sometimes fail to initialize problem due to a lock contention with management hardware. However, in order to deliver that fix it was necessary to take a shared code update as a whole, and this required scattered changes in the core code to be compatible. The em driver now has VLAN HW support added as the igb driver had previously. MFC after: ASAP - in time for 7.1 RELEASE
2008-11-26 23:57:23 +00:00
INIT_DEBUGOUT("RX LRO Initialized\n");
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
rxr->lro_enabled = TRUE;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
lro->ifp = adapter->ifp;
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
IGB_RX_UNLOCK(rxr);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
return (0);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
fail:
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
igb_free_receive_ring(rxr);
IGB_RX_UNLOCK(rxr);
return (error);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
* Initialize all receive rings.
*
**********************************************************************/
static int
igb_setup_receive_structures(struct adapter *adapter)
{
struct rx_ring *rxr = adapter->rx_rings;
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
int i;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (i = 0; i < adapter->num_queues; i++, rxr++)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (igb_setup_receive_ring(rxr))
goto fail;
return (0);
fail:
/*
* Free RX buffers allocated so far, we will only handle
* the rings that completed, the failing case will have
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
* cleaned up for itself. 'i' is the endpoint.
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*/
Fixed mbuf free when receive structures fail to allocate. This prevents quad igb card on high core machines, without any nmbcluster or igb queue tuning wedging the boot process if all nics are configured. Reviewed by: jfv Approved by: pjd (mentor) MFC after: 1 week
2013-01-12 16:05:55 +00:00
for (int j = 0; j < i; ++j) {
rxr = &adapter->rx_rings[j];
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
IGB_RX_LOCK(rxr);
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
igb_free_receive_ring(rxr);
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
IGB_RX_UNLOCK(rxr);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
return (ENOBUFS);
}
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
/*
* Initialise the RSS mapping for NICs that support multiple transmit/
* receive rings.
*/
static void
igb_initialise_rss_mapping(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
int i;
int queue_id;
Fix the igb(4) redirection table to correctly populate. This is similar to the ixgbe(4) fix. Tested: * Intel I350 gigabit adapter
2014-07-23 05:40:28 +00:00
u32 reta;
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
u32 rss_key[10], mrqc, shift = 0;
/* XXX? */
if (adapter->hw.mac.type == e1000_82575)
shift = 6;
/*
* The redirection table controls which destination
* queue each bucket redirects traffic to.
* Each DWORD represents four queues, with the LSB
* being the first queue in the DWORD.
*
* This just allocates buckets to queues using round-robin
* allocation.
*
* NOTE: It Just Happens to line up with the default
* RSS allocation method.
*/
/* Warning FM follows */
Fix the igb(4) redirection table to correctly populate. This is similar to the ixgbe(4) fix. Tested: * Intel I350 gigabit adapter
2014-07-23 05:40:28 +00:00
reta = 0;
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
for (i = 0; i < 128; i++) {
#ifdef RSS
queue_id = rss_get_indirection_to_bucket(i);
/*
* If we have more queues than buckets, we'll
* end up mapping buckets to a subset of the
* queues.
*
* If we have more buckets than queues, we'll
* end up instead assigning multiple buckets
* to queues.
*
* Both are suboptimal, but we need to handle
* the case so we don't go out of bounds
* indexing arrays and such.
*/
queue_id = queue_id % adapter->num_queues;
#else
queue_id = (i % adapter->num_queues);
#endif
Fix the igb(4) redirection table to correctly populate. This is similar to the ixgbe(4) fix. Tested: * Intel I350 gigabit adapter
2014-07-23 05:40:28 +00:00
/* Adjust if required */
queue_id = queue_id << shift;
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
Fix the igb(4) redirection table to correctly populate. This is similar to the ixgbe(4) fix. Tested: * Intel I350 gigabit adapter
2014-07-23 05:40:28 +00:00
/*
* The low 8 bits are for hash value (n+0);
* The next 8 bits are for hash value (n+1), etc.
*/
reta = reta >> 8;
reta = reta | ( ((uint32_t) queue_id) << 24);
if ((i & 3) == 3) {
E1000_WRITE_REG(hw, E1000_RETA(i >> 2), reta);
reta = 0;
}
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
}
/* Now fill in hash table */
/* XXX This means RSS enable + 8 queues for my igb (82580.) */
mrqc = E1000_MRQC_ENABLE_RSS_4Q;
#ifdef RSS
/* XXX ew typecasting */
rss_getkey((uint8_t *) &rss_key);
#else
arc4rand(&rss_key, sizeof(rss_key), 0);
#endif
for (i = 0; i < 10; i++)
E1000_WRITE_REG_ARRAY(hw,
E1000_RSSRK(0), i, rss_key[i]);
/*
* Configure the RSS fields to hash upon.
*/
mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
E1000_MRQC_RSS_FIELD_IPV4_TCP);
mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
E1000_MRQC_RSS_FIELD_IPV6_TCP);
mrqc |=( E1000_MRQC_RSS_FIELD_IPV4_UDP |
E1000_MRQC_RSS_FIELD_IPV6_UDP);
mrqc |=( E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
E1000_WRITE_REG(hw, E1000_MRQC, mrqc);
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
* Enable receive unit.
*
**********************************************************************/
static void
igb_initialize_receive_units(struct adapter *adapter)
{
struct rx_ring *rxr = adapter->rx_rings;
struct ifnet *ifp = adapter->ifp;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct e1000_hw *hw = &adapter->hw;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
u32 rctl, rxcsum, psize, srrctl = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
INIT_DEBUGOUT("igb_initialize_receive_unit: begin");
/*
* Make sure receives are disabled while setting
* up the descriptor ring
*/
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rctl = E1000_READ_REG(hw, E1000_RCTL);
E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/*
** Set up for header split
*/
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
if (igb_header_split) {
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/* Use a standard mbuf for the header */
srrctl |= IGB_HDR_BUF << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
} else
srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
/*
** Set up for jumbo frames
*/
if (ifp->if_mtu > ETHERMTU) {
rctl |= E1000_RCTL_LPE;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
if (adapter->rx_mbuf_sz == MJUMPAGESIZE) {
srrctl |= 4096 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX;
} else if (adapter->rx_mbuf_sz > MJUMPAGESIZE) {
srrctl |= 8192 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX;
}
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/* Set maximum packet len */
psize = adapter->max_frame_size;
/* are we on a vlan? */
if (adapter->ifp->if_vlantrunk != NULL)
psize += VLAN_TAG_SIZE;
E1000_WRITE_REG(&adapter->hw, E1000_RLPML, psize);
} else {
rctl &= ~E1000_RCTL_LPE;
srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
rctl |= E1000_RCTL_SZ_2048;
}
Set DROP_EN on each RX queue if transmit flow-control is disabled. This allows the NIC to drop frames on the receive queue and not cause the MAC to block on receiving to _any_ queue. Tested: igb0@pci0:5:0:0: class=0x020000 card=0x152115d9 chip=0x15218086 rev=0x01 hdr=0x00 vendor = 'Intel Corporation' device = 'I350 Gigabit Network Connection' class = network subclass = ethernet Discussed with: Eric Joyner <eric.joyner@intel.com> MFC after: 1 week Sponsored by: Norse Corp, Inc.
2014-09-15 19:53:49 +00:00
/*
* If TX flow control is disabled and there's >1 queue defined,
* enable DROP.
*
* This drops frames rather than hanging the RX MAC for all queues.
*/
if ((adapter->num_queues > 1) &&
(adapter->fc == e1000_fc_none ||
adapter->fc == e1000_fc_rx_pause)) {
srrctl |= E1000_SRRCTL_DROP_EN;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Setup the Base and Length of the Rx Descriptor Rings */
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++, rxr++) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
u64 bus_addr = rxr->rxdma.dma_paddr;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
u32 rxdctl;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_RDLEN(i),
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
adapter->num_rx_desc * sizeof(struct e1000_rx_desc));
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_RDBAH(i),
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
(uint32_t)(bus_addr >> 32));
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_RDBAL(i),
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
(uint32_t)bus_addr);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_SRRCTL(i), srrctl);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Enable this Queue */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
rxdctl &= 0xFFF00000;
rxdctl |= IGB_RX_PTHRESH;
rxdctl |= IGB_RX_HTHRESH << 8;
rxdctl |= IGB_RX_WTHRESH << 16;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
/*
** Setup for RX MultiQueue
*/
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if (adapter->num_queues >1) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
/* rss setup */
igb_initialise_rss_mapping(adapter);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
** NOTE: Receive Full-Packet Checksum Offload
** is mutually exclusive with Multiqueue. However
** this is not the same as TCP/IP checksums which
** still work.
*/
rxcsum |= E1000_RXCSUM_PCSD;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
#if __FreeBSD_version >= 800000
/* For SCTP Offload */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if ((hw->mac.type == e1000_82576)
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
&& (ifp->if_capenable & IFCAP_RXCSUM))
rxcsum |= E1000_RXCSUM_CRCOFL;
#endif
} else {
/* Non RSS setup */
if (ifp->if_capenable & IFCAP_RXCSUM) {
rxcsum |= E1000_RXCSUM_IPPCSE;
#if __FreeBSD_version >= 800000
if (adapter->hw.mac.type == e1000_82576)
rxcsum |= E1000_RXCSUM_CRCOFL;
#endif
} else
rxcsum &= ~E1000_RXCSUM_TUOFL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Setup the Receive Control Register */
rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
E1000_RCTL_RDMTS_HALF |
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
(hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
/* Strip CRC bytes. */
rctl |= E1000_RCTL_SECRC;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Make sure VLAN Filters are off */
rctl &= ~E1000_RCTL_VFE;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/* Don't store bad packets */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
rctl &= ~E1000_RCTL_SBP;
/* Enable Receives */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
E1000_WRITE_REG(hw, E1000_RCTL, rctl);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*
* Setup the HW Rx Head and Tail Descriptor Pointers
* - needs to be after enable
*/
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++) {
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
rxr = &adapter->rx_rings[i];
E1000_WRITE_REG(hw, E1000_RDH(i), rxr->next_to_check);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
/*
* an init() while a netmap client is active must
* preserve the rx buffers passed to userspace.
* In this driver it means we adjust RDT to
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
* something different from next_to_refresh
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
* (which is not used in netmap mode).
*/
if (ifp->if_capenable & IFCAP_NETMAP) {
struct netmap_adapter *na = NA(adapter->ifp);
struct netmap_kring *kring = &na->rx_rings[i];
It is 2014 and we have a new version of netmap. Most relevant features: - netmap emulation on any NIC, even those without native netmap support. On the ixgbe we have measured about 4Mpps/core/queue in this mode, which is still a lot more than with sockets/bpf. - seamless interconnection of VALE switch, NICs and host stack. If you disable accelerations on your NIC (say em0) ifconfig em0 -txcsum -txcsum you can use the VALE switch to connect the NIC and the host stack: vale-ctl -h valeXX:em0 allowing sharing the NIC with other netmap clients. - THE USER API HAS SLIGHTLY CHANGED (head/cur/tail pointers instead of pointers/count as before). This was unavoidable to support, in the future, multiple threads operating on the same rings. Netmap clients require very small source code changes to compile again. On the plus side, the new API should be easier to understand and the internals are a lot simpler. The manual page has been updated extensively to reflect the current features and give some examples. This is the result of work of several people including Giuseppe Lettieri, Vincenzo Maffione, Michio Honda and myself, and has been financially supported by EU projects CHANGE and OPENLAB, from NetApp University Research Fund, NEC, and of course the Universita` di Pisa.
2014-01-06 12:53:15 +00:00
int t = rxr->next_to_refresh - nm_kr_rxspace(kring);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
if (t >= adapter->num_rx_desc)
t -= adapter->num_rx_desc;
else if (t < 0)
t += adapter->num_rx_desc;
E1000_WRITE_REG(hw, E1000_RDT(i), t);
} else
#endif /* DEV_NETMAP */
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
E1000_WRITE_REG(hw, E1000_RDT(i), rxr->next_to_refresh);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
return;
}
/*********************************************************************
*
* Free receive rings.
*
**********************************************************************/
static void
igb_free_receive_structures(struct adapter *adapter)
{
struct rx_ring *rxr = adapter->rx_rings;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
for (int i = 0; i < adapter->num_queues; i++, rxr++) {
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
struct lro_ctrl *lro = &rxr->lro;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
igb_free_receive_buffers(rxr);
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
tcp_lro_free(lro);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
igb_dma_free(adapter, &rxr->rxdma);
}
free(adapter->rx_rings, M_DEVBUF);
}
/*********************************************************************
*
* Free receive ring data structures.
*
**********************************************************************/
static void
igb_free_receive_buffers(struct rx_ring *rxr)
{
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
struct adapter *adapter = rxr->adapter;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct igb_rx_buf *rxbuf;
int i;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
INIT_DEBUGOUT("free_receive_structures: begin");
/* Cleanup any existing buffers */
if (rxr->rx_buffers != NULL) {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
for (i = 0; i < adapter->num_rx_desc; i++) {
rxbuf = &rxr->rx_buffers[i];
if (rxbuf->m_head != NULL) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
bus_dmamap_sync(rxr->htag, rxbuf->hmap,
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
BUS_DMASYNC_POSTREAD);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
bus_dmamap_unload(rxr->htag, rxbuf->hmap);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxbuf->m_head->m_flags |= M_PKTHDR;
m_freem(rxbuf->m_head);
}
if (rxbuf->m_pack != NULL) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
bus_dmamap_sync(rxr->ptag, rxbuf->pmap,
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
BUS_DMASYNC_POSTREAD);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
bus_dmamap_unload(rxr->ptag, rxbuf->pmap);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxbuf->m_pack->m_flags |= M_PKTHDR;
m_freem(rxbuf->m_pack);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxbuf->m_head = NULL;
rxbuf->m_pack = NULL;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (rxbuf->hmap != NULL) {
bus_dmamap_destroy(rxr->htag, rxbuf->hmap);
rxbuf->hmap = NULL;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (rxbuf->pmap != NULL) {
bus_dmamap_destroy(rxr->ptag, rxbuf->pmap);
rxbuf->pmap = NULL;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
}
}
if (rxr->rx_buffers != NULL) {
free(rxr->rx_buffers, M_DEVBUF);
rxr->rx_buffers = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (rxr->htag != NULL) {
bus_dma_tag_destroy(rxr->htag);
rxr->htag = NULL;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (rxr->ptag != NULL) {
bus_dma_tag_destroy(rxr->ptag);
rxr->ptag = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
static __inline void
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
igb_rx_discard(struct rx_ring *rxr, int i)
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
{
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
struct igb_rx_buf *rbuf;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
rbuf = &rxr->rx_buffers[i];
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
/* Partially received? Free the chain */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (rxr->fmp != NULL) {
rxr->fmp->m_flags |= M_PKTHDR;
m_freem(rxr->fmp);
rxr->fmp = NULL;
rxr->lmp = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
/*
** With advanced descriptors the writeback
** clobbers the buffer addrs, so its easier
** to just free the existing mbufs and take
** the normal refresh path to get new buffers
** and mapping.
*/
if (rbuf->m_head) {
m_free(rbuf->m_head);
rbuf->m_head = NULL;
Fix several issues with the busdma(9) KPI use in the e1000 drivers. The problems do not affect bouncing busdma in a visible way, but are critical for the dmar backend. - The bus_dmamap_create(9) is not documented to take BUS_DMA_NOWAIT flag. - Unload descriptor map after receive. - Do not reset descriptor map to NULL, bus_dmamap_load(9) requires valid map, and also this leaks the map. Reported and tested by: pho Approved by: jfv Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-11-02 09:16:11 +00:00
bus_dmamap_unload(rxr->htag, rbuf->hmap);
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
if (rbuf->m_pack) {
m_free(rbuf->m_pack);
rbuf->m_pack = NULL;
Fix several issues with the busdma(9) KPI use in the e1000 drivers. The problems do not affect bouncing busdma in a visible way, but are critical for the dmar backend. - The bus_dmamap_create(9) is not documented to take BUS_DMA_NOWAIT flag. - Unload descriptor map after receive. - Do not reset descriptor map to NULL, bus_dmamap_load(9) requires valid map, and also this leaks the map. Reported and tested by: pho Approved by: jfv Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-11-02 09:16:11 +00:00
bus_dmamap_unload(rxr->ptag, rbuf->pmap);
Update code from Intel: - Sync shared code with Intel internal - New client chipset support added - em driver - fixes to 82574, limit queues to 1 but use MSIX - em driver - large changes in TX checksum offload and tso code, thanks to yongari. - some small changes for watchdog issues. - igb driver - local timer watchdog code was missing locking this and a couple other watchdog related fixes. - bug in rx discard found by Andrew Boyer, check for null pointer MFC: a week
2010-09-28 00:13:15 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
return;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
static __inline void
igb_rx_input(struct rx_ring *rxr, struct ifnet *ifp, struct mbuf *m, u32 ptype)
{
/*
* ATM LRO is only for IPv4/TCP packets and TCP checksum of the packet
* should be computed by hardware. Also it should not have VLAN tag in
* ethernet header.
*/
if (rxr->lro_enabled &&
(ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 &&
(ptype & E1000_RXDADV_PKTTYPE_ETQF) == 0 &&
(ptype & (E1000_RXDADV_PKTTYPE_IPV4 | E1000_RXDADV_PKTTYPE_TCP)) ==
(E1000_RXDADV_PKTTYPE_IPV4 | E1000_RXDADV_PKTTYPE_TCP) &&
(m->m_pkthdr.csum_flags & (CSUM_DATA_VALID | CSUM_PSEUDO_HDR)) ==
(CSUM_DATA_VALID | CSUM_PSEUDO_HDR)) {
/*
* Send to the stack if:
** - LRO not enabled, or
** - no LRO resources, or
** - lro enqueue fails
*/
if (rxr->lro.lro_cnt != 0)
if (tcp_lro_rx(&rxr->lro, m, 0) == 0)
return;
}
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
IGB_RX_UNLOCK(rxr);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
(*ifp->if_input)(ifp, m);
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
IGB_RX_LOCK(rxr);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
* This routine executes in interrupt context. It replenishes
* the mbufs in the descriptor and sends data which has been
* dma'ed into host memory to upper layer.
*
* We loop at most count times if count is > 0, or until done if
* count < 0.
*
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
* Return TRUE if more to clean, FALSE otherwise
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*********************************************************************/
static bool
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
igb_rxeof(struct igb_queue *que, int count, int *done)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
struct adapter *adapter = que->adapter;
struct rx_ring *rxr = que->rxr;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
struct ifnet *ifp = adapter->ifp;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
struct lro_ctrl *lro = &rxr->lro;
struct lro_entry *queued;
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
int i, processed = 0, rxdone = 0;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
u32 ptype, staterr = 0;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
union e1000_adv_rx_desc *cur;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
IGB_RX_LOCK(rxr);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Sync the ring. */
bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
update to the latest netmap snapshot. This includes the following: - use separate memory regions for VALE ports - locking fixes - some simplifications in the NIC-specific routines - performance improvements for the VALE switch - some new features in the pkt-gen test program - documentation updates There are small API changes that require programs to be recompiled (NETMAP_API has been bumped so you will detect old binaries at runtime). In particular: - struct netmap_slot now is 16 bytes to support an extra pointer, which may save one data copy when using VALE ports or VMs; - the struct netmap_if has two extra fields; MFC after: 3 days
2013-11-01 21:21:14 +00:00
if (netmap_rx_irq(ifp, rxr->me, &processed)) {
IGB_RX_UNLOCK(rxr);
use netmap_rx_irq() / netmap_tx_irq() to handle interrupts in netmap mode, removing the logic from individual drivers. (note: if_lem.c not updated yet due to some other pending modifications)
2013-04-30 16:18:29 +00:00
return (FALSE);
update to the latest netmap snapshot. This includes the following: - use separate memory regions for VALE ports - locking fixes - some simplifications in the NIC-specific routines - performance improvements for the VALE switch - some new features in the pkt-gen test program - documentation updates There are small API changes that require programs to be recompiled (NETMAP_API has been bumped so you will detect old binaries at runtime). In particular: - struct netmap_slot now is 16 bytes to support an extra pointer, which may save one data copy when using VALE ports or VMs; - the struct netmap_if has two extra fields; MFC after: 3 days
2013-11-01 21:21:14 +00:00
}
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#endif /* DEV_NETMAP */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* Main clean loop */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
for (i = rxr->next_to_check; count != 0;) {
struct mbuf *sendmp, *mh, *mp;
struct igb_rx_buf *rxbuf;
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
u16 hlen, plen, hdr, vtag, pkt_info;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
bool eop = FALSE;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
cur = &rxr->rx_base[i];
staterr = le32toh(cur->wb.upper.status_error);
if ((staterr & E1000_RXD_STAT_DD) == 0)
break;
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
break;
count--;
sendmp = mh = mp = NULL;
cur->wb.upper.status_error = 0;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
rxbuf = &rxr->rx_buffers[i];
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
plen = le16toh(cur->wb.upper.length);
ptype = le32toh(cur->wb.lower.lo_dword.data) & IGB_PKTTYPE_MASK;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
if (((adapter->hw.mac.type == e1000_i350) ||
(adapter->hw.mac.type == e1000_i354)) &&
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
(staterr & E1000_RXDEXT_STATERR_LB))
vtag = be16toh(cur->wb.upper.vlan);
else
vtag = le16toh(cur->wb.upper.vlan);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
hdr = le16toh(cur->wb.lower.lo_dword.hs_rss.hdr_info);
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
pkt_info = le16toh(cur->wb.lower.lo_dword.hs_rss.pkt_info);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
eop = ((staterr & E1000_RXD_STAT_EOP) == E1000_RXD_STAT_EOP);
Fix the handling of EOP in status descriptors for if_igb(4) and don't double-free mbufs. Like ixgbe(4) chipsets, EOP is only set on the final descriptor in a chain of descriptors. So, to free the whole list of descriptors, we should free the current slot _and_ the assembled list of descriptors that make up the fragment list. The existing code was setting discard once it saw EOP + an error status; it then freed all the subsequent descriptors until the next EOP. That's totally the wrong order.
2014-09-18 16:20:17 +00:00
/*
* Free the frame (all segments) if we're at EOP and
* it's an error.
*
* The datasheet states that EOP + status is only valid for
* the final segment in a multi-segment frame.
*/
if (eop && ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) != 0)) {
This patch fixes a nit in the em, lem, and igb driver statistics. Increment adapter->dropped_pkts instead of if_ierrors because if_ierrors is overwritten by hw stats collection. Submitted by: Andrew Boyer <aboyer@averesystems.com> Reviewed by: Jack F Vogel <jfv@freebsd.org> MFC after: 2 weeks
2012-09-23 22:53:39 +00:00
adapter->dropped_pkts++;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
++rxr->rx_discarded;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
igb_rx_discard(rxr, i);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
goto next_desc;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
}
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/*
** The way the hardware is configured to
** split, it will ONLY use the header buffer
** when header split is enabled, otherwise we
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
** get normal behavior, ie, both header and
** payload are DMA'd into the payload buffer.
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
**
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
** The fmp test is to catch the case where a
** packet spans multiple descriptors, in that
** case only the first header is valid.
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
*/
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (rxr->hdr_split && rxr->fmp == NULL) {
Fix several issues with the busdma(9) KPI use in the e1000 drivers. The problems do not affect bouncing busdma in a visible way, but are critical for the dmar backend. - The bus_dmamap_create(9) is not documented to take BUS_DMA_NOWAIT flag. - Unload descriptor map after receive. - Do not reset descriptor map to NULL, bus_dmamap_load(9) requires valid map, and also this leaks the map. Reported and tested by: pho Approved by: jfv Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-11-02 09:16:11 +00:00
bus_dmamap_unload(rxr->htag, rxbuf->hmap);
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
hlen = (hdr & E1000_RXDADV_HDRBUFLEN_MASK) >>
E1000_RXDADV_HDRBUFLEN_SHIFT;
if (hlen > IGB_HDR_BUF)
hlen = IGB_HDR_BUF;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
mh = rxr->rx_buffers[i].m_head;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
mh->m_len = hlen;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
/* clear buf pointer for refresh */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
rxbuf->m_head = NULL;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/*
** Get the payload length, this
** could be zero if its a small
** packet.
*/
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (plen > 0) {
mp = rxr->rx_buffers[i].m_pack;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
mp->m_len = plen;
mh->m_next = mp;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
/* clear buf pointer */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
rxbuf->m_pack = NULL;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
rxr->rx_split_packets++;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
}
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
} else {
/*
** Either no header split, or a
** secondary piece of a fragmented
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
** split packet.
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
*/
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
mh = rxr->rx_buffers[i].m_pack;
mh->m_len = plen;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* clear buf info for refresh */
rxbuf->m_pack = NULL;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
}
Fix several issues with the busdma(9) KPI use in the e1000 drivers. The problems do not affect bouncing busdma in a visible way, but are critical for the dmar backend. - The bus_dmamap_create(9) is not documented to take BUS_DMA_NOWAIT flag. - Unload descriptor map after receive. - Do not reset descriptor map to NULL, bus_dmamap_load(9) requires valid map, and also this leaks the map. Reported and tested by: pho Approved by: jfv Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-11-02 09:16:11 +00:00
bus_dmamap_unload(rxr->ptag, rxbuf->pmap);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
++processed; /* So we know when to refresh */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* Initial frame - setup */
if (rxr->fmp == NULL) {
mh->m_pkthdr.len = mh->m_len;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
/* Save the head of the chain */
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxr->fmp = mh;
rxr->lmp = mh;
if (mp != NULL) {
/* Add payload if split */
mh->m_pkthdr.len += mp->m_len;
rxr->lmp = mh->m_next;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
} else {
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* Chain mbuf's together */
rxr->lmp->m_next = mh;
rxr->lmp = rxr->lmp->m_next;
rxr->fmp->m_pkthdr.len += mh->m_len;
}
if (eop) {
rxr->fmp->m_pkthdr.rcvif = ifp;
- Provide igb_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-19 11:49:41 +00:00
if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxr->rx_packets++;
/* capture data for AIM */
rxr->packets++;
rxr->bytes += rxr->fmp->m_pkthdr.len;
rxr->rx_bytes += rxr->fmp->m_pkthdr.len;
if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
igb_rx_checksum(staterr, rxr->fmp, ptype);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 &&
(staterr & E1000_RXD_STAT_VP) != 0) {
rxr->fmp->m_pkthdr.ether_vtag = vtag;
rxr->fmp->m_flags |= M_VLANTAG;
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
}
Add initial RSS awareness to the igb(4) driver. The igb(4) hardware is capable of RSS hashing RX packets and doing RSS queue selection for up to 8 queues. (I believe some hardware is limited to 4 queues, but I haven't tested on that.) However, even if multi-queue is enabled for igb(4), the RX path doesn't use the RSS flowid from the received descriptor. It just uses the MSIX queue id. This patch does a handful of things if RSS is enabled: * Instead of using a random key at boot, fetch the RSS key from the RSS code and program that in to the RSS redirection table. That whole chunk of code should be double checked for endian correctness. * Use the RSS queue mapping to CPU ID to figure out where to thread pin the RX swi thread and the taskqueue threads for each queue. * The software queue is now really an "RSS bucket". * When programming the RSS indirection table, use the RSS code to figure out which RSS bucket each slot in the indirection table maps to. * When transmitting, use the flowid RSS mapping if the mbuf has an RSS aware hash. The existing method wasn't guaranteed to align correctly with the destination RSS bucket (and thus CPU ID.) This code warns if the number of RSS buckets isn't the same as the automatically configured number of hardware queues. The administrator will have to tweak one of them for better performance. There's currently no way to re-balance the RSS indirection table after startup. I'll worry about that later. Additionally, it may be worthwhile to always use the full 32 bit flowid if multi-queue is enabled. It'll make things like lagg(4) behave better with respect to traffic distribution.
2014-06-30 04:34:59 +00:00
#ifdef RSS
/* XXX set flowtype once this works right */
rxr->fmp->m_pkthdr.flowid =
le32toh(cur->wb.lower.hi_dword.rss);
rxr->fmp->m_flags |= M_FLOWID;
switch (pkt_info & E1000_RXDADV_RSSTYPE_MASK) {
case E1000_RXDADV_RSSTYPE_IPV4_TCP:
M_HASHTYPE_SET(rxr->fmp, M_HASHTYPE_RSS_TCP_IPV4);
break;
case E1000_RXDADV_RSSTYPE_IPV4:
M_HASHTYPE_SET(rxr->fmp, M_HASHTYPE_RSS_IPV4);
break;
case E1000_RXDADV_RSSTYPE_IPV6_TCP:
M_HASHTYPE_SET(rxr->fmp, M_HASHTYPE_RSS_TCP_IPV6);
break;
case E1000_RXDADV_RSSTYPE_IPV6_EX:
M_HASHTYPE_SET(rxr->fmp, M_HASHTYPE_RSS_IPV6_EX);
break;
case E1000_RXDADV_RSSTYPE_IPV6:
M_HASHTYPE_SET(rxr->fmp, M_HASHTYPE_RSS_IPV6);
break;
case E1000_RXDADV_RSSTYPE_IPV6_TCP_EX:
M_HASHTYPE_SET(rxr->fmp, M_HASHTYPE_RSS_TCP_IPV6_EX);
break;
/* XXX no UDP support in RSS just yet */
#ifdef notyet
case E1000_RXDADV_RSSTYPE_IPV4_UDP:
case E1000_RXDADV_RSSTYPE_IPV6_UDP:
case E1000_RXDADV_RSSTYPE_IPV6_UDP_EX:
#endif
default:
/* XXX fallthrough */
M_HASHTYPE_SET(rxr->fmp, M_HASHTYPE_NONE);
}
#elif !defined(IGB_LEGACY_TX)
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
rxr->fmp->m_pkthdr.flowid = que->msix;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
rxr->fmp->m_flags |= M_FLOWID;
#endif
sendmp = rxr->fmp;
/* Make sure to set M_PKTHDR. */
sendmp->m_flags |= M_PKTHDR;
rxr->fmp = NULL;
rxr->lmp = NULL;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
next_desc:
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
/* Advance our pointers to the next descriptor. */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (++i == adapter->num_rx_desc)
i = 0;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/*
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
** Send to the stack or LRO
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
*/
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
if (sendmp != NULL) {
rxr->next_to_check = i;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
igb_rx_input(rxr, ifp, sendmp, ptype);
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
i = rxr->next_to_check;
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
rxdone++;
Add a couple fixes from Michael Tuexen. Remove unneeded rxtx handler, make que handler generic. Do not allocate header mbufs in rx ring if not doing hdr split. Release the lock in rxeof call to stack. MFC for 8.1 asap
2010-06-11 20:54:27 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Every 8 descriptors we go to refresh mbufs */
if (processed == 8) {
igb_refresh_mbufs(rxr, i);
processed = 0;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Catch any remainders */
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
if (igb_rx_unrefreshed(rxr))
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
igb_refresh_mbufs(rxr, i);
Resync with Intel versions of both the em and igb drivers. These add new hardware support, most importantly the pch (i5 chipset) in the em driver. Also, both drivers now have the simplified (and I hope improved) watchdog code. The igb driver uses the new RX cleanup that I first implemented in ixgbe. em - version 6.9.24 igb - version 1.8.4
2009-12-08 01:07:44 +00:00
rxr->next_to_check = i;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
/*
Update to igb driver: - changes in support of the VLAN filter fix to 126850 - removal of a bunch of legacy code that was cruft, if not possibly harmful. - removal of POLLING from this driver, with multiqueue and MSIX it just makes no sense here. - Fix an LRO bug that I've been working on internally, intermittent panics under stress, the problem was releasing the RX ring lock before the LRO flushing. - Following the above fix I now enable LRO by default - For performance reasons increase the default number of RX queues to 4. - Add AIM - "Adaptive Interrupt Moderation", a fancy way of saying that the EITR value is dynamically changed based on the size of packets in the last interrupt interval. - Much goodness to try, enjoy!!
2008-08-28 22:28:28 +00:00
* Flush any outstanding LRO work
*/
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
SLIST_REMOVE_HEAD(&lro->lro_active, next);
tcp_lro_flush(lro, queued);
}
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
if (done != NULL)
Make the polling interface in igb able to handle multiqueue, and correct the rxdone handling. Update the polling man page to include igb as well. Thanks to Mark Johnston for these changes.
2012-08-06 22:43:49 +00:00
*done += rxdone;
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
IGB_RX_UNLOCK(rxr);
return ((staterr & E1000_RXD_STAT_DD) ? TRUE : FALSE);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/*********************************************************************
*
* Verify that the hardware indicated that the checksum is valid.
* Inform the stack about the status of checksum so that stack
* doesn't spend time verifying the checksum.
*
*********************************************************************/
static void
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
igb_rx_checksum(u32 staterr, struct mbuf *mp, u32 ptype)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
u16 status = (u16)staterr;
u8 errors = (u8) (staterr >> 24);
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
int sctp;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Ignore Checksum bit is set */
if (status & E1000_RXD_STAT_IXSM) {
mp->m_pkthdr.csum_flags = 0;
return;
}
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if ((ptype & E1000_RXDADV_PKTTYPE_ETQF) == 0 &&
(ptype & E1000_RXDADV_PKTTYPE_SCTP) != 0)
sctp = 1;
else
sctp = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (status & E1000_RXD_STAT_IPCS) {
/* Did it pass? */
if (!(errors & E1000_RXD_ERR_IPE)) {
/* IP Checksum Good */
mp->m_pkthdr.csum_flags = CSUM_IP_CHECKED;
mp->m_pkthdr.csum_flags |= CSUM_IP_VALID;
} else
mp->m_pkthdr.csum_flags = 0;
}
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
if (status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)) {
Restructure the mbuf pkthdr to make it fit for upcoming capabilities and features. The changes in particular are: o Remove rarely used "header" pointer and replace it with a 64bit protocol/ layer specific union PH_loc for local use. Protocols can flexibly overlay their own 8 to 64 bit fields to store information while the packet is worked on. o Mechanically convert IP reassembly, IGMP/MLD and ATM to use pkthdr.PH_loc instead of pkthdr.header. o Extend csum_flags to 64bits to allow for additional future offload information to be carried (e.g. iSCSI, IPsec offload, and others). o Move the RSS hash type enumerator from abusing m_flags to its own 8bit rsstype field. Adjust accessor macros. o Add cosqos field to store Class of Service / Quality of Service information with the packet. It is not yet supported in any drivers but allows us to get on par with Cisco/Juniper in routing applications (plus MPLS QoS) with a modernized ALTQ. o Add four 8 bit fields l[2-5]hlen to store the relative header offsets from the start of the packet. This is important for various offload capabilities and to relieve the drivers from having to parse the packet and protocol headers to find out location of checksums and other information. Header parsing in drivers is a lot of copy-paste and unhandled corner cases which we want to avoid. o Add another flexible 64bit union to map various additional persistent packet information, like ether_vtag, tso_segsz and csum fields. Depending on the csum_flags settings some fields may have different usage making it very flexible and adaptable to future capabilities. o Restructure the CSUM flags to better signify their outbound (down the stack) and inbound (up the stack) use. The CSUM flags used to be a bit chaotic and rather poorly documented leading to incorrect use in many places. Bring clarity into their use through better naming. Compatibility mappings are provided to preserve the API. The drivers can be corrected one by one and MFC'd without issue. o The size of pkthdr stays the same at 48/56bytes (32/64bit architectures). Sponsored by: The FreeBSD Foundation
2013-08-24 19:51:18 +00:00
u64 type = (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
#if __FreeBSD_version >= 800000
if (sctp) /* reassign */
type = CSUM_SCTP_VALID;
#endif
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Did it pass? */
if (!(errors & E1000_RXD_ERR_TCPE)) {
Thanks to Michael Tuexen and Randall Scott for providing a few important bug fixes found while they were doing SCTP development, and that I somehow lost during the scramble. Thanks guys!!
2009-04-14 17:14:35 +00:00
mp->m_pkthdr.csum_flags |= type;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (sctp == 0)
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
mp->m_pkthdr.csum_data = htons(0xffff);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
}
return;
}
/*
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
* This routine is run via an vlan
* config EVENT
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
*/
static void
Improvement on the last change, this gives a precise way to tell the one and only interface that a vlan event is for. Thanks to John Baldwin for the patch. Approved by: re
2009-07-24 21:35:52 +00:00
igb_register_vlan(void *arg, struct ifnet *ifp, u16 vtag)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
struct adapter *adapter = ifp->if_softc;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
u32 index, bit;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (ifp->if_softc != arg) /* Not our event */
This delta fixes two bugs: - When a vlan event occurs a check was not made that the event was actually for the interface, thus resulting in a panic. All three drivers have this vulnerability. Add a check for this condition. - Secondly, there was a duplicate buf_ring free in the em driver resulting in a panic on unload. Remove. Approved by: re
2009-07-24 16:57:49 +00:00
return;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if ((vtag == 0) || (vtag > 4095)) /* Invalid */
return;
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
IGB_CORE_LOCK(adapter);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
index = (vtag >> 5) & 0x7F;
bit = vtag & 0x1F;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
adapter->shadow_vfta[index] |= (1 << bit);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
++adapter->num_vlans;
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
/* Change hw filter setting */
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
igb_setup_vlan_hw_support(adapter);
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
IGB_CORE_UNLOCK(adapter);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
/*
* This routine is run via an vlan
* unconfig EVENT
*/
static void
Improvement on the last change, this gives a precise way to tell the one and only interface that a vlan event is for. Thanks to John Baldwin for the patch. Approved by: re
2009-07-24 21:35:52 +00:00
igb_unregister_vlan(void *arg, struct ifnet *ifp, u16 vtag)
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
{
struct adapter *adapter = ifp->if_softc;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
u32 index, bit;
Update the 1G drivers, shared code sync with Intel, igb now has a queue notion that has a single interrupt with an RX/TX pair, this will reduce the total interrupts seen on a system. Both em and igb have a new watchdog method. igb has fixes from Pyun Yong-Hyeon that have improved stability, thank you :) I wish to MFC this for 7.3 asap, please test if able.
2010-01-26 22:32:22 +00:00
if (ifp->if_softc != arg)
This delta fixes two bugs: - When a vlan event occurs a check was not made that the event was actually for the interface, thus resulting in a panic. All three drivers have this vulnerability. Add a check for this condition. - Secondly, there was a duplicate buf_ring free in the em driver resulting in a panic on unload. Remove. Approved by: re
2009-07-24 16:57:49 +00:00
return;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
if ((vtag == 0) || (vtag > 4095)) /* Invalid */
return;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
IGB_CORE_LOCK(adapter);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
index = (vtag >> 5) & 0x7F;
bit = vtag & 0x1F;
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
adapter->shadow_vfta[index] &= ~(1 << bit);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
--adapter->num_vlans;
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
/* Change hw filter setting */
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
igb_setup_vlan_hw_support(adapter);
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
IGB_CORE_UNLOCK(adapter);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
}
static void
igb_setup_vlan_hw_support(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
struct ifnet *ifp = adapter->ifp;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
u32 reg;
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->vf_ifp) {
e1000_rlpml_set_vf(hw,
adapter->max_frame_size + VLAN_TAG_SIZE);
return;
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
reg = E1000_READ_REG(hw, E1000_CTRL);
reg |= E1000_CTRL_VME;
E1000_WRITE_REG(hw, E1000_CTRL, reg);
/* Enable the Filter Table */
if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) {
reg = E1000_READ_REG(hw, E1000_RCTL);
reg &= ~E1000_RCTL_CFIEN;
reg |= E1000_RCTL_VFE;
E1000_WRITE_REG(hw, E1000_RCTL, reg);
}
/* Update the frame size */
E1000_WRITE_REG(&adapter->hw, E1000_RLPML,
adapter->max_frame_size + VLAN_TAG_SIZE);
/* Don't bother with table if no vlans */
if ((adapter->num_vlans == 0) ||
((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0))
return;
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
/*
** A soft reset zero's out the VFTA, so
** we need to repopulate it now.
*/
for (int i = 0; i < IGB_VFTA_SIZE; i++)
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
if (adapter->shadow_vfta[i] != 0) {
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->vf_ifp)
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
e1000_vfta_set_vf(hw,
adapter->shadow_vfta[i], TRUE);
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
else
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
e1000_write_vfta(hw,
i, adapter->shadow_vfta[i]);
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
}
Merge of the source for igb and em into dev/e1000, this proved to be necessary to make the static drivers work in EITHER/OR or BOTH configurations. Modules will still build in sys/modules/igb or em as before. This also updates the igb driver for support for the 82576 adapter, adds shared code fixes, and etc.... MFC after: ASAP
2008-07-30 21:56:53 +00:00
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static void
igb_enable_intr(struct adapter *adapter)
{
/* With RSS set up what to auto clear */
if (adapter->msix_mem) {
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
u32 mask = (adapter->que_mask | adapter->link_mask);
E1000_WRITE_REG(&adapter->hw, E1000_EIAC, mask);
E1000_WRITE_REG(&adapter->hw, E1000_EIAM, mask);
E1000_WRITE_REG(&adapter->hw, E1000_EIMS, mask);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_IMS,
E1000_IMS_LSC);
} else {
E1000_WRITE_REG(&adapter->hw, E1000_IMS,
IMS_ENABLE_MASK);
}
E1000_WRITE_FLUSH(&adapter->hw);
return;
}
static void
igb_disable_intr(struct adapter *adapter)
{
if (adapter->msix_mem) {
E1000_WRITE_REG(&adapter->hw, E1000_EIMC, ~0);
E1000_WRITE_REG(&adapter->hw, E1000_EIAC, 0);
}
This delta has a few important items: PR 122839 is fixed in both em and in igb Second, the issue on building modules since the static kernel build changes is now resolved. I was not able to get the fancier directory hierarchy working, but this works, both em and igb build as modules now. Third, there is now support in em for two new NICs, Hartwell (or 82574) is a low cost PCIE dual port adapter that has MSIX, for this release it uses 3 vectors only, RX, TX, and LINK. In the next release I will add a second TX and RX queue. Also, there is support here for ICH10, the followon to ICH9. Both of these are early releases, general availability will follow soon. Fourth: On Hartwell and ICH10 we now have IEEE 1588 PTP support, I have implemented this in a provisional way so that early adopters may try and comment on the functionality. The IOCTL structure may change. This feature is off by default, you need to edit the Makefile and add the EM_TIMESYNC define to get the code. Enjoy all!!
2008-04-25 21:19:41 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_IMC, ~0);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
E1000_WRITE_FLUSH(&adapter->hw);
return;
}
/*
* Bit of a misnomer, what this really means is
* to enable OS management of the system... aka
* to disable special hardware management features
*/
static void
igb_init_manageability(struct adapter *adapter)
{
if (adapter->has_manage) {
int manc2h = E1000_READ_REG(&adapter->hw, E1000_MANC2H);
int manc = E1000_READ_REG(&adapter->hw, E1000_MANC);
/* disable hardware interception of ARP */
manc &= ~(E1000_MANC_ARP_EN);
/* enable receiving management packets to the host */
manc |= E1000_MANC_EN_MNG2HOST;
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
manc2h |= 1 << 5; /* Mng Port 623 */
manc2h |= 1 << 6; /* Mng Port 664 */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_MANC2H, manc2h);
E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc);
}
}
/*
* Give control back to hardware management
* controller if there is one.
*/
static void
igb_release_manageability(struct adapter *adapter)
{
if (adapter->has_manage) {
int manc = E1000_READ_REG(&adapter->hw, E1000_MANC);
/* re-enable hardware interception of ARP */
manc |= E1000_MANC_ARP_EN;
manc &= ~E1000_MANC_EN_MNG2HOST;
E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc);
}
}
/*
* igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
* For ASF and Pass Through versions of f/w this means that
* the driver is loaded.
*
*/
static void
igb_get_hw_control(struct adapter *adapter)
{
u32 ctrl_ext;
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->vf_ifp)
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
return;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Let firmware know the driver has taken over */
ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT,
ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
}
/*
* igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
* For ASF and Pass Through versions of f/w this means that the
* driver is no longer loaded.
*
*/
static void
igb_release_hw_control(struct adapter *adapter)
{
u32 ctrl_ext;
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->vf_ifp)
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
return;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Let firmware taken over control of h/w */
ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT,
ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
}
static int
igb_is_valid_ether_addr(uint8_t *addr)
{
char zero_addr[6] = { 0, 0, 0, 0, 0, 0 };
if ((addr[0] & 1) || (!bcmp(addr, zero_addr, ETHER_ADDR_LEN))) {
return (FALSE);
}
return (TRUE);
}
/*
* Enable PCI Wake On Lan capability
*/
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
static void
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
igb_enable_wakeup(device_t dev)
{
u16 cap, status;
u8 id;
/* First find the capabilities pointer*/
cap = pci_read_config(dev, PCIR_CAP_PTR, 2);
/* Read the PM Capabilities */
id = pci_read_config(dev, cap, 1);
if (id != PCIY_PMG) /* Something wrong */
return;
/* OK, we have the power capabilities, so
now get the status register */
cap += PCIR_POWER_STATUS;
status = pci_read_config(dev, cap, 2);
status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
pci_write_config(dev, cap, status, 2);
return;
}
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
static void
igb_led_func(void *arg, int onoff)
{
struct adapter *adapter = arg;
IGB_CORE_LOCK(adapter);
if (onoff) {
e1000_setup_led(&adapter->hw);
e1000_led_on(&adapter->hw);
} else {
e1000_led_off(&adapter->hw);
e1000_cleanup_led(&adapter->hw);
}
IGB_CORE_UNLOCK(adapter);
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
- Provide igb_get_counter() to return counters that are not collected, but taken from hardware. - Mechanically convert to if_inc_counter() the rest of counters.
2014-09-19 11:49:41 +00:00
static uint64_t
igb_get_counter(if_t ifp, ift_counter cnt)
{
struct adapter *adapter;
struct e1000_hw_stats *stats;
adapter = if_getsoftc(ifp);
stats = (struct e1000_hw_stats *)adapter->stats;
switch (cnt) {
case IFCOUNTER_IERRORS:
return (adapter->dropped_pkts + stats->rxerrc +
stats->crcerrs + stats->algnerrc +
stats->ruc + stats->roc + stats->mpc + stats->cexterr);
case IFCOUNTER_OERRORS:
return (stats->ecol + stats->latecol +
adapter->watchdog_events);
case IFCOUNTER_COLLISIONS:
return (stats->colc);
default:
return (if_get_counter_default(ifp, cnt));
}
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/**********************************************************************
*
* Update the board statistics counters.
*
**********************************************************************/
static void
igb_update_stats_counters(struct adapter *adapter)
{
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
struct e1000_hw *hw = &adapter->hw;
struct e1000_hw_stats *stats;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
/*
** The virtual function adapter has only a
** small controlled set of stats, do only
** those and return.
*/
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->vf_ifp) {
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
igb_update_vf_stats_counters(adapter);
return;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
stats = (struct e1000_hw_stats *)adapter->stats;
if(adapter->hw.phy.media_type == e1000_media_type_copper ||
(E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
stats->symerrs +=
E1000_READ_REG(hw,E1000_SYMERRS);
stats->sec += E1000_READ_REG(hw, E1000_SEC);
}
stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
stats->mpc += E1000_READ_REG(hw, E1000_MPC);
stats->scc += E1000_READ_REG(hw, E1000_SCC);
stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
stats->mcc += E1000_READ_REG(hw, E1000_MCC);
stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
stats->colc += E1000_READ_REG(hw, E1000_COLC);
stats->dc += E1000_READ_REG(hw, E1000_DC);
stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
Update code from Intel: - Sync shared code with Intel internal - New client chipset support added - em driver - fixes to 82574, limit queues to 1 but use MSIX - em driver - large changes in TX checksum offload and tso code, thanks to yongari. - some small changes for watchdog issues. - igb driver - local timer watchdog code was missing locking this and a couple other watchdog related fixes. - bug in rx discard found by Andrew Boyer, check for null pointer MFC: a week
2010-09-28 00:13:15 +00:00
/*
** For watchdog management we need to know if we have been
** paused during the last interval, so capture that here.
*/
adapter->pause_frames = E1000_READ_REG(&adapter->hw, E1000_XOFFRXC);
stats->xoffrxc += adapter->pause_frames;
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* For the 64-bit byte counters the low dword must be read first. */
/* Both registers clear on the read of the high dword */
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
stats->gorc += E1000_READ_REG(hw, E1000_GORCL) +
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
((u64)E1000_READ_REG(hw, E1000_GORCH) << 32);
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
stats->gotc += E1000_READ_REG(hw, E1000_GOTCL) +
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
((u64)E1000_READ_REG(hw, E1000_GOTCH) << 32);
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
stats->ruc += E1000_READ_REG(hw, E1000_RUC);
stats->rfc += E1000_READ_REG(hw, E1000_RFC);
stats->roc += E1000_READ_REG(hw, E1000_ROC);
stats->rjc += E1000_READ_REG(hw, E1000_RJC);
Various fixes to stats: - Read the counts of received, dropped, and transmitted management packets and add sysctl nodes for them. - Fix the total octets received/transmitted to read all 64 bits of the counters. - Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot. - Remove spurious spaces. Reviewed by: Eric Joyner @ Intel MFC after: 1 week
2014-10-10 16:36:25 +00:00
stats->mgprc += E1000_READ_REG(hw, E1000_MGTPRC);
stats->mgpdc += E1000_READ_REG(hw, E1000_MGTPDC);
stats->mgptc += E1000_READ_REG(hw, E1000_MGTPTC);
stats->tor += E1000_READ_REG(hw, E1000_TORL) +
((u64)E1000_READ_REG(hw, E1000_TORH) << 32);
stats->tot += E1000_READ_REG(hw, E1000_TOTL) +
((u64)E1000_READ_REG(hw, E1000_TOTH) << 32);
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
stats->tpr += E1000_READ_REG(hw, E1000_TPR);
stats->tpt += E1000_READ_REG(hw, E1000_TPT);
stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
/* Interrupt Counts */
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
stats->iac += E1000_READ_REG(hw, E1000_IAC);
stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
/* Host to Card Statistics */
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
stats->hgorc += (E1000_READ_REG(hw, E1000_HGORCL) +
((u64)E1000_READ_REG(hw, E1000_HGORCH) << 32));
stats->hgotc += (E1000_READ_REG(hw, E1000_HGOTCL) +
((u64)E1000_READ_REG(hw, E1000_HGOTCH) << 32));
stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Make sure that all the exposed counters and variables are actually being updated. Pointed out by: jfv
2010-06-24 21:17:58 +00:00
/* Driver specific counters */
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
adapter->device_control = E1000_READ_REG(hw, E1000_CTRL);
adapter->rx_control = E1000_READ_REG(hw, E1000_RCTL);
adapter->int_mask = E1000_READ_REG(hw, E1000_IMS);
adapter->eint_mask = E1000_READ_REG(hw, E1000_EIMS);
adapter->packet_buf_alloc_tx =
((E1000_READ_REG(hw, E1000_PBA) & 0xffff0000) >> 16);
adapter->packet_buf_alloc_rx =
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
(E1000_READ_REG(hw, E1000_PBA) & 0xffff);
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
}
/**********************************************************************
*
* Initialize the VF board statistics counters.
*
**********************************************************************/
static void
igb_vf_init_stats(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
struct e1000_vf_stats *stats;
Make sure that all the exposed counters and variables are actually being updated. Pointed out by: jfv
2010-06-24 21:17:58 +00:00
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
stats = (struct e1000_vf_stats *)adapter->stats;
Make sure not to access unallocated stats memory. Reviewed by: jfv
2010-08-27 23:50:13 +00:00
if (stats == NULL)
return;
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
stats->last_gprc = E1000_READ_REG(hw, E1000_VFGPRC);
stats->last_gorc = E1000_READ_REG(hw, E1000_VFGORC);
stats->last_gptc = E1000_READ_REG(hw, E1000_VFGPTC);
stats->last_gotc = E1000_READ_REG(hw, E1000_VFGOTC);
stats->last_mprc = E1000_READ_REG(hw, E1000_VFMPRC);
}
/**********************************************************************
*
* Update the VF board statistics counters.
*
**********************************************************************/
static void
igb_update_vf_stats_counters(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
struct e1000_vf_stats *stats;
if (adapter->link_speed == 0)
return;
stats = (struct e1000_vf_stats *)adapter->stats;
UPDATE_VF_REG(E1000_VFGPRC,
stats->last_gprc, stats->gprc);
UPDATE_VF_REG(E1000_VFGORC,
stats->last_gorc, stats->gorc);
UPDATE_VF_REG(E1000_VFGPTC,
stats->last_gptc, stats->gptc);
UPDATE_VF_REG(E1000_VFGOTC,
stats->last_gotc, stats->gotc);
UPDATE_VF_REG(E1000_VFMPRC,
stats->last_mprc, stats->mprc);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
/* Export a single 32-bit register via a read-only sysctl. */
static int
igb_sysctl_reg_handler(SYSCTL_HANDLER_ARGS)
Fix a bug in the statistics code for tracking the head and tail pointers of the tx and rx queues. We needed a SYSCTL_PROC to correctly get the values at run time. Submitted by: Andrew Boyer aboyer at averesystems.com MFC after: 1 week
2010-07-23 17:53:39 +00:00
{
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
struct adapter *adapter;
u_int val;
Fix a bug in the statistics code for tracking the head and tail pointers of the tx and rx queues. We needed a SYSCTL_PROC to correctly get the values at run time. Submitted by: Andrew Boyer aboyer at averesystems.com MFC after: 1 week
2010-07-23 17:53:39 +00:00
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
adapter = oidp->oid_arg1;
val = E1000_READ_REG(&adapter->hw, oidp->oid_arg2);
return (sysctl_handle_int(oidp, &val, 0, req));
Fix a bug in the statistics code for tracking the head and tail pointers of the tx and rx queues. We needed a SYSCTL_PROC to correctly get the values at run time. Submitted by: Andrew Boyer aboyer at averesystems.com MFC after: 1 week
2010-07-23 17:53:39 +00:00
}
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
/*
** Tuneable interrupt rate handler
*/
static int
igb_sysctl_interrupt_rate_handler(SYSCTL_HANDLER_ARGS)
{
struct igb_queue *que = ((struct igb_queue *)oidp->oid_arg1);
int error;
u32 reg, usec, rate;
reg = E1000_READ_REG(&que->adapter->hw, E1000_EITR(que->msix));
usec = ((reg & 0x7FFC) >> 2);
if (usec > 0)
rate = 1000000 / usec;
else
rate = 0;
error = sysctl_handle_int(oidp, &rate, 0, req);
if (error || !req->newptr)
return error;
return 0;
}
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
/*
* Add sysctl variables, one per statistic, to the system.
*/
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
static void
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
igb_add_hw_stats(struct adapter *adapter)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
device_t dev = adapter->dev;
This delta has a few important items: PR 122839 is fixed in both em and in igb Second, the issue on building modules since the static kernel build changes is now resolved. I was not able to get the fancier directory hierarchy working, but this works, both em and igb build as modules now. Third, there is now support in em for two new NICs, Hartwell (or 82574) is a low cost PCIE dual port adapter that has MSIX, for this release it uses 3 vectors only, RX, TX, and LINK. In the next release I will add a second TX and RX queue. Also, there is support here for ICH10, the followon to ICH9. Both of these are early releases, general availability will follow soon. Fourth: On Hartwell and ICH10 we now have IEEE 1588 PTP support, I have implemented this in a provisional way so that early adopters may try and comment on the functionality. The IOCTL structure may change. This feature is off by default, you need to edit the Makefile and add the EM_TIMESYNC define to get the code. Enjoy all!!
2008-04-25 21:19:41 +00:00
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
struct tx_ring *txr = adapter->tx_rings;
struct rx_ring *rxr = adapter->rx_rings;
This delta has a few important items: PR 122839 is fixed in both em and in igb Second, the issue on building modules since the static kernel build changes is now resolved. I was not able to get the fancier directory hierarchy working, but this works, both em and igb build as modules now. Third, there is now support in em for two new NICs, Hartwell (or 82574) is a low cost PCIE dual port adapter that has MSIX, for this release it uses 3 vectors only, RX, TX, and LINK. In the next release I will add a second TX and RX queue. Also, there is support here for ICH10, the followon to ICH9. Both of these are early releases, general availability will follow soon. Fourth: On Hartwell and ICH10 we now have IEEE 1588 PTP support, I have implemented this in a provisional way so that early adopters may try and comment on the functionality. The IOCTL structure may change. This feature is off by default, you need to edit the Makefile and add the EM_TIMESYNC define to get the code. Enjoy all!!
2008-04-25 21:19:41 +00:00
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(dev);
struct sysctl_oid *tree = device_get_sysctl_tree(dev);
struct sysctl_oid_list *child = SYSCTL_CHILDREN(tree);
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
struct e1000_hw_stats *stats = adapter->stats;
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
struct sysctl_oid *stat_node, *queue_node, *int_node, *host_node;
struct sysctl_oid_list *stat_list, *queue_list, *int_list, *host_list;
#define QUEUE_NAME_LEN 32
char namebuf[QUEUE_NAME_LEN];
/* Driver Statistics */
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "link_irq",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &adapter->link_irq, 0,
"Link MSIX IRQ Handled");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "dropped",
CTLFLAG_RD, &adapter->dropped_pkts,
"Driver dropped packets");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_dma_fail",
CTLFLAG_RD, &adapter->no_tx_dma_setup,
"Driver tx dma failure in xmit");
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_overruns",
CTLFLAG_RD, &adapter->rx_overruns,
"RX overruns");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "watchdog_timeouts",
CTLFLAG_RD, &adapter->watchdog_events,
"Watchdog timeouts");
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "device_control",
CTLFLAG_RD, &adapter->device_control,
"Device Control Register");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_control",
CTLFLAG_RD, &adapter->rx_control,
"Receiver Control Register");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "interrupt_mask",
CTLFLAG_RD, &adapter->int_mask,
"Interrupt Mask");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "extended_int_mask",
CTLFLAG_RD, &adapter->eint_mask,
"Extended Interrupt Mask");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_buf_alloc",
CTLFLAG_RD, &adapter->packet_buf_alloc_tx,
"Transmit Buffer Packet Allocation");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_buf_alloc",
CTLFLAG_RD, &adapter->packet_buf_alloc_rx,
"Receive Buffer Packet Allocation");
SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "fc_high_water",
CTLFLAG_RD, &adapter->hw.fc.high_water, 0,
"Flow Control High Watermark");
SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "fc_low_water",
CTLFLAG_RD, &adapter->hw.fc.low_water, 0,
"Flow Control Low Watermark");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
for (int i = 0; i < adapter->num_queues; i++, rxr++, txr++) {
struct lro_ctrl *lro = &rxr->lro;
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
snprintf(namebuf, QUEUE_NAME_LEN, "queue%d", i);
queue_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
CTLFLAG_RD, NULL, "Queue Name");
queue_list = SYSCTL_CHILDREN(queue_node);
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "interrupt_rate",
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
CTLFLAG_RD, &adapter->queues[i],
- New 82580 devices supported - Fixes from John Baldwin: vlan shadow tables made per/interface, make vlan hw setup only happen when capability enabled, and finally, make a tuneable interrupt rate. Thanks John! - Tweaked watchdog handling to avoid any false positives, now detection is in the TX clean path, with only the final check and init happening in the local timer. - limit queues to 8 for all devices, with 82576 or 82580 on larger machines it can get greater than this, and it seems mostly a resource waste to do so. Even 8 might be high but it can be manually reduced. - use 2k, 4k and now 9k clusters based on the MTU size. - rework the igb_refresh_mbuf() code, its important to make sure the descriptor is rewritten even when reusing mbufs since writeback clobbers things. MFC: in a few days, this delta needs to get to 8.2
2010-11-23 22:12:02 +00:00
sizeof(&adapter->queues[i]),
igb_sysctl_interrupt_rate_handler,
"IU", "Interrupt Rate");
Fix a bug in the statistics code for tracking the head and tail pointers of the tx and rx queues. We needed a SYSCTL_PROC to correctly get the values at run time. Submitted by: Andrew Boyer aboyer at averesystems.com MFC after: 1 week
2010-07-23 17:53:39 +00:00
SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_head",
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
CTLFLAG_RD, adapter, E1000_TDH(txr->me),
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
igb_sysctl_reg_handler, "IU",
Fix a bug in the statistics code for tracking the head and tail pointers of the tx and rx queues. We needed a SYSCTL_PROC to correctly get the values at run time. Submitted by: Andrew Boyer aboyer at averesystems.com MFC after: 1 week
2010-07-23 17:53:39 +00:00
"Transmit Descriptor Head");
SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_tail",
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
CTLFLAG_RD, adapter, E1000_TDT(txr->me),
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
igb_sysctl_reg_handler, "IU",
Fix a bug in the statistics code for tracking the head and tail pointers of the tx and rx queues. We needed a SYSCTL_PROC to correctly get the values at run time. Submitted by: Andrew Boyer aboyer at averesystems.com MFC after: 1 week
2010-07-23 17:53:39 +00:00
"Transmit Descriptor Tail");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, queue_list, OID_AUTO, "no_desc_avail",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &txr->no_desc_avail,
The description is a bit misleading. Trying to make it more obvious. Phabric: https://phabric.freebsd.org/D435 Reviewed by: gnn
2014-07-18 16:25:35 +00:00
"Queue Descriptors Unavailable");
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "tx_packets",
CTLFLAG_RD, &txr->total_packets,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Queue Packets Transmitted");
Fix a bug in the statistics code for tracking the head and tail pointers of the tx and rx queues. We needed a SYSCTL_PROC to correctly get the values at run time. Submitted by: Andrew Boyer aboyer at averesystems.com MFC after: 1 week
2010-07-23 17:53:39 +00:00
SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "rxd_head",
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
CTLFLAG_RD, adapter, E1000_RDH(rxr->me),
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
igb_sysctl_reg_handler, "IU",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Receive Descriptor Head");
Fix a bug in the statistics code for tracking the head and tail pointers of the tx and rx queues. We needed a SYSCTL_PROC to correctly get the values at run time. Submitted by: Andrew Boyer aboyer at averesystems.com MFC after: 1 week
2010-07-23 17:53:39 +00:00
SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "rxd_tail",
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
CTLFLAG_RD, adapter, E1000_RDT(rxr->me),
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
igb_sysctl_reg_handler, "IU",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Receive Descriptor Tail");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, queue_list, OID_AUTO, "rx_packets",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &rxr->rx_packets,
"Queue Packets Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, queue_list, OID_AUTO, "rx_bytes",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &rxr->rx_bytes,
"Queue Bytes Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_UINT(ctx, queue_list, OID_AUTO, "lro_queued",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &lro->lro_queued, 0,
"LRO Queued");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_UINT(ctx, queue_list, OID_AUTO, "lro_flushed",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &lro->lro_flushed, 0,
"LRO Flushed");
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
/* MAC stats get their own sub node */
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
stat_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "mac_stats",
CTLFLAG_RD, NULL, "MAC Statistics");
stat_list = SYSCTL_CHILDREN(stat_node);
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
/*
** VF adapter has a very limited set of stats
** since its not managing the metal, so to speak.
*/
Add support for the new I350 family of 1G interfaces. - this also includes virtualization support on these devices Correct some vlan issues we were seeing in test, jumbo frames on vlans did not work correctly, this was all due to confused logic around HW filters, the new code should now work for all uses. Important fix: when mbuf resources are depeleted, it was possible to completely empty the RX ring, and then the RX engine would stall forever. This is fixed by a flag being set whenever the refresh code fails due to an mbuf shortage, also the local timer now makes sure that all queues get an interrupt when it runs, the interrupt code will then always call rxeof, and in that routine the first thing done is now to check the refresh flag and call refresh_mbufs. This has been verified to fix this type 'hang'. Similar code will follow in the other drivers. Finally, sync up shared code for the I350 support. Thanks to everyone that has been reporting issues, and helping in the debug/test process!!
2011-02-11 01:00:26 +00:00
if (adapter->vf_ifp) {
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_pkts_recvd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->gprc,
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
"Good Packets Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_pkts_txd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->gptc,
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
"Good Packets Transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_octets_recvd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->gorc,
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
"Good Octets Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_octets_txd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->gotc,
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
"Good Octets Transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_recvd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->mprc,
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
"Multicast Packets Received");
return;
}
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "excess_coll",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &stats->ecol,
"Excessive collisions");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "single_coll",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &stats->scc,
"Single collisions");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "multiple_coll",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &stats->mcc,
"Multiple collisions");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "late_coll",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &stats->latecol,
"Late collisions");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "collision_count",
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
CTLFLAG_RD, &stats->colc,
"Collision Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "symbol_errors",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->symerrs,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Symbol Errors");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "sequence_errors",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->sec,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Sequence Errors");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "defer_count",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->dc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Defer Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "missed_packets",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->mpc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Missed Packets");
Various fixes to stats: - Read the counts of received, dropped, and transmitted management packets and add sysctl nodes for them. - Fix the total octets received/transmitted to read all 64 bits of the counters. - Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot. - Remove spurious spaces. Reviewed by: Eric Joyner @ Intel MFC after: 1 week
2014-10-10 16:36:25 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_length_errors",
CTLFLAG_RD, &stats->rlec,
"Receive Length Errors");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_no_buff",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->rnbc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Receive No Buffers");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_undersize",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ruc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Receive Undersize");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_fragmented",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->rfc,
Various fixes to stats: - Read the counts of received, dropped, and transmitted management packets and add sysctl nodes for them. - Fix the total octets received/transmitted to read all 64 bits of the counters. - Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot. - Remove spurious spaces. Reviewed by: Eric Joyner @ Intel MFC after: 1 week
2014-10-10 16:36:25 +00:00
"Fragmented Packets Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_oversize",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->roc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Oversized Packets Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_jabber",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->rjc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Recevied Jabber");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_errs",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->rxerrc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Receive Errors");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "crc_errs",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->crcerrs,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"CRC errors");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "alignment_errs",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->algnerrc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Alignment Errors");
Various fixes to stats: - Read the counts of received, dropped, and transmitted management packets and add sysctl nodes for them. - Fix the total octets received/transmitted to read all 64 bits of the counters. - Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot. - Remove spurious spaces. Reviewed by: Eric Joyner @ Intel MFC after: 1 week
2014-10-10 16:36:25 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_no_crs",
CTLFLAG_RD, &stats->tncrs,
"Transmit with No CRS");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* On 82575 these are collision counts */
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "coll_ext_errs",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->cexterr,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Collision/Carrier extension errors");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "xon_recvd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->xonrxc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"XON Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "xon_txd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->xontxc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"XON Transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "xoff_recvd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->xoffrxc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"XOFF Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "xoff_txd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->xofftxc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"XOFF Transmitted");
Various fixes to stats: - Read the counts of received, dropped, and transmitted management packets and add sysctl nodes for them. - Fix the total octets received/transmitted to read all 64 bits of the counters. - Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot. - Remove spurious spaces. Reviewed by: Eric Joyner @ Intel MFC after: 1 week
2014-10-10 16:36:25 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "unsupported_fc_recvd",
CTLFLAG_RD, &stats->fcruc,
"Unsupported Flow Control Received");
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "mgmt_pkts_recvd",
CTLFLAG_RD, &stats->mgprc,
"Management Packets Received");
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "mgmt_pkts_drop",
CTLFLAG_RD, &stats->mgpdc,
"Management Packets Dropped");
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "mgmt_pkts_txd",
CTLFLAG_RD, &stats->mgptc,
"Management Packets Transmitted");
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
/* Packet Reception Stats */
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "total_pkts_recvd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->tpr,
Various fixes to stats: - Read the counts of received, dropped, and transmitted management packets and add sysctl nodes for them. - Fix the total octets received/transmitted to read all 64 bits of the counters. - Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot. - Remove spurious spaces. Reviewed by: Eric Joyner @ Intel MFC after: 1 week
2014-10-10 16:36:25 +00:00
"Total Packets Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_pkts_recvd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->gprc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Good Packets Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_recvd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->bprc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Broadcast Packets Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_recvd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->mprc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Multicast Packets Received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_64",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->prc64,
Various fixes to stats: - Read the counts of received, dropped, and transmitted management packets and add sysctl nodes for them. - Fix the total octets received/transmitted to read all 64 bits of the counters. - Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot. - Remove spurious spaces. Reviewed by: Eric Joyner @ Intel MFC after: 1 week
2014-10-10 16:36:25 +00:00
"64 byte frames received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_65_127",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->prc127,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"65-127 byte frames received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_128_255",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->prc255,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"128-255 byte frames received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_256_511",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->prc511,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"256-511 byte frames received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_512_1023",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->prc1023,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"512-1023 byte frames received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_1024_1522",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->prc1522,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"1023-1522 byte frames received");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_octets_recvd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->gorc,
Various fixes to stats: - Read the counts of received, dropped, and transmitted management packets and add sysctl nodes for them. - Fix the total octets received/transmitted to read all 64 bits of the counters. - Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot. - Remove spurious spaces. Reviewed by: Eric Joyner @ Intel MFC after: 1 week
2014-10-10 16:36:25 +00:00
"Good Octets Received");
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "total_octets_recvd",
CTLFLAG_RD, &stats->tor,
"Total Octets Received");
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
/* Packet Transmission Stats */
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_octets_txd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->gotc,
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
"Good Octets Transmitted");
Various fixes to stats: - Read the counts of received, dropped, and transmitted management packets and add sysctl nodes for them. - Fix the total octets received/transmitted to read all 64 bits of the counters. - Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot. - Remove spurious spaces. Reviewed by: Eric Joyner @ Intel MFC after: 1 week
2014-10-10 16:36:25 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "total_octets_txd",
CTLFLAG_RD, &stats->tot,
"Total Octets Transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "total_pkts_txd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->tpt,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Total Packets Transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_pkts_txd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->gptc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Good Packets Transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_txd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->bptc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Broadcast Packets Transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_txd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->mptc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Multicast Packets Transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_64",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ptc64,
Various fixes to stats: - Read the counts of received, dropped, and transmitted management packets and add sysctl nodes for them. - Fix the total octets received/transmitted to read all 64 bits of the counters. - Add missing sysctl nodes for rlec, tncrs, fcruc, tor, and tot. - Remove spurious spaces. Reviewed by: Eric Joyner @ Intel MFC after: 1 week
2014-10-10 16:36:25 +00:00
"64 byte frames transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_65_127",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ptc127,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"65-127 byte frames transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_128_255",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ptc255,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"128-255 byte frames transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_256_511",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ptc511,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"256-511 byte frames transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_512_1023",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ptc1023,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"512-1023 byte frames transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_1024_1522",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ptc1522,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"1024-1522 byte frames transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tso_txd",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->tsctc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"TSO Contexts Transmitted");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tso_ctx_fail",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->tsctfc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"TSO Contexts Failed");
/* Interrupt Stats */
int_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "interrupts",
CTLFLAG_RD, NULL, "Interrupt Statistics");
int_list = SYSCTL_CHILDREN(int_node);
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "asserts",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->iac,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Interrupt Assertion Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "rx_pkt_timer",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->icrxptc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Interrupt Cause Rx Pkt Timer Expire Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "rx_abs_timer",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->icrxatc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Interrupt Cause Rx Abs Timer Expire Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "tx_pkt_timer",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ictxptc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Interrupt Cause Tx Pkt Timer Expire Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "tx_abs_timer",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ictxatc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Interrupt Cause Tx Abs Timer Expire Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "tx_queue_empty",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ictxqec,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Interrupt Cause Tx Queue Empty Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "tx_queue_min_thresh",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->ictxqmtc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Interrupt Cause Tx Queue Min Thresh Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "rx_desc_min_thresh",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->icrxdmtc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Interrupt Cause Rx Desc Min Thresh Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "rx_overrun",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->icrxoc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Interrupt Cause Receiver Overrun Count");
/* Host to Card Stats */
host_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "host",
CTLFLAG_RD, NULL,
"Host to Card Statistics");
host_list = SYSCTL_CHILDREN(host_node);
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "breaker_tx_pkt",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->cbtmpc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Circuit Breaker Tx Packet Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "host_tx_pkt_discard",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->htdpmc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Host Transmit Discarded Packets");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "rx_pkt",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->rpthc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Rx Packets To Host");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "breaker_rx_pkts",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->cbrmpc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Circuit Breaker Rx Packet Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "breaker_rx_pkt_drop",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->cbrdpc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Circuit Breaker Rx Dropped Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "tx_good_pkt",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->hgptc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Host Good Packets Tx Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "breaker_tx_pkt_drop",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->htcbdpc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Host Tx Circuit Breaker Dropped Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "rx_good_bytes",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->hgorc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Host Good Octets Received Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "tx_good_bytes",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->hgotc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Host Good Octets Transmit Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "length_errors",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->lenerrs,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Length Errors");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "serdes_violation_pkt",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->scvpc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"SerDes/SGMII Code Violation Pkt Count");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
SYSCTL_ADD_QUAD(ctx, host_list, OID_AUTO, "header_redir_missed",
OK, I was a bit sleep this morning and checked in the core changes but left out the shared code, lol. Well, and a couple fixes to the core... hopefully this will all be complete now. Happy happy joy joy :)
2010-06-30 21:05:51 +00:00
CTLFLAG_RD, &stats->hrmpc,
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
"Header Redirection Missed Packet Count");
SR-IOV support added to igb What this provides is support for the 'virtual function' interface that a FreeBSD VM may be assigned from a host like KVM on Linux, or newer versions of Xen with such support. When the guest is set up with the capability, a special limited function 82576 PCI device is present in its virtual PCI space, so with this driver installed in the guest that device will be detected and function nearly like the bare metal, as it were. The interface is only allowed a single queue in this configuration however initial performance tests have looked very good. Enjoy!!
2010-06-30 17:26:47 +00:00
}
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/**********************************************************************
*
* This routine provides a way to dump out the adapter eeprom,
* often a useful debug/service tool. This only dumps the first
* 32 words, stuff that matters is in that extent.
*
**********************************************************************/
static int
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
igb_sysctl_nvm_info(SYSCTL_HANDLER_ARGS)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
struct adapter *adapter;
int error;
int result;
result = -1;
error = sysctl_handle_int(oidp, &result, 0, req);
if (error || !req->newptr)
return (error);
/*
* This value will cause a hex dump of the
* first 32 16-bit words of the EEPROM to
* the screen.
*/
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
if (result == 1) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
adapter = (struct adapter *)arg1;
igb_print_nvm_info(adapter);
}
return (error);
}
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
static void
igb_print_nvm_info(struct adapter *adapter)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
{
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
u16 eeprom_data;
int i, j, row = 0;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
/* Its a bit crude, but it gets the job done */
printf("\nInterface EEPROM Dump:\n");
printf("Offset\n0x0000 ");
for (i = 0, j = 0; i < 32; i++, j++) {
if (j == 8) { /* Make the offset block */
j = 0; ++row;
printf("\n0x00%x0 ",row);
}
e1000_read_nvm(&adapter->hw, i, 1, &eeprom_data);
printf("%04x ", eeprom_data);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
Move statistics into the sysctl tree making it easier to find and use them. Add previously hidden statistics, some of which include interrupt and host/card communication counters.
2010-06-16 17:36:53 +00:00
printf("\n");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
}
static void
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
igb_set_sysctl_value(struct adapter *adapter, const char *name,
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
const char *description, int *limit, int value)
{
*limit = value;
SYSCTL_ADD_INT(device_get_sysctl_ctx(adapter->dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(adapter->dev)),
OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW, limit, value, description);
}
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
/*
** Set flow control using sysctl:
** Flow control values:
** 0 - off
** 1 - rx pause
** 2 - tx pause
** 3 - full
*/
static int
igb_set_flowcntl(SYSCTL_HANDLER_ARGS)
{
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
int error;
static int input = 3; /* default is full */
struct adapter *adapter = (struct adapter *) arg1;
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
error = sysctl_handle_int(oidp, &input, 0, req);
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
if ((error) || (req->newptr == NULL))
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
return (error);
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
switch (input) {
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
case e1000_fc_rx_pause:
case e1000_fc_tx_pause:
case e1000_fc_full:
case e1000_fc_none:
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
adapter->hw.fc.requested_mode = input;
adapter->fc = input;
break;
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
default:
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
/* Do nothing */
return (error);
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
}
adapter->hw.fc.current_mode = adapter->hw.fc.requested_mode;
e1000_force_mac_fc(&adapter->hw);
Set DROP_EN on each RX queue if transmit flow-control is disabled. This allows the NIC to drop frames on the receive queue and not cause the MAC to block on receiving to _any_ queue. Tested: igb0@pci0:5:0:0: class=0x020000 card=0x152115d9 chip=0x15218086 rev=0x01 hdr=0x00 vendor = 'Intel Corporation' device = 'I350 Gigabit Network Connection' class = network subclass = ethernet Discussed with: Eric Joyner <eric.joyner@intel.com> MFC after: 1 week Sponsored by: Norse Corp, Inc.
2014-09-15 19:53:49 +00:00
/* XXX TODO: update DROP_EN on each RX queue if appropriate */
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
return (error);
}
/*
** Manage DMA Coalesce:
** Control values:
** 0/1 - off/on
** Legal timer values are:
** 250,500,1000-10000 in thousands
*/
static int
igb_sysctl_dmac(SYSCTL_HANDLER_ARGS)
{
struct adapter *adapter = (struct adapter *) arg1;
int error;
error = sysctl_handle_int(oidp, &adapter->dmac, 0, req);
if ((error) || (req->newptr == NULL))
return (error);
switch (adapter->dmac) {
case 0:
/*Disabling */
break;
case 1: /* Just enable and use default */
adapter->dmac = 1000;
break;
case 250:
case 500:
case 1000:
case 2000:
case 3000:
case 4000:
case 5000:
case 6000:
case 7000:
case 8000:
case 9000:
case 10000:
/* Legal values - allow */
break;
default:
/* Do nothing, illegal value */
adapter->dmac = 0;
Update the Intel igb driver to version 2.4.0 - This version has support for the new Intel Avoton systems, including 2.5Gb support, further it now has IPv6/TSO6 support as well. Shared code has been updated where necessary as well. Thanks to my new assistant Eric Joyner for doing the transmit path changes to bring in the IPv6/TSO6 support. Thanks to Gleb for catching the one bug and change needed in NETMAP. Approved by: re
2013-10-09 17:32:52 +00:00
return (EINVAL);
Eliminate some global tuneables in favor of adapter-specific, particular flow control and dma coalesce. Also improve the sysctl operation on those too. Add IPv6 detection in the ioctl code, this was done for ixgbe first, carrying that over. Add resource ability to disable particular adapter. Add HW TSO capability so vlans can make use of TSO
2011-06-20 22:59:29 +00:00
}
/* Reinit the interface */
igb_init(adapter);
return (error);
Important update for the igb driver: - Add the change made in em to the actual unrefreshed number of descriptors is used as a basis in rxeof on the way out to determine if more refresh is needed. NOTE: there is a difference in the ring setup in igb, this is not accidental, it is necessitated by hardware behavior, when you reset the newer adapters it will not let you write RDH, it ALWAYS sets it to 0. Thus the way em does it is not possible. - Change the sysctl handling of flow control, it will now make the change dynamically when the variable setting changes rather than requiring a reset. - Change the eee sysctl naming, validation found the old unintuitive :) - Last but not least, some important performance tweaks in the TX path, I found that UDP behavior could be drastically hindered or improved with just small changes in the start loop. What I have here is what testing has shown to be the best overall. Its interesting to note that changing the clean threshold to start at a full half of the ring, made a BIG difference in performance. I hope that this will prove to be advantageous for most workloads. MFC in a week.
2011-04-05 21:55:43 +00:00
}
Change the interface to the Energy Efficient Ethernet (EEE) setting in the igb and em driver. This was necessitated by a shared code change that I was given late in the game, a data type changed from bool to int, in the last update I dealt with it by a cast, but it was pointed out (thanks jhb) that there was a potential problem with this. John suggested this safer approach, and it is fine with me... MFC after:2 days (to catch the 9.1 update)
2012-07-07 20:21:05 +00:00
/*
** Manage Energy Efficient Ethernet:
** Control values:
** 0/1 - enabled/disabled
*/
static int
igb_sysctl_eee(SYSCTL_HANDLER_ARGS)
{
struct adapter *adapter = (struct adapter *) arg1;
int error, value;
value = adapter->hw.dev_spec._82575.eee_disable;
error = sysctl_handle_int(oidp, &value, 0, req);
if (error || req->newptr == NULL)
return (error);
IGB_CORE_LOCK(adapter);
adapter->hw.dev_spec._82575.eee_disable = (value != 0);
igb_init_locked(adapter);
IGB_CORE_UNLOCK(adapter);
return (0);
}
Reference in New Issue Copy Permalink