d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
6893e74dc0
freebsd-skq/sys/dev/e1000/if_igb.c

5837 lines
163 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
/******************************************************************************
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
Copyright (c) 2001-2011, 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
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
#ifdef HAVE_KERNEL_OPTION_HEADERS
#include "opt_device_polling.h"
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"
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>
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 __FreeBSD_version >= 800000
#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>
#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>
#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:
*********************************************************************/
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
char igb_driver_version[] = "version - 2.3.1";
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
/*********************************************************************
* 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},
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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);
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
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
#if __FreeBSD_version >= 800000
static int igb_mq_start(struct ifnet *, struct mbuf *);
static int igb_mq_start_locked(struct ifnet *,
struct tx_ring *, struct mbuf *);
static void igb_qflush(struct ifnet *);
#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);
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);
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 bool igb_tx_ctx_setup(struct tx_ring *, struct mbuf *);
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
static bool igb_tso_setup(struct tx_ring *, struct mbuf *, int,
struct ip *, struct tcphdr *);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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),
{0, 0}
};
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;
TUNABLE_INT("hw.igb.rxd", &igb_rxd);
TUNABLE_INT("hw.igb.txd", &igb_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;
TUNABLE_INT("hw.igb.enable_aim", &igb_enable_aim);
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_aim, CTLFLAG_RW, &igb_enable_aim, 0,
"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;
TUNABLE_INT("hw.igb.enable_msix", &igb_enable_msix);
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;
TUNABLE_INT("hw.igb.max_interrupt_rate", &igb_max_interrupt_rate);
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
/*
** 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;
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
TUNABLE_INT("hw.igb.hdr_split", &igb_header_split);
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
/*
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 will autoconfigure based on
** the number of CPUs 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;
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
TUNABLE_INT("hw.igb.num_queues", &igb_num_queues);
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");
/* How many packets rxeof tries to clean at a time */
static int igb_rx_process_limit = 100;
TUNABLE_INT("hw.igb.rx_process_limit", &igb_rx_process_limit);
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");
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 (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;
adapter->min_frame_size = ETH_ZLEN + 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");
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_set_sysctl_value(adapter, "eee_disabled",
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
"enable Energy Efficient Ethernet",
&adapter->hw.dev_spec._82575.eee_disable,
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
TRUE);
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_set_eee_i350(&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
/*
** 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 */
if (e1000_validate_nvm_checksum(&adapter->hw) < 0) {
/*
** 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);
}
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);
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
ether_ifdetach(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
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);
struct ifnet *ifp = adapter->ifp;
IGB_CORE_LOCK(adapter);
igb_init_locked(adapter);
igb_init_manageability(adapter);
if ((ifp->if_flags & IFF_UP) &&
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))
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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_start(ifp);
IGB_CORE_UNLOCK(adapter);
return bus_generic_resume(dev);
}
/*********************************************************************
* 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
}
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 __FreeBSD_version >= 800000
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
/*
** Multiqueue Transmit driver
**
*/
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;
bool moveable = TRUE;
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 */
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->m_flags & M_FLOWID) != 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
i = m->m_pkthdr.flowid % adapter->num_queues;
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
moveable = FALSE;
} else
i = curcpu % adapter->num_queues;
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
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];
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->queue_status & IGB_QUEUE_DEPLETED) == 0) &&
IGB_TX_TRYLOCK(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
err = igb_mq_start_locked(ifp, txr, m);
IGB_TX_UNLOCK(txr);
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
} else {
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
err = drbr_enqueue(ifp, txr->br, m);
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
taskqueue_enqueue(que->tq, &que->que_task);
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
}
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);
}
static int
igb_mq_start_locked(struct ifnet *ifp, struct tx_ring *txr, struct mbuf *m)
{
struct adapter *adapter = txr->adapter;
struct mbuf *next;
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 err = 0, enq;
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) ||
(txr->queue_status == IGB_QUEUE_DEPLETED) ||
adapter->link_active == 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
if (m != NULL)
err = drbr_enqueue(ifp, txr->br, m);
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);
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
}
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
enq = 0;
Fix drbr and altq interaction: - introduce drbr_needs_enqueue that returns whether the interface/br needs an enqueue operation: returns true if altq is enabled or there are already packets in the ring (as we need to maintain packet order) - update all drbr consumers - fix drbr_flush - avoid using the driver queue (IFQ_DRV_*) in the altq case as the multiqueue consumer does not provide enough protection, serialize altq interaction with the main queue lock - make drbr_dequeue_cond work with altq Discussed with: kmacy, yongari, jfv MFC after: 4 weeks
2010-02-13 16:04:58 +00:00
if (m == 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
next = drbr_dequeue(ifp, txr->br);
Fix drbr and altq interaction: - introduce drbr_needs_enqueue that returns whether the interface/br needs an enqueue operation: returns true if altq is enabled or there are already packets in the ring (as we need to maintain packet order) - update all drbr consumers - fix drbr_flush - avoid using the driver queue (IFQ_DRV_*) in the altq case as the multiqueue consumer does not provide enough protection, serialize altq interaction with the main queue lock - make drbr_dequeue_cond work with altq Discussed with: kmacy, yongari, jfv MFC after: 4 weeks
2010-02-13 16:04:58 +00:00
} else if (drbr_needs_enqueue(ifp, txr->br)) {
if ((err = drbr_enqueue(ifp, txr->br, m)) != 0)
return (err);
next = drbr_dequeue(ifp, txr->br);
} 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
next = m;
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 */
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 (next != NULL) {
if ((err = igb_xmit(txr, &next)) != 0) {
if (next != NULL)
err = drbr_enqueue(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
}
enq++;
drbr_stats_update(ifp, next->m_pkthdr.len, next->m_flags);
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
next = drbr_dequeue(ifp, txr->br);
}
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);
}
/*
** 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);
}
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
#endif /* __FreeBSD_version >= 800000 */
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;
}
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
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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_capenable & IFCAP_TSO4)
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 */
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_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);
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 (igb_txeof(txr))
more = TRUE;
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 __FreeBSD_version >= 800000
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))
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_mq_start_locked(ifp, txr, NULL);
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
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_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? */
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 (more || (ifp->if_drv_flags & IFF_DRV_OACTIVE)) {
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;
adapter->hw.mac.get_link_status = 1;
igb_update_link_status(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
/*********************************************************************
*
* 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
/*********************************************************************
*
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
* Legacy polling routine : if using this code you MUST be sure that
* multiqueue is not defined, ie, set igb_num_queues to 1.
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 __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;
struct igb_queue *que = adapter->queues;
struct tx_ring *txr = adapter->tx_rings;
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))
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);
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);
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(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
IGB_TX_LOCK(txr);
do {
more = igb_txeof(txr);
} while (loop-- && more);
#if __FreeBSD_version >= 800000
if (!drbr_empty(ifp, txr->br))
igb_mq_start_locked(ifp, txr, NULL);
#else
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_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
IGB_TX_UNLOCK(txr);
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;
struct tx_ring *txr = que->txr;
struct rx_ring *rxr = que->rxr;
u32 newitr = 0;
bool more_tx, 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
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);
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
more_tx = igb_txeof(txr);
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*/
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 (more_tx || 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;
u_char fiber_type = IFM_1000_SX;
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;
if ((adapter->hw.phy.media_type == e1000_media_type_fiber) ||
(adapter->hw.phy.media_type == e1000_media_type_internal_serdes))
ifmr->ifm_active |= fiber_type | IFM_FDX;
else {
switch (adapter->link_speed) {
case 10:
ifmr->ifm_active |= IFM_10_T;
break;
case 100:
ifmr->ifm_active |= IFM_100_TX;
break;
case 1000:
ifmr->ifm_active |= IFM_1000_T;
break;
}
if (adapter->link_duplex == FULL_DUPLEX)
ifmr->ifm_active |= IFM_FDX;
else
ifmr->ifm_active |= IFM_HDX;
}
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)
{
struct adapter *adapter = txr->adapter;
bus_dma_segment_t segs[IGB_MAX_SCATTER];
bus_dmamap_t map;
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
struct igb_tx_buffer *tx_buffer, *tx_buffer_mapped;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
union e1000_adv_tx_desc *txd = NULL;
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 mbuf *m_head = *m_headp;
struct ether_vlan_header *eh = NULL;
struct ip *ip = NULL;
struct tcphdr *th = NULL;
u32 hdrlen, cmd_type_len, olinfo_status = 0;
int ehdrlen, poff;
int nsegs, i, first, last = 0;
int error, do_tso, remap = 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
/* Set basic descriptor constants */
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
cmd_type_len = E1000_ADVTXD_DTYP_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
cmd_type_len |= E1000_ADVTXD_DCMD_IFCS | E1000_ADVTXD_DCMD_DEXT;
if (m_head->m_flags & M_VLANTAG)
cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
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
retry:
m_head = *m_headp;
do_tso = ((m_head->m_pkthdr.csum_flags & CSUM_TSO) != 0);
hdrlen = ehdrlen = poff = 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
* Intel recommends entire IP/TCP header length reside in a single
* buffer. If multiple descriptors are used to describe the IP and
* TCP header, each descriptor should describe one or more
* complete headers; descriptors referencing only parts of headers
* are not supported. If all layer headers are not coalesced into
* a single buffer, each buffer should not cross a 4KB boundary,
* or be larger than the maximum read request size.
* Controller also requires modifing IP/TCP header to make TSO work
* so we firstly get a writable mbuf chain then coalesce ethernet/
* IP/TCP header into a single buffer to meet the requirement of
* controller. This also simplifies IP/TCP/UDP checksum offloading
* which also has similiar restrictions.
*/
if (do_tso || m_head->m_pkthdr.csum_flags & CSUM_OFFLOAD) {
if (do_tso || (m_head->m_next != NULL &&
m_head->m_pkthdr.csum_flags & CSUM_OFFLOAD)) {
if (M_WRITABLE(*m_headp) == 0) {
m_head = m_dup(*m_headp, M_DONTWAIT);
m_freem(*m_headp);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
*m_headp = m_head;
}
}
/*
* Assume IPv4, we don't have TSO/checksum offload support
* for IPv6 yet.
*/
ehdrlen = sizeof(struct ether_header);
m_head = m_pullup(m_head, ehdrlen);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
eh = mtod(m_head, struct ether_vlan_header *);
if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
ehdrlen = sizeof(struct ether_vlan_header);
m_head = m_pullup(m_head, ehdrlen);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
}
m_head = m_pullup(m_head, ehdrlen + sizeof(struct ip));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
ip = (struct ip *)(mtod(m_head, char *) + ehdrlen);
poff = ehdrlen + (ip->ip_hl << 2);
if (do_tso) {
m_head = m_pullup(m_head, poff + sizeof(struct tcphdr));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
/*
* The pseudo TCP checksum does not include TCP payload
* length so driver should recompute the checksum here
* what hardware expect to see. This is adherence of
* Microsoft's Large Send specification.
*/
th = (struct tcphdr *)(mtod(m_head, char *) + poff);
th->th_sum = in_pseudo(ip->ip_src.s_addr,
ip->ip_dst.s_addr, htons(IPPROTO_TCP));
/* Keep track of the full header length */
hdrlen = poff + (th->th_off << 2);
} else if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
m_head = m_pullup(m_head, poff + sizeof(struct tcphdr));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
th = (struct tcphdr *)(mtod(m_head, char *) + poff);
m_head = m_pullup(m_head, poff + (th->th_off << 2));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
ip = (struct ip *)(mtod(m_head, char *) + ehdrlen);
th = (struct tcphdr *)(mtod(m_head, char *) + poff);
} else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) {
m_head = m_pullup(m_head, poff + sizeof(struct udphdr));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
ip = (struct ip *)(mtod(m_head, char *) + ehdrlen);
}
*m_headp = m_head;
}
/*
* 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
*
* Capture the first descriptor index,
* this descriptor will have the index
* of the EOP which is the only one that
* now gets a DONE bit writeback.
*/
first = txr->next_avail_desc;
tx_buffer = &txr->tx_buffers[first];
tx_buffer_mapped = tx_buffer;
map = tx_buffer->map;
error = bus_dmamap_load_mbuf_sg(txr->txtag, map,
*m_headp, segs, &nsegs, BUS_DMA_NOWAIT);
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
/*
* There are two types of errors we can (try) to handle:
* - EFBIG means the mbuf chain was too long and bus_dma ran
* out of segments. Defragment the mbuf chain and try again.
* - ENOMEM means bus_dma could not obtain enough bounce buffers
* at this point in time. Defer sending and try again later.
* All other errors, in particular EINVAL, are fatal and prevent the
* mbuf chain from ever going through. Drop it and report error.
*/
if (error == EFBIG && remap) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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;
m = m_defrag(*m_headp, M_DONTWAIT);
if (m == 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
adapter->mbuf_defrag_failed++;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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;
return (ENOBUFS);
}
*m_headp = m;
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
/* Try it again, but only once */
remap = 0;
goto 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
} else if (error == ENOMEM) {
adapter->no_tx_dma_setup++;
return (error);
} else if (error != 0) {
adapter->no_tx_dma_setup++;
m_freem(*m_headp);
*m_headp = NULL;
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
/*
** Make sure we don't overrun the ring,
** we need nsegs descriptors and one for
** the context descriptor used for the
** offloads.
*/
if ((nsegs + 1) > (txr->tx_avail - 2)) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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->no_desc_avail++;
bus_dmamap_unload(txr->txtag, map);
return (ENOBUFS);
}
m_head = *m_headp;
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 hardware assists:
* Set up the context descriptor, used
* when any hardware offload is done.
* This includes CSUM, VLAN, and TSO.
* It will use the first 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
*/
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->m_pkthdr.csum_flags & CSUM_TSO) {
if (igb_tso_setup(txr, m_head, ehdrlen, ip, th)) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
} 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
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
} else if (igb_tx_ctx_setup(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
olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
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
/* Calculate payload length */
olinfo_status |= ((m_head->m_pkthdr.len - hdrlen)
<< E1000_ADVTXD_PAYLEN_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)
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
/* Set up our transmit descriptors */
i = txr->next_avail_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
for (int j = 0; j < nsegs; j++) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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_size_t seg_len;
bus_addr_t seg_addr;
tx_buffer = &txr->tx_buffers[i];
txd = (union e1000_adv_tx_desc *)&txr->tx_base[i];
seg_addr = segs[j].ds_addr;
seg_len = segs[j].ds_len;
txd->read.buffer_addr = htole64(seg_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
txd->read.cmd_type_len = htole32(cmd_type_len | seg_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
txd->read.olinfo_status = htole32(olinfo_status);
last = i;
if (++i == adapter->num_tx_desc)
i = 0;
tx_buffer->m_head = NULL;
tx_buffer->next_eop = -1;
}
txr->next_avail_desc = i;
txr->tx_avail -= nsegs;
tx_buffer->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.
*/
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
tx_buffer_mapped->map = tx_buffer->map;
tx_buffer->map = map;
bus_dmamap_sync(txr->txtag, map, BUS_DMASYNC_PREWRITE);
/*
* Last Descriptor of Packet
* needs End Of Packet (EOP)
* and Report Status (RS)
*/
txd->read.cmd_type_len |=
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
htole32(E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_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
/*
* Keep track in the first buffer which
* descriptor will be written back
*/
tx_buffer = &txr->tx_buffers[first];
tx_buffer->next_eop = last;
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 watchdog time early and often */
Missing a fix for the new watchdog handling.
2010-01-26 23:04:13 +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
/*
* Advance the Transmit Descriptor Tail (TDT), this tells the E1000
* that this frame is available to transmit.
*/
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
E1000_WRITE_REG(&adapter->hw, E1000_TDT(txr->me), i);
++txr->tx_packets;
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;
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_disabled);
return;
}
reg = E1000_READ_REG(hw, E1000_RCTL);
reg &= (~E1000_RCTL_UPE);
reg &= (~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
}
/*********************************************************************
* 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)
{
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
struct e1000_hw *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
struct ifnet *ifp = adapter->ifp;
device_t dev = adapter->dev;
struct tx_ring *txr = adapter->tx_rings;
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
u32 link_check, thstat, ctrl;
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);
}
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)
device_printf(dev, "Link is up %d Mbps %s\n",
adapter->link_speed,
((adapter->link_duplex == FULL_DUPLEX) ?
"Full Duplex" : "Half Duplex"));
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 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 */
adapter->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
if (!((adapter->hw.bus.pci_cmd_word & PCIM_CMD_BUSMASTEREN) &&
(adapter->hw.bus.pci_cmd_word & PCIM_CMD_MEMEN))) {
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("Memory Access and/or Bus Master "
"bits were not set!\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
adapter->hw.bus.pci_cmd_word |=
(PCIM_CMD_BUSMASTEREN | PCIM_CMD_MEMEN);
pci_write_config(dev, PCIR_COMMAND,
adapter->hw.bus.pci_cmd_word, 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;
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);
}
/*
* 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;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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
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;
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
/*
** Bind the msix vector, and thus the
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
** rings to the corresponding cpu.
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
*/
Change intr_bind to bus_bind_intr, also limit this to multiqueue setup which is not the shipping default for igb (its set to 1).
2009-06-25 17:21:12 +00:00
if (adapter->num_queues > 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
bus_bind_intr(dev, que->res, i);
/* Make tasklet for deferred handling */
TASK_INIT(&que->que_task, 0, igb_handle_que, que);
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
que->tq = taskqueue_create_fast("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);
taskqueue_start_threads(&que->tq, 1, PI_NET, "%s que",
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
device_get_nameunit(adapter->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
}
/* 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:
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);
if (adapter->tag != NULL) {
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);
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,
PCIR_BAR(IGB_MSIX_BAR), adapter->msix_mem);
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)
{
device_t dev = adapter->dev;
int rid, want, queues, msgs;
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 */
rid = PCIR_BAR(IGB_MSIX_BAR);
adapter->msix_mem = bus_alloc_resource_any(dev,
SYS_RES_MEMORY, &rid, RF_ACTIVE);
if (!adapter->msix_mem) {
/* May not be enabled */
device_printf(adapter->dev,
"Unable to map MSIX table \n");
goto msi;
}
msgs = pci_msix_count(dev);
if (msgs == 0) { /* system has msix disabled */
bus_release_resource(dev, SYS_RES_MEMORY,
PCIR_BAR(IGB_MSIX_BAR), adapter->msix_mem);
adapter->msix_mem = NULL;
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;
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;
- 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 (queues > 8) /* max queues */
queues = 8;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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
/* Can have max of 4 queues on 82575 */
if ((adapter->hw.mac.type == e1000_82575) && (queues > 4))
queues = 4;
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 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
/* Limit the VF devices to one queue */
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
queues = 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
/*
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);
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
}
if ((msgs) && pci_alloc_msix(dev, &msgs) == 0) {
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);
}
msi:
msgs = pci_msi_count(dev);
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
if (msgs == 1 && pci_alloc_msi(dev, &msgs) == 0) {
device_printf(adapter->dev," Using MSI interrupt\n");
return (msgs);
}
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
}
/*********************************************************************
*
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:
case e1000_vfadapt_i350:
pba = E1000_READ_REG(hw, E1000_RXPBS);
pba = e1000_rxpbs_adjust_82580(pba);
break;
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
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 */
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 (hw->mac.type == e1000_i350) {
u32 reg = ~E1000_DMACR_DMAC_EN;
if (adapter->dmac == 0) { /* Disabling it */
E1000_WRITE_REG(hw, E1000_DMACR, reg);
goto reset_out;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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
hwm = (pba - 4) << 10;
reg = (((pba-6) << 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);
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
/* timer = value in adapter->dmac in 32usec intervals */
reg |= (adapter->dmac >> 5);
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_DMACR, reg);
/* No lower threshold */
E1000_WRITE_REG(hw, E1000_DMCRTRH, 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
/* set hwm to PBA - 2 * max frame size */
E1000_WRITE_REG(hw, E1000_FCRTC, hwm);
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
/* Set the interval before transition */
reg = E1000_READ_REG(hw, E1000_DMCTLX);
reg |= 0x800000FF; /* 255 usec */
E1000_WRITE_REG(hw, E1000_DMCTLX, reg);
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
/* free space in tx packet buffer to wake from DMA coal */
E1000_WRITE_REG(hw, E1000_DMCTXTH,
(20480 - (2 * adapter->max_frame_size)) >> 6);
/* make low power state decision controlled by DMA coal */
reg = E1000_READ_REG(hw, E1000_PCIEMISC);
E1000_WRITE_REG(hw, E1000_PCIEMISC,
reg | E1000_PCIEMISC_LX_DECISION);
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
device_printf(dev, "DMA Coalescing enabled\n");
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
}
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
reset_out:
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;
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_start = igb_start;
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 __FreeBSD_version >= 800000
ifp->if_transmit = igb_mq_start;
ifp->if_qflush = igb_qflush;
#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
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);
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;
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_capabilities |= IFCAP_TSO4;
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
*/
ifp->if_data.ifi_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_map = NULL;
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;
if (dma->dma_map != NULL) {
bus_dmamap_sync(dma->dma_tag, dma->dma_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(dma->dma_tag, dma->dma_map);
bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
dma->dma_map = NULL;
}
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;
/* 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;
}
txr->tx_base = (struct e1000_tx_desc *)txr->txdma.dma_vaddr;
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;
}
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 __FreeBSD_version >= 800000
/* Allocate a buf ring */
txr->br = buf_ring_alloc(IGB_BR_SIZE, M_DEVBUF,
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:
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
#if __FreeBSD_version >= 800000
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;
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
struct igb_tx_buffer *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 =
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
(struct igb_tx_buffer *) malloc(sizeof(struct igb_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
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;
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
struct igb_tx_buffer *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 */
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
netmap_load_map(txr->txtag, txbuf->map, NMB(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 */
txbuf->next_eop = -1;
}
/* 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;
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
struct igb_tx_buffer *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;
}
}
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 __FreeBSD_version >= 800000
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 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 work for hardware segmentation offload (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
*
**********************************************************************/
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 bool
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
igb_tso_setup(struct tx_ring *txr, struct mbuf *mp, int ehdrlen,
struct ip *ip, struct tcphdr *th)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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;
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
struct igb_tx_buffer *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
u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
u32 mss_l4len_idx = 0;
u16 vtag = 0;
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 ctxd, 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
ctxd = txr->next_avail_desc;
tx_buffer = &txr->tx_buffers[ctxd];
TXD = (struct e1000_adv_tx_context_desc *) &txr->tx_base[ctxd];
ip->ip_sum = 0;
ip_hlen = ip->ip_hl << 2;
tcp_hlen = th->th_off << 2;
/* VLAN MACLEN IPLEN */
if (mp->m_flags & M_VLANTAG) {
vtag = htole16(mp->m_pkthdr.ether_vtag);
vlan_macip_lens |= (vtag << E1000_ADVTXD_VLAN_SHIFT);
}
vlan_macip_lens |= (ehdrlen << E1000_ADVTXD_MACLEN_SHIFT);
vlan_macip_lens |= ip_hlen;
TXD->vlan_macip_lens |= htole32(vlan_macip_lens);
/* ADV DTYPE TUCMD */
type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP;
type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4;
TXD->type_tucmd_mlhl |= htole32(type_tucmd_mlhl);
/* 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);
tx_buffer->m_head = NULL;
tx_buffer->next_eop = -1;
if (++ctxd == adapter->num_tx_desc)
ctxd = 0;
txr->tx_avail--;
txr->next_avail_desc = ctxd;
return TRUE;
}
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
/*********************************************************************
*
* Context Descriptor setup for VLAN or CSUM
*
**********************************************************************/
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 bool
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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_ctx_setup(struct tx_ring *txr, struct mbuf *mp)
{
struct adapter *adapter = txr->adapter;
struct e1000_adv_tx_context_desc *TXD;
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
struct igb_tx_buffer *tx_buffer;
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
u32 vlan_macip_lens, type_tucmd_mlhl, 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
struct ether_vlan_header *eh;
struct ip *ip = NULL;
struct ip6_hdr *ip6;
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 ehdrlen, ctxd, ip_hlen = 0;
u16 etype, vtag = 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
u8 ipproto = 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
bool offload = TRUE;
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
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
vlan_macip_lens = type_tucmd_mlhl = mss_l4len_idx = 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
ctxd = txr->next_avail_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
tx_buffer = &txr->tx_buffers[ctxd];
TXD = (struct e1000_adv_tx_context_desc *) &txr->tx_base[ctxd];
/*
** In advanced descriptors the vlan tag must
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
** be placed into the context descriptor, thus
** we need to be here just for that 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
*/
if (mp->m_flags & M_VLANTAG) {
vtag = htole16(mp->m_pkthdr.ether_vtag);
vlan_macip_lens |= (vtag << E1000_ADVTXD_VLAN_SHIFT);
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 if (offload == FALSE)
return 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
/*
* 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;
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_len < ehdrlen + ip_hlen) {
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
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);
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;
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
}
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 */
TXD->vlan_macip_lens |= htole32(vlan_macip_lens);
TXD->type_tucmd_mlhl |= htole32(type_tucmd_mlhl);
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
tx_buffer->m_head = NULL;
tx_buffer->next_eop = -1;
/* We've consumed the first desc, adjust counters */
if (++ctxd == adapter->num_tx_desc)
ctxd = 0;
txr->next_avail_desc = ctxd;
--txr->tx_avail;
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 (offload);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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)
{
struct adapter *adapter = txr->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
int first, last, done, processed;
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
struct igb_tx_buffer *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
struct e1000_tx_desc *tx_desc, *eop_desc;
struct ifnet *ifp = adapter->ifp;
IGB_TX_LOCK_ASSERT(txr);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
#ifdef DEV_NETMAP
if (ifp->if_capenable & IFCAP_NETMAP) {
struct netmap_adapter *na = NA(ifp);
selwakeuppri(&na->tx_rings[txr->me].si, PI_NET);
IGB_TX_UNLOCK(txr);
IGB_CORE_LOCK(adapter);
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
selwakeuppri(&na->tx_si, PI_NET);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
IGB_CORE_UNLOCK(adapter);
IGB_TX_LOCK(txr);
return FALSE;
}
#endif /* DEV_NETMAP */
- 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 (txr->tx_avail == adapter->num_tx_desc) {
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
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
- 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
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
first = txr->next_to_clean;
tx_desc = &txr->tx_base[first];
tx_buffer = &txr->tx_buffers[first];
last = tx_buffer->next_eop;
eop_desc = &txr->tx_base[last];
/*
* What this does is get the index of the
* first descriptor AFTER the EOP of the
* first packet, that way we can do the
* simple comparison on the inner while loop.
*/
if (++last == adapter->num_tx_desc)
last = 0;
done = last;
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);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
while (eop_desc->upper.fields.status & E1000_TXD_STAT_DD) {
/* We clean the range of the packet */
while (first != done) {
tx_desc->upper.data = 0;
tx_desc->lower.data = 0;
tx_desc->buffer_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
++txr->tx_avail;
- 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
++processed;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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 (tx_buffer->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 +=
tx_buffer->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,
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;
}
tx_buffer->next_eop = -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
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
if (++first == adapter->num_tx_desc)
first = 0;
tx_buffer = &txr->tx_buffers[first];
tx_desc = &txr->tx_base[first];
}
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;
++ifp->if_opackets;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* See if we can continue to the next packet */
last = tx_buffer->next_eop;
if (last != -1) {
eop_desc = &txr->tx_base[last];
/* Get new done point */
if (++last == adapter->num_tx_desc) last = 0;
done = last;
} else
break;
}
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
txr->next_to_clean = first;
- 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;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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
* If we have a minimum free,
* clear depleted state 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
*/
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_QUEUE_THRESHOLD)
txr->queue_status &= ~IGB_QUEUE_DEPLETED;
/* All clean, turn off the watchdog */
if (txr->tx_avail == adapter->num_tx_desc) {
txr->queue_status = IGB_QUEUE_IDLE;
return (FALSE);
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
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) {
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
mh = m_gethdr(M_DONTWAIT, 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) {
mp = m_getjcl(M_DONTWAIT, MT_DATA,
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];
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
error = bus_dmamap_create(rxr->htag,
BUS_DMA_NOWAIT, &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;
}
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
error = bus_dmamap_create(rxr->ptag,
BUS_DMA_NOWAIT, &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) {
/* slot sj is mapped to the i-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;
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
addr = PNMB(slot + sj, &paddr);
netmap_load_map(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 */
rxbuf->m_head = m_gethdr(M_DONTWAIT, 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 */
rxbuf->m_pack = m_getjcl(M_DONTWAIT, MT_DATA,
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;
rxr->discard = 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
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
*/
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
for (int j = 0; j > i; ++j) {
rxr = &adapter->rx_rings[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
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);
}
/*********************************************************************
*
* 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;
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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
u32 random[10], mrqc, shift = 0;
union igb_reta {
u32 dword;
u8 bytes[4];
} reta;
arc4rand(&random, sizeof(random), 0);
if (adapter->hw.mac.type == e1000_82575)
shift = 6;
/* Warning FM follows */
for (int i = 0; i < 128; i++) {
reta.bytes[i & 3] =
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
(i % adapter->num_queues) << 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
if ((i & 3) == 3)
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,
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_RETA(i >> 2), reta.dword);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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 fill in hash table */
mrqc = E1000_MRQC_ENABLE_RSS_4Q;
for (int i = 0; i < 10; 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
E1000_WRITE_REG_ARRAY(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
E1000_RSSRK(0), i, random[i]);
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);
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_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
/*
** 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
* somthing different from next_to_refresh
* (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];
int t = rxr->next_to_refresh - kring->nr_hwavail;
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;
}
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;
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
if (ifp->if_capenable & IFCAP_NETMAP) {
struct netmap_adapter *na = NA(ifp);
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
na->rx_rings[rxr->me].nr_kflags |= NKR_PENDINTR;
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
selwakeuppri(&na->rx_rings[rxr->me].si, PI_NET);
IGB_RX_UNLOCK(rxr);
IGB_CORE_LOCK(adapter);
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
selwakeuppri(&na->rx_si, PI_NET);
put back netmap support, deleted by mistake in a previous commit
2011-12-22 15:33:41 +00:00
IGB_CORE_UNLOCK(adapter);
return (0);
}
#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;
u16 hlen, plen, hdr, vtag;
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;
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 ((adapter->hw.mac.type == e1000_i350) &&
(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);
eop = ((staterr & E1000_RXD_STAT_EOP) == E1000_RXD_STAT_EOP);
/* Make sure all segments of a bad packet are discarded */
if (((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) != 0) ||
(rxr->discard)) {
ifp->if_ierrors++;
++rxr->rx_discarded;
if (!eop) /* Catch subsequent segs */
rxr->discard = TRUE;
else
rxr->discard = FALSE;
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) {
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
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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;
ifp->if_ipackets++;
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
}
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 __FreeBSD_version >= 800000
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)
*done = rxdone;
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)) {
u16 type = (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
#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
/**********************************************************************
*
* 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 ifnet *ifp;
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);
stats->tor += E1000_READ_REG(hw, E1000_TORH);
stats->tot += E1000_READ_REG(hw, E1000_TOTH);
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
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
ifp = adapter->ifp;
ifp->if_collisions = stats->colc;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Rx Errors */
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
ifp->if_ierrors = adapter->dropped_pkts + stats->rxerrc +
stats->crcerrs + stats->algnerrc +
stats->ruc + stats->roc + stats->mpc + stats->cexterr;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Tx Errors */
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
ifp->if_oerrors = stats->ecol +
stats->latecol + adapter->watchdog_events;
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,
"Queue No Descriptor Available");
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, "tx_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, &txr->tx_packets,
"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");
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,
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
"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");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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");
/* 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,
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 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,
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
"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,
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 Octets Received");
/* 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");
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,
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
"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);
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;
return (error);
}
/* 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
}
Reference in New Issue Copy Permalink