d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
f9b11500c2
freebsd-nq/sys/dev/e1000/if_em.c

6020 lines
172 KiB
C
Raw Normal View History

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
/******************************************************************************
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
Copyright (c) 2001-2015, Intel Corporation
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
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.
******************************************************************************/
/*$FreeBSD$*/
- Added support for receive in multiple descriptors. This simplifies code for jumbo frames. - Cleaned up coding conventions to make code more unix-like. - Cleaned up code in if_em_fxhw.c and if_em_phy.c. Added relevant comments. MFC after: 1 week
2002-02-13 18:19:27 +00:00
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#include "opt_em.h"
#include "opt_ddb.h"
Restore opt_inet.h include which was lost in the last commit.
2008-12-07 21:32:56 +00:00
#include "opt_inet.h"
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
#include "opt_inet6.h"
Change the E1000 driver option header handling to match the ixgbe driver. As it was, when building them as a module INET and INET6 are not defined. In these drivers it does not cause a panic, however it does result in different behavior in the ioctl routine when you are using a module vs static, and I think the behavior should be the same. MFC after: 3 days
2013-07-12 22:36:26 +00:00
#ifdef HAVE_KERNEL_OPTION_HEADERS
#include "opt_device_polling.h"
- Don't pollute opt_global.h with DEVICE_POLLING and introduce opt_device_polling.h - Include opt_device_polling.h into appropriate files. - Embrace with HAVE_KERNEL_OPTION_HEADERS the include in the files that can be compiled as loadable modules. Reviewed by: bde
2005-10-05 10:09:17 +00:00
#endif
Move includes from if_em.h to if_em.c and sort them.
2006-02-14 13:11:36 +00:00
#include <sys/param.h>
#include <sys/systm.h>
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#ifdef DDB
#include <sys/types.h>
#include <ddb/ddb.h>
#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
#if __FreeBSD_version >= 800000
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
#include <sys/buf_ring.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
#endif
Move includes from if_em.h to if_em.c and sort them.
2006-02-14 13:11:36 +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>
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#include <sys/smp.h>
Move includes from if_em.h to if_em.c and sort them.
2006-02-14 13:11:36 +00:00
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>
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
#include <sys/eventhandler.h>
Move includes from if_em.h to if_em.c and sort them.
2006-02-14 13:11:36 +00:00
#include <machine/bus.h>
#include <machine/resource.h>
#include <net/bpf.h>
#include <net/ethernet.h>
#include <net/if.h>
The r48589 promised to remove implicit inclusion of if_var.h soon. Prepare to this event, adding if_var.h to files that do need it. Also, include all includes that now are included due to implicit pollution via if_var.h Sponsored by: Netflix Sponsored by: Nginx, Inc.
2013-10-26 17:58:36 +00:00
#include <net/if_var.h>
Move includes from if_em.h to if_em.c and sort them.
2006-02-14 13:11:36 +00:00
#include <net/if_arp.h>
#include <net/if_dl.h>
#include <net/if_media.h>
#include <net/if_types.h>
#include <net/if_vlan_var.h>
#include <netinet/in_systm.h>
#include <netinet/in.h>
Since resetting hardware takes a very long time and results in link renegotiation, we only initialize the hardware only when it is absolutely required. Process SIOCGIFADDR ioctl in em(4) when we know an IPv4 address is added. Handling SIOCGIFADDR in a driver is layering violation but it seems that there is no easy way without rewritting hardware initialization code to reduce settle time after reset. This should fix a long standing bug which didn't send ARP packet when interface address is changed or an alias address is added. Another effect of this fix is it doesn't need additional delays anymore when adding an alias address to the interface. While I'm here add a new if_flags into softc which remembers current prgroammed interface flags and make use of it when we have to program promiscuous mode. Tested by: Atanas <atanas AT asd DOT aplus DOT net> Analyzed by: rwatson Discussed with: -stable
2006-07-20 04:18:45 +00:00
#include <netinet/if_ether.h>
Move includes from if_em.h to if_em.c and sort them.
2006-02-14 13:11:36 +00:00
#include <netinet/ip.h>
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
#include <netinet/ip6.h>
Move includes from if_em.h to if_em.c and sort them.
2006-02-14 13:11:36 +00:00
#include <netinet/tcp.h>
#include <netinet/udp.h>
Second attempt at fixing module build Pointyhat: pdeuskar
2006-09-09 20:05:24 +00:00
#include <machine/in_cksum.h>
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
#include <dev/led/led.h>
Move includes from if_em.h to if_em.c and sort them.
2006-02-14 13:11:36 +00:00
#include <dev/pci/pcivar.h>
#include <dev/pci/pcireg.h>
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
#include "e1000_api.h"
Thanks to report from Neil Hoggarth I found a missing UNLOCK in the watchdog code. This delta also incorporates some missing PCI IDs that got added. PR 122928 - might be fixed by this, no verification at this point.
2008-05-21 21:06:17 +00:00
#include "e1000_82571.h"
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
#include "if_em.h"
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* Set this to one to display debug statistics
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*********************************************************************/
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
int em_display_debug_stats = 0;
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
* Driver version:
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*********************************************************************/
Sync the E1000 shared code with Intel internal, this adds fixes, and more importantly, new I218 adapter support to the em driver. MFC after: 1 week
2014-06-26 21:33:32 +00:00
char em_driver_version[] = "7.4.2";
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
* PCI Device ID Table
*
* Used by probe to select devices to load on
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
* Last field stores an index into e1000_strings
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
* Last entry must be all 0s
*
* { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index }
*********************************************************************/
Added support for 82541 and 82547 based adapters. - These have Intel gigabit PHY - 82547 uses CSA interface MFC after: 1 week
2003-03-21 21:47:31 +00:00
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
static em_vendor_info_t em_vendor_info_array[] =
{
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
/* Intel(R) PRO/1000 Network Connection */
{ 0x8086, E1000_DEV_ID_82571EB_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82571EB_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82571EB_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0},
Thanks to report from Neil Hoggarth I found a missing UNLOCK in the watchdog code. This delta also incorporates some missing PCI IDs that got added. PR 122928 - might be fixed by this, no verification at this point.
2008-05-21 21:06:17 +00:00
{ 0x8086, E1000_DEV_ID_82571EB_SERDES_DUAL,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82571EB_SERDES_QUAD,
PCI_ANY_ID, PCI_ANY_ID, 0},
Merge in new driver from Intel, version 6.1.4. It adds support for 82571EB quad port copper NIC and has few minor fixes. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15.
2006-08-11 10:58:24 +00:00
{ 0x8086, E1000_DEV_ID_82571EB_QUAD_COPPER,
PCI_ANY_ID, PCI_ANY_ID, 0},
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
{ 0x8086, E1000_DEV_ID_82571EB_QUAD_COPPER_LP,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82571EB_QUAD_FIBER,
Merge new vendor release - 6.2.9. Details: o if_em.c changes: - Added several new PCI ids. - Check em_check_phy_reset_block() before doing SIOCSIFMEDIA ioctl. - Don't touch TARC registers, they are now handled in shared code in if_em_hw.c. - Move RDH and RDT setting to the end of em_initialize_receive_unit(). - Declare em_read_pcie_cap_reg(), now empty. o if_em_hw.c dropped in from vendor, then restored rev. 1.15. o if_em_hw.h dropped in from vendor, then modified: - Added RX overrun interrupt flag to interrupt enable mask. - Remove declarations of em_io_read(), em_io_write(). Approved by: jfv
2006-10-31 15:00:14 +00:00
PCI_ANY_ID, PCI_ANY_ID, 0},
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
{ 0x8086, E1000_DEV_ID_82571PT_QUAD_COPPER,
PCI_ANY_ID, PCI_ANY_ID, 0},
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
{ 0x8086, E1000_DEV_ID_82572EI_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82572EI_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82572EI_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0},
Merge in new driver from Intel, version 5.1.5. Adds support for some new chips and improves support for already supported ones. Some details, important for future merges: - if_em.c merged manually, viewing diff between new vendor driver and previous one. - if_em_hw.h dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c dropped in from vendor, and then two liner change made, that restores support for two rare chips.
2006-04-06 17:09:03 +00:00
{ 0x8086, E1000_DEV_ID_82572EI, PCI_ANY_ID, PCI_ANY_ID, 0},
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
{ 0x8086, E1000_DEV_ID_82573E, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82573E_IAMT, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82573L, 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_82583V, PCI_ANY_ID, PCI_ANY_ID, 0},
Merge in new driver from Intel, version 6.0.5. It adds support for 80003 NICs and NICs found on ICH8 mobos, and improves support for already known chips. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. This was an easy task, because most changes between 5.1.5 and 6.0.5 are bugfixes taken from FreeBSD. - if_em_hw.h. Dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15. - if_em_osdep.h. Added new required macros from vendor file and add a hack against define namespace mangling in if_em_hw.h. Intel made another hack, but I prefer mine.
2006-08-03 09:20:11 +00:00
{ 0x8086, E1000_DEV_ID_80003ES2LAN_COPPER_SPT,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_80003ES2LAN_SERDES_SPT,
PCI_ANY_ID, PCI_ANY_ID, 0},
Merge in new driver from Intel, version 5.1.5. Adds support for some new chips and improves support for already supported ones. Some details, important for future merges: - if_em.c merged manually, viewing diff between new vendor driver and previous one. - if_em_hw.h dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c dropped in from vendor, and then two liner change made, that restores support for two rare chips.
2006-04-06 17:09:03 +00:00
{ 0x8086, E1000_DEV_ID_80003ES2LAN_COPPER_DPT,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_80003ES2LAN_SERDES_DPT,
PCI_ANY_ID, PCI_ANY_ID, 0},
Merge new vendor release - 6.2.9. Details: o if_em.c changes: - Added several new PCI ids. - Check em_check_phy_reset_block() before doing SIOCSIFMEDIA ioctl. - Don't touch TARC registers, they are now handled in shared code in if_em_hw.c. - Move RDH and RDT setting to the end of em_initialize_receive_unit(). - Declare em_read_pcie_cap_reg(), now empty. o if_em_hw.c dropped in from vendor, then restored rev. 1.15. o if_em_hw.h dropped in from vendor, then modified: - Added RX overrun interrupt flag to interrupt enable mask. - Remove declarations of em_io_read(), em_io_write(). Approved by: jfv
2006-10-31 15:00:14 +00:00
{ 0x8086, E1000_DEV_ID_ICH8_IGP_M_AMT, PCI_ANY_ID, PCI_ANY_ID, 0},
Merge in new driver from Intel, version 6.0.5. It adds support for 80003 NICs and NICs found on ICH8 mobos, and improves support for already known chips. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. This was an easy task, because most changes between 5.1.5 and 6.0.5 are bugfixes taken from FreeBSD. - if_em_hw.h. Dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15. - if_em_osdep.h. Added new required macros from vendor file and add a hack against define namespace mangling in if_em_hw.h. Intel made another hack, but I prefer mine.
2006-08-03 09:20:11 +00:00
{ 0x8086, E1000_DEV_ID_ICH8_IGP_AMT, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH8_IGP_C, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH8_IFE, PCI_ANY_ID, PCI_ANY_ID, 0},
Merge new vendor release - 6.2.9. Details: o if_em.c changes: - Added several new PCI ids. - Check em_check_phy_reset_block() before doing SIOCSIFMEDIA ioctl. - Don't touch TARC registers, they are now handled in shared code in if_em_hw.c. - Move RDH and RDT setting to the end of em_initialize_receive_unit(). - Declare em_read_pcie_cap_reg(), now empty. o if_em_hw.c dropped in from vendor, then restored rev. 1.15. o if_em_hw.h dropped in from vendor, then modified: - Added RX overrun interrupt flag to interrupt enable mask. - Remove declarations of em_io_read(), em_io_write(). Approved by: jfv
2006-10-31 15:00:14 +00:00
{ 0x8086, E1000_DEV_ID_ICH8_IFE_GT, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH8_IFE_G, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH8_IGP_M, 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_ICH8_82567V_3, PCI_ANY_ID, PCI_ANY_ID, 0},
Thanks to report from Neil Hoggarth I found a missing UNLOCK in the watchdog code. This delta also incorporates some missing PCI IDs that got added. PR 122928 - might be fixed by this, no verification at this point.
2008-05-21 21:06:17 +00:00
{ 0x8086, E1000_DEV_ID_ICH9_IGP_M_AMT, PCI_ANY_ID, PCI_ANY_ID, 0},
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
{ 0x8086, E1000_DEV_ID_ICH9_IGP_AMT, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH9_IGP_C, PCI_ANY_ID, PCI_ANY_ID, 0},
Thanks to report from Neil Hoggarth I found a missing UNLOCK in the watchdog code. This delta also incorporates some missing PCI IDs that got added. PR 122928 - might be fixed by this, no verification at this point.
2008-05-21 21:06:17 +00:00
{ 0x8086, E1000_DEV_ID_ICH9_IGP_M, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH9_IGP_M_V, PCI_ANY_ID, PCI_ANY_ID, 0},
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
{ 0x8086, E1000_DEV_ID_ICH9_IFE, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH9_IFE_GT, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH9_IFE_G, PCI_ANY_ID, PCI_ANY_ID, 0},
This small change will allow this driver in HEAD to build on 6.3 as well as 7 :)
2008-05-19 23:01:11 +00:00
{ 0x8086, E1000_DEV_ID_ICH9_BM, PCI_ANY_ID, PCI_ANY_ID, 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
{ 0x8086, E1000_DEV_ID_82574L, 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_82574LA, PCI_ANY_ID, PCI_ANY_ID, 0},
Thanks to report from Neil Hoggarth I found a missing UNLOCK in the watchdog code. This delta also incorporates some missing PCI IDs that got added. PR 122928 - might be fixed by this, no verification at this point.
2008-05-21 21:06:17 +00:00
{ 0x8086, E1000_DEV_ID_ICH10_R_BM_LM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH10_R_BM_LF, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH10_R_BM_V, PCI_ANY_ID, PCI_ANY_ID, 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
{ 0x8086, E1000_DEV_ID_ICH10_D_BM_LM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_ICH10_D_BM_LF, 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_ICH10_D_BM_V, 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_PCH_M_HV_LM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_PCH_M_HV_LC, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_PCH_D_HV_DM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_PCH_D_HV_DC, PCI_ANY_ID, PCI_ANY_ID, 0},
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
{ 0x8086, E1000_DEV_ID_PCH2_LV_LM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_PCH2_LV_V, PCI_ANY_ID, PCI_ANY_ID, 0},
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
{ 0x8086, E1000_DEV_ID_PCH_LPT_I217_LM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_PCH_LPT_I217_V, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_PCH_LPTLP_I218_LM,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_PCH_LPTLP_I218_V,
PCI_ANY_ID, PCI_ANY_ID, 0},
Sync the E1000 shared code with Intel internal, this adds fixes, and more importantly, new I218 adapter support to the em driver. MFC after: 1 week
2014-06-26 21:33:32 +00:00
{ 0x8086, E1000_DEV_ID_PCH_I218_LM2, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_PCH_I218_V2, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_PCH_I218_LM3, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_PCH_I218_V3, PCI_ANY_ID, PCI_ANY_ID, 0},
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
/* required last entry */
{ 0, 0, 0, 0, 0}
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
};
/*********************************************************************
* Table of branding strings for all supported NICs.
*********************************************************************/
static char *em_strings[] = {
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
"Intel(R) PRO/1000 Network Connection"
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
};
/*********************************************************************
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* Function prototypes
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*********************************************************************/
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
static int em_probe(device_t);
static int em_attach(device_t);
static int em_detach(device_t);
static int em_shutdown(device_t);
static int em_suspend(device_t);
static int em_resume(device_t);
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
#ifdef EM_MULTIQUEUE
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
static int em_mq_start(if_t, struct mbuf *);
static int em_mq_start_locked(if_t,
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
struct tx_ring *);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
static void em_qflush(if_t);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#else
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
static void em_start(if_t);
static void em_start_locked(if_t, 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
#endif
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
static int em_ioctl(if_t, u_long, caddr_t);
- Use if_inc_counter() to increment various counters. - Do not ever set a counter to a value. For those counters that we don't increment, but return directly from hardware create cases in if_get_counter() method. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-09-18 15:56:14 +00:00
static uint64_t em_get_counter(if_t, ift_counter);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
static void em_init(void *);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static void em_init_locked(struct adapter *);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
static void em_stop(void *);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
static void em_media_status(if_t, struct ifmediareq *);
static int em_media_change(if_t);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static void em_identify_hardware(struct adapter *);
static int em_allocate_pci_resources(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 int em_allocate_legacy(struct adapter *);
static int em_allocate_msix(struct adapter *);
static int em_allocate_queues(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 em_setup_msix(struct adapter *);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static void em_free_pci_resources(struct adapter *);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
static void em_local_timer(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
static void em_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 em_setup_interface(device_t, 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
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
static void em_setup_transmit_structures(struct adapter *);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static void em_initialize_transmit_unit(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 int em_allocate_transmit_buffers(struct tx_ring *);
static void em_free_transmit_structures(struct adapter *);
static void em_free_transmit_buffers(struct tx_ring *);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static int em_setup_receive_structures(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 int em_allocate_receive_buffers(struct rx_ring *);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static void em_initialize_receive_unit(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 em_free_receive_structures(struct adapter *);
static void em_free_receive_buffers(struct rx_ring *);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static void em_enable_intr(struct adapter *);
static void em_disable_intr(struct adapter *);
static void em_update_stats_counters(struct adapter *);
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 20:57:41 +00:00
static void em_add_hw_stats(struct adapter *adapter);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
static void em_txeof(struct tx_ring *);
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 em_rxeof(struct rx_ring *, int, int *);
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
#ifndef __NO_STRICT_ALIGNMENT
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 int em_fixup_rx(struct rx_ring *);
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
#endif
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
static void em_receive_checksum(struct e1000_rx_desc *, struct mbuf *);
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
static void em_transmit_checksum_setup(struct tx_ring *, struct mbuf *, int,
struct ip *, u32 *, u32 *);
static void em_tso_setup(struct tx_ring *, struct mbuf *, int, struct ip *,
struct tcphdr *, u32 *, u32 *);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static void em_set_promisc(struct adapter *);
static void em_disable_promisc(struct adapter *);
static void em_set_multi(struct adapter *);
static void em_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 em_refresh_mbufs(struct rx_ring *, int);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
static void em_register_vlan(void *, if_t, u16);
static void em_unregister_vlan(void *, if_t, 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 em_setup_vlan_hw_support(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 int em_xmit(struct tx_ring *, struct mbuf **);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static int em_dma_malloc(struct adapter *, bus_size_t,
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
struct em_dma_alloc *, int);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static void em_dma_free(struct adapter *, struct em_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 20:57:41 +00:00
static int em_sysctl_nvm_info(SYSCTL_HANDLER_ARGS);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
static void em_print_nvm_info(struct adapter *);
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
static int em_sysctl_debug_info(SYSCTL_HANDLER_ARGS);
static void em_print_debug_info(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 em_is_valid_ether_addr(u8 *);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
static int em_sysctl_int_delay(SYSCTL_HANDLER_ARGS);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
static void em_add_int_delay_sysctl(struct adapter *, const char *,
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
const char *, struct em_int_delay_info *, int, int);
/* Management and WOL Support */
static void em_init_manageability(struct adapter *);
static void em_release_manageability(struct adapter *);
static void em_get_hw_control(struct adapter *);
static void em_release_hw_control(struct 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
static void em_get_wakeup(device_t);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
static void em_enable_wakeup(device_t);
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 em_enable_phy_wakeup(struct adapter *);
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 em_led_func(void *, int);
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
static void em_disable_aspm(struct adapter *);
The procedure of raceless switching between polling mode and taskqueued interrupt mode is going to be quite complex. Since the polling mode is considered legacy feature for em(4) driver, the decision is made to make polling and new interrupt handler mutually exclusive, selected at compile time. If kernel is compiled with DEVICE_POLLING, the fast taskqueued interrupt handler code is disabled and the em_poll() and legacy em_intr() functions are enabled. Otherwise, legacy functions are disabled and only em_intr_fast() code is compiled. Discussed with: scottl
2006-06-06 08:03:49 +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 int em_irq_fast(void *);
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 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
/* MSIX handlers */
static void em_msix_tx(void *);
static void em_msix_rx(void *);
static void em_msix_link(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
static void em_handle_tx(void *context, int pending);
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 em_handle_rx(void *context, int pending);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
static void em_handle_link(void *context, int pending);
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
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#ifdef EM_MULTIQUEUE
static void em_enable_vectors_82574(struct adapter *);
#endif
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 em_set_sysctl_value(struct adapter *, const char *,
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
const char *, int *, int);
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 int em_set_flowcntl(SYSCTL_HANDLER_ARGS);
Change the interface to the Energy Efficient Ethernet (EEE) setting in the igb and em driver. This was necessitated by a shared code change that I was given late in the game, a data type changed from bool to int, in the last update I dealt with it by a cast, but it was pointed out (thanks jhb) that there was a potential problem with this. John suggested this safer approach, and it is fine with me... MFC after:2 days (to catch the 9.1 update)
2012-07-07 20:21:05 +00:00
static int em_sysctl_eee(SYSCTL_HANDLER_ARGS);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Tighten up the rx mbuf refresh code, there were some discrepencies from the igb version which was the target. Change the message when neither MSI or MSIX are enabled and a fallback to Legacy interrupts happen, the existing message was confusing.
2010-09-07 20:13:08 +00:00
static __inline void em_rx_discard(struct rx_ring *, int);
Take out em_poll() prototype from under EM_FAST_IRQ control. Reported by: tindebox compiling a LINT kernel
2007-11-21 12:55:33 +00:00
#ifdef DEVICE_POLLING
- Remove empty wrappers ether_poll_[de]register_drv(). [1] - Move polling(9) declarations out of ifq.h back to if_var.h they are absolutely unrelated to queues. Submitted by: Mikhail <mp lenta.ru> [1]
2014-09-28 14:05:18 +00:00
static poll_handler_t em_poll;
Take out em_poll() prototype from under EM_FAST_IRQ control. Reported by: tindebox compiling a LINT kernel
2007-11-21 12:55:33 +00:00
#endif /* POLLING */
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* FreeBSD Device Interface Entry Points
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*********************************************************************/
static device_method_t em_methods[] = {
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
/* Device interface */
DEVMETHOD(device_probe, em_probe),
DEVMETHOD(device_attach, em_attach),
DEVMETHOD(device_detach, em_detach),
DEVMETHOD(device_shutdown, em_shutdown),
Add simple suspend and resume methods. We call em_stop() on suspend and em_init() on resume. With this change the network is ready right after resume, without half minute lag. Tested by: Jacques Garrigue
2005-12-26 10:39:21 +00:00
DEVMETHOD(device_suspend, em_suspend),
DEVMETHOD(device_resume, em_resume),
Use DEVMETHOD_END macro defined in sys/bus.h instead of {0, 0} sentinel on device_method_t arrays Reviewed by: cognet Approved by: cognet
2013-01-30 18:01:20 +00:00
DEVMETHOD_END
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
};
static driver_t em_driver = {
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
"em", em_methods, sizeof(struct adapter),
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
devclass_t em_devclass;
- Express hard dependencies on bus (pci, isa, pccard) and network layer (ether). - Don't abuse module names to facilitate ifconfig module loading; such abuse isn't really needed. (And if we do need type information associated with a module then we should make it explicit and not use hacks.)
2003-04-15 06:37:30 +00:00
DRIVER_MODULE(em, pci, em_driver, em_devclass, 0, 0);
MODULE_DEPEND(em, pci, 1, 1, 1);
MODULE_DEPEND(em, ether, 1, 1, 1);
Sync netmap sources with the version in our private tree. This commit contains large contributions from Giuseppe Lettieri and Stefano Garzarella, is partly supported by grants from Verisign and Cisco, and brings in the following: - fix zerocopy monitor ports and introduce copying monitor ports (the latter are lower performance but give access to all traffic in parallel with the application) - exclusive open mode, useful to implement solutions that recover from crashes of the main netmap client (suggested by Patrick Kelsey) - revised memory allocator in preparation for the 'passthrough mode' (ptnetmap) recently presented at bsdcan. ptnetmap is described in S. Garzarella, G. Lettieri, L. Rizzo; Virtual device passthrough for high speed VM networking, ACM/IEEE ANCS 2015, Oakland (CA) May 2015 http://info.iet.unipi.it/~luigi/research.html - fix rx CRC handing on ixl - add module dependencies for netmap when building drivers as modules - minor simplifications to device-specific routines (*txsync, *rxsync) - general code cleanup (remove unused variables, introduce macros to access rings and remove duplicate code, Applications do not need to be recompiled, unless of course they want to use the new features (monitors and exclusive open). Those willing to try this code on stable/10 can just update the sys/dev/netmap/*, sys/net/netmap* with the version in HEAD and apply the small patches to individual device drivers. MFC after: 1 month Sponsored by: (partly) Verisign, Cisco
2015-07-10 05:51:36 +00:00
#ifdef DEV_NETMAP
MODULE_DEPEND(em, netmap, 1, 1, 1);
#endif /* DEV_NETMAP */
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
/*********************************************************************
* Tunable default values.
*********************************************************************/
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
#define EM_TICKS_TO_USECS(ticks) ((1024 * (ticks) + 500) / 1000)
#define EM_USECS_TO_TICKS(usecs) ((1000 * (usecs) + 512) / 1024)
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
#define M_TSO_LEN 66
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
if_lem.c: make sure that lem_rxeof() can drain the entire rx queue irrespective of the setting of lem_rx_process_limit, while giving a chance to the taskqueue scheduler to act after each chunk. This makes lem_rxeof similar to the one in if_em.c and if_igb.c . if_lem.c and if_em.c: add a sysctl to manually configure the 'itr' moderation register. Approved by: Jack Vogel
2013-05-09 17:07:30 +00:00
#define MAX_INTS_PER_SEC 8000
#define DEFAULT_ITR (1000000000/(MAX_INTS_PER_SEC * 256))
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
/* Allow common code without TSO */
#ifndef CSUM_TSO
#define CSUM_TSO 0
#endif
Mark all SYSCTL_NODEs static that have no corresponding SYSCTL_DECLs. The SYSCTL_NODE macro defines a list that stores all child-elements of that node. If there's no SYSCTL_DECL macro anywhere else, there's no reason why it shouldn't be static.
2011-11-07 15:43:11 +00:00
static SYSCTL_NODE(_hw, OID_AUTO, em, CTLFLAG_RD, 0, "EM driver parameters");
- 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. Reviewed by: jfv (earlier version) MFC after: 1 week
2011-06-29 16:20:52 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
static int em_tx_int_delay_dflt = EM_TICKS_TO_USECS(EM_TIDV);
static int em_rx_int_delay_dflt = EM_TICKS_TO_USECS(EM_RDTR);
- 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. Reviewed by: jfv (earlier version) MFC after: 1 week
2011-06-29 16:20:52 +00:00
SYSCTL_INT(_hw_em, OID_AUTO, tx_int_delay, CTLFLAG_RDTUN, &em_tx_int_delay_dflt,
0, "Default transmit interrupt delay in usecs");
SYSCTL_INT(_hw_em, OID_AUTO, rx_int_delay, CTLFLAG_RDTUN, &em_rx_int_delay_dflt,
0, "Default receive interrupt delay in usecs");
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 int em_tx_abs_int_delay_dflt = EM_TICKS_TO_USECS(EM_TADV);
static int em_rx_abs_int_delay_dflt = EM_TICKS_TO_USECS(EM_RADV);
- 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. Reviewed by: jfv (earlier version) MFC after: 1 week
2011-06-29 16:20:52 +00:00
SYSCTL_INT(_hw_em, OID_AUTO, tx_abs_int_delay, CTLFLAG_RDTUN,
&em_tx_abs_int_delay_dflt, 0,
"Default transmit interrupt delay limit in usecs");
SYSCTL_INT(_hw_em, OID_AUTO, rx_abs_int_delay, CTLFLAG_RDTUN,
&em_rx_abs_int_delay_dflt, 0,
"Default receive interrupt delay limit in usecs");
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 int em_rxd = EM_DEFAULT_RXD;
static int em_txd = EM_DEFAULT_TXD;
- 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. Reviewed by: jfv (earlier version) MFC after: 1 week
2011-06-29 16:20:52 +00:00
SYSCTL_INT(_hw_em, OID_AUTO, rxd, CTLFLAG_RDTUN, &em_rxd, 0,
"Number of receive descriptors per queue");
SYSCTL_INT(_hw_em, OID_AUTO, txd, CTLFLAG_RDTUN, &em_txd, 0,
"Number of transmit descriptors per queue");
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 int em_smart_pwr_down = FALSE;
- 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. Reviewed by: jfv (earlier version) MFC after: 1 week
2011-06-29 16:20:52 +00:00
SYSCTL_INT(_hw_em, OID_AUTO, smart_pwr_down, CTLFLAG_RDTUN, &em_smart_pwr_down,
0, "Set to true to leave smart power down enabled on newer adapters");
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
/* Controls whether promiscuous also shows bad packets */
static int em_debug_sbp = FALSE;
- 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. Reviewed by: jfv (earlier version) MFC after: 1 week
2011-06-29 16:20:52 +00:00
SYSCTL_INT(_hw_em, OID_AUTO, sbp, CTLFLAG_RDTUN, &em_debug_sbp, 0,
"Show bad packets in promiscuous mode");
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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 int em_enable_msix = TRUE;
- 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. Reviewed by: jfv (earlier version) MFC after: 1 week
2011-06-29 16:20:52 +00:00
SYSCTL_INT(_hw_em, OID_AUTO, enable_msix, CTLFLAG_RDTUN, &em_enable_msix, 0,
"Enable MSI-X interrupts");
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
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#ifdef EM_MULTIQUEUE
static int em_num_queues = 1;
SYSCTL_INT(_hw_em, OID_AUTO, num_queues, CTLFLAG_RDTUN, &em_num_queues, 0,
"82574 only: Number of queues to configure, 0 indicates autoconfigure");
#endif
/*
** Global variable to store last used CPU when binding queues
** to CPUs in igb_allocate_msix. Starts at CPU_FIRST and increments when a
** queue is bound to a cpu.
*/
static int em_last_bind_cpu = -1;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* How many packets rxeof tries to clean at a time */
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
static int em_rx_process_limit = 100;
- 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. Reviewed by: jfv (earlier version) MFC after: 1 week
2011-06-29 16:20:52 +00:00
SYSCTL_INT(_hw_em, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN,
&em_rx_process_limit, 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
"Maximum number of received packets to process "
"at a time, -1 means unlimited");
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 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
/* Energy efficient ethernet - default to OFF */
Change the interface to the Energy Efficient Ethernet (EEE) setting in the igb and em driver. This was necessitated by a shared code change that I was given late in the game, a data type changed from bool to int, in the last update I dealt with it by a cast, but it was pointed out (thanks jhb) that there was a potential problem with this. John suggested this safer approach, and it is fine with me... MFC after:2 days (to catch the 9.1 update)
2012-07-07 20:21:05 +00:00
static int eee_setting = 1;
- 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. Reviewed by: jfv (earlier version) MFC after: 1 week
2011-06-29 16:20:52 +00:00
SYSCTL_INT(_hw_em, OID_AUTO, eee_setting, CTLFLAG_RDTUN, &eee_setting, 0,
"Enable Energy Efficient Ethernet");
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
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* Global used in WOL setup with multiport cards */
static int global_quad_port_a = 0;
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP /* see ixgbe.c for details */
#include <dev/netmap/if_em_netmap.h>
#endif /* DEV_NETMAP */
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
* Device identification routine
*
* em_probe determines if the driver should be loaded on
* adapter based on PCI vendor/device id of the adapter.
*
Use BUS_PROBE_DEFAULT for pci probe return value
2005-03-05 18:30:12 +00:00
* return BUS_PROBE_DEFAULT on success, positive on failure
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*********************************************************************/
static int
em_probe(device_t dev)
{
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
char adapter_name[60];
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
uint16_t pci_vendor_id = 0;
uint16_t pci_device_id = 0;
uint16_t pci_subvendor_id = 0;
uint16_t pci_subdevice_id = 0;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
em_vendor_info_t *ent;
INIT_DEBUGOUT("em_probe: begin");
pci_vendor_id = pci_get_vendor(dev);
if (pci_vendor_id != EM_VENDOR_ID)
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (ENXIO);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
pci_device_id = pci_get_device(dev);
pci_subvendor_id = pci_get_subvendor(dev);
pci_subdevice_id = pci_get_subdevice(dev);
ent = em_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) ||
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
(ent->subvendor_id == PCI_ANY_ID)) &&
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
((pci_subdevice_id == ent->subdevice_id) ||
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
(ent->subdevice_id == PCI_ANY_ID))) {
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
sprintf(adapter_name, "%s %s",
em_strings[ent->index],
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
em_driver_version);
device_set_desc_copy(dev, adapter_name);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (BUS_PROBE_DEFAULT);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
}
ent++;
}
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (ENXIO);
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
/*********************************************************************
* Device initialization routine
*
* The attach entry point is called when the driver is being loaded.
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* This routine identifies the type of hardware, allocates all resources
* and initializes the hardware.
*
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
* return 0 on success, positive on failure
*********************************************************************/
static int
em_attach(device_t dev)
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
struct adapter *adapter;
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
struct e1000_hw *hw;
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
int error = 0;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
INIT_DEBUGOUT("em_attach: begin");
This isn't functionally identical. In some cases a hint to disable unit 0 would in fact disable all units. This reverts r241856 Approved by: cperciva (implicit)
2012-10-22 13:06:09 +00:00
if (resource_disabled("em", device_get_unit(dev))) {
device_printf(dev, "Disabled by device hint\n");
return (ENXIO);
}
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter = device_get_softc(dev);
adapter->dev = adapter->osdep.dev = dev;
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
hw = &adapter->hw;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK_INIT(adapter, device_get_nameunit(dev));
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
/* SYSCTL stuff */
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
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 20:57:41 +00:00
OID_AUTO, "nvm", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
em_sysctl_nvm_info, "I", "NVM Information");
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
OID_AUTO, "debug", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
em_sysctl_debug_info, "I", "Debug Information");
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
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
OID_AUTO, "fc", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
em_set_flowcntl, "I", "Flow Control");
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
callout_init_mtx(&adapter->timer, &adapter->core_mtx, 0);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Couple of changes, back down on last TSO change, instead make old adapter list still capable, but only PCI-E adapters are now enabled. The user can enable older PCI-X or PCI adapters using ifconfig. Secondly, Arthur Hartwig pointed out my MSI change was not working correctly, changed to something that now does. Thanks Arthur. There was also a fundamental bug in the 82575 MSIX code, the MSIX registers had to be mapped, opps :) Rubber-stamped by: Pdeuskar
2007-05-17 00:14:03 +00:00
/* Determine hardware and mac info */
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_identify_hardware(adapter);
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* Setup PCI resources */
if (em_allocate_pci_resources(adapter)) {
device_printf(dev, "Allocation of PCI resources failed\n");
error = ENXIO;
goto err_pci;
}
/*
** For ICH8 and family we need to
** map the flash memory, and this
** must happen after the MAC is
** identified
*/
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 ((hw->mac.type == e1000_ich8lan) ||
(hw->mac.type == e1000_ich9lan) ||
(hw->mac.type == e1000_ich10lan) ||
(hw->mac.type == e1000_pchlan) ||
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
(hw->mac.type == e1000_pch2lan) ||
(hw->mac.type == e1000_pch_lpt)) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
int rid = EM_BAR_TYPE_FLASH;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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->flash = bus_alloc_resource_any(dev,
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
SYS_RES_MEMORY, &rid, RF_ACTIVE);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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->flash == NULL) {
device_printf(dev, "Mapping of Flash failed\n");
error = ENXIO;
goto err_pci;
}
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* This is used in the shared code */
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
hw->flash_address = (u8 *)adapter->flash;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->osdep.flash_bus_space_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
rman_get_bustag(adapter->flash);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->osdep.flash_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
rman_get_bushandle(adapter->flash);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
}
/* Do Shared Code initialization */
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 (e1000_setup_init_funcs(hw, TRUE)) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
device_printf(dev, "Setup of Shared code failed\n");
error = ENXIO;
goto err_pci;
}
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
/*
* Setup MSI/X or MSI if PCI Express
*/
adapter->msix = em_setup_msix(adapter);
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_get_bus_info(hw);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
/* Set up some sysctls for the tunable interrupt delays */
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_add_int_delay_sysctl(adapter, "rx_int_delay",
"receive interrupt delay in usecs", &adapter->rx_int_delay,
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_REGISTER(hw, E1000_RDTR), em_rx_int_delay_dflt);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_add_int_delay_sysctl(adapter, "tx_int_delay",
"transmit interrupt delay in usecs", &adapter->tx_int_delay,
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_REGISTER(hw, E1000_TIDV), em_tx_int_delay_dflt);
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
em_add_int_delay_sysctl(adapter, "rx_abs_int_delay",
"receive interrupt delay limit in usecs",
&adapter->rx_abs_int_delay,
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_REGISTER(hw, E1000_RADV),
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
em_rx_abs_int_delay_dflt);
em_add_int_delay_sysctl(adapter, "tx_abs_int_delay",
"transmit interrupt delay limit in usecs",
&adapter->tx_abs_int_delay,
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_REGISTER(hw, E1000_TADV),
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
em_tx_abs_int_delay_dflt);
if_lem.c: make sure that lem_rxeof() can drain the entire rx queue irrespective of the setting of lem_rx_process_limit, while giving a chance to the taskqueue scheduler to act after each chunk. This makes lem_rxeof similar to the one in if_em.c and if_igb.c . if_lem.c and if_em.c: add a sysctl to manually configure the 'itr' moderation register. Approved by: Jack Vogel
2013-05-09 17:07:30 +00:00
em_add_int_delay_sysctl(adapter, "itr",
"interrupt delay limit in usecs/4",
&adapter->tx_itr,
E1000_REGISTER(hw, E1000_ITR),
DEFAULT_ITR);
- Backout last change, since it is memory overkill for a non busy host or for a notebook with em(4) adapter. - Introduce tunables em.hw.txd and em.hw.rxd, which allow administrator to configure number of transmit and receive descriptors. - Check em.hw.txd and em.hw.rxd against hardware limits [*] and require them to be multiple of 128. [*] According to comments in if_em.h the 82540EM/82541ER chips can handle more than 256 descriptors. Since we don't have this hardware to test, we decided to mimic NetBSD wm(4) driver, that limits these chips to 256 descriptors. In collaboration with: yongari
2005-11-17 10:13:18 +00:00
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
/* Sysctl for limiting the amount of work done in the taskqueue */
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
em_set_sysctl_value(adapter, "rx_processing_limit",
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
"max number of rx packets to process", &adapter->rx_process_limit,
em_rx_process_limit);
- Backout last change, since it is memory overkill for a non busy host or for a notebook with em(4) adapter. - Introduce tunables em.hw.txd and em.hw.rxd, which allow administrator to configure number of transmit and receive descriptors. - Check em.hw.txd and em.hw.rxd against hardware limits [*] and require them to be multiple of 128. [*] According to comments in if_em.h the 82540EM/82541ER chips can handle more than 256 descriptors. Since we don't have this hardware to test, we decided to mimic NetBSD wm(4) driver, that limits these chips to 256 descriptors. In collaboration with: yongari
2005-11-17 10:13:18 +00:00
/*
* Validate number of transmit and receive descriptors. It
* must not exceed hardware maximum, and must be multiple
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
* of E1000_DBA_ALIGN.
- Backout last change, since it is memory overkill for a non busy host or for a notebook with em(4) adapter. - Introduce tunables em.hw.txd and em.hw.rxd, which allow administrator to configure number of transmit and receive descriptors. - Check em.hw.txd and em.hw.rxd against hardware limits [*] and require them to be multiple of 128. [*] According to comments in if_em.h the 82540EM/82541ER chips can handle more than 256 descriptors. Since we don't have this hardware to test, we decided to mimic NetBSD wm(4) driver, that limits these chips to 256 descriptors. In collaboration with: yongari
2005-11-17 10:13:18 +00:00
*/
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (((em_txd * sizeof(struct e1000_tx_desc)) % EM_DBA_ALIGN) != 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
(em_txd > EM_MAX_TXD) || (em_txd < EM_MIN_TXD)) {
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
device_printf(dev, "Using %d TX descriptors instead of %d!\n",
EM_DEFAULT_TXD, em_txd);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter->num_tx_desc = EM_DEFAULT_TXD;
- Backout last change, since it is memory overkill for a non busy host or for a notebook with em(4) adapter. - Introduce tunables em.hw.txd and em.hw.rxd, which allow administrator to configure number of transmit and receive descriptors. - Check em.hw.txd and em.hw.rxd against hardware limits [*] and require them to be multiple of 128. [*] According to comments in if_em.h the 82540EM/82541ER chips can handle more than 256 descriptors. Since we don't have this hardware to test, we decided to mimic NetBSD wm(4) driver, that limits these chips to 256 descriptors. In collaboration with: yongari
2005-11-17 10:13:18 +00:00
} else
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter->num_tx_desc = em_txd;
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
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (((em_rxd * sizeof(struct e1000_rx_desc)) % EM_DBA_ALIGN) != 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
(em_rxd > EM_MAX_RXD) || (em_rxd < EM_MIN_RXD)) {
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
device_printf(dev, "Using %d RX descriptors instead of %d!\n",
EM_DEFAULT_RXD, em_rxd);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter->num_rx_desc = EM_DEFAULT_RXD;
- Backout last change, since it is memory overkill for a non busy host or for a notebook with em(4) adapter. - Introduce tunables em.hw.txd and em.hw.rxd, which allow administrator to configure number of transmit and receive descriptors. - Check em.hw.txd and em.hw.rxd against hardware limits [*] and require them to be multiple of 128. [*] According to comments in if_em.h the 82540EM/82541ER chips can handle more than 256 descriptors. Since we don't have this hardware to test, we decided to mimic NetBSD wm(4) driver, that limits these chips to 256 descriptors. In collaboration with: yongari
2005-11-17 10:13:18 +00:00
} else
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter->num_rx_desc = em_rxd;
- Backout last change, since it is memory overkill for a non busy host or for a notebook with em(4) adapter. - Introduce tunables em.hw.txd and em.hw.rxd, which allow administrator to configure number of transmit and receive descriptors. - Check em.hw.txd and em.hw.rxd against hardware limits [*] and require them to be multiple of 128. [*] According to comments in if_em.h the 82540EM/82541ER chips can handle more than 256 descriptors. Since we don't have this hardware to test, we decided to mimic NetBSD wm(4) driver, that limits these chips to 256 descriptors. In collaboration with: yongari
2005-11-17 10:13:18 +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
hw->mac.autoneg = DO_AUTO_NEG;
hw->phy.autoneg_wait_to_complete = FALSE;
hw->phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* Copper options */
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 (hw->phy.media_type == e1000_media_type_copper) {
hw->phy.mdix = AUTO_ALL_MODES;
hw->phy.disable_polarity_correction = FALSE;
hw->phy.ms_type = EM_MASTER_SLAVE;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
}
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
/*
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
* Set the frame limits assuming
* standard ethernet sized frames.
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
*/
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
adapter->hw.mac.max_frame_size =
ETHERMTU + ETHER_HDR_LEN + ETHERNET_FCS_SIZE;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
/*
* This controls when hardware reports transmit completion
* status.
Added support for 82541 and 82547 based adapters. - These have Intel gigabit PHY - 82547 uses CSA interface MFC after: 1 week
2003-03-21 21:47:31 +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
hw->mac.report_tx_early = 1;
Added support for 82541 and 82547 based adapters. - These have Intel gigabit PHY - 82547 uses CSA interface MFC after: 1 week
2003-03-21 21:47:31 +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
/*
** Get queue/ring memory
*/
if (em_allocate_queues(adapter)) {
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
error = ENOMEM;
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 err_pci;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
}
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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;
}
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
/* Check SOL/IDER usage */
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 (e1000_check_reset_block(hw))
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
device_printf(dev, "PHY reset is blocked"
" due to SOL/IDER session.\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
/* Sysctl for setting Energy Efficient Ethernet */
Change the interface to the Energy Efficient Ethernet (EEE) setting in the igb and em driver. This was necessitated by a shared code change that I was given late in the game, a data type changed from bool to int, in the last update I dealt with it by a cast, but it was pointed out (thanks jhb) that there was a potential problem with this. John suggested this safer approach, and it is fine with me... MFC after:2 days (to catch the 9.1 update)
2012-07-07 20:21:05 +00:00
hw->dev_spec.ich8lan.eee_disable = eee_setting;
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
OID_AUTO, "eee_control", CTLTYPE_INT|CTLFLAG_RW,
adapter, 0, em_sysctl_eee, "I",
"Disable Energy Efficient Ethernet");
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
This delta syncs the em and igb drivers with Intel, adds header split and SCTP support into the igb driver. Various small improvements and fixes. MFC after: 2 weeks
2009-04-10 00:05:46 +00:00
/*
** Start from a known state, this is
** important in reading the nvm and
** mac from that.
*/
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_reset_hw(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
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* Make sure we have a good EEPROM before we read from it */
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 (e1000_validate_nvm_checksum(hw) < 0) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/*
** Some PCI-E parts fail the first check due to
** the link being in sleep state, call it again,
** if it fails a second time its a real issue.
*/
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 (e1000_validate_nvm_checksum(hw) < 0) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
device_printf(dev,
"The EEPROM Checksum Is Not Valid\n");
error = EIO;
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 err_late;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
}
}
- Use IO mode to reset the controller (82544 and beyond) - Read the Mac address only once during attach. (This fixes the failover issue observed using the bonding driver) MFC after: 3 days
2002-07-16 16:55:03 +00:00
/* Copy the permanent MAC address out of the EEPROM */
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 (e1000_read_mac_addr(hw) < 0) {
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
device_printf(dev, "EEPROM read error while reading MAC"
" address\n");
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
error = EIO;
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 err_late;
- Use IO mode to reset the controller (82544 and beyond) - Read the Mac address only once during attach. (This fixes the failover issue observed using the bonding driver) MFC after: 3 days
2002-07-16 16:55:03 +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
if (!em_is_valid_ether_addr(hw->mac.addr)) {
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
device_printf(dev, "Invalid MAC address\n");
error = EIO;
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 err_late;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
}
Sync the E1000 shared code with Intel internal, this adds fixes, and more importantly, new I218 adapter support to the em driver. MFC after: 1 week
2014-06-26 21:33:32 +00:00
/* Disable ULP support */
e1000_disable_ulp_lpt_lp(hw, 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
/*
** Do interrupt configuration
*/
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 (adapter->msix > 1) /* Do MSIX */
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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 = em_allocate_msix(adapter);
else /* MSI or Legacy */
error = em_allocate_legacy(adapter);
if (error)
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 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
/*
* Get Wake-on-Lan and Management info for later use
*/
em_get_wakeup(dev);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
/* 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 (em_setup_interface(dev, adapter) != 0)
goto err_late;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
em_reset(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
/* Initialize statistics */
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_update_stats_counters(adapter);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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
hw->mac.get_link_status = 1;
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_update_link_status(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
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
/* 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
em_register_vlan, adapter, EVENTHANDLER_PRI_FIRST);
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
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
em_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST);
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
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 20:57:41 +00:00
em_add_hw_stats(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
/* Non-AMT based hardware can now take control from firmware */
if (adapter->has_manage && !adapter->has_amt)
em_get_hw_control(adapter);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* Tell the stack that the interface is not active */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setdrvflagbits(adapter->ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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(em_led_func, adapter,
device_get_nameunit(dev));
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
em_netmap_attach(adapter);
#endif /* DEV_NETMAP */
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
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
INIT_DEBUGOUT("em_attach: end");
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (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
err_late:
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
em_free_transmit_structures(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
em_free_receive_structures(adapter);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
em_release_hw_control(adapter);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (adapter->ifp != (void *)NULL)
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
if_free(adapter->ifp);
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
err_pci:
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_free_pci_resources(adapter);
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);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK_DESTROY(adapter);
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (error);
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
/*********************************************************************
* 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.
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
*
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
* return 0 on success, positive on failure
*********************************************************************/
static int
em_detach(device_t dev)
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
struct adapter *adapter = device_get_softc(dev);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
INIT_DEBUGOUT("em_detach: begin");
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
/* Make sure VLANS are not using driver */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_vlantrunkinuse(ifp)) {
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
device_printf(dev,"Vlan in use, detach first\n");
return (EBUSY);
}
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
#ifdef DEVICE_POLLING
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getcapenable(ifp) & IFCAP_POLLING)
- Remove empty wrappers ether_poll_[de]register_drv(). [1] - Move polling(9) declarations out of ifq.h back to if_var.h they are absolutely unrelated to queues. Submitted by: Mikhail <mp lenta.ru> [1]
2014-09-28 14:05:18 +00:00
ether_poll_deregister(ifp);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
#endif
Fix a mismerge in r206001. PR: 146614 Approved by: jfv (implicit) MFC afer: 3 days
2010-05-15 19:46:16 +00:00
if (adapter->led_dev != NULL)
led_destroy(adapter->led_dev);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter->in_detach = 1;
em_stop(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
EM_CORE_UNLOCK(adapter);
EM_CORE_LOCK_DESTROY(adapter);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
e1000_phy_hw_reset(&adapter->hw);
em_release_manageability(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
em_release_hw_control(adapter);
A few small but significant fixes: - Coverity Prevent(tm) CID 1906 a bogus use of bzero where unneeded. - ICH8 systems autoneg to 100 rather than 1000, this can also be seen in 82573, the logic was backwards. - On new 82575 quadports half duplex tx speed is slow... this was due to overwriting TCTL reg rather than adding bits.
2007-05-30 23:32:21 +00:00
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
/* 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);
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
ether_ifdetach(adapter->ifp);
- Use callout_init_mtx() to close various callout-related races. - Drain the two timers in detach. - Check IFF_DRV_RUNNING in the link task and bail w/o doing anything if it is clear. Reviewed by: jfv, scottl
2006-10-31 17:05:02 +00:00
callout_drain(&adapter->timer);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
netmap_detach(ifp);
#endif /* DEV_NETMAP */
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_free_pci_resources(adapter);
Only reset the phy when it is absolutely required. This should fix the issues with long *init* times when you do ifconfig em0 alias. MFC after: 3 days
2004-02-10 21:31:09 +00:00
bus_generic_detach(dev);
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
if_free(ifp);
Only reset the phy when it is absolutely required. This should fix the issues with long *init* times when you do ifconfig em0 alias. MFC after: 3 days
2004-02-10 21:31:09 +00:00
Allocate receive and transmit data structures during attach() and free them during detach() similar to other NIC drivers rather than allocating them during init() and freeing them during stop(): - Move creation of tx bus_dma tag amd maps and tx_buffer_area from em_setup_transmit_structures() to em_allocate_transmit_structures(). - Call em_allocate_xxx_structures() in em_attach(). - Only call em_free_xxx_structures() in em_detach(). - Change em_setup_xxx_structures() to free any existing tx or rx buffers and in the case of rx repopulate the ring with newer buffers. Reviewed by: jfv
2006-10-31 17:21:14 +00:00
em_free_transmit_structures(adapter);
em_free_receive_structures(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
em_release_hw_control(adapter);
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);
locking fixups: o correct recursive locking when polling and in em_82547_move_tail o destroy mutex on detach o add EM_LOCK_ASSERT and similar macros for creating+deleteing the mtx Submitted by: Daniel Eischen <eischen@vigrid.com>
2003-10-10 23:14:21 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (0);
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
/*********************************************************************
*
* Shutdown entry point
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +00:00
*
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
**********************************************************************/
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
static int
em_shutdown(device_t dev)
{
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
return em_suspend(dev);
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
Add simple suspend and resume methods. We call em_stop() on suspend and em_init() on resume. With this change the network is ready right after resume, without half minute lag. Tested by: Jacques Garrigue
2005-12-26 10:39:21 +00:00
/*
* Suspend/resume device methods.
*/
static int
em_suspend(device_t dev)
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
struct adapter *adapter = device_get_softc(dev);
Add simple suspend and resume methods. We call em_stop() on suspend and em_init() on resume. With this change the network is ready right after resume, without half minute lag. Tested by: Jacques Garrigue
2005-12-26 10:39:21 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Fix for a reported panic in certain circumstances. When calling em_stop() now make sure the TX lock is held as well as CORE.
2007-11-26 19:47:03 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
em_release_manageability(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
em_release_hw_control(adapter);
em_enable_wakeup(dev);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
Add simple suspend and resume methods. We call em_stop() on suspend and em_init() on resume. With this change the network is ready right after resume, without half minute lag. Tested by: Jacques Garrigue
2005-12-26 10:39:21 +00:00
return bus_generic_suspend(dev);
}
static int
em_resume(device_t dev)
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
struct adapter *adapter = device_get_softc(dev);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
struct tx_ring *txr = adapter->tx_rings;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
Add simple suspend and resume methods. We call em_stop() on suspend and em_init() on resume. With this change the network is ready right after resume, without half minute lag. Tested by: Jacques Garrigue
2005-12-26 10:39:21 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(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->hw.mac.type == e1000_pch2lan)
e1000_resume_workarounds_pchlan(&adapter->hw);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_init_locked(adapter);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
em_init_manageability(adapter);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if ((if_getflags(ifp) & IFF_UP) &&
(if_getdrvflags(ifp) & IFF_DRV_RUNNING) && adapter->link_active) {
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++) {
EM_TX_LOCK(txr);
#ifdef EM_MULTIQUEUE
if (!drbr_empty(ifp, txr->br))
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
em_mq_start_locked(ifp, txr);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#else
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (!if_sendq_empty(ifp))
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
em_start_locked(ifp, txr);
#endif
EM_TX_UNLOCK(txr);
}
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
Add simple suspend and resume methods. We call em_stop() on suspend and em_init() on resume. With this change the network is ready right after resume, without half minute lag. Tested by: Jacques Garrigue
2005-12-26 10:39:21 +00:00
return bus_generic_resume(dev);
}
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +00:00
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#ifndef EM_MULTIQUEUE
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
static void
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
em_start_locked(if_t ifp, struct tx_ring *txr)
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
struct adapter *adapter = if_getsoftc(ifp);
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
struct mbuf *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
EM_TX_LOCK_ASSERT(txr);
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) !=
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
IFF_DRV_RUNNING)
return;
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
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
if (!adapter->link_active)
return;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
while (!if_sendq_empty(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
/* Call cleanup if number of TX descriptors low */
if (txr->tx_avail <= EM_TX_CLEANUP_THRESHOLD)
em_txeof(txr);
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
if (txr->tx_avail < EM_MAX_SCATTER) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setdrvflagbits(ifp,IFF_DRV_OACTIVE, 0);
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
break;
}
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
m_head = if_dequeue(ifp);
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
if (m_head == NULL)
break;
/*
* Encapsulation can modify our pointer, and or make it
* NULL on failure. In that event, we can't requeue.
*/
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 (em_xmit(txr, &m_head)) {
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
if (m_head == NULL)
break;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_sendq_prepend(ifp, m_head);
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
break;
}
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
/* Mark the queue as having work */
if (txr->busy == EM_TX_IDLE)
txr->busy = EM_TX_BUSY;
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
/* Send a copy of the frame to the BPF listener */
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
ETHER_BPF_MTAP(ifp, m_head);
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
}
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
return;
- define em_transmit and em_qflush - make buF_ring usage conditional but enabled by default Reviewed by: jfv
2009-04-14 03:36:59 +00:00
}
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
static void
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
em_start(if_t ifp)
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
struct adapter *adapter = if_getsoftc(ifp);
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 tx_ring *txr = adapter->tx_rings;
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
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
EM_TX_LOCK(txr);
em_start_locked(ifp, txr);
EM_TX_UNLOCK(txr);
}
return;
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
}
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#else /* EM_MULTIQUEUE */
/*********************************************************************
* Multiqueue Transmit routines
*
* em_mq_start is called by the stack to initiate a transmit.
* however, if busy the driver can queue the request rather
* than do an immediate send. It is this that is an advantage
* in this driver, rather than also having multiple tx queues.
**********************************************************************/
/*
** Multiqueue capable stack interface
*/
static int
em_mq_start(if_t ifp, struct mbuf *m)
{
struct adapter *adapter = if_getsoftc(ifp);
struct tx_ring *txr = adapter->tx_rings;
unsigned int i, error;
if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
i = m->m_pkthdr.flowid % adapter->num_queues;
else
i = curcpu % adapter->num_queues;
txr = &adapter->tx_rings[i];
error = drbr_enqueue(ifp, txr->br, m);
if (error)
return (error);
if (EM_TX_TRYLOCK(txr)) {
em_mq_start_locked(ifp, txr);
EM_TX_UNLOCK(txr);
} else
taskqueue_enqueue(txr->tq, &txr->tx_task);
return (0);
}
static int
em_mq_start_locked(if_t ifp, struct tx_ring *txr)
{
struct adapter *adapter = txr->adapter;
struct mbuf *next;
int err = 0, enq = 0;
EM_TX_LOCK_ASSERT(txr);
if (((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) ||
adapter->link_active == 0) {
return (ENETDOWN);
}
/* Process the queue */
while ((next = drbr_peek(ifp, txr->br)) != NULL) {
if ((err = em_xmit(txr, &next)) != 0) {
if (next == NULL) {
/* It was freed, move forward */
drbr_advance(ifp, txr->br);
} else {
/*
* Still have one left, it may not be
* the same since the transmit function
* may have changed it.
*/
drbr_putback(ifp, txr->br, next);
}
break;
}
drbr_advance(ifp, txr->br);
enq++;
if_inc_counter(ifp, IFCOUNTER_OBYTES, next->m_pkthdr.len);
if (next->m_flags & M_MCAST)
if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
ETHER_BPF_MTAP(ifp, next);
if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
break;
}
/* Mark the queue as having work */
if ((enq > 0) && (txr->busy == EM_TX_IDLE))
txr->busy = EM_TX_BUSY;
if (txr->tx_avail < EM_MAX_SCATTER)
em_txeof(txr);
if (txr->tx_avail < EM_MAX_SCATTER) {
if_setdrvflagbits(ifp, IFF_DRV_OACTIVE,0);
}
return (err);
}
/*
** Flush all ring buffers
*/
static void
em_qflush(if_t ifp)
{
struct adapter *adapter = if_getsoftc(ifp);
struct tx_ring *txr = adapter->tx_rings;
struct mbuf *m;
for (int i = 0; i < adapter->num_queues; i++, txr++) {
EM_TX_LOCK(txr);
while ((m = buf_ring_dequeue_sc(txr->br)) != NULL)
m_freem(m);
EM_TX_UNLOCK(txr);
}
if_qflush(ifp);
}
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#endif /* EM_MULTIQUEUE */
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
* Ioctl entry point
*
* em_ioctl is called when the user wants to configure the
* interface.
*
* return 0 on success, positive on failure
**********************************************************************/
static int
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
em_ioctl(if_t ifp, u_long command, caddr_t data)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
struct adapter *adapter = if_getsoftc(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
struct ifreq *ifr = (struct ifreq *)data;
#if defined(INET) || defined(INET6)
struct ifaddr *ifa = (struct ifaddr *)data;
Correct LINT build issues in the ioctl code.
2011-12-11 09:37:25 +00:00
#endif
Last change still had an issue, one more time...
2011-12-11 18:46:14 +00:00
bool avoid_reset = FALSE;
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 = 0;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
if (adapter->in_detach)
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (error);
- Code cleanup - In the receive routine handle the case where last descriptor could have less than 4 bytes of data. - Handle race between detach/ioctl routine. MFC after: 3 days
2003-11-14 18:02:25 +00:00
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
switch (command) {
case SIOCSIFADDR:
Hide AF_INET specific ioctl handling under #ifdef INET. MFC after: 2 months
2008-11-06 11:00:57 +00:00
#ifdef INET
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 (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
if (ifa->ifa_addr->sa_family == AF_INET6)
avoid_reset = TRUE;
#endif
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
/*
** Calling init results in link renegotiation,
** so we avoid doing it when possible.
*/
if (avoid_reset) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setflagbits(ifp,IFF_UP,0);
if (!(if_getdrvflags(ifp)& IFF_DRV_RUNNING))
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
em_init(adapter);
Last change still had an issue, one more time...
2011-12-11 18:46:14 +00:00
#ifdef INET
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (!(if_getflags(ifp) & IFF_NOARP))
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
arp_ifinit(ifp, ifa);
Last change still had an issue, one more time...
2011-12-11 18:46:14 +00:00
#endif
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
} else
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
error = ether_ioctl(ifp, command, data);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
break;
case SIOCSIFMTU:
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +00:00
{
int max_frame_size;
IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFMTU (Set Interface MTU)");
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
switch (adapter->hw.mac.type) {
case e1000_82571:
case e1000_82572:
case e1000_ich9lan:
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
case e1000_ich10lan:
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
case e1000_pch2lan:
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
case e1000_pch_lpt:
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
case e1000_82574:
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
case e1000_82583:
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
case e1000_80003es2lan: /* 9K Jumbo Frame size */
Merge in new driver from Intel, version 5.1.5. Adds support for some new chips and improves support for already supported ones. Some details, important for future merges: - if_em.c merged manually, viewing diff between new vendor driver and previous one. - if_em_hw.h dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c dropped in from vendor, and then two liner change made, that restores support for two rare chips.
2006-04-06 17:09:03 +00:00
max_frame_size = 9234;
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +00:00
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
case e1000_pchlan:
max_frame_size = 4096;
break;
Two minor fixes, keep old 82542 from using jumbo frames, and add missing htole64 in encap code. Reviewed by:Pdeuskar Approved by:Pdeuskar
2007-05-23 20:41:20 +00:00
/* Adapters that do not support jumbo frames */
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
case e1000_ich8lan:
Merge in new driver from Intel, version 6.0.5. It adds support for 80003 NICs and NICs found on ICH8 mobos, and improves support for already known chips. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. This was an easy task, because most changes between 5.1.5 and 6.0.5 are bugfixes taken from FreeBSD. - if_em_hw.h. Dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15. - if_em_osdep.h. Added new required macros from vendor file and add a hack against define namespace mangling in if_em_hw.h. Intel made another hack, but I prefer mine.
2006-08-03 09:20:11 +00:00
max_frame_size = ETHER_MAX_LEN;
break;
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +00:00
default:
max_frame_size = MAX_JUMBO_FRAME_SIZE;
}
if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
ETHER_CRC_LEN) {
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
error = EINVAL;
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +00:00
break;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
}
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +00:00
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setmtu(ifp, ifr->ifr_mtu);
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
adapter->hw.mac.max_frame_size =
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_getmtu(ifp) + ETHER_HDR_LEN + ETHER_CRC_LEN;
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_init_locked(adapter);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
break;
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +00:00
}
Added support for 82541 and 82547 based adapters. - These have Intel gigabit PHY - 82547 uses CSA interface MFC after: 1 week
2003-03-21 21:47:31 +00:00
case SIOCSIFFLAGS:
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
IOCTL_DEBUGOUT("ioctl rcv'd:\
SIOCSIFFLAGS (Set Interface Flags)");
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getflags(ifp) & IFF_UP) {
if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
if ((if_getflags(ifp) ^ 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)) {
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_disable_promisc(adapter);
em_set_promisc(adapter);
Since resetting hardware takes a very long time and results in link renegotiation, we only initialize the hardware only when it is absolutely required. Process SIOCGIFADDR ioctl in em(4) when we know an IPv4 address is added. Handling SIOCGIFADDR in a driver is layering violation but it seems that there is no easy way without rewritting hardware initialization code to reduce settle time after reset. This should fix a long standing bug which didn't send ARP packet when interface address is changed or an alias address is added. Another effect of this fix is it doesn't need additional delays anymore when adding an alias address to the interface. While I'm here add a new if_flags into softc which remembers current prgroammed interface flags and make use of it when we have to program promiscuous mode. Tested by: Atanas <atanas AT asd DOT aplus DOT net> Analyzed by: rwatson Discussed with: -stable
2006-07-20 04:18:45 +00:00
}
} else
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_init_locked(adapter);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
} else
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_stop(adapter);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
adapter->if_flags = if_getflags(ifp);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
Added support for 82541 and 82547 based adapters. - These have Intel gigabit PHY - 82547 uses CSA interface MFC after: 1 week
2003-03-21 21:47:31 +00:00
break;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
case SIOCADDMULTI:
case SIOCDELMULTI:
IOCTL_DEBUGOUT("ioctl rcv'd: SIOC(ADD|DEL)MULTI");
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_disable_intr(adapter);
em_set_multi(adapter);
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
#ifdef DEVICE_POLLING
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (!(if_getcapenable(ifp) & IFCAP_POLLING))
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
#endif
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_enable_intr(adapter);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
}
break;
case SIOCSIFMEDIA:
Merge new vendor release - 6.2.9. Details: o if_em.c changes: - Added several new PCI ids. - Check em_check_phy_reset_block() before doing SIOCSIFMEDIA ioctl. - Don't touch TARC registers, they are now handled in shared code in if_em_hw.c. - Move RDH and RDT setting to the end of em_initialize_receive_unit(). - Declare em_read_pcie_cap_reg(), now empty. o if_em_hw.c dropped in from vendor, then restored rev. 1.15. o if_em_hw.h dropped in from vendor, then modified: - Added RX overrun interrupt flag to interrupt enable mask. - Remove declarations of em_io_read(), em_io_write(). Approved by: jfv
2006-10-31 15:00:14 +00:00
/* Check SOL/IDER usage */
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (e1000_check_reset_block(&adapter->hw)) {
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
Merge new vendor release - 6.2.9. Details: o if_em.c changes: - Added several new PCI ids. - Check em_check_phy_reset_block() before doing SIOCSIFMEDIA ioctl. - Don't touch TARC registers, they are now handled in shared code in if_em_hw.c. - Move RDH and RDT setting to the end of em_initialize_receive_unit(). - Declare em_read_pcie_cap_reg(), now empty. o if_em_hw.c dropped in from vendor, then restored rev. 1.15. o if_em_hw.h dropped in from vendor, then modified: - Added RX overrun interrupt flag to interrupt enable mask. - Remove declarations of em_io_read(), em_io_write(). Approved by: jfv
2006-10-31 15:00:14 +00:00
device_printf(adapter->dev, "Media change is"
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
" blocked due to SOL/IDER session.\n");
Merge new vendor release - 6.2.9. Details: o if_em.c changes: - Added several new PCI ids. - Check em_check_phy_reset_block() before doing SIOCSIFMEDIA ioctl. - Don't touch TARC registers, they are now handled in shared code in if_em_hw.c. - Move RDH and RDT setting to the end of em_initialize_receive_unit(). - Declare em_read_pcie_cap_reg(), now empty. o if_em_hw.c dropped in from vendor, then restored rev. 1.15. o if_em_hw.h dropped in from vendor, then modified: - Added RX overrun interrupt flag to interrupt enable mask. - Remove declarations of em_io_read(), em_io_write(). Approved by: jfv
2006-10-31 15:00:14 +00:00
break;
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
/* falls thru */
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
case SIOCGIFMEDIA:
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
IOCTL_DEBUGOUT("ioctl rcv'd: \
SIOCxIFMEDIA (Get/Set Interface Media)");
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
error = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
break;
case SIOCSIFCAP:
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +00:00
{
int mask, reinit;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFCAP (Set Capabilities)");
Implement support for controlling VLAN_HWTAGGING through ioctl(SIOCSIFCAP). This includes not only toggling the flag in if_capenable, but also really reconfiguring the hardware. Approved by: tackerman (as the em(4) maintainer)
2004-06-04 16:57:18 +00:00
reinit = 0;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
mask = ifr->ifr_reqcap ^ if_getcapenable(ifp);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
#ifdef DEVICE_POLLING
if (mask & IFCAP_POLLING) {
if (ifr->ifr_reqcap & IFCAP_POLLING) {
- Remove empty wrappers ether_poll_[de]register_drv(). [1] - Move polling(9) declarations out of ifq.h back to if_var.h they are absolutely unrelated to queues. Submitted by: Mikhail <mp lenta.ru> [1]
2014-09-28 14:05:18 +00:00
error = ether_poll_register(em_poll, ifp);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
if (error)
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (error);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_disable_intr(adapter);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setcapenablebit(ifp, IFCAP_POLLING, 0);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
} else {
- Remove empty wrappers ether_poll_[de]register_drv(). [1] - Move polling(9) declarations out of ifq.h back to if_var.h they are absolutely unrelated to queues. Submitted by: Mikhail <mp lenta.ru> [1]
2014-09-28 14:05:18 +00:00
error = ether_poll_deregister(ifp);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
/* Enable interrupt even in error case */
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_enable_intr(adapter);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setcapenablebit(ifp, 0, IFCAP_POLLING);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
}
}
#endif
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
if (mask & IFCAP_HWCSUM) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_togglecapenable(ifp,IFCAP_HWCSUM);
Implement support for controlling VLAN_HWTAGGING through ioctl(SIOCSIFCAP). This includes not only toggling the flag in if_capenable, but also really reconfiguring the hardware. Approved by: tackerman (as the em(4) maintainer)
2004-06-04 16:57:18 +00:00
reinit = 1;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
}
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
if (mask & IFCAP_TSO4) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_togglecapenable(ifp,IFCAP_TSO4);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
reinit = 1;
}
Implement support for controlling VLAN_HWTAGGING through ioctl(SIOCSIFCAP). This includes not only toggling the flag in if_capenable, but also really reconfiguring the hardware. Approved by: tackerman (as the em(4) maintainer)
2004-06-04 16:57:18 +00:00
if (mask & IFCAP_VLAN_HWTAGGING) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_togglecapenable(ifp,IFCAP_VLAN_HWTAGGING);
Implement support for controlling VLAN_HWTAGGING through ioctl(SIOCSIFCAP). This includes not only toggling the flag in if_capenable, but also really reconfiguring the hardware. Approved by: tackerman (as the em(4) maintainer)
2004-06-04 16:57:18 +00:00
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) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER);
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
reinit = 1;
}
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 (mask & IFCAP_VLAN_HWTSO) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
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
reinit = 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
if ((mask & IFCAP_WOL) &&
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
(if_getcapabilities(ifp) & IFCAP_WOL) != 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
if (mask & IFCAP_WOL_MCAST)
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_togglecapenable(ifp, IFCAP_WOL_MCAST);
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 (mask & IFCAP_WOL_MAGIC)
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_togglecapenable(ifp, IFCAP_WOL_MAGIC);
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
}
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (reinit && (if_getdrvflags(ifp) & IFF_DRV_RUNNING))
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_init(adapter);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_vlancap(ifp);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
break;
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +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
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
default:
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
error = ether_ioctl(ifp, command, data);
Since resetting hardware takes a very long time and results in link renegotiation, we only initialize the hardware only when it is absolutely required. Process SIOCGIFADDR ioctl in em(4) when we know an IPv4 address is added. Handling SIOCGIFADDR in a driver is layering violation but it seems that there is no easy way without rewritting hardware initialization code to reduce settle time after reset. This should fix a long standing bug which didn't send ARP packet when interface address is changed or an alias address is added. Another effect of this fix is it doesn't need additional delays anymore when adding an alias address to the interface. While I'm here add a new if_flags into softc which remembers current prgroammed interface flags and make use of it when we have to program promiscuous mode. Tested by: Atanas <atanas AT asd DOT aplus DOT net> Analyzed by: rwatson Discussed with: -stable
2006-07-20 04:18:45 +00:00
break;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
}
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (error);
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
/*********************************************************************
* 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
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* by the driver as a hw/sw initialization routine to get to a
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
* consistent state.
*
* return 0 on success, positive on failure
**********************************************************************/
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_init_locked(struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
device_t dev = adapter->dev;
Implement support for controlling VLAN_HWTAGGING through ioctl(SIOCSIFCAP). This includes not only toggling the flag in if_capenable, but also really reconfiguring the hardware. Approved by: tackerman (as the em(4) maintainer)
2004-06-04 16:57:18 +00:00
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
INIT_DEBUGOUT("em_init: begin");
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK_ASSERT(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
em_disable_intr(adapter);
callout_stop(&adapter->timer);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
- Code cleanup - In the receive routine handle the case where last descriptor could have less than 4 bytes of data. - Handle race between detach/ioctl routine. MFC after: 3 days
2003-11-14 18:02:25 +00:00
/* Get the latest mac address, User can use a LAA */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
bcopy(if_getlladdr(adapter->ifp), adapter->hw.mac.addr,
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
ETHER_ADDR_LEN);
Back out 1.112,1.113. I don't have enough resources to fix breakages introduced by this change.
2006-02-22 14:11:16 +00:00
A couple bug fixes that I've had internally at Intel. First is a long time workaround for problems with 82571 adapters and LAAs, one port getting reset can cause the other to have its RAR[0] also reset, thus overwriting an LAA. This fix works around it by also keeping the address in the last array member. The other bug is specific to the new 575 adapter, its transmit code logic in handling hwassists was too crude, it broken when doing bridges. I am much happier with the new logic,we may want to change the legacy path at some point to something similar. Reviewed by: pdeuskar Approved by: pdeuskar
2007-05-09 00:41:45 +00:00
/* Put the address into the Receive Address Array */
e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
/*
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
* With the 82571 adapter, RAR[0] may be overwritten
* when the other port is reset, we make a duplicate
* in RAR[14] for that eventuality, this assures
* the interface continues to function.
A couple bug fixes that I've had internally at Intel. First is a long time workaround for problems with 82571 adapters and LAAs, one port getting reset can cause the other to have its RAR[0] also reset, thus overwriting an LAA. This fix works around it by also keeping the address in the last array member. The other bug is specific to the new 575 adapter, its transmit code logic in handling hwassists was too crude, it broken when doing bridges. I am much happier with the new logic,we may want to change the legacy path at some point to something similar. Reviewed by: pdeuskar Approved by: pdeuskar
2007-05-09 00:41:45 +00:00
*/
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
if (adapter->hw.mac.type == e1000_82571) {
e1000_set_laa_state_82571(&adapter->hw, TRUE);
e1000_rar_set(&adapter->hw, adapter->hw.mac.addr,
E1000_RAR_ENTRIES - 1);
}
A couple bug fixes that I've had internally at Intel. First is a long time workaround for problems with 82571 adapters and LAAs, one port getting reset can cause the other to have its RAR[0] also reset, thus overwriting an LAA. This fix works around it by also keeping the address in the last array member. The other bug is specific to the new 575 adapter, its transmit code logic in handling hwassists was too crude, it broken when doing bridges. I am much happier with the new logic,we may want to change the legacy path at some point to something similar. Reviewed by: pdeuskar Approved by: pdeuskar
2007-05-09 00:41:45 +00:00
Back out 1.112,1.113. I don't have enough resources to fix breakages introduced by this change.
2006-02-22 14:11:16 +00:00
/* Initialize the hardware */
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
em_reset(adapter);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_update_link_status(adapter);
- Code cleanup - In the receive routine handle the case where last descriptor could have less than 4 bytes of data. - Handle race between detach/ioctl routine. MFC after: 3 days
2003-11-14 18:02:25 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
/* Setup VLAN support, basic and offload if available */
E1000_WRITE_REG(&adapter->hw, E1000_VET, ETHERTYPE_VLAN);
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
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* Set hardware offload abilities */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_clearhwassist(ifp);
if (if_getcapenable(ifp) & IFCAP_TXCSUM)
if_sethwassistbits(ifp, CSUM_TCP | CSUM_UDP, 0);
if (if_getcapenable(ifp) & IFCAP_TSO4)
if_sethwassistbits(ifp, CSUM_TSO, 0);
Move the initialization of the hardware capabilities in em_init_locked() before em_setup_transmit_structures() as it needs this information to properly set up TSO parameters. Reviewed by: jfv
2006-09-21 19:14:51 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* Configure for OS presence */
em_init_manageability(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
/* Prepare transmit descriptors and buffers */
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
em_setup_transmit_structures(adapter);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_initialize_transmit_unit(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
/* Setup Multicast table */
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_set_multi(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
/*
** Figure out the desired mbuf
** pool for doing jumbos
*/
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
if (adapter->hw.mac.max_frame_size <= 2048)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
adapter->rx_mbuf_sz = MCLBYTES;
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
else if (adapter->hw.mac.max_frame_size <= 4096)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
adapter->rx_mbuf_sz = MJUMPAGESIZE;
else
adapter->rx_mbuf_sz = MJUM9BYTES;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
/* Prepare receive descriptors and buffers */
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
if (em_setup_receive_structures(adapter)) {
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
device_printf(dev, "Could not setup receive structures\n");
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_stop(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
return;
}
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_initialize_receive_unit(adapter);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
/* Use real VLAN Filter support? */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) {
if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
/* Use real VLAN Filter support */
em_setup_vlan_hw_support(adapter);
else {
u32 ctrl;
ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL);
ctrl |= E1000_CTRL_VME;
E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl);
}
}
Commit the results of the typo hunt by Darren Pilgrim. This change affects documentation and comments only, no real code involved. PR: misc/101245 Submitted by: Darren Pilgrim <darren pilgrim bitfreak org> Tested by: md5(1) MFC after: 1 week
2006-08-04 07:56:35 +00:00
/* Don't lose promiscuous settings */
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_set_promisc(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
/* Set the interface as ACTIVE */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setdrvflagbits(ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
callout_reset(&adapter->timer, hz, em_local_timer, adapter);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
e1000_clear_hw_cntrs_base_generic(&adapter->hw);
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
/* MSI/X configuration for 82574 */
if (adapter->hw.mac.type == e1000_82574) {
int tmp;
tmp = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
tmp |= E1000_CTRL_EXT_PBA_CLR;
E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, tmp);
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
/* Set the IVAR - interrupt vector routing. */
E1000_WRITE_REG(&adapter->hw, E1000_IVAR, adapter->ivars);
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
}
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
Add polling support to the "em" driver. Reviewed by: pdeuskar (maintainer) Approved by: re
2002-11-25 19:56:16 +00:00
#ifdef DEVICE_POLLING
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
/*
* Only enable interrupts if we are not polling, make sure
* they are off otherwise.
*/
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getcapenable(ifp) & IFCAP_POLLING)
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_disable_intr(adapter);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
else
Add polling support to the "em" driver. Reviewed by: pdeuskar (maintainer) Approved by: re
2002-11-25 19:56:16 +00:00
#endif /* DEVICE_POLLING */
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_enable_intr(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
/* AMT based hardware can now take control from firmware */
if (adapter->has_manage && adapter->has_amt)
em_get_hw_control(adapter);
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
}
static void
em_init(void *arg)
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
struct adapter *adapter = arg;
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_init_locked(adapter);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
Added support for 82541 and 82547 based adapters. - These have Intel gigabit PHY - 82547 uses CSA interface MFC after: 1 week
2003-03-21 21:47:31 +00:00
Add polling support to the "em" driver. Reviewed by: pdeuskar (maintainer) Approved by: re
2002-11-25 19:56:16 +00:00
#ifdef DEVICE_POLLING
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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
* Legacy polling routine: note this only works with single queue
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
*
*********************************************************************/
When user_frac in the polling subsystem is low it is going to busy the CPU for too long period than necessary. Additively, interfaces are kept polled (in the tick) even if no more packets are available. In order to avoid such situations a new generic mechanism can be implemented in proactive way, keeping track of the time spent on any packet and fragmenting the time for any tick, stopping the processing as soon as possible. In order to implement such mechanism, the polling handler needs to change, returning the number of packets processed. While the intended logic is not part of this patch, the polling KPI is broken by this commit, adding an int return value and the new flag IFCAP_POLLING_NOCOUNT (which will signal that the return value is meaningless for the installed handler and checking should be skipped). Bump __FreeBSD_version in order to signal such situation. Reviewed by: emaste Sponsored by: Sandvine Incorporated
2009-05-30 15:14:44 +00:00
static int
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
em_poll(if_t ifp, enum poll_cmd cmd, int count)
Add polling support to the "em" driver. Reviewed by: pdeuskar (maintainer) Approved by: re
2002-11-25 19:56:16 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
struct adapter *adapter = if_getsoftc(ifp);
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 tx_ring *txr = adapter->tx_rings;
struct rx_ring *rxr = adapter->rx_rings;
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
u32 reg_icr;
int rx_done;
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) {
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_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
return (0);
The procedure of raceless switching between polling mode and taskqueued interrupt mode is going to be quite complex. Since the polling mode is considered legacy feature for em(4) driver, the decision is made to make polling and new interrupt handler mutually exclusive, selected at compile time. If kernel is compiled with DEVICE_POLLING, the fast taskqueued interrupt handler code is disabled and the em_poll() and legacy em_intr() functions are enabled. Otherwise, legacy functions are disabled and only em_intr_fast() code is compiled. Discussed with: scottl
2006-06-06 08:03:49 +00:00
}
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
if (cmd == POLL_AND_CHECK_STATUS) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
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
callout_stop(&adapter->timer);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.mac.get_link_status = 1;
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_update_link_status(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
callout_reset(&adapter->timer, hz,
em_local_timer, adapter);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
}
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_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
em_rxeof(rxr, count, &rx_done);
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 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
EM_TX_LOCK(txr);
em_txeof(txr);
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
#ifdef EM_MULTIQUEUE
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 (!drbr_empty(ifp, txr->br))
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
em_mq_start_locked(ifp, txr);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
#else
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (!if_sendq_empty(ifp))
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
em_start_locked(ifp, txr);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
#endif
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
EM_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 (rx_done);
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
#endif /* DEVICE_POLLING */
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +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
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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
* Fast Legacy/MSI Combined Interrupt Service routine
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
*
*********************************************************************/
o break newbus api: add a new argument of type driver_filter_t to bus_setup_intr() o add an int return code to all fast handlers o retire INTR_FAST/IH_FAST For more info: http://docs.freebsd.org/cgi/getmsg.cgi?fetch=465712+0+current/freebsd-current Reviewed by: many Approved by: re@
2007-02-23 12:19:07 +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
em_irq_fast(void *arg)
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
{
struct adapter *adapter = arg;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t 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
u32 reg_icr;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
ifp = adapter->ifp;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
/* Hot eject? */
if (reg_icr == 0xffffffff)
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
return FILTER_STRAY;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
/* Definitely not our interrupt. */
if (reg_icr == 0x0)
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
return FILTER_STRAY;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
/*
* Starting with the 82571 chip, bit 31 should be used to
* determine whether the interrupt belongs to us.
*/
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (adapter->hw.mac.type >= e1000_82571 &&
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
(reg_icr & E1000_ICR_INT_ASSERTED) == 0)
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
return FILTER_STRAY;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
em_disable_intr(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
taskqueue_enqueue(adapter->tq, &adapter->que_task);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
/* Link status change */
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 (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
adapter->hw.mac.get_link_status = 1;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
taskqueue_enqueue(taskqueue_fast, &adapter->link_task);
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
}
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
if (reg_icr & E1000_ICR_RXO)
adapter->rx_overruns++;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
return FILTER_HANDLED;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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
/* Combined RX/TX handler, used by Legacy and MSI */
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
static 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
em_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 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 = context;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
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 tx_ring *txr = adapter->tx_rings;
struct rx_ring *rxr = adapter->rx_rings;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
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
bool more = em_rxeof(rxr, adapter->rx_process_limit, NULL);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +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
EM_TX_LOCK(txr);
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
em_txeof(txr);
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
#ifdef EM_MULTIQUEUE
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 (!drbr_empty(ifp, txr->br))
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
em_mq_start_locked(ifp, txr);
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
#else
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (!if_sendq_empty(ifp))
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
em_start_locked(ifp, txr);
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
#endif
EM_TX_UNLOCK(txr);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
if (more) {
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
taskqueue_enqueue(adapter->tq, &adapter->que_task);
A few more changes from yongari: - code flow in handler could let interrupt be reenabled when not wanted. - change where the RX lock is taken to improve performance. - adapter->msix is true for MSI systems also, it needs to explicitly test for 82574, good one :)
2010-04-09 23:15:37 +00:00
return;
}
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 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
em_enable_intr(adapter);
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
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
/*********************************************************************
*
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
* MSIX Interrupt Service Routines
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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 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
static 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
em_msix_tx(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 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 tx_ring *txr = arg;
struct adapter *adapter = txr->adapter;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
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
++txr->tx_irq;
EM_TX_LOCK(txr);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
em_txeof(txr);
#ifdef EM_MULTIQUEUE
if (!drbr_empty(ifp, txr->br))
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
em_mq_start_locked(ifp, txr);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#else
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (!if_sendq_empty(ifp))
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
em_start_locked(ifp, txr);
#endif
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
/* Reenable this interrupt */
E1000_WRITE_REG(&adapter->hw, E1000_IMS, txr->ims);
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
EM_TX_UNLOCK(txr);
return;
}
/*********************************************************************
*
* MSIX RX 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
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
em_msix_rx(void *arg)
{
struct rx_ring *rxr = arg;
struct adapter *adapter = rxr->adapter;
bool more;
++rxr->rx_irq;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (!(if_getdrvflags(adapter->ifp) & IFF_DRV_RUNNING))
Customer report of a panic on boot due to the old "m_getjcl:invalid cluster type" that occurred some time back with the igb driver. This happens often when booting over the net. I believe the NIC hardware is left in a warm state when handed over to the driver, and a stray RX interrupt happens earlier than the code is prepared for it to happen. This change was verified to fix the problem, its kind of a bandaid... but it is similar to what was done in the igb code.
2012-08-15 17:12:40 +00:00
return;
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 = em_rxeof(rxr, adapter->rx_process_limit, 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 (more)
taskqueue_enqueue(rxr->tq, &rxr->rx_task);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
else {
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
/* Reenable this interrupt */
E1000_WRITE_REG(&adapter->hw, E1000_IMS, rxr->ims);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +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
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
}
/*********************************************************************
*
* MSIX Link Fast Interrupt Service routine
*
**********************************************************************/
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
static void
em_msix_link(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
{
struct adapter *adapter = arg;
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
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
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_irq;
reg_icr = E1000_READ_REG(&adapter->hw, E1000_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
Bump rx_overruns when indicated by the ICR mask. PR: 199716 MFC after: 3 days Sponsored by: Limelight Networks
2015-05-22 17:01:43 +00:00
if (reg_icr & E1000_ICR_RXO)
adapter->rx_overruns++;
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
if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
adapter->hw.mac.get_link_status = 1;
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
em_handle_link(adapter, 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
} else
E1000_WRITE_REG(&adapter->hw, E1000_IMS,
EM_MSIX_LINK | E1000_IMS_LSC);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
/*
** Because we must read the ICR for this interrupt
** it may clear other causes using autoclear, for
** this reason we simply create a soft interrupt
** for all these vectors.
*/
if (reg_icr) {
E1000_WRITE_REG(&adapter->hw,
E1000_ICS, adapter->ims);
}
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
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
}
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
static void
em_handle_rx(void *context, int pending)
{
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 rx_ring *rxr = context;
struct adapter *adapter = rxr->adapter;
bool more;
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 = em_rxeof(rxr, adapter->rx_process_limit, NULL);
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
if (more)
taskqueue_enqueue(rxr->tq, &rxr->rx_task);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
else {
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
/* Reenable this interrupt */
E1000_WRITE_REG(&adapter->hw, E1000_IMS, rxr->ims);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +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
}
static void
em_handle_tx(void *context, int pending)
{
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 tx_ring *txr = context;
struct adapter *adapter = txr->adapter;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
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 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
EM_TX_LOCK(txr);
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
em_txeof(txr);
#ifdef EM_MULTIQUEUE
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 (!drbr_empty(ifp, txr->br))
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
em_mq_start_locked(ifp, txr);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
#else
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (!if_sendq_empty(ifp))
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
em_start_locked(ifp, txr);
Updates for both the em and igb drivers, add support for multiqueue tx, shared code updates, new device support, and some bug fixes.
2009-06-24 17:41:29 +00:00
#endif
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_IMS, txr->ims);
EM_TX_UNLOCK(txr);
}
static void
em_handle_link(void *context, int pending)
{
struct adapter *adapter = context;
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
struct tx_ring *txr = adapter->tx_rings;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
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
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING))
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;
EM_CORE_LOCK(adapter);
callout_stop(&adapter->timer);
em_update_link_status(adapter);
callout_reset(&adapter->timer, hz, em_local_timer, adapter);
E1000_WRITE_REG(&adapter->hw, E1000_IMS,
EM_MSIX_LINK | E1000_IMS_LSC);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
if (adapter->link_active) {
for (int i = 0; i < adapter->num_queues; i++, txr++) {
EM_TX_LOCK(txr);
#ifdef EM_MULTIQUEUE
if (!drbr_empty(ifp, txr->br))
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
em_mq_start_locked(ifp, txr);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#else
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_sendq_empty(ifp))
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
em_start_locked(ifp, txr);
#endif
EM_TX_UNLOCK(txr);
}
}
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
EM_CORE_UNLOCK(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
}
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
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
*
* Media Ioctl callback
*
* This routine is called whenever the user queries the status of
* the interface using ifconfig.
*
**********************************************************************/
static void
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
em_media_status(if_t ifp, struct ifmediareq *ifmr)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
struct adapter *adapter = if_getsoftc(ifp);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
u_char fiber_type = IFM_1000_SX;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
INIT_DEBUGOUT("em_media_status: begin");
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_update_link_status(adapter);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
ifmr->ifm_status = IFM_AVALID;
ifmr->ifm_active = IFM_ETHER;
Add locking to the ifmedia callouts. Reviewed by: jvogel, yongari
2006-09-01 16:08:36 +00:00
if (!adapter->link_active) {
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
return;
Add locking to the ifmedia callouts. Reviewed by: jvogel, yongari
2006-09-01 16:08:36 +00:00
}
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
ifmr->ifm_status |= IFM_ACTIVE;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
if ((adapter->hw.phy.media_type == e1000_media_type_fiber) ||
(adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
ifmr->ifm_active |= fiber_type | IFM_FDX;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
} else {
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
switch (adapter->link_speed) {
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
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;
}
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
if (adapter->link_duplex == FULL_DUPLEX)
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
ifmr->ifm_active |= IFM_FDX;
else
ifmr->ifm_active |= IFM_HDX;
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
}
/*********************************************************************
*
* Media Ioctl callback
*
* This routine is called when the user changes speed/duplex using
* media/mediopt option with ifconfig.
*
**********************************************************************/
static int
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
em_media_change(if_t ifp)
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
struct adapter *adapter = if_getsoftc(ifp);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
struct ifmedia *ifm = &adapter->media;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
INIT_DEBUGOUT("em_media_change: begin");
if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (EINVAL);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
switch (IFM_SUBTYPE(ifm->ifm_media)) {
case IFM_AUTO:
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.mac.autoneg = DO_AUTO_NEG;
adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
break;
Merge in new driver from Intel, version 6.1.4. It adds support for 82571EB quad port copper NIC and has few minor fixes. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15.
2006-08-11 10:58:24 +00:00
case IFM_1000_LX:
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
case IFM_1000_SX:
case IFM_1000_T:
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.mac.autoneg = DO_AUTO_NEG;
adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
break;
case IFM_100_TX:
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.mac.autoneg = FALSE;
adapter->hw.phy.autoneg_advertised = 0;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
else
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
break;
case IFM_10_T:
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.mac.autoneg = FALSE;
adapter->hw.phy.autoneg_advertised = 0;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
else
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
break;
default:
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
device_printf(adapter->dev, "Unsupported media type\n");
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
}
Add locking to the ifmedia callouts. Reviewed by: jvogel, yongari
2006-09-01 16:08:36 +00:00
em_init_locked(adapter);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (0);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
}
/*********************************************************************
*
* This routine maps the mbufs to tx descriptors.
*
* return 0 on success, positive on failure
**********************************************************************/
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
static int
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
em_xmit(struct tx_ring *txr, struct mbuf **m_headp)
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +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 = txr->adapter;
Convert if_em to the new bus_dmamap_load_sg() interface. The old callback was really just a waste of cycles, so this streamlines it considerably.
2005-01-15 20:52:15 +00:00
bus_dma_segment_t segs[EM_MAX_SCATTER];
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
bus_dmamap_t map;
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
struct em_buffer *tx_buffer, *tx_buffer_mapped;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
struct e1000_tx_desc *ctxd = NULL;
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
struct mbuf *m_head;
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
struct ether_header *eh;
struct ip *ip = NULL;
struct tcphdr *tp = NULL;
Remove unused txd_saved. Intialize txd_upper, txd_lower and txd_used at declaration. Differential Revision: D3174 Reviewed by: erj hiren MFC after: 2 weeks Sponsored by: Limelight Networks
2015-07-25 19:24:33 +00:00
u32 txd_upper = 0, txd_lower = 0, txd_used = 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
int ip_off, poff;
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
int nsegs, i, j, first, last = 0;
Fix to an error condition case, when an mbuf chain get's defragged due to a mapping failure the header pointers will be invalidated and can result in a TSO or other failure down the line. So, when the remapping occurs force a retry thru the offload calculation code. Thanks to Andrew Boyer for discovering this and cooking up the fix!!
2011-04-01 20:24:51 +00:00
int error, do_tso, tso_desc = 0, remap = 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
Correct a bug in the if_em driver relating to the use of vlans with promiscuous mode introduced in 1.45, which programs the em card not to strip or prepend tags when in promiscuous mode without also modifying behavior to manually prepend a vlan header in the event that the card isn't doing it on transmit. Due to a feature of card operation, if the global VLAN prepend/strip register isn't set, setting the VLAN tag flag on individual packet descriptors will cause the packet to be transmitted using ISL encapsulation rather than 802.1Q VLAN encapsulation. This fix causes em_encap() to prepend the header by tracking whether the card is configured to temporarily disable prepending/stripping due to promiscuous mode. As a result, entering promiscuous mode on the parent em interface no longer causes vlans to appear to "wedge" or transmit ISL-encapsulated frames, which typically will not be configured/spoken by the other endpoints on the VLAN trunk. This bug may also exist in other drivers, and the additional vlan encapsulation logic should be abstracted and centralized in if_vlan.c if so. RELENG_5_3 candidate. MFC after: 1 week Tested by: pjd, rwatson Reported by: astesin at ukrtelecom dot net Reported by: Mike Tancsa <mike at sentex dot net> Reported by: Iasen Kostov <tbyte at OTEL dot net>
2004-11-12 11:03:07 +00:00
m_head = *m_headp;
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
do_tso = ((m_head->m_pkthdr.csum_flags & CSUM_TSO) != 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
ip_off = poff = 0;
Correct a bug in the if_em driver relating to the use of vlans with promiscuous mode introduced in 1.45, which programs the em card not to strip or prepend tags when in promiscuous mode without also modifying behavior to manually prepend a vlan header in the event that the card isn't doing it on transmit. Due to a feature of card operation, if the global VLAN prepend/strip register isn't set, setting the VLAN tag flag on individual packet descriptors will cause the packet to be transmitted using ISL encapsulation rather than 802.1Q VLAN encapsulation. This fix causes em_encap() to prepend the header by tracking whether the card is configured to temporarily disable prepending/stripping due to promiscuous mode. As a result, entering promiscuous mode on the parent em interface no longer causes vlans to appear to "wedge" or transmit ISL-encapsulated frames, which typically will not be configured/spoken by the other endpoints on the VLAN trunk. This bug may also exist in other drivers, and the additional vlan encapsulation logic should be abstracted and centralized in if_vlan.c if so. RELENG_5_3 candidate. MFC after: 1 week Tested by: pjd, rwatson Reported by: astesin at ukrtelecom dot net Reported by: Mike Tancsa <mike at sentex dot net> Reported by: Iasen Kostov <tbyte at OTEL dot net>
2004-11-12 11:03:07 +00:00
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
/*
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
* 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.
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
*/
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
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) {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m_head = m_dup(*m_headp, M_NOWAIT);
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
m_freem(*m_headp);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
*m_headp = m_head;
}
}
/*
* XXX
* Assume IPv4, we don't have TSO/checksum offload support
* for IPv6 yet.
*/
ip_off = sizeof(struct ether_header);
m_head = m_pullup(m_head, ip_off);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
eh = mtod(m_head, struct ether_header *);
if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
ip_off = sizeof(struct ether_vlan_header);
m_head = m_pullup(m_head, ip_off);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
}
m_head = m_pullup(m_head, ip_off + sizeof(struct ip));
if (m_head == NULL) {
*m_headp = NULL;
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
return (ENOBUFS);
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
}
ip = (struct ip *)(mtod(m_head, char *) + ip_off);
poff = ip_off + (ip->ip_hl << 2);
if (do_tso) {
Fix for kern/152853, pullup at the wrong point is breaking UDP. Thanks to Petr Lampa for the patch.
2011-01-19 18:20:11 +00:00
m_head = m_pullup(m_head, poff + sizeof(struct tcphdr));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
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
tp = (struct tcphdr *)(mtod(m_head, char *) + poff);
/*
* TSO workaround:
* pull 4 more bytes of data into it.
*/
m_head = m_pullup(m_head, poff + (tp->th_off << 2) + 4);
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
In the data setup code for doing offloads the ip and tcp pointers were not reset after some pullups. In practice this led to an NFS mount failure when using UDP reported by Kevin Lo, thanks Kevin. Fix from yongari, thank you!
2010-10-28 00:16:54 +00:00
ip = (struct ip *)(mtod(m_head, char *) + ip_off);
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
ip->ip_len = 0;
ip->ip_sum = 0;
/*
* 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.
*/
In the data setup code for doing offloads the ip and tcp pointers were not reset after some pullups. In practice this led to an NFS mount failure when using UDP reported by Kevin Lo, thanks Kevin. Fix from yongari, thank you!
2010-10-28 00:16:54 +00:00
tp = (struct tcphdr *)(mtod(m_head, char *) + poff);
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
tp->th_sum = in_pseudo(ip->ip_src.s_addr,
ip->ip_dst.s_addr, htons(IPPROTO_TCP));
} else if (m_head->m_pkthdr.csum_flags & CSUM_TCP) {
Fix for kern/152853, pullup at the wrong point is breaking UDP. Thanks to Petr Lampa for the patch.
2011-01-19 18:20:11 +00:00
m_head = m_pullup(m_head, poff + sizeof(struct tcphdr));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
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
tp = (struct tcphdr *)(mtod(m_head, char *) + poff);
m_head = m_pullup(m_head, poff + (tp->th_off << 2));
if (m_head == NULL) {
*m_headp = NULL;
return (ENOBUFS);
}
In the data setup code for doing offloads the ip and tcp pointers were not reset after some pullups. In practice this led to an NFS mount failure when using UDP reported by Kevin Lo, thanks Kevin. Fix from yongari, thank you!
2010-10-28 00:16:54 +00:00
ip = (struct ip *)(mtod(m_head, char *) + ip_off);
tp = (struct tcphdr *)(mtod(m_head, char *) + poff);
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
} 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);
}
In the data setup code for doing offloads the ip and tcp pointers were not reset after some pullups. In practice this led to an NFS mount failure when using UDP reported by Kevin Lo, thanks Kevin. Fix from yongari, thank you!
2010-10-28 00:16:54 +00:00
ip = (struct ip *)(mtod(m_head, char *) + ip_off);
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
}
*m_headp = m_head;
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
}
Add support for TSO. Thanks to Andre for adding support in the stack and Jack Vogel for driver changes. Submitted by: Jack Vogel
2006-09-09 06:19:20 +00:00
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
/*
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
* Map the packet for DMA
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +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.
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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
first = txr->next_avail_desc;
tx_buffer = &txr->tx_buffers[first];
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
tx_buffer_mapped = tx_buffer;
Prepending an mbuf after loading a DMA map results in unexpected result. So, modify mbuf chains before loading a DMA map.
2006-07-27 00:43:34 +00:00
map = tx_buffer->map;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
Move the "retry:" label so that the calls to m_pullup() are not done after the call to m_defrag(). This fixes a problem where m_pullup() would prepend an mbuf to the list created by m_defrag() making the chain greater than 32 again. Tested by: rcarter@pinyon.org Reviewed by: yongari, jfv MFC after: 2 weeks
2014-07-15 23:32:13 +00:00
retry:
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_load_mbuf_sg(txr->txtag, map,
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
*m_headp, segs, &nsegs, BUS_DMA_NOWAIT);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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.
*/
Fix to an error condition case, when an mbuf chain get's defragged due to a mapping failure the header pointers will be invalidated and can result in a TSO or other failure down the line. So, when the remapping occurs force a retry thru the offload calculation code. Thanks to Andrew Boyer for discovering this and cooking up the fix!!
2011-04-01 20:24:51 +00:00
if (error == EFBIG && remap) {
Make em(4) handle too many fragmented frame with m_defrag(9). Previously em(4) requeued the failed mbuf chains from bus_dmamap_load_mbuf_sg(9) failure to resend it later. However, bus_dmamap_load_mbuf_sg(9) may never complete its request as the fragmented frames can have more than EM_MAX_SCATTER segments. To handle the above EFBIG case, defragment the frame with m_defrag(9) and free the mbuf chain if it can't deframent the chain due to resource shortage. Reviewed by glebius (with improvements)
2006-08-14 02:21:26 +00:00
struct mbuf *m;
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_defrag(*m_headp, M_NOWAIT);
Make em(4) handle too many fragmented frame with m_defrag(9). Previously em(4) requeued the failed mbuf chains from bus_dmamap_load_mbuf_sg(9) failure to resend it later. However, bus_dmamap_load_mbuf_sg(9) may never complete its request as the fragmented frames can have more than EM_MAX_SCATTER segments. To handle the above EFBIG case, defragment the frame with m_defrag(9) and free the mbuf chain if it can't deframent the chain due to resource shortage. Reviewed by glebius (with improvements)
2006-08-14 02:21:26 +00:00
if (m == NULL) {
adapter->mbuf_alloc_failed++;
m_freem(*m_headp);
*m_headp = NULL;
return (ENOBUFS);
}
*m_headp = m;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
Fix to an error condition case, when an mbuf chain get's defragged due to a mapping failure the header pointers will be invalidated and can result in a TSO or other failure down the line. So, when the remapping occurs force a retry thru the offload calculation code. Thanks to Andrew Boyer for discovering this and cooking up the fix!!
2011-04-01 20:24:51 +00:00
/* Try it again, but only once */
remap = 0;
goto retry;
Handle all error cases from bus_dmamap_load_mbuf_sg(). Those are: - EFBIG means the mbuf chain was too long and bus_dma ran out of segments. Defragment the mbuf chain and try again. (Already existed, not changed.) - 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. - Checking (nsegs == 0) is unnecessary as bus_dmamap_load_mbuf_sg() always reports an error if it is < 1. This prevents broken packets from clogging the interface queue indefinately. Discussed with: scottl Reviewed by: jfv
2006-09-29 13:12:38 +00:00
} else if (error != 0) {
adapter->no_tx_dma_setup++;
A couple problems discovered by Andrew Boyer: - failure code in em_xmit got mangled along the way and was not properly handling errors. - local timer code had a leftover UNLOCK call that should be removed. MFC after 3 days
2011-01-12 00:23:47 +00:00
m_freem(*m_headp);
*m_headp = NULL;
Handle all error cases from bus_dmamap_load_mbuf_sg(). Those are: - EFBIG means the mbuf chain was too long and bus_dma ran out of segments. Defragment the mbuf chain and try again. (Already existed, not changed.) - 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. - Checking (nsegs == 0) is unnecessary as bus_dmamap_load_mbuf_sg() always reports an error if it is < 1. This prevents broken packets from clogging the interface queue indefinately. Discussed with: scottl Reviewed by: jfv
2006-09-29 13:12:38 +00:00
return (error);
Make em(4) handle too many fragmented frame with m_defrag(9). Previously em(4) requeued the failed mbuf chains from bus_dmamap_load_mbuf_sg(9) failure to resend it later. However, bus_dmamap_load_mbuf_sg(9) may never complete its request as the fragmented frames can have more than EM_MAX_SCATTER segments. To handle the above EFBIG case, defragment the frame with m_defrag(9) and free the mbuf chain if it can't deframent the chain due to resource shortage. Reviewed by glebius (with improvements)
2006-08-14 02:21:26 +00:00
}
Prepending an mbuf after loading a DMA map results in unexpected result. So, modify mbuf chains before loading a DMA map.
2006-07-27 00:43:34 +00:00
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
/*
* TSO Hardware workaround, if this packet is not
* TSO, and is only a single descriptor long, and
* it follows a TSO burst, then we need to add a
* sentinel descriptor to prevent premature writeback.
*/
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 ((do_tso == 0) && (txr->tx_tso == TRUE)) {
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
if (nsegs == 1)
tso_desc = 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
txr->tx_tso = FALSE;
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
}
Add support for TSO. Thanks to Andre for adding support in the stack and Jack Vogel for driver changes. Submitted by: Jack Vogel
2006-09-09 06:19:20 +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 (nsegs > (txr->tx_avail - 2)) {
txr->no_desc_avail++;
bus_dmamap_unload(txr->txtag, map);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
return (ENOBUFS);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
}
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
m_head = *m_headp;
Prepending an mbuf after loading a DMA map results in unexpected result. So, modify mbuf chains before loading a DMA map.
2006-07-27 00:43:34 +00:00
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
/* Do hardware assists */
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
em_tso_setup(txr, m_head, ip_off, ip, tp,
&txd_upper, &txd_lower);
This delta adds two bug fixes: one that makes HW Offload logic in legacy codepath match the 82575, without this we were seeing bridging fail on 82546 adapters. Secondly, I have limited TSO to PCI Express adapters, I meant to do this and it got dropped in the earlier delta. Next, I am dropping in the latest shared code from our development team, consensus was that this should be done frequently, so I am :) Approved by: pdeuskar
2007-05-16 00:14:23 +00:00
/* we need to make a final sentinel transmit desc */
tso_desc = TRUE;
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
} else if (m_head->m_pkthdr.csum_flags & CSUM_OFFLOAD)
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
em_transmit_checksum_setup(txr, m_head,
ip_off, ip, &txd_upper, &txd_lower);
Prepending an mbuf after loading a DMA map results in unexpected result. So, modify mbuf chains before loading a DMA map.
2006-07-27 00:43:34 +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_flags & M_VLANTAG) {
/* Set the vlan id. */
Correct the setting of the VID in transmit descriptors when hardware VLAN tagging is enabled. This was broken in r266978. Reported by: gjb Tested by: gjb
2014-07-10 16:46:46 +00:00
txd_upper |= htole16(if_getvtag(m_head)) << 16;
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
/* Tell hardware to add tag */
txd_lower |= htole32(E1000_TXD_CMD_VLE);
}
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
i = txr->next_avail_desc;
Add support for TSO. Thanks to Andre for adding support in the stack and Jack Vogel for driver changes. Submitted by: Jack Vogel
2006-09-09 06:19:20 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* Set up our transmit descriptors */
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
for (j = 0; j < nsegs; j++) {
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
bus_size_t seg_len;
bus_addr_t seg_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
tx_buffer = &txr->tx_buffers[i];
ctxd = &txr->tx_base[i];
seg_addr = segs[j].ds_addr;
seg_len = segs[j].ds_len;
/*
** TSO Workaround:
** If this is the last descriptor, we want to
** split it so we have a small final sentinel
*/
if (tso_desc && (j == (nsegs -1)) && (seg_len > 8)) {
seg_len -= 4;
ctxd->buffer_addr = htole64(seg_addr);
ctxd->lower.data = htole32(
adapter->txd_cmd | txd_lower | seg_len);
ctxd->upper.data =
htole32(txd_upper);
if (++i == adapter->num_tx_desc)
i = 0;
/* Now make the sentinel */
++txd_used; /* using an extra txd */
ctxd = &txr->tx_base[i];
tx_buffer = &txr->tx_buffers[i];
ctxd->buffer_addr =
htole64(seg_addr + seg_len);
ctxd->lower.data = htole32(
adapter->txd_cmd | txd_lower | 4);
ctxd->upper.data =
htole32(txd_upper);
last = i;
if (++i == adapter->num_tx_desc)
i = 0;
Add support for new devices. Bug Fixes: - Allow users to use LAA - Remember promiscuous mode settings while bridging - Allow gratuitous arp's to be sent PR: 52966/54488 MFC after: 1 week
2003-08-27 21:52:37 +00:00
} else {
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
ctxd->buffer_addr = htole64(seg_addr);
ctxd->lower.data = htole32(
adapter->txd_cmd | txd_lower | seg_len);
ctxd->upper.data =
htole32(txd_upper);
last = i;
if (++i == adapter->num_tx_desc)
i = 0;
Add support for new devices. Bug Fixes: - Allow users to use LAA - Remember promiscuous mode settings while bridging - Allow gratuitous arp's to be sent PR: 52966/54488 MFC after: 1 week
2003-08-27 21:52:37 +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
tx_buffer->m_head = NULL;
tx_buffer->next_eop = -1;
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
}
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +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
txr->next_avail_desc = i;
txr->tx_avail -= nsegs;
if (tso_desc) /* TSO used an extra for sentinel */
txr->tx_avail -= txd_used;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
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.
*/
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
tx_buffer_mapped->map = tx_buffer->map;
Make sure to use the same DMA map in DMA map load/unload operations by remembering a map used in bus_dmamap_load_mbuf_sg(9). I have no idea how it could ever worked before. This fixes a warning generated by a diagnostic check in sun4v iommu driver. Reported by: jb Tested by: jb(sun4v)
2006-07-27 00:26:20 +00:00
tx_buffer->map = map;
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(txr->txtag, map, BUS_DMASYNC_PREWRITE);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/*
* Last Descriptor of Packet
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
* needs End Of Packet (EOP)
* and Report Status (RS)
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
*/
ctxd->lower.data |=
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
htole32(E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS);
/*
* Keep track in the first buffer which
* descriptor will be written back
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +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
tx_buffer = &txr->tx_buffers[first];
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
tx_buffer->next_eop = last;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
/*
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
* Advance the Transmit Descriptor Tail (TDT), this tells the E1000
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* that this frame is available to transmit.
*/
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(txr->txdma.dma_tag, txr->txdma.dma_map,
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
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, E1000_TDT(txr->me), i);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
return (0);
}
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_set_promisc(struct adapter *adapter)
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t 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
u32 reg_rctl;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getflags(ifp) & IFF_PROMISC) {
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
reg_rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
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
/* Turn this on if you want to see bad packets */
if (em_debug_sbp)
reg_rctl |= E1000_RCTL_SBP;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
} else if (if_getflags(ifp) & IFF_ALLMULTI) {
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
reg_rctl |= E1000_RCTL_MPE;
reg_rctl &= ~E1000_RCTL_UPE;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
Implement new ETHER_BPF_MTAP macro. Roll back the various changes made to accommodate the chip being in promiscuous mode while offloading VLAN tag processing to the hardware. We can now properly handle the absence of VLAN tags from hardware stripping. Reviewed by: rwatson, andre MFC after: 1 month
2006-11-18 23:18:43 +00:00
}
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
}
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_disable_promisc(struct adapter *adapter)
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
Correct the multicast handling in the E1000 drivers as was done in ixgbe, thanks to Mike Karels for this fix. When exiting promiscuous mode MPE bit was being unconditionally cleared, this should not be done if we are in MAX multicast groups.
2013-04-03 23:39:54 +00:00
u32 reg_rctl;
int mcnt = 0;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
reg_rctl &= (~E1000_RCTL_UPE);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getflags(ifp) & IFF_ALLMULTI)
Correct the multicast handling in the E1000 drivers as was done in ixgbe, thanks to Mike Karels for this fix. When exiting promiscuous mode MPE bit was being unconditionally cleared, this should not be done if we are in MAX multicast groups.
2013-04-03 23:39:54 +00:00
mcnt = MAX_NUM_MULTICAST_ADDRESSES;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
else
mcnt = if_multiaddr_count(ifp, MAX_NUM_MULTICAST_ADDRESSES);
Correct the multicast handling in the E1000 drivers as was done in ixgbe, thanks to Mike Karels for this fix. When exiting promiscuous mode MPE bit was being unconditionally cleared, this should not be done if we are in MAX multicast groups.
2013-04-03 23:39:54 +00:00
/* Don't disable if in MAX groups */
if (mcnt < MAX_NUM_MULTICAST_ADDRESSES)
reg_rctl &= (~E1000_RCTL_MPE);
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
reg_rctl &= (~E1000_RCTL_SBP);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
}
/*********************************************************************
* Multicast Update
*
* This routine is called whenever multicast address list is updated.
*
**********************************************************************/
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_set_multi(struct adapter *adapter)
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t 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
u32 reg_rctl = 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
u8 *mta; /* Multicast array memory */
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
int mcnt = 0;
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
IOCTL_DEBUGOUT("em_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(u8) * ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (adapter->hw.mac.type == e1000_82542 &&
adapter->hw.revision_id == E1000_REVISION_2) {
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
e1000_pci_clear_mwi(&adapter->hw);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
reg_rctl |= E1000_RCTL_RST;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
msec_delay(5);
}
Modify device drivers supporting multicast addresses to lock if_addr_mtx over iteration of their multicast address lists when synchronizing the hardware address filter with the network stack-maintained list. Problem reported by: Ed Maste (emaste at phaedrus dot sandvine dot ca> MFC after: 1 week
2005-08-03 00:18:35 +00:00
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_multiaddr_array(ifp, mta, &mcnt, MAX_NUM_MULTICAST_ADDRESSES);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
reg_rctl |= E1000_RCTL_MPE;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
} 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);
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (adapter->hw.mac.type == e1000_82542 &&
adapter->hw.revision_id == E1000_REVISION_2) {
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
reg_rctl &= ~E1000_RCTL_RST;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
msec_delay(5);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
e1000_pci_set_mwi(&adapter->hw);
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
}
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
}
/*********************************************************************
* Timer routine
*
* This routine checks for link status and updates statistics.
*
**********************************************************************/
static void
em_local_timer(void *arg)
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
struct adapter *adapter = arg;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
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 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
struct rx_ring *rxr = adapter->rx_rings;
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
u32 trigger = 0;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK_ASSERT(adapter);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_update_link_status(adapter);
em_update_stats_counters(adapter);
Couple of the fixes needed revising. The ICH8 autoneg was still broken, this change both simplifies the code and plugs a hole where the devise was reset without keeping the management controller at bay :) Second, the 82571 LAA reset problem was incomplete, this addition is necessary. Just one of those days :)
2007-05-31 23:36:21 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
/* Reset LAA into RAR[0] on 82571 */
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
if ((adapter->hw.mac.type == e1000_82571) &&
e1000_get_laa_state_82571(&adapter->hw))
Couple of the fixes needed revising. The ICH8 autoneg was still broken, this change both simplifies the code and plugs a hole where the devise was reset without keeping the management controller at bay :) Second, the 82571 LAA reset problem was incomplete, this addition is necessary. Just one of those days :)
2007-05-31 23:36:21 +00:00
e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
Change the refresh_mbuf logic slightly, add an inline to calculate the outstanding descriptors that need to be refreshed at any time, and use THAT in rxeof to determine if refreshing needs to be done. Also change the local_timer to simply fire off the appropriate interrupt rather than schedule a tasklet, its simpler. MFC in two weeks
2011-04-01 18:48:31 +00:00
/* Mask to use in the irq trigger */
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
if (adapter->msix_mem) {
for (int i = 0; i < adapter->num_queues; i++, rxr++)
trigger |= rxr->ims;
rxr = adapter->rx_rings;
} else
Change the refresh_mbuf logic slightly, add an inline to calculate the outstanding descriptors that need to be refreshed at any time, and use THAT in rxeof to determine if refreshing needs to be done. Also change the local_timer to simply fire off the appropriate interrupt rather than schedule a tasklet, its simpler. MFC in two weeks
2011-04-01 18:48:31 +00:00
trigger = E1000_ICS_RXDMT0;
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
Instead of using the legacy if_timer/if_watchdog interface create our own watchdog that piggybacks on the em_local_timer() routine. We suppose that the if_timer/if_watchdog interface should be obsoleted, since it doesn't fit the modern SMP network stack. NIC drivers should create their own watchdogs, that check and clear the timers always holding driver's lock. In collaboration with: jfv, scottl
2006-11-09 16:00:18 +00:00
/*
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
** Check on the state of the TX queue(s), this
** can be done without the lock because its RO
** and the HUNG state will be static if set.
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
*/
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++, txr++) {
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
if (txr->busy == EM_TX_HUNG)
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 hung;
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
if (txr->busy >= EM_TX_MAXTRIES)
txr->busy = EM_TX_HUNG;
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
/* Schedule a TX tasklet if needed */
if (txr->tx_avail <= EM_MAX_SCATTER)
taskqueue_enqueue(txr->tq, &txr->tx_task);
}
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
callout_reset(&adapter->timer, hz, em_local_timer, adapter);
Change the refresh_mbuf logic slightly, add an inline to calculate the outstanding descriptors that need to be refreshed at any time, and use THAT in rxeof to determine if refreshing needs to be done. Also change the local_timer to simply fire off the appropriate interrupt rather than schedule a tasklet, its simpler. MFC in two weeks
2011-04-01 18:48:31 +00:00
#ifndef DEVICE_POLLING
/* Trigger an RX interrupt to guarantee mbuf refresh */
E1000_WRITE_REG(&adapter->hw, E1000_ICS, trigger);
#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;
hung:
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
/* Looks like we're hung */
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
device_printf(adapter->dev, "Watchdog timeout Queue[%d]-- resetting\n",
txr->me);
em_print_debug_info(adapter);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
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
adapter->watchdog_events++;
em_init_locked(adapter);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +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
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_update_link_status(struct adapter *adapter)
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
{
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;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
device_t dev = adapter->dev;
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
struct tx_ring *txr = adapter->tx_rings;
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
u32 link_check = 0;
/* Get the cached link value or read phy 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;
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
if (link_check) /* ESB2 fix */
e1000_cfg_on_link_up(hw);
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
} 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;
default:
case e1000_media_type_unknown:
break;
}
Call if_link_state_change() when link status changes. PR: kern/76890 Reviewed by: rwatson, sam
2005-02-04 18:36:04 +00:00
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 check for a transition */
if (link_check && (adapter->link_active == 0)) {
e1000_get_speed_and_duplex(hw, &adapter->link_speed,
&adapter->link_duplex);
/* Check if we must disable SPEED_MODE bit on PCI-E */
if ((adapter->link_speed != SPEED_1000) &&
((hw->mac.type == e1000_82571) ||
(hw->mac.type == e1000_82572))) {
int tarc0;
tarc0 = E1000_READ_REG(hw, E1000_TARC(0));
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
tarc0 &= ~TARC_SPEED_MODE_BIT;
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
E1000_WRITE_REG(hw, E1000_TARC(0), tarc0);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
}
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 (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;
adapter->smartspeed = 0;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setbaudrate(ifp, adapter->link_speed * 1000000);
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
if_link_state_change(ifp, LINK_STATE_UP);
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
} else if (!link_check && (adapter->link_active == 1)) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setbaudrate(ifp, 0);
adapter->link_speed = 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
adapter->link_duplex = 0;
if (bootverbose)
device_printf(dev, "Link is Down\n");
adapter->link_active = 0;
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
/* Link down, disable hang detection */
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 (int i = 0; i < adapter->num_queues; i++, txr++)
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
txr->busy = EM_TX_IDLE;
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
if_link_state_change(ifp, LINK_STATE_DOWN);
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
}
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
/*********************************************************************
*
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
* This routine disables all traffic on the adapter by issuing a
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* global reset on the MAC and deallocates TX/RX buffers.
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*
Fix for a reported panic in certain circumstances. When calling em_stop() now make sure the TX lock is held as well as CORE.
2007-11-26 19:47:03 +00:00
* This routine should always be called with BOTH the CORE
* and TX locks.
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
**********************************************************************/
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
static void
em_stop(void *arg)
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
struct adapter *adapter = arg;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
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 tx_ring *txr = adapter->tx_rings;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK_ASSERT(adapter);
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
Add support for new devices. Bug Fixes: - Allow users to use LAA - Remember promiscuous mode settings while bridging - Allow gratuitous arp's to be sent PR: 52966/54488 MFC after: 1 week
2003-08-27 21:52:37 +00:00
INIT_DEBUGOUT("em_stop: begin");
Big polling(4) cleanup. o Axe poll in trap. o Axe IFF_POLLING flag from if_flags. o Rework revision 1.21 (Giant removal), in such a way that poll_mtx is not dropped during call to polling handler. This fixes problem with idle polling. o Make registration and deregistration from polling in a functional way, insted of next tick/interrupt. o Obsolete kern.polling.enable. Polling is turned on/off with ifconfig. Detailed kern_poll.c changes: - Remove polling handler flags, introduced in 1.21. The are not needed now. - Forget and do not check if_flags, if_capenable and if_drv_flags. - Call all registered polling handlers unconditionally. - Do not drop poll_mtx, when entering polling handlers. - In ether_poll() NET_LOCK_GIANT prior to locking poll_mtx. - In netisr_poll() axe the block, where polling code asks drivers to unregister. - In netisr_poll() and ether_poll() do polling always, if any handlers are present. - In ether_poll_[de]register() remove a lot of error hiding code. Assert that arguments are correct, instead. - In ether_poll_[de]register() use standard return values in case of error or success. - Introduce poll_switch() that is a sysctl handler for kern.polling.enable. poll_switch() goes through interface list and enabled/disables polling. A message that kern.polling.enable is deprecated is printed. Detailed driver changes: - On attach driver announces IFCAP_POLLING in if_capabilities, but not in if_capenable. - On detach driver calls ether_poll_deregister() if polling is enabled. - In polling handler driver obtains its lock and checks IFF_DRV_RUNNING flag. If there is no, then unlocks and returns. - In ioctl handler driver checks for IFCAP_POLLING flag requested to be set or cleared. Driver first calls ether_poll_[de]register(), then obtains driver lock and [dis/en]ables interrupts. - In interrupt handler driver checks IFCAP_POLLING flag in if_capenable. If present, then returns.This is important to protect from spurious interrupts. Reviewed by: ru, sam, jhb
2005-10-01 18:56:19 +00:00
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_disable_intr(adapter);
callout_stop(&adapter->timer);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
/* Tell the stack that the interface is no longer active */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
/* Disarm Hang Detection. */
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 (int i = 0; i < adapter->num_queues; i++, txr++) {
EM_TX_LOCK(txr);
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
txr->busy = EM_TX_IDLE;
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
EM_TX_UNLOCK(txr);
}
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
e1000_reset_hw(&adapter->hw);
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, 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 is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/*********************************************************************
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*
* Determine hardware revision.
*
**********************************************************************/
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_identify_hardware(struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
device_t dev = adapter->dev;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
/* Make sure our PCI config space has the necessary stuff set */
Update PCI drivers to no longer look at the MEMIO-enabled bit in the PCI command register. The lazy BAR allocation code in FreeBSD sometimes disables this bit when it detects a range conflict, and will re-enable it on demand when a driver allocates the BAR. Thus, the bit is no longer a reliable indication of capability, and should not be checked. This results in the elimination of a lot of code from drivers, and also gives the opportunity to simplify a lot of drivers to use a helper API to set the busmaster enable bit. This changes fixes some recent reports of disk controllers and their associated drives/enclosures disappearing during boot. Submitted by: jhb Reviewed by: jfv, marius, achadd, achim MFC after: 1 day
2013-08-12 23:30:01 +00:00
pci_enable_busmaster(dev);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
/* Save off the information about this board */
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter->hw.vendor_id = pci_get_vendor(dev);
adapter->hw.device_id = pci_get_device(dev);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1);
adapter->hw.subsystem_vendor_id =
pci_read_config(dev, PCIR_SUBVEND_0, 2);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
adapter->hw.subsystem_device_id =
pci_read_config(dev, PCIR_SUBDEV_0, 2);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/* Do Shared Code Init and Setup */
if (e1000_set_mac_type(&adapter->hw)) {
device_printf(dev, "Setup init failure\n");
return;
}
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
static int
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_allocate_pci_resources(struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
device_t dev = adapter->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
int rid;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Make em(4) work on big-endian architectures. - disable jumbo frame support on strict alignment architectures due to the limitation of hardware. The driver needs a fix-up code for RX side. The fix will show up in near future. - fix endian issue for 82544 on PCI-X bus. I couldn't test this as I don't have the NIC/hardware. - prefer PCIR_BAR to hardcoded EM_MMBA. - Properly checks for for 64bit BAR [1] - replace inl/outl with bus_space(9) [1] - fix endian issue on VLAN handling. - reorder header files and remove unnecessary one. Reviewed by: cognet No response from: pdeuskar, tackerman Obtained from: OpenBSD [1]
2005-11-09 08:43:18 +00:00
rid = PCIR_BAR(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
adapter->memory = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
&rid, RF_ACTIVE);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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->memory == NULL) {
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
device_printf(dev, "Unable to allocate bus resource: memory\n");
return (ENXIO);
}
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter->osdep.mem_bus_space_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
rman_get_bustag(adapter->memory);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
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
rman_get_bushandle(adapter->memory);
adapter->hw.hw_addr = (u8 *)&adapter->osdep.mem_bus_space_handle;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter->hw.back = &adapter->osdep;
Fix the interrupt race for real. Don't register the interrupt until after the the interface has been configured. I'm not sure how this could ever have worked before, but it should be fixed now. Also break out the interrupt degresitration function into it's own step.
2006-01-13 08:18:04 +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 (0);
Fix the interrupt race for real. Don't register the interrupt until after the the interface has been configured. I'm not sure how this could ever have worked before, but it should be fixed now. Also break out the interrupt degresitration function into it's own step.
2006-01-13 08:18:04 +00:00
}
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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
* Setup the Legacy or MSI Interrupt handler
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
*
**********************************************************************/
Revert previous commit and add myself to the list of people who should know better than to commit with a cat in the area.
2009-09-08 13:19:05 +00:00
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
em_allocate_legacy(struct adapter *adapter)
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
{
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
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 tx_ring *txr = adapter->tx_rings;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
int error, rid = 0;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
/* Manually turn off all interrupts */
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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 (adapter->msix == 1) /* using MSI */
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 */
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->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);
}
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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
* Allocate a fast interrupt and the associated
* deferred processing contexts.
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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
TASK_INIT(&adapter->que_task, 0, em_handle_que, adapter);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
adapter->tq = taskqueue_create_fast("em_taskq", M_NOWAIT,
taskqueue_thread_enqueue, &adapter->tq);
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_start_threads(&adapter->tq, 1, PI_NET, "%s que",
device_get_nameunit(adapter->dev));
/* Use a TX only tasklet for local timer */
TASK_INIT(&txr->tx_task, 0, em_handle_tx, txr);
txr->tq = taskqueue_create_fast("em_txq", M_NOWAIT,
taskqueue_thread_enqueue, &txr->tq);
taskqueue_start_threads(&txr->tq, 1, PI_NET, "%s txq",
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
device_get_nameunit(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
TASK_INIT(&adapter->link_task, 0, em_handle_link, 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
if ((error = bus_setup_intr(dev, adapter->res, INTR_TYPE_NET,
em_irq_fast, NULL, adapter, &adapter->tag)) != 0) {
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
device_printf(dev, "Failed to register fast interrupt "
"handler: %d\n", error);
taskqueue_free(adapter->tq);
adapter->tq = NULL;
return (error);
}
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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 MSIX Interrupt handlers
* This is not really Multiqueue, rather
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
* its just seperate interrupt vectors
* for TX, RX, and Link.
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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
em_allocate_msix(struct adapter *adapter)
Fix the interrupt race for real. Don't register the interrupt until after the the interface has been configured. I'm not sure how this could ever have worked before, but it should be fixed now. Also break out the interrupt degresitration function into it's own step.
2006-01-13 08:18:04 +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
device_t dev = adapter->dev;
struct tx_ring *txr = adapter->tx_rings;
struct rx_ring *rxr = adapter->rx_rings;
int error, rid, vector = 0;
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
int cpu_id = 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
Fix the interrupt race for real. Don't register the interrupt until after the the interface has been configured. I'm not sure how this could ever have worked before, but it should be fixed now. Also break out the interrupt degresitration function into it's own step.
2006-01-13 08:18:04 +00:00
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
/* Make sure all interrupts are disabled */
E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff);
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 set up ring resources */
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
for (int i = 0; i < adapter->num_queues; i++, rxr++, vector++) {
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
/* RX ring */
rid = vector + 1;
rxr->res = bus_alloc_resource_any(dev,
SYS_RES_IRQ, &rid, RF_ACTIVE);
if (rxr->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 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
"RX MSIX Interrupt %d\n", 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 (ENXIO);
}
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 ((error = bus_setup_intr(dev, rxr->res,
INTR_TYPE_NET | INTR_MPSAFE, NULL, em_msix_rx,
rxr, &rxr->tag)) != 0) {
device_printf(dev, "Failed to register RX handler");
return (error);
}
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
#if __FreeBSD_version >= 800504
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
bus_describe_intr(dev, rxr->res, rxr->tag, "rx%d", i);
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
#endif
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
rxr->msix = vector;
if (em_last_bind_cpu < 0)
em_last_bind_cpu = CPU_FIRST();
cpu_id = em_last_bind_cpu;
bus_bind_intr(dev, rxr->res, cpu_id);
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
TASK_INIT(&rxr->rx_task, 0, em_handle_rx, rxr);
rxr->tq = taskqueue_create_fast("em_rxq", M_NOWAIT,
taskqueue_thread_enqueue, &rxr->tq);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
taskqueue_start_threads(&rxr->tq, 1, PI_NET, "%s rxq (cpuid %d)",
device_get_nameunit(adapter->dev), cpu_id);
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
/*
** Set the bit to enable interrupt
** in E1000_IMS -- bits 20 and 21
** are for RX0 and RX1, note this has
** NOTHING to do with the MSIX vector
*/
rxr->ims = 1 << (20 + i);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
adapter->ims |= rxr->ims;
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->ivars |= (8 | rxr->msix) << (i * 4);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
em_last_bind_cpu = CPU_NEXT(em_last_bind_cpu);
}
for (int i = 0; i < adapter->num_queues; i++, txr++, vector++) {
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
/* TX ring */
rid = vector + 1;
txr->res = bus_alloc_resource_any(dev,
SYS_RES_IRQ, &rid, RF_ACTIVE);
if (txr->res == NULL) {
device_printf(dev,
"Unable to allocate bus resource: "
"TX MSIX Interrupt %d\n", i);
return (ENXIO);
}
if ((error = bus_setup_intr(dev, txr->res,
INTR_TYPE_NET | INTR_MPSAFE, NULL, em_msix_tx,
txr, &txr->tag)) != 0) {
device_printf(dev, "Failed to register TX handler");
return (error);
}
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
#if __FreeBSD_version >= 800504
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
bus_describe_intr(dev, txr->res, txr->tag, "tx%d", i);
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
#endif
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
txr->msix = vector;
if (em_last_bind_cpu < 0)
em_last_bind_cpu = CPU_FIRST();
cpu_id = em_last_bind_cpu;
bus_bind_intr(dev, txr->res, cpu_id);
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
TASK_INIT(&txr->tx_task, 0, em_handle_tx, txr);
txr->tq = taskqueue_create_fast("em_txq", M_NOWAIT,
taskqueue_thread_enqueue, &txr->tq);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
taskqueue_start_threads(&txr->tq, 1, PI_NET, "%s txq (cpuid %d)",
device_get_nameunit(adapter->dev), cpu_id);
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
/*
** Set the bit to enable interrupt
** in E1000_IMS -- bits 22 and 23
** are for TX0 and TX1, note this has
** NOTHING to do with the MSIX vector
*/
txr->ims = 1 << (22 + i);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
adapter->ims |= txr->ims;
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->ivars |= (8 | txr->msix) << (8 + (i * 4));
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
em_last_bind_cpu = CPU_NEXT(em_last_bind_cpu);
This is the merge of the Intel 6.2.9 driver. It provides all new shared code, new device support, and it is hoped a more stable driver for 6.2. RELEASE. This checkin was discussed and approved today by RE, scottl, jhb, and pdeuskar
2006-10-28 00:47:55 +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
/* Link interrupt */
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
rid = vector + 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
adapter->res = bus_alloc_resource_any(dev,
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE);
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 (!adapter->res) {
device_printf(dev,"Unable to allocate "
"bus resource: Link interrupt [%d]\n", rid);
return (ENXIO);
}
/* Set the link handler function */
error = bus_setup_intr(dev, adapter->res,
INTR_TYPE_NET | INTR_MPSAFE, NULL,
em_msix_link, adapter, &adapter->tag);
if (error) {
adapter->res = NULL;
device_printf(dev, "Failed to register LINK handler");
return (error);
}
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
#if __FreeBSD_version >= 800504
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
bus_describe_intr(dev, adapter->res, adapter->tag, "link");
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
#endif
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
adapter->linkvec = vector;
adapter->ivars |= (8 | vector) << 16;
adapter->ivars |= 0x80000000;
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
static void
em_free_pci_resources(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 tx_ring *txr;
struct rx_ring *rxr;
int rid;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
Couple of changes, back down on last TSO change, instead make old adapter list still capable, but only PCI-E adapters are now enabled. The user can enable older PCI-X or PCI adapters using ifconfig. Secondly, Arthur Hartwig pointed out my MSI change was not working correctly, changed to something that now does. Thanks Arthur. There was also a fundamental bug in the 82575 MSIX code, the MSIX registers had to be mapped, opps :) Rubber-stamped by: Pdeuskar
2007-05-17 00:14:03 +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
** Release all the queue interrupt resources:
*/
for (int i = 0; i < adapter->num_queues; i++) {
txr = &adapter->tx_rings[i];
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
/* an early abort? */
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
if (txr == NULL)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
break;
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
rid = txr->msix +1;
if (txr->tag != NULL) {
bus_teardown_intr(dev, txr->res, txr->tag);
txr->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 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 (txr->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
bus_release_resource(dev, SYS_RES_IRQ,
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
rid, txr->res);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
rxr = &adapter->rx_rings[i];
/* an early abort? */
if (rxr == NULL)
break;
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
rid = rxr->msix +1;
if (rxr->tag != NULL) {
bus_teardown_intr(dev, rxr->res, rxr->tag);
rxr->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 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->res != NULL)
bus_release_resource(dev, SYS_RES_IRQ,
rid, rxr->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
}
Add MSI support to em(4), bce(4), and mpt(4). For now, we only support devices that support a maximum of 1 message, and we use that 1 message instead of the INTx rid 0 IRQ with the same interrupt handler, etc.
2006-11-15 20:04:57 +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 (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);
if (adapter->msix)
Add MSI support to em(4), bce(4), and mpt(4). For now, we only support devices that support a maximum of 1 message, and we use that 1 message instead of the INTx rid 0 IRQ with the same interrupt handler, etc.
2006-11-15 20:04:57 +00:00
pci_release_msi(dev);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +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 (adapter->msix_mem != NULL)
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
bus_release_resource(dev, SYS_RES_MEMORY,
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
PCIR_BAR(EM_MSIX_BAR), adapter->msix_mem);
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared code that is up to the latest we have released. Some stylistic changes as well. Enjoy :)
2008-02-29 21:50:11 +00:00
if (adapter->memory != NULL)
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
bus_release_resource(dev, SYS_RES_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
PCIR_BAR(0), adapter->memory);
- Use IO mode to reset the controller (82544 and beyond) - Read the Mac address only once during attach. (This fixes the failover issue observed using the bonding driver) MFC after: 3 days
2002-07-16 16:55:03 +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 (adapter->flash != NULL)
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
bus_release_resource(dev, SYS_RES_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
EM_FLASH, adapter->flash);
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +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
* Setup MSI or MSI/X
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +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
static int
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
em_setup_msix(struct adapter *adapter)
{
device_t dev = adapter->dev;
Make the various driver MSIX setup routines fallback to MSI more gracefully. This change was suggested by Marius Strobl, thank you. PR: kern/181016 MFC after: ASAP
2013-08-06 21:01:38 +00:00
int val;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
/* Nearly always going to use one queue */
adapter->num_queues = 1;
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
/*
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
** Try using MSI-X for Hartwell adapters
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
if ((adapter->hw.mac.type == e1000_82574) &&
(em_enable_msix == TRUE)) {
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#ifdef EM_MULTIQUEUE
adapter->num_queues = (em_num_queues == 1) ? 1 : 2;
if (adapter->num_queues > 1)
em_enable_vectors_82574(adapter);
#endif
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
/* Map the MSIX BAR */
int rid = PCIR_BAR(EM_MSIX_BAR);
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->msix_mem = bus_alloc_resource_any(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
SYS_RES_MEMORY, &rid, RF_ACTIVE);
Make the various driver MSIX setup routines fallback to MSI more gracefully. This change was suggested by Marius Strobl, thank you. PR: kern/181016 MFC after: ASAP
2013-08-06 21:01:38 +00:00
if (adapter->msix_mem == NULL) {
This 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
/* May not be enabled */
device_printf(adapter->dev,
"Unable to map MSIX table \n");
goto msi;
}
val = pci_msix_count(dev);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#ifdef EM_MULTIQUEUE
/* We need 5 vectors in the multiqueue case */
if (adapter->num_queues > 1 ) {
if (val >= 5)
val = 5;
else {
adapter->num_queues = 1;
device_printf(adapter->dev,
"Insufficient MSIX vectors for >1 queue, "
"using single queue...\n");
goto msix_one;
}
} else {
msix_one:
#endif
if (val >= 3)
val = 3;
else {
device_printf(adapter->dev,
"Insufficient MSIX vectors, using MSI\n");
goto msi;
}
#ifdef EM_MULTIQUEUE
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
}
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#endif
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
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
if ((pci_alloc_msix(dev, &val) == 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
device_printf(adapter->dev,
"Using MSIX interrupts "
"with %d vectors\n", val);
Make the various driver MSIX setup routines fallback to MSI more gracefully. This change was suggested by Marius Strobl, thank you. PR: kern/181016 MFC after: ASAP
2013-08-06 21:01:38 +00:00
return (val);
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
}
Improve the MSIX setup code in the drivers, thanks to Marius for the changes. Make sure that pci_alloc_msix() does give us the vectors we need and fall back to MSI when it doesn't, also release any that were allocated when insufficient. MFC after: 3 days
2013-08-12 22:54:38 +00:00
/*
** If MSIX alloc failed or provided us with
** less than needed, free and fall through to MSI
*/
pci_release_msi(dev);
This 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
}
msi:
Improve the MSIX setup code in the drivers, thanks to Marius for the changes. Make sure that pci_alloc_msix() does give us the vectors we need and fall back to MSI when it doesn't, also release any that were allocated when insufficient. MFC after: 3 days
2013-08-12 22:54:38 +00:00
if (adapter->msix_mem != NULL) {
bus_release_resource(dev, SYS_RES_MEMORY,
PCIR_BAR(EM_MSIX_BAR), adapter->msix_mem);
adapter->msix_mem = NULL;
}
Make the various driver MSIX setup routines fallback to MSI more gracefully. This change was suggested by Marius Strobl, thank you. PR: kern/181016 MFC after: ASAP
2013-08-06 21:01:38 +00:00
val = 1;
if (pci_alloc_msi(dev, &val) == 0) {
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
device_printf(adapter->dev, "Using an MSI 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 (val);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
}
Tighten up the rx mbuf refresh code, there were some discrepencies from the igb version which was the target. Change the message when neither MSI or MSIX are enabled and a fallback to Legacy interrupts happen, the existing message was confusing.
2010-09-07 20:13:08 +00:00
/* Should only happen due to manual configuration */
device_printf(adapter->dev,"No MSI/MSIX using a Legacy IRQ\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 (0);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +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
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
*
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
* Initialize the hardware to a configuration
* as specified by the adapter structure.
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
em_reset(struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
device_t dev = adapter->dev;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
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 e1000_hw *hw = &adapter->hw;
u16 rx_buffer_size;
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
u32 pba;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
INIT_DEBUGOUT("em_reset: begin");
Added support for 82541 and 82547 based adapters. - These have Intel gigabit PHY - 82547 uses CSA interface MFC after: 1 week
2003-03-21 21:47:31 +00:00
Merge in new driver from Intel, version 6.0.5. It adds support for 80003 NICs and NICs found on ICH8 mobos, and improves support for already known chips. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. This was an easy task, because most changes between 5.1.5 and 6.0.5 are bugfixes taken from FreeBSD. - if_em_hw.h. Dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15. - if_em_osdep.h. Added new required macros from vendor file and add a hack against define namespace mangling in if_em_hw.h. Intel made another hack, but I prefer mine.
2006-08-03 09:20:11 +00:00
/* Set up smart power down as default off on newer adapters. */
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 (!em_smart_pwr_down && (hw->mac.type == e1000_82571 ||
hw->mac.type == e1000_82572)) {
This change introduces a split to the Intel E1000 driver, now rather than just em, there is an igb driver (this follows behavior with our Linux drivers). All adapters up to the 82575 are supported in em, and new client/desktop support will continue to be in that adapter. The igb driver is for new server NICs like the 82575 and its followons. Advanced features for virtualization and performance will be in this driver. Also, both drivers now have shared 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 phy_tmp = 0;
Merge in new driver from Intel, version 6.0.5. It adds support for 80003 NICs and NICs found on ICH8 mobos, and improves support for already known chips. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. This was an easy task, because most changes between 5.1.5 and 6.0.5 are bugfixes taken from FreeBSD. - if_em_hw.h. Dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15. - if_em_osdep.h. Added new required macros from vendor file and add a hack against define namespace mangling in if_em_hw.h. Intel made another hack, but I prefer mine.
2006-08-03 09:20:11 +00:00
/* Speed up time to link by disabling smart power down. */
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_read_phy_reg(hw, IGP02E1000_PHY_POWER_MGMT, &phy_tmp);
Merge in new driver from Intel, version 6.0.5. It adds support for 80003 NICs and NICs found on ICH8 mobos, and improves support for already known chips. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. This was an easy task, because most changes between 5.1.5 and 6.0.5 are bugfixes taken from FreeBSD. - if_em_hw.h. Dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15. - if_em_osdep.h. Added new required macros from vendor file and add a hack against define namespace mangling in if_em_hw.h. Intel made another hack, but I prefer mine.
2006-08-03 09:20:11 +00:00
phy_tmp &= ~IGP02E1000_PM_SPD;
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_phy_reg(hw, IGP02E1000_PHY_POWER_MGMT, phy_tmp);
Merge in new driver from Intel, version 6.0.5. It adds support for 80003 NICs and NICs found on ICH8 mobos, and improves support for already known chips. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. This was an easy task, because most changes between 5.1.5 and 6.0.5 are bugfixes taken from FreeBSD. - if_em_hw.h. Dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15. - if_em_osdep.h. Added new required macros from vendor file and add a hack against define namespace mangling in if_em_hw.h. Intel made another hack, but I prefer mine.
2006-08-03 09:20:11 +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
/*
* 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) {
/* Total Packet Buffer on these is 48K */
case e1000_82571:
case e1000_82572:
case e1000_80003es2lan:
pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
break;
case e1000_82573: /* 82573: Total Packet Buffer is 32K */
pba = E1000_PBA_12K; /* 12K for Rx, 20K for Tx */
break;
case e1000_82574:
case e1000_82583:
pba = E1000_PBA_20K; /* 20K for Rx, 20K for Tx */
break;
case e1000_ich8lan:
pba = E1000_PBA_8K;
break;
case e1000_ich9lan:
case e1000_ich10lan:
/* Boost Receive side for jumbo frames */
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
if (adapter->hw.mac.max_frame_size > 4096)
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
pba = E1000_PBA_14K;
else
pba = E1000_PBA_10K;
break;
case e1000_pchlan:
case e1000_pch2lan:
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
case e1000_pch_lpt:
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
pba = E1000_PBA_26K;
break;
default:
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
if (adapter->hw.mac.max_frame_size > 8192)
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
pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */
else
pba = E1000_PBA_48K; /* 48K for Rx, 16K for Tx */
}
E1000_WRITE_REG(&adapter->hw, E1000_PBA, pba);
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +00:00
/*
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* These parameters control the automatic generation (Tx) and
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +00:00
* 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.
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
* This allows the receiver to restart by sending XON when it has
* drained a bit. Here we use an arbitary value of 1500 which will
* restart after one full frame is pulled from the buffer. There
* could be several smaller frames in the buffer and if so they will
* not trigger the XON until their total number reduces the buffer
* by 1500.
Merge in new driver version from Intel - 3.2.18. The most important change is support for adapters based on 82571 and 82572 chips. Tested on: 82547EI on i386 Tested on: 82540EM on sparc64
2005-11-24 01:44:49 +00:00
* - The pause time is fairly large at 1000 x 512ns = 512 usec.
*/
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
rx_buffer_size = ((E1000_READ_REG(hw, E1000_PBA) & 0xffff) << 10 );
hw->fc.high_water = rx_buffer_size -
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
roundup2(adapter->hw.mac.max_frame_size, 1024);
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
hw->fc.low_water = hw->fc.high_water - 1500;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +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 (adapter->fc) /* locally set flow control value? */
hw->fc.requested_mode = adapter->fc;
else
hw->fc.requested_mode = e1000_fc_full;
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 (hw->mac.type == e1000_80003es2lan)
hw->fc.pause_time = 0xFFFF;
Merge in new driver from Intel, version 5.1.5. Adds support for some new chips and improves support for already supported ones. Some details, important for future merges: - if_em.c merged manually, viewing diff between new vendor driver and previous one. - if_em_hw.h dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c dropped in from vendor, and then two liner change made, that restores support for two rare chips.
2006-04-06 17:09:03 +00:00
else
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
hw->fc.pause_time = EM_FC_PAUSE_TIME;
hw->fc.send_xon = 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
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
/* Device specific overrides/settings */
switch (hw->mac.type) {
case e1000_pchlan:
/* Workaround: no TX flow ctrl for PCH */
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
hw->fc.requested_mode = e1000_fc_rx_pause;
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
hw->fc.pause_time = 0xFFFF; /* override */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getmtu(ifp) > ETHERMTU) {
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
hw->fc.high_water = 0x3500;
hw->fc.low_water = 0x1500;
} else {
hw->fc.high_water = 0x5000;
hw->fc.low_water = 0x3000;
}
hw->fc.refresh_time = 0x1000;
break;
case e1000_pch2lan:
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
case e1000_pch_lpt:
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
hw->fc.high_water = 0x5C20;
hw->fc.low_water = 0x5048;
hw->fc.pause_time = 0x0650;
hw->fc.refresh_time = 0x0400;
/* Jumbos need adjusted PBA */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getmtu(ifp) > ETHERMTU)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
E1000_WRITE_REG(hw, E1000_PBA, 12);
else
E1000_WRITE_REG(hw, E1000_PBA, 26);
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
break;
case e1000_ich9lan:
case e1000_ich10lan:
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getmtu(ifp) > ETHERMTU) {
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
hw->fc.high_water = 0x2800;
hw->fc.low_water = hw->fc.high_water - 8;
break;
}
/* else fall thru */
default:
if (hw->mac.type == e1000_80003es2lan)
hw->fc.pause_time = 0xFFFF;
break;
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +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
/* Issue a global reset */
e1000_reset_hw(hw);
E1000_WRITE_REG(hw, E1000_WUC, 0);
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
em_disable_aspm(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
/* and a re-init */
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 (e1000_init_hw(hw) < 0) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
device_printf(dev, "Hardware Initialization Failed\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
return;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
E1000_WRITE_REG(hw, E1000_VET, ETHERTYPE_VLAN);
e1000_get_phy_info(hw);
e1000_check_for_link(hw);
return;
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_setup_interface(device_t dev, struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
INIT_DEBUGOUT("em_setup_interface: begin");
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
ifp = adapter->ifp = if_gethandle(IFT_ETHER);
if (ifp == 0) {
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
device_printf(dev, "can not allocate ifnet structure\n");
return (-1);
}
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
if_initname(ifp, device_get_name(dev), device_get_unit(dev));
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setdev(ifp, dev);
if_setinitfn(ifp, em_init);
if_setsoftc(ifp, adapter);
if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
if_setioctlfn(ifp, em_ioctl);
- Use if_inc_counter() to increment various counters. - Do not ever set a counter to a value. For those counters that we don't increment, but return directly from hardware create cases in if_get_counter() method. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-09-18 15:56:14 +00:00
if_setgetcounterfn(ifp, em_get_counter);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#ifdef EM_MULTIQUEUE
/* Multiqueue stack interface */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_settransmitfn(ifp, em_mq_start);
if_setqflushfn(ifp, em_qflush);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#else
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setstartfn(ifp, em_start);
if_setsendqlen(ifp, adapter->num_tx_desc - 1);
if_setsendqready(ifp);
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
#endif
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +00:00
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
ether_ifattach(ifp, adapter->hw.mac.addr);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setcapabilities(ifp, 0);
if_setcapenable(ifp, 0);
Stylistic changes around the previous commit: - since the number of supported capabilities is growing, set bits in if_cap* in a consistent way; - unexpand(1) leading SPACE characters.
2004-05-20 11:04:09 +00:00
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +00:00
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setcapabilitiesbit(ifp, IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM |
IFCAP_TSO4, 0);
Tell the upper layer(s) that we support long frames. Not doing this caused the vlan mtu to be reduced by 4 bytes. Submitted by: Doug Ambrisko (ambrisko) MFC after: 1 day
2003-04-18 17:36:13 +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
Tell the upper layer(s) that we support long frames. Not doing this caused the vlan mtu to be reduced by 4 bytes. Submitted by: Doug Ambrisko (ambrisko) MFC after: 1 day
2003-04-18 17:36:13 +00:00
*/
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWTSO |
IFCAP_VLAN_MTU, 0);
if_setcapenable(ifp, if_getcapabilities(ifp));
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
/*
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
** 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
** using vlans directly on the em driver you can
** enable this and get full hardware tag filtering.
*/
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER,0);
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
Implemented per-interface polling(4) control.
2004-04-11 21:01:12 +00:00
#ifdef DEVICE_POLLING
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setcapabilitiesbit(ifp, IFCAP_POLLING,0);
Implemented per-interface polling(4) control.
2004-04-11 21:01:12 +00:00
#endif
Added support for 82541 and 82547 based adapters. - These have Intel gigabit PHY - 82547 uses CSA interface MFC after: 1 week
2003-03-21 21:47:31 +00:00
Change default WOL back to MAGIC only, having multicast enabled causes problems in man environments.
2010-04-28 17:37:30 +00:00
/* Enable only WOL MAGIC by default */
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 (adapter->wol) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setcapabilitiesbit(ifp, IFCAP_WOL, 0);
if_setcapenablebit(ifp, IFCAP_WOL_MAGIC, 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
}
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
/*
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
* Specify the media types supported by this adapter and register
* callbacks to update media and link information
*/
It is actually possible to have if_t a typedef to non-void type, and keep both converted to drvapi and non-converted drivers compilable. o Make if_t typedef to struct ifnet *. o Remove shim functions. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-08-31 12:48:13 +00:00
ifmedia_init(&adapter->media, IFM_IMASK,
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
em_media_change, em_media_status);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
if ((adapter->hw.phy.media_type == e1000_media_type_fiber) ||
(adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
u_char fiber_type = IFM_1000_SX; /* default type */
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
ifmedia_add(&adapter->media, IFM_ETHER | fiber_type | IFM_FDX,
0, NULL);
Merge in new driver from Intel, version 6.1.4. It adds support for 82571EB quad port copper NIC and has few minor fixes. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15.
2006-08-11 10:58:24 +00:00
ifmedia_add(&adapter->media, IFM_ETHER | fiber_type, 0, NULL);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
} else {
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T, 0, NULL);
ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T | IFM_FDX,
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
0, NULL);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX,
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
0, NULL);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
0, NULL);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (adapter->hw.phy.type != e1000_phy_ife) {
Merge in new driver from Intel, version 6.1.4. It adds support for 82571EB quad port copper NIC and has few minor fixes. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15.
2006-08-11 10:58:24 +00:00
ifmedia_add(&adapter->media,
10/100 PHY shouldn't support gigabit media types. Submitted by: brad (brad@comstyle.com) Obtained from: OpenBSD MFC after: 1 week
2006-08-09 20:10:35 +00:00
IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
Merge in new driver from Intel, version 6.1.4. It adds support for 82571EB quad port copper NIC and has few minor fixes. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15.
2006-08-11 10:58:24 +00:00
ifmedia_add(&adapter->media,
10/100 PHY shouldn't support gigabit media types. Submitted by: brad (brad@comstyle.com) Obtained from: OpenBSD MFC after: 1 week
2006-08-09 20:10:35 +00:00
IFM_ETHER | IFM_1000_T, 0, NULL);
}
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
}
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
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 is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +00:00
/*
* Manage DMA'able memory.
*/
static void
em_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
{
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
if (error)
return;
*(bus_addr_t *) arg = segs[0].ds_addr;
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +00:00
}
static int
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
em_dma_malloc(struct adapter *adapter, bus_size_t size,
struct em_dma_alloc *dma, int mapflags)
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
{
int error;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
error = bus_dma_tag_create(bus_get_dma_tag(adapter->dev), /* parent */
Merge in new driver from Intel, version 6.0.5. It adds support for 80003 NICs and NICs found on ICH8 mobos, and improves support for already known chips. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. This was an easy task, because most changes between 5.1.5 and 6.0.5 are bugfixes taken from FreeBSD. - if_em_hw.h. Dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15. - if_em_osdep.h. Added new required macros from vendor file and add a hack against define namespace mangling in if_em_hw.h. Intel made another hack, but I prefer mine.
2006-08-03 09:20:11 +00:00
EM_DBA_ALIGN, 0, /* alignment, bounds */
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +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) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
device_printf(adapter->dev,
"%s: bus_dma_tag_create failed: %d\n",
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
__func__, error);
goto fail_0;
}
error = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr,
Add COHERENT to descriptor mem allocation for the benefit of ARM (request from Olivier Houchard), its a noop on most architectures and goodness on those that use it.
2007-11-28 23:04:11 +00:00
BUS_DMA_NOWAIT | BUS_DMA_COHERENT, &dma->dma_map);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
if (error) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
device_printf(adapter->dev,
"%s: bus_dmamem_alloc(%ju) failed: %d\n",
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
__func__, (uintmax_t)size, error);
goto fail_2;
}
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +00:00
Make em(4) work on big-endian architectures. - disable jumbo frame support on strict alignment architectures due to the limitation of hardware. The driver needs a fix-up code for RX side. The fix will show up in near future. - fix endian issue for 82544 on PCI-X bus. I couldn't test this as I don't have the NIC/hardware. - prefer PCIR_BAR to hardcoded EM_MMBA. - Properly checks for for 64bit BAR [1] - replace inl/outl with bus_space(9) [1] - fix endian issue on VLAN handling. - reorder header files and remove unnecessary one. Reviewed by: cognet No response from: pdeuskar, tackerman Obtained from: OpenBSD [1]
2005-11-09 08:43:18 +00:00
dma->dma_paddr = 0;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
error = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr,
size, em_dmamap_cb, &dma->dma_paddr, mapflags | BUS_DMA_NOWAIT);
if (error || dma->dma_paddr == 0) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
device_printf(adapter->dev,
"%s: bus_dmamap_load failed: %d\n",
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
__func__, error);
goto fail_3;
}
return (0);
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +00:00
fail_3:
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
bus_dmamap_unload(dma->dma_tag, dma->dma_map);
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +00:00
fail_2:
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
bus_dma_tag_destroy(dma->dma_tag);
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +00:00
fail_0:
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
dma->dma_tag = NULL;
return (error);
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +00:00
}
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_dma_free(struct adapter *adapter, struct em_dma_alloc *dma)
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +00:00
{
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
if (dma->dma_tag == NULL)
return;
- Don't compare bus_dma map pointers for static DMA allocations against NULL to determine if bus_dmamap_unload() or bus_dmamem_free() should be called. Instead, check the associated bus and virtual addresses. - Don't clear static DMA maps to NULL. Reviewed by: jfv
2014-06-12 11:15:19 +00:00
if (dma->dma_paddr != 0) {
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
bus_dmamap_sync(dma->dma_tag, dma->dma_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(dma->dma_tag, dma->dma_map);
- Don't compare bus_dma map pointers for static DMA allocations against NULL to determine if bus_dmamap_unload() or bus_dmamem_free() should be called. Instead, check the associated bus and virtual addresses. - Don't clear static DMA maps to NULL. Reviewed by: jfv
2014-06-12 11:15:19 +00:00
dma->dma_paddr = 0;
}
if (dma->dma_vaddr != NULL) {
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
- Don't compare bus_dma map pointers for static DMA allocations against NULL to determine if bus_dmamap_unload() or bus_dmamem_free() should be called. Instead, check the associated bus and virtual addresses. - Don't clear static DMA maps to NULL. Reviewed by: jfv
2014-06-12 11:15:19 +00:00
dma->dma_vaddr = NULL;
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
}
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
bus_dma_tag_destroy(dma->dma_tag);
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
dma->dma_tag = NULL;
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +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
/*********************************************************************
*
* Allocate memory for the transmit and receive rings, and then
* the descriptors associated with each, called only once at attach.
*
**********************************************************************/
static int
em_allocate_queues(struct adapter *adapter)
{
device_t dev = adapter->dev;
struct tx_ring *txr = NULL;
struct rx_ring *rxr = NULL;
int rsize, tsize, error = E1000_SUCCESS;
int txconf = 0, rxconf = 0;
/* Allocate the TX ring struct memory */
if (!(adapter->tx_rings =
(struct tx_ring *) malloc(sizeof(struct tx_ring) *
adapter->num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) {
device_printf(dev, "Unable to allocate TX ring memory\n");
error = ENOMEM;
goto fail;
}
/* Now allocate the RX */
if (!(adapter->rx_rings =
(struct rx_ring *) malloc(sizeof(struct rx_ring) *
adapter->num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) {
device_printf(dev, "Unable to allocate RX ring memory\n");
error = ENOMEM;
goto rx_fail;
}
tsize = roundup2(adapter->num_tx_desc *
sizeof(struct e1000_tx_desc), EM_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
*/
for (int i = 0; i < adapter->num_queues; i++, txconf++) {
/* Set up some basics */
txr = &adapter->tx_rings[i];
txr->adapter = adapter;
txr->me = i;
/* Initialize the TX lock */
snprintf(txr->mtx_name, sizeof(txr->mtx_name), "%s:tx(%d)",
device_get_nameunit(dev), txr->me);
mtx_init(&txr->tx_mtx, txr->mtx_name, NULL, MTX_DEF);
if (em_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);
if (em_allocate_transmit_buffers(txr)) {
device_printf(dev,
"Critical Failure setting up transmit buffers\n");
error = ENOMEM;
goto err_tx_desc;
}
#if __FreeBSD_version >= 800000
/* Allocate a buf ring */
txr->br = buf_ring_alloc(4096, M_DEVBUF,
M_WAITOK, &txr->tx_mtx);
#endif
}
/*
* Next the RX queues...
*/
rsize = roundup2(adapter->num_rx_desc *
sizeof(struct e1000_rx_desc), EM_DBA_ALIGN);
for (int i = 0; i < adapter->num_queues; i++, rxconf++) {
rxr = &adapter->rx_rings[i];
rxr->adapter = adapter;
rxr->me = i;
/* Initialize the RX lock */
snprintf(rxr->mtx_name, sizeof(rxr->mtx_name), "%s:rx(%d)",
device_get_nameunit(dev), txr->me);
mtx_init(&rxr->rx_mtx, rxr->mtx_name, NULL, MTX_DEF);
if (em_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 = (struct e1000_rx_desc *)rxr->rxdma.dma_vaddr;
bzero((void *)rxr->rx_base, rsize);
/* Allocate receive buffers for the ring*/
if (em_allocate_receive_buffers(rxr)) {
device_printf(dev,
"Critical Failure setting up receive buffers\n");
error = ENOMEM;
goto err_rx_desc;
}
}
return (0);
err_rx_desc:
for (rxr = adapter->rx_rings; rxconf > 0; rxr++, rxconf--)
em_dma_free(adapter, &rxr->rxdma);
err_tx_desc:
for (txr = adapter->tx_rings; txconf > 0; txr++, txconf--)
em_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 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
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
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
free(adapter->tx_rings, M_DEVBUF);
fail:
return (error);
}
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
*
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* Allocate memory for tx_buffer structures. The tx_buffer stores all
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
* the information needed to transmit a packet on the wire. This is
* called only once at attach, setup is done every reset.
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*
**********************************************************************/
static int
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
em_allocate_transmit_buffers(struct tx_ring *txr)
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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 = txr->adapter;
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
device_t dev = adapter->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 em_buffer *txbuf;
int error, i;
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +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
* Setup DMA descriptor areas.
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +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 ((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 */
EM_TSO_SIZE, /* maxsize */
EM_MAX_SCATTER, /* nsegments */
PAGE_SIZE, /* maxsegsize */
0, /* flags */
NULL, /* lockfunc */
NULL, /* lockfuncarg */
&txr->txtag))) {
device_printf(dev,"Unable to allocate TX DMA tag\n");
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
goto fail;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
}
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +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 (!(txr->tx_buffers =
(struct em_buffer *) malloc(sizeof(struct em_buffer) *
adapter->num_tx_desc, M_DEVBUF, M_NOWAIT | M_ZERO))) {
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
device_printf(dev, "Unable to allocate tx_buffer memory\n");
error = ENOMEM;
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
goto fail;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
}
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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
/* 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);
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
if (error != 0) {
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
device_printf(dev, "Unable to create TX DMA map\n");
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
goto fail;
}
}
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +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 0;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
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
/* We free all, it handles case where we are in the middle */
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
em_free_transmit_structures(adapter);
return (error);
}
/*********************************************************************
*
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
* Initialize a transmit ring.
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
*
**********************************************************************/
static 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
em_setup_transmit_ring(struct tx_ring *txr)
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +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 adapter *adapter = txr->adapter;
struct em_buffer *txbuf;
int i;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
struct netmap_slot *slot;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
struct netmap_adapter *na = netmap_getna(adapter->ifp);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#endif /* DEV_NETMAP */
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +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
/* Clear the old descriptor contents */
EM_TX_LOCK(txr);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
slot = netmap_reset(na, NR_TX, txr->me, 0);
#endif /* DEV_NETMAP */
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
bzero((void *)txr->tx_base,
(sizeof(struct e1000_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;
}
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +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);
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;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
void *addr;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
addr = PNMB(na, slot + si, &paddr);
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
txr->tx_base[i].buffer_addr = htole64(paddr);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
/* reload the map for netmap mode */
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
netmap_load_map(na, txr->txtag, txbuf->map, addr);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
}
#endif /* DEV_NETMAP */
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 the watch index */
txbuf->next_eop = -1;
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +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
/* Set number of descriptors available */
txr->tx_avail = adapter->num_tx_desc;
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
txr->busy = EM_TX_IDLE;
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
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
/* Clear checksum offload context. */
txr->last_hw_offload = 0;
txr->last_hw_ipcss = 0;
txr->last_hw_ipcso = 0;
txr->last_hw_tucss = 0;
txr->last_hw_tucso = 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
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
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
EM_TX_UNLOCK(txr);
}
/*********************************************************************
*
* Initialize all transmit rings.
*
**********************************************************************/
static void
em_setup_transmit_structures(struct adapter *adapter)
{
struct tx_ring *txr = adapter->tx_rings;
for (int i = 0; i < adapter->num_queues; i++, txr++)
em_setup_transmit_ring(txr);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
return;
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
/*********************************************************************
*
* Enable transmit unit.
*
**********************************************************************/
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_initialize_transmit_unit(struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 tx_ring *txr = adapter->tx_rings;
struct e1000_hw *hw = &adapter->hw;
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
u32 tctl, txdctl = 0, tarc, tipg = 0;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
INIT_DEBUGOUT("em_initialize_transmit_unit: begin");
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 (int i = 0; i < adapter->num_queues; i++, txr++) {
u64 bus_addr = txr->txdma.dma_paddr;
/* Base and Len of TX Ring */
E1000_WRITE_REG(hw, E1000_TDLEN(i),
adapter->num_tx_desc * sizeof(struct e1000_tx_desc));
E1000_WRITE_REG(hw, E1000_TDBAH(i),
(u32)(bus_addr >> 32));
E1000_WRITE_REG(hw, E1000_TDBAL(i),
(u32)bus_addr);
/* Init the HEAD/TAIL indices */
E1000_WRITE_REG(hw, E1000_TDT(i), 0);
E1000_WRITE_REG(hw, E1000_TDH(i), 0);
HW_DEBUGOUT2("Base = %x, Length = %x\n",
E1000_READ_REG(&adapter->hw, E1000_TDBAL(i)),
E1000_READ_REG(&adapter->hw, E1000_TDLEN(i)));
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
txr->busy = EM_TX_IDLE;
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
txdctl = 0; /* clear txdctl */
txdctl |= 0x1f; /* PTHRESH */
txdctl |= 1 << 8; /* HTHRESH */
txdctl |= 1 << 16;/* WTHRESH */
txdctl |= 1 << 22; /* Reserved bit 22 must always be 1 */
txdctl |= E1000_TXDCTL_GRAN;
txdctl |= 1 << 25; /* LWTHRESH */
E1000_WRITE_REG(hw, E1000_TXDCTL(i), txdctl);
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
}
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
/* Set the default values for the Tx Inter Packet Gap timer */
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
switch (adapter->hw.mac.type) {
case e1000_80003es2lan:
tipg = DEFAULT_82543_TIPG_IPGR1;
tipg |= DEFAULT_80003ES2LAN_TIPG_IPGR2 <<
Merge in new driver from Intel, version 5.1.5. Adds support for some new chips and improves support for already supported ones. Some details, important for future merges: - if_em.c merged manually, viewing diff between new vendor driver and previous one. - if_em_hw.h dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c dropped in from vendor, and then two liner change made, that restores support for two rare chips.
2006-04-06 17:09:03 +00:00
E1000_TIPG_IPGR2_SHIFT;
break;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
default:
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
if ((adapter->hw.phy.media_type == e1000_media_type_fiber) ||
(adapter->hw.phy.media_type ==
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
e1000_media_type_internal_serdes))
tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
else
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
tipg |= DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
tipg |= DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
}
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_TIPG, tipg);
E1000_WRITE_REG(&adapter->hw, E1000_TIDV, adapter->tx_int_delay.value);
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
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if(adapter->hw.mac.type >= e1000_82540)
E1000_WRITE_REG(&adapter->hw, E1000_TADV,
adapter->tx_abs_int_delay.value);
if ((adapter->hw.mac.type == e1000_82571) ||
(adapter->hw.mac.type == e1000_82572)) {
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0));
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
tarc |= TARC_SPEED_MODE_BIT;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
} else if (adapter->hw.mac.type == e1000_80003es2lan) {
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
/* errata: program both queues to unweighted RR */
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0));
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
tarc |= 1;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc);
tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(1));
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
tarc |= 1;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_TARC(1), tarc);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
} else if (adapter->hw.mac.type == e1000_82574) {
tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0));
tarc |= TARC_ERRATA_BIT;
if ( adapter->num_queues > 1) {
tarc |= (TARC_COMPENSATION_MODE | TARC_MQ_FIX);
E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc);
E1000_WRITE_REG(&adapter->hw, E1000_TARC(1), tarc);
} else
E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
}
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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
adapter->txd_cmd = E1000_TXD_CMD_IFCS;
if (adapter->tx_int_delay.value > 0)
adapter->txd_cmd |= E1000_TXD_CMD_IDE;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
/* Program the Transmit Control Register */
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
tctl = E1000_READ_REG(&adapter->hw, E1000_TCTL);
tctl &= ~E1000_TCTL_CT;
A few small but significant fixes: - Coverity Prevent(tm) CID 1906 a bogus use of bzero where unneeded. - ICH8 systems autoneg to 100 rather than 1000, this can also be seen in 82573, the logic was backwards. - On new 82575 quadports half duplex tx speed is slow... this was due to overwriting TCTL reg rather than adding bits.
2007-05-30 23:32:21 +00:00
tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT));
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (adapter->hw.mac.type >= e1000_82571)
tctl |= E1000_TCTL_MULR;
Merge in new driver from Intel, version 6.0.5. It adds support for 80003 NICs and NICs found on ICH8 mobos, and improves support for already known chips. Details: - if_em.c. Merged manually, viewing diff between new vendor driver and previous one. This was an easy task, because most changes between 5.1.5 and 6.0.5 are bugfixes taken from FreeBSD. - if_em_hw.h. Dropped in from vendor, and then restored revisions 1.16, 1.17, 1.18. - if_em_hw.c. Dropped in from vendor, and then restored revision 1.15. - if_em_osdep.h. Added new required macros from vendor file and add a hack against define namespace mangling in if_em_hw.h. Intel made another hack, but I prefer mine.
2006-08-03 09:20:11 +00:00
/* This write will effectively turn on the transmit unit. */
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_TCTL, tctl);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
* Free all transmit rings.
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*
**********************************************************************/
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_free_transmit_structures(struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 tx_ring *txr = adapter->tx_rings;
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +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
for (int i = 0; i < adapter->num_queues; i++, txr++) {
EM_TX_LOCK(txr);
em_free_transmit_buffers(txr);
em_dma_free(adapter, &txr->txdma);
EM_TX_UNLOCK(txr);
EM_TX_LOCK_DESTROY(txr);
}
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +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
free(adapter->tx_rings, M_DEVBUF);
}
/*********************************************************************
*
* Free transmit ring related data structures.
*
**********************************************************************/
static void
em_free_transmit_buffers(struct tx_ring *txr)
{
struct adapter *adapter = txr->adapter;
struct em_buffer *txbuf;
INIT_DEBUGOUT("free_transmit_ring: begin");
if (txr->tx_buffers == NULL)
return;
for (int i = 0; i < adapter->num_tx_desc; i++) {
txbuf = &txr->tx_buffers[i];
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;
if (txbuf->map != NULL) {
bus_dmamap_destroy(txr->txtag,
txbuf->map);
txbuf->map = NULL;
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +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
} else if (txbuf->map != NULL) {
bus_dmamap_unload(txr->txtag,
txbuf->map);
bus_dmamap_destroy(txr->txtag,
txbuf->map);
txbuf->map = NULL;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +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
#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
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
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 (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;
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
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
* The offload context is protocol specific (TCP/UDP) and thus
* only needs to be set when the protocol changes. The occasion
* of a context change can be a performance detriment, and
* might be better just disabled. The reason arises in the way
* in which the controller supports pipelined requests from the
* Tx data DMA. Up to four requests can be pipelined, and they may
* belong to the same packet or to multiple packets. However all
* requests for one packet are issued before a request is issued
* for a subsequent packet and if a request for the next packet
* requires a context change, that request will be stalled
* until the previous request completes. This means setting up
* a new context effectively disables pipelined Tx data DMA which
* in turn greatly slow down performance to send small sized
* frames.
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
**********************************************************************/
Back out TX/RX descriptor/buffer management changes from earier commit. We are having panics with the driver under stress with jumbo frames. Unfortunately we didnot catch it during our regular test cycle. I am going to MFC the backout immediately.
2002-08-26 18:59:16 +00:00
static void
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
em_transmit_checksum_setup(struct tx_ring *txr, struct mbuf *mp, int ip_off,
struct ip *ip, u32 *txd_upper, u32 *txd_lower)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 = txr->adapter;
struct e1000_context_desc *TXD = 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
struct em_buffer *tx_buffer;
int cur, hdr_len;
u32 cmd = 0;
u16 offload = 0;
u8 ipcso, ipcss, tucso, tucss;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
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
ipcss = ipcso = tucss = tucso = 0;
hdr_len = ip_off + (ip->ip_hl << 2);
cur = txr->next_avail_desc;
Fix for a panic when TX checksum offload is done and a packet has only a header in the first mbuf, the checksum code will dereference a pointer into the non-existing IP header. Do a check for the size and pullup if needed. Thanks to Michael Tuexen for this fix. MFC: asap - should be in 8.1 IMHO
2010-07-12 21:47:30 +00:00
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
/* Setup of IP header checksum. */
if (mp->m_pkthdr.csum_flags & CSUM_IP) {
*txd_upper |= E1000_TXD_POPTS_IXSM << 8;
offload |= CSUM_IP;
ipcss = ip_off;
ipcso = ip_off + offsetof(struct ip, ip_sum);
/*
* Start offset for header checksum calculation.
* End offset for header checksum calculation.
* Offset of place to put the checksum.
*/
TXD = (struct e1000_context_desc *)&txr->tx_base[cur];
TXD->lower_setup.ip_fields.ipcss = ipcss;
TXD->lower_setup.ip_fields.ipcse = htole16(hdr_len);
TXD->lower_setup.ip_fields.ipcso = ipcso;
cmd |= E1000_TXD_CMD_IP;
Change em_transmit_checksum_setup() to deal with already inserted vlan headers, IP options and add skeleton IPv6 support. The new code structure can also be easily enhanced to support new/more protocols (SCTP) in the future. Reviewed by: jfv
2006-09-29 13:37:26 +00:00
}
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
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
if (mp->m_pkthdr.csum_flags & CSUM_TCP) {
*txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
*txd_upper |= E1000_TXD_POPTS_TXSM << 8;
offload |= CSUM_TCP;
tucss = hdr_len;
tucso = hdr_len + offsetof(struct tcphdr, th_sum);
/*
* Setting up new checksum offload context for every frames
* takes a lot of processing time for hardware. This also
* reduces performance a lot for small sized frames so avoid
* it if driver can use previously configured checksum
* offload context.
*/
if (txr->last_hw_offload == offload) {
if (offload & CSUM_IP) {
if (txr->last_hw_ipcss == ipcss &&
txr->last_hw_ipcso == ipcso &&
txr->last_hw_tucss == tucss &&
txr->last_hw_tucso == tucso)
return;
} else {
if (txr->last_hw_tucss == tucss &&
txr->last_hw_tucso == tucso)
return;
}
}
txr->last_hw_offload = offload;
txr->last_hw_tucss = tucss;
txr->last_hw_tucso = tucso;
/*
* Start offset for payload checksum calculation.
* End offset for payload checksum calculation.
* Offset of place to put the checksum.
*/
TXD = (struct e1000_context_desc *)&txr->tx_base[cur];
TXD->upper_setup.tcp_fields.tucss = hdr_len;
TXD->upper_setup.tcp_fields.tucse = htole16(0);
TXD->upper_setup.tcp_fields.tucso = tucso;
cmd |= E1000_TXD_CMD_TCP;
} else if (mp->m_pkthdr.csum_flags & CSUM_UDP) {
*txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
*txd_upper |= E1000_TXD_POPTS_TXSM << 8;
tucss = hdr_len;
tucso = hdr_len + offsetof(struct udphdr, uh_sum);
/*
* Setting up new checksum offload context for every frames
* takes a lot of processing time for hardware. This also
* reduces performance a lot for small sized frames so avoid
* it if driver can use previously configured checksum
* offload context.
*/
if (txr->last_hw_offload == offload) {
if (offload & CSUM_IP) {
if (txr->last_hw_ipcss == ipcss &&
txr->last_hw_ipcso == ipcso &&
txr->last_hw_tucss == tucss &&
txr->last_hw_tucso == tucso)
return;
} else {
if (txr->last_hw_tucss == tucss &&
txr->last_hw_tucso == tucso)
return;
}
}
txr->last_hw_offload = offload;
txr->last_hw_tucss = tucss;
txr->last_hw_tucso = tucso;
/*
* Start offset for header checksum calculation.
* End offset for header checksum calculation.
* Offset of place to put the checksum.
*/
TXD = (struct e1000_context_desc *)&txr->tx_base[cur];
TXD->upper_setup.tcp_fields.tucss = tucss;
TXD->upper_setup.tcp_fields.tucse = htole16(0);
TXD->upper_setup.tcp_fields.tucso = tucso;
}
if (offload & CSUM_IP) {
txr->last_hw_ipcss = ipcss;
txr->last_hw_ipcso = ipcso;
}
This is the merge of the Intel 6.2.9 driver. It provides all new shared code, new device support, and it is hoped a more stable driver for 6.2. RELEASE. This checkin was discussed and approved today by RE, scottl, jhb, and pdeuskar
2006-10-28 00:47:55 +00:00
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
TXD->tcp_seg_setup.data = htole32(0);
TXD->cmd_and_length =
htole32(adapter->txd_cmd | E1000_TXD_CMD_DEXT | cmd);
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
tx_buffer = &txr->tx_buffers[cur];
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +00:00
tx_buffer->m_head = NULL;
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
tx_buffer->next_eop = -1;
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +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 (++cur == adapter->num_tx_desc)
cur = 0;
Back out TX/RX descriptor/buffer management changes from earier commit. We are having panics with the driver under stress with jumbo frames. Unfortunately we didnot catch it during our regular test cycle. I am going to MFC the backout immediately.
2002-08-26 18:59:16 +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
txr->tx_avail--;
txr->next_avail_desc = cur;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
Add support for TSO. Thanks to Andre for adding support in the stack and Jack Vogel for driver changes. Submitted by: Jack Vogel
2006-09-09 06:19:20 +00:00
/**********************************************************************
*
* Setup work for hardware segmentation offload (TSO)
*
**********************************************************************/
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
static void
em_tso_setup(struct tx_ring *txr, struct mbuf *mp, int ip_off,
struct ip *ip, struct tcphdr *tp, u32 *txd_upper, u32 *txd_lower)
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +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
struct adapter *adapter = txr->adapter;
struct e1000_context_desc *TXD;
struct em_buffer *tx_buffer;
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
int cur, hdr_len;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
/*
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
* In theory we can use the same TSO context if and only if
* frame is the same type(IP/TCP) and the same MSS. However
* checking whether a frame has the same IP/TCP structure is
* hard thing so just ignore that and always restablish a
* new TSO context.
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
*/
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
hdr_len = ip_off + (ip->ip_hl << 2) + (tp->th_off << 2);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
*txd_lower = (E1000_TXD_CMD_DEXT | /* Extended descr type */
E1000_TXD_DTYP_D | /* Data descr type */
E1000_TXD_CMD_TSE); /* Do TSE on this packet */
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
/* IP and/or TCP header checksum calculation and insertion. */
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
*txd_upper = (E1000_TXD_POPTS_IXSM | E1000_TXD_POPTS_TXSM) << 8;
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +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
cur = txr->next_avail_desc;
tx_buffer = &txr->tx_buffers[cur];
TXD = (struct e1000_context_desc *) &txr->tx_base[cur];
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
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
/*
* Start offset for header checksum calculation.
* End offset for header checksum calculation.
* Offset of place put the checksum.
*/
TXD->lower_setup.ip_fields.ipcss = ip_off;
TXD->lower_setup.ip_fields.ipcse =
htole16(ip_off + (ip->ip_hl << 2) - 1);
TXD->lower_setup.ip_fields.ipcso = ip_off + offsetof(struct ip, ip_sum);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
/*
* Start offset for payload checksum calculation.
* End offset for payload checksum calculation.
* Offset of place to put the checksum.
*/
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
TXD->upper_setup.tcp_fields.tucss = ip_off + (ip->ip_hl << 2);
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
TXD->upper_setup.tcp_fields.tucse = 0;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
TXD->upper_setup.tcp_fields.tucso =
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
ip_off + (ip->ip_hl << 2) + offsetof(struct tcphdr, th_sum);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
/*
* Payload size per packet w/o any headers.
* Length of all headers up to payload.
*/
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
TXD->tcp_seg_setup.fields.mss = htole16(mp->m_pkthdr.tso_segsz);
TXD->tcp_seg_setup.fields.hdr_len = hdr_len;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
TXD->cmd_and_length = htole32(adapter->txd_cmd |
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
E1000_TXD_CMD_DEXT | /* Extended descr */
E1000_TXD_CMD_TSE | /* TSE context */
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
E1000_TXD_CMD_IP | /* Do IP csum */
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
E1000_TXD_CMD_TCP | /* Do TCP checksum */
(mp->m_pkthdr.len - (hdr_len))); /* Total len */
Add support for TSO. Thanks to Andre for adding support in the stack and Jack Vogel for driver changes. Submitted by: Jack Vogel
2006-09-09 06:19:20 +00:00
Fix style(9) issues in the TSO specific changes. Pointed out by: jmallett
2006-09-10 19:23:27 +00:00
tx_buffer->m_head = NULL;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
tx_buffer->next_eop = -1;
Add support for TSO. Thanks to Andre for adding support in the stack and Jack Vogel for driver changes. Submitted by: Jack Vogel
2006-09-09 06:19:20 +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 (++cur == adapter->num_tx_desc)
cur = 0;
Add support for TSO. Thanks to Andre for adding support in the stack and Jack Vogel for driver changes. Submitted by: Jack Vogel
2006-09-09 06:19:20 +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
txr->tx_avail--;
txr->next_avail_desc = cur;
txr->tx_tso = TRUE;
Add support for TSO. Thanks to Andre for adding support in the stack and Jack Vogel for driver changes. Submitted by: Jack Vogel
2006-09-09 06:19:20 +00:00
}
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
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
/**********************************************************************
*
* Examine each tx_buffer in the used queue. If the hardware is done
* processing the packet then free associated resources. The
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +00:00
* tx_buffer is put back on the free queue.
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
*
**********************************************************************/
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
static void
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
em_txeof(struct tx_ring *txr)
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +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 = txr->adapter;
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
int first, last, done, processed;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
struct em_buffer *tx_buffer;
struct e1000_tx_desc *tx_desc, *eop_desc;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +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
EM_TX_LOCK_ASSERT(txr);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
update to the latest netmap snapshot. This includes the following: - use separate memory regions for VALE ports - locking fixes - some simplifications in the NIC-specific routines - performance improvements for the VALE switch - some new features in the pkt-gen test program - documentation updates There are small API changes that require programs to be recompiled (NETMAP_API has been bumped so you will detect old binaries at runtime). In particular: - struct netmap_slot now is 16 bytes to support an extra pointer, which may save one data copy when using VALE ports or VMs; - the struct netmap_if has two extra fields; MFC after: 3 days
2013-11-01 21:21:14 +00:00
if (netmap_tx_irq(ifp, txr->me))
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
return;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#endif /* DEV_NETMAP */
add locking Reviewed by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> Sponsored by: FreeBSD Foundation
2003-09-23 00:18:25 +00:00
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
/* No work, make sure hang detection is disabled */
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
if (txr->tx_avail == adapter->num_tx_desc) {
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
txr->busy = EM_TX_IDLE;
Fix a few issues with transmit handling in em(4) and igb(4): - Do not define the foo_start() methods or set if_start in the ifnet if multiq transmit is enabled. Also, set if_transmit and if_qflush before ether_ifattach rather than after when multiq transmit is enabled. This helps to ensure that the drivers never try to mix different transmit methods. - Properly restart transmit during resume. igb(4) was not restarting it at all, and em(4) was restarting even if the link was down and was calling the wrong method if multiq transmit was enabled. - Remove all the 'more' handling for transmit completions. Transmit completion processing does not have a processing limit, so it always runs to completion and never has more work to do when it returns. Instead, the previous code was returning 'true' anytime there were packets in the queue that weren't still in the process of being transmitted. The effect was that the driver would continuously reschedule a task to process TX completions in effect running at 100% CPU polling the hardware until it finished transmitting all of the packets in the ring. Now it will just wait for the next TX completion interrupt. - Restart packet transmission when the link becomes active. - Fix the MSI-X queue interrupt handlers to restart packet transmission if there are pending packets in the relevant software queue (IFQ or buf_ring) after processing TX completions. This is the root cause for the OACTIVE hangs as if the MSI-X queue handler drained all the pending packets from the TX ring, nothing would ever restart it. As such, remove some previously-added workarounds to reschedule a task to poll the TX ring anytime OACTIVE was set. Tested by: sbruno Reviewed by: jfv MFC after: 1 week
2012-03-30 19:54:48 +00:00
return;
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
}
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
processed = 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
first = txr->next_to_clean;
tx_desc = &txr->tx_base[first];
tx_buffer = &txr->tx_buffers[first];
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
last = tx_buffer->next_eop;
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
eop_desc = &txr->tx_base[last];
- Set RS (Report Status) bit on all descriptors of a packet instead of just the last one. - Set RDTR to zero by default instead of 28. - Fixed a problem with TX hangs with jumbo frames when number of fragments in the mbuf chain is large. - Added support for 82540EP based cards. MFC after: 3 days
2002-11-08 18:14:17 +00:00
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
/*
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
* 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.
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
*/
if (++last == adapter->num_tx_desc)
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
last = 0;
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
done = last;
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +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
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
BUS_DMASYNC_POSTREAD);
while (eop_desc->upper.fields.status & E1000_TXD_STAT_DD) {
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
/* We clean the range of the packet */
while (first != done) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
tx_desc->upper.data = 0;
tx_desc->lower.data = 0;
tx_desc->buffer_addr = 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
++txr->tx_avail;
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
++processed;
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
if (tx_buffer->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
bus_dmamap_sync(txr->txtag,
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
tx_buffer->map,
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
BUS_DMASYNC_POSTWRITE);
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(txr->txtag,
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
tx_buffer->map);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
m_freem(tx_buffer->m_head);
tx_buffer->m_head = NULL;
}
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
tx_buffer->next_eop = -1;
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (++first == adapter->num_tx_desc)
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
first = 0;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +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
tx_buffer = &txr->tx_buffers[first];
tx_desc = &txr->tx_base[first];
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
}
- Use if_inc_counter() to increment various counters. - Do not ever set a counter to a value. For those counters that we don't increment, but return directly from hardware create cases in if_get_counter() method. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-09-18 15:56:14 +00:00
if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
/* See if we can continue to the next packet */
last = tx_buffer->next_eop;
if (last != -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
eop_desc = &txr->tx_base[last];
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
/* Get new done point */
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (++last == adapter->num_tx_desc) last = 0;
Rework the transmit register handling. In em_encap() store index of the EOP descriptor in the first descriptor of the packet. And then in em_txeof() search for DD bits set only in the EOP descriptors, embedding the cleanup of all packet's descriptors into inner loop. This change is important for future chips, where DD bit is going to be set only on the EOP descriptors. Submitted by: jfv
2006-10-31 16:19:21 +00:00
done = last;
} else
break;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
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
txr->next_to_clean = first;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
/*
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
** Hang detection: we know there's work outstanding
** or the entry return would have been taken, so no
** descriptor processed here indicates a potential hang.
** The local timer will examine this and do a reset if needed.
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
*/
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
if (processed == 0) {
if (txr->busy != EM_TX_HUNG)
++txr->busy;
} else /* At least one descriptor was cleaned */
txr->busy = EM_TX_BUSY; /* note this clears HUNG */
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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
* If we have a minimum free, clear IFF_DRV_OACTIVE
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
* to tell the stack that it is OK to send packets.
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
* Notice that all writes of OACTIVE happen under the
* TX lock which, with a single queue, guarantees
* sanity.
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
*/
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
if (txr->tx_avail >= EM_MAX_SCATTER) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +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
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
/* Disable hang detection if all clean */
if (txr->tx_avail == adapter->num_tx_desc)
txr->busy = EM_TX_IDLE;
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 RX descriptor mbufs from system mbuf buffer pool.
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
em_refresh_mbufs(struct rx_ring *rxr, int limit)
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +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 = rxr->adapter;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
struct mbuf *m;
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
bus_dma_segment_t segs[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 em_buffer *rxbuf;
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, error, nsegs;
bool cleaned = FALSE;
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;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
while (j != limit) {
Tighten up the rx mbuf refresh code, there were some discrepencies from the igb version which was the target. Change the message when neither MSI or MSIX are enabled and a fallback to Legacy interrupts happen, the existing message was confusing.
2010-09-07 20:13:08 +00:00
rxbuf = &rxr->rx_buffers[i];
Small cut and paste bug in flow control string fixed. Second, correct the discard/refresh_mbufs code to behave more like igb, there have been panics due to discards and this should fix them. MFC after: 3 days
2010-12-04 01:59:58 +00:00
if (rxbuf->m_head == NULL) {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_getjcl(M_NOWAIT, MT_DATA,
Small cut and paste bug in flow control string fixed. Second, correct the discard/refresh_mbufs code to behave more like igb, there have been panics due to discards and this should fix them. MFC after: 3 days
2010-12-04 01:59:58 +00:00
M_PKTHDR, adapter->rx_mbuf_sz);
/*
** If we have a temporary resource shortage
** that causes a failure, just abort refresh
** for now, we will return to this point when
** reinvoked from em_rxeof.
*/
if (m == NULL)
goto update;
} else
m = rxbuf->m_head;
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
m->m_len = m->m_pkthdr.len = adapter->rx_mbuf_sz;
Small cut and paste bug in flow control string fixed. Second, correct the discard/refresh_mbufs code to behave more like igb, there have been panics due to discards and this should fix them. MFC after: 3 days
2010-12-04 01:59:58 +00:00
m->m_flags |= M_PKTHDR;
m->m_data = m->m_ext.ext_buf;
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
Tighten up the rx mbuf refresh code, there were some discrepencies from the igb version which was the target. Change the message when neither MSI or MSIX are enabled and a fallback to Legacy interrupts happen, the existing message was confusing.
2010-09-07 20:13:08 +00:00
/* Use bus_dma machinery to setup the memory mapping */
error = bus_dmamap_load_mbuf_sg(rxr->rxtag, rxbuf->map,
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, segs, &nsegs, BUS_DMA_NOWAIT);
if (error != 0) {
Small cut and paste bug in flow control string fixed. Second, correct the discard/refresh_mbufs code to behave more like igb, there have been panics due to discards and this should fix them. MFC after: 3 days
2010-12-04 01:59:58 +00:00
printf("Refresh mbufs: hdr dmamap load"
" failure - %d\n", error);
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_free(m);
Small cut and paste bug in flow control string fixed. Second, correct the discard/refresh_mbufs code to behave more like igb, there have been panics due to discards and this should fix them. MFC after: 3 days
2010-12-04 01:59:58 +00:00
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
goto update;
}
Small cut and paste bug in flow control string fixed. Second, correct the discard/refresh_mbufs code to behave more like igb, there have been panics due to discards and this should fix them. MFC after: 3 days
2010-12-04 01:59:58 +00:00
rxbuf->m_head = m;
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->rxtag,
rxbuf->map, BUS_DMASYNC_PREREAD);
rxr->rx_base[i].buffer_addr = htole64(segs[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
cleaned = TRUE;
Small cut and paste bug in flow control string fixed. Second, correct the discard/refresh_mbufs code to behave more like igb, there have been panics due to discards and this should fix them. MFC after: 3 days
2010-12-04 01:59:58 +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; /* Next is precalulated for us */
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
/* Calculate next controlling index */
if (++j == adapter->num_rx_desc)
j = 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
}
update:
Tighten up the rx mbuf refresh code, there were some discrepencies from the igb version which was the target. Change the message when neither MSI or MSIX are enabled and a fallback to Legacy interrupts happen, the existing message was confusing.
2010-09-07 20:13:08 +00:00
/*
** Update the tail pointer only if,
** and as far as we have refreshed.
*/
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 (cleaned)
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);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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 is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
*
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +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
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
* that we've allocated.
*
**********************************************************************/
static int
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
em_allocate_receive_buffers(struct rx_ring *rxr)
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +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 = rxr->adapter;
device_t dev = adapter->dev;
struct em_buffer *rxbuf;
int error;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +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
rxr->rx_buffers = malloc(sizeof(struct em_buffer) *
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
adapter->num_rx_desc, M_DEVBUF, M_NOWAIT | M_ZERO);
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->rx_buffers == NULL) {
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
device_printf(dev, "Unable to allocate rx_buffer memory\n");
return (ENOMEM);
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
error = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
1, 0, /* alignment, bounds */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
MJUM9BYTES, /* maxsize */
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
1, /* nsegments */
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
MJUM9BYTES, /* maxsegsize */
Nuke invalid use of BUS_DMA_ALLOCNOW.
2006-07-27 00:29:44 +00:00
0, /* flags */
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
NULL, /* lockfunc */
NULL, /* lockarg */
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->rxtag);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
if (error) {
device_printf(dev, "%s: bus_dma_tag_create failed %d\n",
__func__, error);
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
rxbuf = rxr->rx_buffers;
for (int i = 0; i < adapter->num_rx_desc; i++, rxbuf++) {
rxbuf = &rxr->rx_buffers[i];
Fix several issues with the busdma(9) KPI use in the e1000 drivers. The problems do not affect bouncing busdma in a visible way, but are critical for the dmar backend. - The bus_dmamap_create(9) is not documented to take BUS_DMA_NOWAIT flag. - Unload descriptor map after receive. - Do not reset descriptor map to NULL, bus_dmamap_load(9) requires valid map, and also this leaks the map. Reported and tested by: pho Approved by: jfv Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-11-02 09:16:11 +00:00
error = bus_dmamap_create(rxr->rxtag, 0, &rxbuf->map);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
if (error) {
device_printf(dev, "%s: bus_dmamap_create failed: %d\n",
__func__, error);
goto fail;
}
}
return (0);
- Bus DMA'fy the driver - Use htole* macros where appropriate so that the driver could work on non-x86 architectures - Use m_getcl() instead of MGETHDR/MCLGET macros Submitted by: sam (Sam Leffler)
2003-05-02 21:17:08 +00:00
busdma cleanup for em(4). - don't force busdma to pre-allocate bounce pages for parent tag. - use system supplied roundup2 macro instead of rolling its own version. - TX/RX decriptor length should be multiple of 128. There is no no need to expand the size with the multiple of 4096. - don't create/destroy DMA maps in TX/RX handlers. Use pre-allocated DMA maps. Since creating DMA maps on sparc64 is time consuming operations(resource mananger overhead), this change should boost performance on sparc64. I could get > 2x speedup on Ultra60. - TX/RX descriptors could be aligned on 128 boundary. Aligning them on PAGE_SIZE is waste of resource. - don't blindly create TX DMA tag with size of MCLBYTES * 8. The size is only valid under jumbo frame environments. Instead of using the hardcoded value, re-compute necessary size on the fly. - RX side bus_dmamap_load_mbuf_sg(9) support. - remove unused macro EM_ROUNDUP and constant EM_MMBA. Reviewed by: scottl Tested by: glebius
2005-11-21 04:17:43 +00:00
fail:
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_free_receive_structures(adapter);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (error);
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
* Initialize a receive ring and its buffers.
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
*
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
**********************************************************************/
static int
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
em_setup_receive_ring(struct rx_ring *rxr)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 = rxr->adapter;
struct em_buffer *rxbuf;
bus_dma_segment_t seg[1];
Initialize "error" to zero when it's declared in em_setup_receive_ring()
2012-05-11 03:15:22 +00:00
int rsize, nsegs, error = 0;
fix the initialization of the rings when netmap is used, to adapt it to the changes in 228387 . Now the code is similar to the one used in other drivers. Not applicable to stable/9 and stable/8
2012-01-12 17:28:00 +00:00
#ifdef DEV_NETMAP
struct netmap_slot *slot;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
struct netmap_adapter *na = netmap_getna(adapter->ifp);
fix the initialization of the rings when netmap is used, to adapt it to the changes in 228387 . Now the code is similar to the one used in other drivers. Not applicable to stable/9 and stable/8
2012-01-12 17:28:00 +00:00
#endif
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Clear the ring contents */
EM_RX_LOCK(rxr);
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
rsize = roundup2(adapter->num_rx_desc *
sizeof(struct e1000_rx_desc), EM_DBA_ALIGN);
bzero((void *)rxr->rx_base, rsize);
fix the initialization of the rings when netmap is used, to adapt it to the changes in 228387 . Now the code is similar to the one used in other drivers. Not applicable to stable/9 and stable/8
2012-01-12 17:28:00 +00:00
#ifdef DEV_NETMAP
update to the latest netmap snapshot. This includes the following: - use separate memory regions for VALE ports - locking fixes - some simplifications in the NIC-specific routines - performance improvements for the VALE switch - some new features in the pkt-gen test program - documentation updates There are small API changes that require programs to be recompiled (NETMAP_API has been bumped so you will detect old binaries at runtime). In particular: - struct netmap_slot now is 16 bytes to support an extra pointer, which may save one data copy when using VALE ports or VMs; - the struct netmap_if has two extra fields; MFC after: 3 days
2013-11-01 21:21:14 +00:00
slot = netmap_reset(na, NR_RX, rxr->me, 0);
fix the initialization of the rings when netmap is used, to adapt it to the changes in 228387 . Now the code is similar to the one used in other drivers. Not applicable to stable/9 and stable/8
2012-01-12 17:28:00 +00:00
#endif
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
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
/*
** Free current RX buffer structs and their mbufs
*/
for (int i = 0; i < adapter->num_rx_desc; i++) {
rxbuf = &rxr->rx_buffers[i];
if (rxbuf->m_head != NULL) {
bus_dmamap_sync(rxr->rxtag, rxbuf->map,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(rxr->rxtag, rxbuf->map);
m_freem(rxbuf->m_head);
clear the pointer after freeing the mbuf. Without that, we risk a double free if the subsequent mbuf allocation fails. This bug is not netmap-related and was introduced in rev. 228387
2012-01-12 17:30:44 +00:00
rxbuf->m_head = NULL; /* mark as freed */
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 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 mbufs */
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 != adapter->num_rx_desc; ++j) {
rxbuf = &rxr->rx_buffers[j];
fix the initialization of the rings when netmap is used, to adapt it to the changes in 228387 . Now the code is similar to the one used in other drivers. Not applicable to stable/9 and stable/8
2012-01-12 17:28:00 +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->rx_rings[rxr->me], j);
fix the initialization of the rings when netmap is used, to adapt it to the changes in 228387 . Now the code is similar to the one used in other drivers. Not applicable to stable/9 and stable/8
2012-01-12 17:28:00 +00:00
uint64_t paddr;
void *addr;
Update to the current version of netmap. Mostly bugfixes or features developed in the past 6 months, so this is a 10.1 candidate. Basically no user API changes (some bugfixes in sys/net/netmap_user.h). In detail: 1. netmap support for virtio-net, including in netmap mode. Under bhyve and with a netmap backend [2] we reach over 1Mpps with standard APIs (e.g. libpcap), and 5-8 Mpps in netmap mode. 2. (kernel) add support for multiple memory allocators, so we can better partition physical and virtual interfaces giving access to separate users. The most visible effect is one additional argument to the various kernel functions to compute buffer addresses. All netmap-supported drivers are affected, but changes are mechanical and trivial 3. (kernel) simplify the prototype for *txsync() and *rxsync() driver methods. All netmap drivers affected, changes mostly mechanical. 4. add support for netmap-monitor ports. Think of it as a mirroring port on a physical switch: a netmap monitor port replicates traffic present on the main port. Restrictions apply. Drive carefully. 5. if_lem.c: support for various paravirtualization features, experimental and disabled by default. Most of these are described in our ANCS'13 paper [1]. Paravirtualized support in netmap mode is new, and beats the numbers in the paper by a large factor (under qemu-kvm, we measured gues-host throughput up to 10-12 Mpps). A lot of refactoring and additional documentation in the files in sys/dev/netmap, but apart from #2 and #3 above, almost nothing of this stuff is visible to other kernel parts. Example programs in tools/tools/netmap have been updated with bugfixes and to support more of the existing features. This is meant to go into 10.1 so we plan an MFC before the Aug.22 deadline. A lot of this code has been contributed by my colleagues at UNIPI, including Giuseppe Lettieri, Vincenzo Maffione, Stefano Garzarella. MFC after: 3 days.
2014-08-16 15:00:01 +00:00
addr = PNMB(na, slot + si, &paddr);
netmap_load_map(na, rxr->rxtag, rxbuf->map, addr);
fix the initialization of the rings when netmap is used, to adapt it to the changes in 228387 . Now the code is similar to the one used in other drivers. Not applicable to stable/9 and stable/8
2012-01-12 17:28:00 +00:00
/* Update descriptor */
rxr->rx_base[j].buffer_addr = htole64(paddr);
continue;
}
#endif /* DEV_NETMAP */
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
rxbuf->m_head = m_getjcl(M_NOWAIT, MT_DATA,
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
M_PKTHDR, adapter->rx_mbuf_sz);
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
if (rxbuf->m_head == NULL) {
error = ENOBUFS;
goto fail;
}
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
rxbuf->m_head->m_len = adapter->rx_mbuf_sz;
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->m_flags &= ~M_HASFCS; /* we strip it */
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
rxbuf->m_head->m_pkthdr.len = adapter->rx_mbuf_sz;
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +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
/* Get the memory mapping */
error = bus_dmamap_load_mbuf_sg(rxr->rxtag,
rxbuf->map, rxbuf->m_head, seg,
&nsegs, BUS_DMA_NOWAIT);
If em(4) failed to allocate RX buffers, do not call panic(9). Just showing some buffer allocation error is more appropriate action for drivers. This should fix occasional panic reported on em(4) when driver encountered resource shortage. Reviewed by: jfv
2010-08-28 00:16:49 +00:00
if (error != 0) {
m_freem(rxbuf->m_head);
rxbuf->m_head = NULL;
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
goto fail;
If em(4) failed to allocate RX buffers, do not call panic(9). Just showing some buffer allocation error is more appropriate action for drivers. This should fix occasional panic reported on em(4) when driver encountered resource shortage. Reviewed by: jfv
2010-08-28 00:16:49 +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
bus_dmamap_sync(rxr->rxtag,
rxbuf->map, BUS_DMASYNC_PREREAD);
/* Update descriptor */
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
rxr->rx_base[j].buffer_addr = htole64(seg[0].ds_addr);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +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
rxr->next_to_check = 0;
rxr->next_to_refresh = 0;
bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
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
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
EM_RX_UNLOCK(rxr);
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
return (error);
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
}
/*********************************************************************
*
* Initialize all receive rings.
*
**********************************************************************/
static int
em_setup_receive_structures(struct adapter *adapter)
{
struct rx_ring *rxr = adapter->rx_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
int q;
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 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
for (q = 0; q < adapter->num_queues; q++, 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
if (em_setup_receive_ring(rxr))
goto fail;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (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
fail:
/*
* Free RX buffers allocated so far, we will only handle
* the rings that completed, the failing case will have
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
* cleaned up for itself. 'q' failed, so its the terminus.
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
*/
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 < q; ++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
rxr = &adapter->rx_rings[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
for (int n = 0; n < adapter->num_rx_desc; 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
struct em_buffer *rxbuf;
rxbuf = &rxr->rx_buffers[n];
if (rxbuf->m_head != NULL) {
bus_dmamap_sync(rxr->rxtag, rxbuf->map,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(rxr->rxtag, rxbuf->map);
m_freem(rxbuf->m_head);
rxbuf->m_head = NULL;
}
}
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
rxr->next_to_check = 0;
rxr->next_to_refresh = 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
}
return (ENOBUFS);
}
/*********************************************************************
*
* Free all receive rings.
*
**********************************************************************/
static void
em_free_receive_structures(struct adapter *adapter)
{
struct rx_ring *rxr = adapter->rx_rings;
for (int i = 0; i < adapter->num_queues; i++, rxr++) {
em_free_receive_buffers(rxr);
/* Free the ring memory as well */
em_dma_free(adapter, &rxr->rxdma);
EM_RX_LOCK_DESTROY(rxr);
}
free(adapter->rx_rings, M_DEVBUF);
}
/*********************************************************************
*
* Free receive ring data structures
*
**********************************************************************/
static void
em_free_receive_buffers(struct rx_ring *rxr)
{
struct adapter *adapter = rxr->adapter;
struct em_buffer *rxbuf = NULL;
INIT_DEBUGOUT("free_receive_buffers: begin");
if (rxr->rx_buffers != 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
for (int i = 0; i < adapter->num_rx_desc; 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
rxbuf = &rxr->rx_buffers[i];
if (rxbuf->map != NULL) {
bus_dmamap_sync(rxr->rxtag, rxbuf->map,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(rxr->rxtag, rxbuf->map);
bus_dmamap_destroy(rxr->rxtag, rxbuf->map);
}
if (rxbuf->m_head != NULL) {
m_freem(rxbuf->m_head);
rxbuf->m_head = NULL;
}
}
free(rxr->rx_buffers, M_DEVBUF);
rxr->rx_buffers = NULL;
This delta updates the em driver to version 7.2.2 which has been undergoing test for some weeks. This improves the RX mbuf handling to avoid system hang due to depletion. Thanks to all those who have been testing the code, and to Beezar Liu for the design changes. Next the igb driver is updated for similar RX changes, but also to add new features support for our upcoming i350 family of adapters. MFC after a week
2011-03-18 18:54:00 +00:00
rxr->next_to_check = 0;
rxr->next_to_refresh = 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
}
if (rxr->rxtag != NULL) {
bus_dma_tag_destroy(rxr->rxtag);
rxr->rxtag = NULL;
}
return;
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
*
* Enable receive unit.
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
*
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_initialize_receive_unit(struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 rx_ring *rxr = adapter->rx_rings;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
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 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
u64 bus_addr;
u32 rctl, rxcsum;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
INIT_DEBUGOUT("em_initialize_receive_units: begin");
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
/*
* Make sure receives are disabled while setting
* up the descriptor ring
*/
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
rctl = E1000_READ_REG(hw, E1000_RCTL);
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 not disable if ever enabled on this hardware */
if ((hw->mac.type != e1000_82574) && (hw->mac.type != e1000_82583))
E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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
E1000_WRITE_REG(&adapter->hw, E1000_RADV,
adapter->rx_abs_int_delay.value);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_RDTR,
adapter->rx_int_delay.value);
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
/*
* Set the interrupt throttling rate. Value is calculated
* as DEFAULT_ITR = 1/(MAX_INTS_PER_SEC * 256ns)
*/
E1000_WRITE_REG(hw, E1000_ITR, DEFAULT_ITR);
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +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
/*
** When using MSIX interrupts we need to throttle
** using the EITR register (82574 only)
*/
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 (hw->mac.type == e1000_82574) {
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
u32 rfctl;
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
for (int i = 0; i < 4; 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
E1000_WRITE_REG(hw, E1000_EITR_82574(i),
DEFAULT_ITR);
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
/* Disable accelerated acknowledge */
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
rfctl = E1000_READ_REG(hw, E1000_RFCTL);
rfctl |= E1000_RFCTL_ACK_DIS;
E1000_WRITE_REG(hw, E1000_RFCTL, rfctl);
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
}
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
Corrections to the RX checksum code, make sure its disabled as well as enabled when necessary. And simplify the checksum routine itself, adding UDP bit to the test. Thanks to Kevin Lo for pointing out the problems and code suggestions.
2013-04-15 17:01:42 +00:00
rxcsum = E1000_READ_REG(hw, E1000_RXCSUM);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
if (if_getcapenable(ifp) & IFCAP_RXCSUM) {
#ifdef EM_MULTIQUEUE
rxcsum |= E1000_RXCSUM_TUOFL |
E1000_RXCSUM_IPOFL |
E1000_RXCSUM_PCSD;
#else
Corrections to the RX checksum code, make sure its disabled as well as enabled when necessary. And simplify the checksum routine itself, adding UDP bit to the test. Thanks to Kevin Lo for pointing out the problems and code suggestions.
2013-04-15 17:01:42 +00:00
rxcsum |= E1000_RXCSUM_TUOFL;
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#endif
} else
Corrections to the RX checksum code, make sure its disabled as well as enabled when necessary. And simplify the checksum routine itself, adding UDP bit to the test. Thanks to Kevin Lo for pointing out the problems and code suggestions.
2013-04-15 17:01:42 +00:00
rxcsum &= ~E1000_RXCSUM_TUOFL;
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
Corrections to the RX checksum code, make sure its disabled as well as enabled when necessary. And simplify the checksum routine itself, adding UDP bit to the test. Thanks to Kevin Lo for pointing out the problems and code suggestions.
2013-04-15 17:01:42 +00:00
E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#ifdef EM_MULTIQUEUE
if (adapter->num_queues > 1) {
uint32_t rss_key[10];
uint32_t reta;
int i;
/*
* Configure RSS key
*/
arc4rand(rss_key, sizeof(rss_key), 0);
for (i = 0; i < 10; ++i)
E1000_WRITE_REG_ARRAY(hw,E1000_RSSRK(0), i, rss_key[i]);
/*
* Configure RSS redirect table in following fashion:
* (hash & ring_cnt_mask) == rdr_table[(hash & rdr_table_mask)]
*/
reta = 0;
for (i = 0; i < 4; ++i) {
uint32_t q;
q = (i % adapter->num_queues) << 7;
reta |= q << (8 * i);
}
for (i = 0; i < 32; ++i)
E1000_WRITE_REG(hw, E1000_RETA(i), reta);
E1000_WRITE_REG(hw, E1000_MRQC, E1000_MRQC_RSS_ENABLE_2Q |
E1000_MRQC_RSS_FIELD_IPV4_TCP |
E1000_MRQC_RSS_FIELD_IPV4 |
E1000_MRQC_RSS_FIELD_IPV6_TCP_EX |
E1000_MRQC_RSS_FIELD_IPV6_EX |
E1000_MRQC_RSS_FIELD_IPV6 |
E1000_MRQC_RSS_FIELD_IPV6_TCP);
}
#endif
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/*
** XXX TEMPORARY WORKAROUND: on some systems with 82573
** long latencies are observed, like Lenovo X60. This
** change eliminates the problem, but since having positive
** values in RDTR is a known source of problems on other
** platforms another solution is being sought.
*/
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 (hw->mac.type == e1000_82573)
E1000_WRITE_REG(hw, E1000_RDTR, 0x20);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
for (int i = 0; i < adapter->num_queues; i++, rxr++) {
/* Setup the Base and Length of the Rx Descriptor Ring */
simplify the code to initialize the RDT while in netmap mode.
2013-05-09 16:57:02 +00:00
u32 rdt = adapter->num_rx_desc - 1; /* default */
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_addr = rxr->rxdma.dma_paddr;
E1000_WRITE_REG(hw, E1000_RDLEN(i),
adapter->num_rx_desc * sizeof(struct e1000_rx_desc));
E1000_WRITE_REG(hw, E1000_RDBAH(i), (u32)(bus_addr >> 32));
E1000_WRITE_REG(hw, E1000_RDBAL(i), (u32)bus_addr);
/* Setup the Head and Tail Descriptor Pointers */
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_REG(hw, E1000_RDH(i), 0);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
/*
* an init() while a netmap client is active must
* preserve the rx buffers passed to userspace.
*/
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getcapenable(ifp) & IFCAP_NETMAP) {
struct netmap_adapter *na = netmap_getna(adapter->ifp);
rdt -= nm_kr_rxspace(&na->rx_rings[i]);
}
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#endif /* DEV_NETMAP */
simplify the code to initialize the RDT while in netmap mode.
2013-05-09 16:57:02 +00:00
E1000_WRITE_REG(hw, E1000_RDT(i), rdt);
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +00:00
}
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
/*
* Set PTHRESH for improved jumbo performance
* According to 10.2.5.11 of Intel 82574 Datasheet,
* RXDCTL(1) is written whenever RXDCTL(0) is written.
* Only write to RXDCTL(1) if there is a need for different
* settings.
*/
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
if (((adapter->hw.mac.type == e1000_ich9lan) ||
(adapter->hw.mac.type == e1000_pch2lan) ||
(adapter->hw.mac.type == e1000_ich10lan)) &&
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
(if_getmtu(ifp) > ETHERMTU)) {
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
u32 rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(0));
E1000_WRITE_REG(hw, E1000_RXDCTL(0), rxdctl | 3);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
} else if ((adapter->hw.mac.type == e1000_82574) &&
(if_getmtu(ifp) > ETHERMTU)) {
for (int i = 0; i < adapter->num_queues; i++) {
u32 rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i));
rxdctl |= 0x20; /* PTHRESH */
rxdctl |= 4 << 8; /* HTHRESH */
rxdctl |= 4 << 16;/* WTHRESH */
rxdctl |= 1 << 24; /* Switch to granularity */
E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl);
}
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
}
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
if (adapter->hw.mac.type >= e1000_pch2lan) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getmtu(ifp) > ETHERMTU)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
e1000_lv_jumbo_workaround_ich8lan(hw, TRUE);
else
e1000_lv_jumbo_workaround_ich8lan(hw, 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
/* 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 |
(hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
Three changes: - add CRC stripping to the RX side, this was handled by some obscure code in rxeof previously, its easier to simply have the hardware strip it now. - Add back an ALTQ change that slipped between the cracks - Add an update to the watchdog_time in the xmit code, not doing this in ixgbe caused problems, think its needed here as well.
2010-04-08 19:13:42 +00:00
/* Strip the CRC */
rctl |= E1000_RCTL_SECRC;
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
/* Make sure VLAN Filters are off */
rctl &= ~E1000_RCTL_VFE;
rctl &= ~E1000_RCTL_SBP;
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
if (adapter->rx_mbuf_sz == MCLBYTES)
rctl |= E1000_RCTL_SZ_2048;
else if (adapter->rx_mbuf_sz == MJUMPAGESIZE)
rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX;
else if (adapter->rx_mbuf_sz > MJUMPAGESIZE)
rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getmtu(ifp) > ETHERMTU)
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
rctl |= E1000_RCTL_LPE;
else
rctl &= ~E1000_RCTL_LPE;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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
/* Write out the settings */
E1000_WRITE_REG(hw, E1000_RCTL, rctl);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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 is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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.
*
- Move to array based indexing for TX/RX descriptor/buffer management - Added support for ITR (interrupt throttle register). This feature is available on adapters based on 82545 and above - Fixed problem with vlan support when traffic has priority bits set. (kern/45907) PR: kern/45907 MFC after: 1 week
2002-12-23 19:11:23 +00:00
* We loop at most count times if count is > 0, or until done if
* count < 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
*
* For polling we also now return the number of cleaned packets
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*********************************************************************/
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
em_rxeof(struct rx_ring *rxr, int count, int *done)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 = rxr->adapter;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t ifp = adapter->ifp;
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 *mp, *sendmp;
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
u8 status = 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
u16 len;
int i, processed, rxdone = 0;
bool eop;
struct e1000_rx_desc *cur;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Address the LOD that some are seeing, put the RX lock back in rxeof (I could see little point in taking it out), and now release it before the stack entry. Also, make it so the 82574 does not configure for multiqueue when its not used in the stack.
2010-04-28 19:22:52 +00:00
EM_RX_LOCK(rxr);
Make em(4) work on big-endian architectures. - disable jumbo frame support on strict alignment architectures due to the limitation of hardware. The driver needs a fix-up code for RX side. The fix will show up in near future. - fix endian issue for 82544 on PCI-X bus. I couldn't test this as I don't have the NIC/hardware. - prefer PCIR_BAR to hardcoded EM_MMBA. - Properly checks for for 64bit BAR [1] - replace inl/outl with bus_space(9) [1] - fix endian issue on VLAN handling. - reorder header files and remove unnecessary one. Reviewed by: cognet No response from: pdeuskar, tackerman Obtained from: OpenBSD [1]
2005-11-09 08:43:18 +00:00
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
/* Sync the ring */
bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
update to the latest netmap snapshot. This includes the following: - use separate memory regions for VALE ports - locking fixes - some simplifications in the NIC-specific routines - performance improvements for the VALE switch - some new features in the pkt-gen test program - documentation updates There are small API changes that require programs to be recompiled (NETMAP_API has been bumped so you will detect old binaries at runtime). In particular: - struct netmap_slot now is 16 bytes to support an extra pointer, which may save one data copy when using VALE ports or VMs; - the struct netmap_if has two extra fields; MFC after: 3 days
2013-11-01 21:21:14 +00:00
if (netmap_rx_irq(ifp, rxr->me, &processed)) {
EM_RX_UNLOCK(rxr);
use netmap_rx_irq() / netmap_tx_irq() to handle interrupts in netmap mode, removing the logic from individual drivers. (note: if_lem.c not updated yet due to some other pending modifications)
2013-04-30 16:18:29 +00:00
return (FALSE);
update to the latest netmap snapshot. This includes the following: - use separate memory regions for VALE ports - locking fixes - some simplifications in the NIC-specific routines - performance improvements for the VALE switch - some new features in the pkt-gen test program - documentation updates There are small API changes that require programs to be recompiled (NETMAP_API has been bumped so you will detect old binaries at runtime). In particular: - struct netmap_slot now is 16 bytes to support an extra pointer, which may save one data copy when using VALE ports or VMs; - the struct netmap_if has two extra fields; MFC after: 3 days
2013-11-01 21:21:14 +00:00
}
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#endif /* DEV_NETMAP */
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, processed = 0; count != 0;) {
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 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
break;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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
cur = &rxr->rx_base[i];
status = cur->status;
mp = sendmp = NULL;
if ((status & E1000_RXD_STAT_DD) == 0)
break;
len = le16toh(cur->length);
eop = (status & E1000_RXD_STAT_EOP) != 0;
Small cut and paste bug in flow control string fixed. Second, correct the discard/refresh_mbufs code to behave more like igb, there have been panics due to discards and this should fix them. MFC after: 3 days
2010-12-04 01:59:58 +00:00
if ((cur->errors & E1000_RXD_ERR_FRAME_ERR_MASK) ||
(rxr->discard == TRUE)) {
This patch fixes a nit in the em, lem, and igb driver statistics. Increment adapter->dropped_pkts instead of if_ierrors because if_ierrors is overwritten by hw stats collection. Submitted by: Andrew Boyer <aboyer@averesystems.com> Reviewed by: Jack F Vogel <jfv@freebsd.org> MFC after: 2 weeks
2012-09-23 22:53:39 +00:00
adapter->dropped_pkts++;
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
++rxr->rx_discarded;
if (!eop) /* Catch subsequent segs */
rxr->discard = TRUE;
else
rxr->discard = FALSE;
em_rx_discard(rxr, i);
goto next_desc;
}
Fix several issues with the busdma(9) KPI use in the e1000 drivers. The problems do not affect bouncing busdma in a visible way, but are critical for the dmar backend. - The bus_dmamap_create(9) is not documented to take BUS_DMA_NOWAIT flag. - Unload descriptor map after receive. - Do not reset descriptor map to NULL, bus_dmamap_load(9) requires valid map, and also this leaks the map. Reported and tested by: pho Approved by: jfv Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-11-02 09:16:11 +00:00
bus_dmamap_unload(rxr->rxtag, rxr->rx_buffers[i].map);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
/* Assign correct length to the current fragment */
mp = rxr->rx_buffers[i].m_head;
mp->m_len = len;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
/* Trigger for refresh */
rxr->rx_buffers[i].m_head = NULL;
Tighten up the rx mbuf refresh code, there were some discrepencies from the igb version which was the target. Change the message when neither MSI or MSIX are enabled and a fallback to Legacy interrupts happen, the existing message was confusing.
2010-09-07 20:13:08 +00:00
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
/* First segment? */
if (rxr->fmp == NULL) {
mp->m_pkthdr.len = len;
rxr->fmp = rxr->lmp = mp;
} else {
/* Chain mbuf's together */
mp->m_flags &= ~M_PKTHDR;
rxr->lmp->m_next = mp;
rxr->lmp = mp;
rxr->fmp->m_pkthdr.len += len;
}
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
if (eop) {
--count;
sendmp = rxr->fmp;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setrcvif(sendmp, ifp);
- Use if_inc_counter() to increment various counters. - Do not ever set a counter to a value. For those counters that we don't increment, but return directly from hardware create cases in if_get_counter() method. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-09-18 15:56:14 +00:00
if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
em_receive_checksum(cur, sendmp);
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
#ifndef __NO_STRICT_ALIGNMENT
Refresh on the shared code for the E1000 drivers. - bear with me, there are lots of white space changes, I would not do them, but I am a mere consumer of this stuff and if these drivers are to stay in shape they need to be taken. em driver changes: support for the new i217/i218 interfaces igb driver changes: - TX mq start has a quick turnaround to the stack - Link/media handling improvement - When link status changes happen the current flow control state will now be displayed. - A few white space/style changes. lem driver changes: - the shared code uncovered a bogus write to the RLPML register (which does not exist in this hardware) in the vlan code,this is removed.
2013-02-21 00:25:45 +00:00
if (adapter->hw.mac.max_frame_size >
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
(MCLBYTES - ETHER_ALIGN) &&
em_fixup_rx(rxr) != 0)
goto skip;
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
#endif
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
if (status & E1000_RXD_STAT_VP) {
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_setvtag(sendmp,
le16toh(cur->special));
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
sendmp->m_flags |= M_VLANTAG;
}
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
#ifndef __NO_STRICT_ALIGNMENT
skip:
#endif
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
rxr->fmp = rxr->lmp = NULL;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
}
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
next_desc:
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
/* Sync the ring */
bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Don't update Rx descriptor status in two different functions. Suggested by: pdeuskar Reviewed by: pdeuskar
2006-08-16 23:55:34 +00:00
/* Zero out the receive descriptors status. */
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
cur->status = 0;
++rxdone; /* cumulative for POLL */
++processed;
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
/* Advance our pointers to the next descriptor. */
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
if (++i == adapter->num_rx_desc)
From the PR: The receive function em_process_receive_interrupts() unlocks the adapter while ether_input() processes the packet, and then locks it back. In the meantime, em_init() may be called, either from em_watchdog() from softclock interrupt or from the ifconfig(8) program. The em_init() resets the card, in particular it sets adapter->next_rx_desc_to_check to 0 and resets hardware RX Head and Tail descriptor pointers. The loop in em_process_receive_interrupts() does not expect these things to change, and a mess may result. This fixes long wedges of em(4) interfaces receive part under high load and IP fastforwarding enabled. PR: kern/87418 Submitted by: Dmitrij Tejblum <tejblum yandex-team.ru>
2005-10-14 11:00:15 +00:00
i = 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
/* Send to the stack */
Address the LOD that some are seeing, put the RX lock back in rxeof (I could see little point in taking it out), and now release it before the stack entry. Also, make it so the 82574 does not configure for multiqueue when its not used in the stack.
2010-04-28 19:22:52 +00:00
if (sendmp != NULL) {
rxr->next_to_check = i;
EM_RX_UNLOCK(rxr);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_input(ifp, sendmp);
Address the LOD that some are seeing, put the RX lock back in rxeof (I could see little point in taking it out), and now release it before the stack entry. Also, make it so the 82574 does not configure for multiqueue when its not used in the stack.
2010-04-28 19:22:52 +00:00
EM_RX_LOCK(rxr);
i = rxr->next_to_check;
}
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
/* Only refresh mbufs every 8 descriptors */
if (processed == 8) {
em_refresh_mbufs(rxr, i);
processed = 0;
From the PR: The receive function em_process_receive_interrupts() unlocks the adapter while ether_input() processes the packet, and then locks it back. In the meantime, em_init() may be called, either from em_watchdog() from softclock interrupt or from the ifconfig(8) program. The em_init() resets the card, in particular it sets adapter->next_rx_desc_to_check to 0 and resets hardware RX Head and Tail descriptor pointers. The loop in em_process_receive_interrupts() does not expect these things to change, and a mess may result. This fixes long wedges of em(4) interfaces receive part under high load and IP fastforwarding enabled. PR: kern/87418 Submitted by: Dmitrij Tejblum <tejblum yandex-team.ru>
2005-10-14 11:00:15 +00:00
}
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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 remaining refresh work */
Change the refresh_mbuf logic slightly, add an inline to calculate the outstanding descriptors that need to be refreshed at any time, and use THAT in rxeof to determine if refreshing needs to be done. Also change the local_timer to simply fire off the appropriate interrupt rather than schedule a tasklet, its simpler. MFC in two weeks
2011-04-01 18:48:31 +00:00
if (e1000_rx_unrefreshed(rxr))
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
em_refresh_mbufs(rxr, 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
rxr->next_to_check = i;
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;
Address the LOD that some are seeing, put the RX lock back in rxeof (I could see little point in taking it out), and now release it before the stack entry. Also, make it so the 82574 does not configure for multiqueue when its not used in the stack.
2010-04-28 19:22:52 +00:00
EM_RX_UNLOCK(rxr);
Incorporate suggested improvements from yongari. Also, from feedback, make the multiqueue code an option (EM_MULTIQUEUE) that is off by default. Problems have been seen with UDP when its on.
2010-04-09 18:42:15 +00:00
return ((status & E1000_RXD_STAT_DD) ? TRUE : FALSE);
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
Tighten up the rx mbuf refresh code, there were some discrepencies from the igb version which was the target. Change the message when neither MSI or MSIX are enabled and a fallback to Legacy interrupts happen, the existing message was confusing.
2010-09-07 20:13:08 +00:00
static __inline void
em_rx_discard(struct rx_ring *rxr, int i)
{
struct em_buffer *rbuf;
rbuf = &rxr->rx_buffers[i];
Fix several issues with the busdma(9) KPI use in the e1000 drivers. The problems do not affect bouncing busdma in a visible way, but are critical for the dmar backend. - The bus_dmamap_create(9) is not documented to take BUS_DMA_NOWAIT flag. - Unload descriptor map after receive. - Do not reset descriptor map to NULL, bus_dmamap_load(9) requires valid map, and also this leaks the map. Reported and tested by: pho Approved by: jfv Sponsored by: The FreeBSD Foundation MFC after: 2 weeks
2013-11-02 09:16:11 +00:00
bus_dmamap_unload(rxr->rxtag, rbuf->map);
Tighten up the rx mbuf refresh code, there were some discrepencies from the igb version which was the target. Change the message when neither MSI or MSIX are enabled and a fallback to Legacy interrupts happen, the existing message was confusing.
2010-09-07 20:13:08 +00:00
/* Free any previous pieces */
if (rxr->fmp != NULL) {
rxr->fmp->m_flags |= M_PKTHDR;
m_freem(rxr->fmp);
rxr->fmp = NULL;
rxr->lmp = NULL;
}
Small cut and paste bug in flow control string fixed. Second, correct the discard/refresh_mbufs code to behave more like igb, there have been panics due to discards and this should fix them. MFC after: 3 days
2010-12-04 01:59:58 +00:00
/*
** Free buffer and allow em_refresh_mbufs()
** to clean up and recharge buffer.
*/
if (rbuf->m_head) {
m_free(rbuf->m_head);
rbuf->m_head = NULL;
}
Tighten up the rx mbuf refresh code, there were some discrepencies from the igb version which was the target. Change the message when neither MSI or MSIX are enabled and a fallback to Legacy interrupts happen, the existing message was confusing.
2010-09-07 20:13:08 +00:00
return;
}
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
#ifndef __NO_STRICT_ALIGNMENT
/*
* When jumbo frames are enabled we should realign entire payload on
* architecures with strict alignment. This is serious design mistake of 8254x
* as it nullifies DMA operations. 8254x just allows RX buffer size to be
* 2048/4096/8192/16384. What we really want is 2048 - ETHER_ALIGN to align its
* payload. On architecures without strict alignment restrictions 8254x still
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* performs unaligned memory access which would reduce the performance too.
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
* To avoid copying over an entire frame to align, we allocate a new mbuf and
* copy ethernet header to the new mbuf. The new mbuf is prepended into the
* existing mbuf chain.
*
* Be aware, best performance of the 8254x is achived only when jumbo frame is
* not used at all on architectures with strict alignment.
*/
static int
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
em_fixup_rx(struct rx_ring *rxr)
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +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 = rxr->adapter;
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
struct mbuf *m, *n;
int error;
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
m = rxr->fmp;
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
if (m->m_len <= (MCLBYTES - ETHER_HDR_LEN)) {
bcopy(m->m_data, m->m_data + ETHER_HDR_LEN, m->m_len);
m->m_data += ETHER_HDR_LEN;
} else {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
MGETHDR(n, M_NOWAIT, MT_DATA);
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
if (n != NULL) {
bcopy(m->m_data, n->m_data, ETHER_HDR_LEN);
m->m_data += ETHER_HDR_LEN;
m->m_len -= ETHER_HDR_LEN;
n->m_len = ETHER_HDR_LEN;
M_MOVE_PKTHDR(n, m);
n->m_next = m;
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 = n;
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
} else {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->dropped_pkts++;
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_freem(rxr->fmp);
rxr->fmp = NULL;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
error = ENOMEM;
Add jumbo frame support for architectures with strict alignment. Reviewed by: glebius
2005-12-16 08:29:43 +00:00
}
}
return (error);
}
#endif
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/*********************************************************************
*
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* Verify that the hardware indicated that the checksum is valid.
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
* Inform the stack about the status of checksum so that stack
* doesn't spend time verifying the checksum.
*
*********************************************************************/
static 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
em_receive_checksum(struct e1000_rx_desc *rx_desc, struct mbuf *mp)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
{
Corrections to the RX checksum code, make sure its disabled as well as enabled when necessary. And simplify the checksum routine itself, adding UDP bit to the test. Thanks to Kevin Lo for pointing out the problems and code suggestions.
2013-04-15 17:01:42 +00:00
mp->m_pkthdr.csum_flags = 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
/* Ignore Checksum bit is set */
Corrections to the RX checksum code, make sure its disabled as well as enabled when necessary. And simplify the checksum routine itself, adding UDP bit to the test. Thanks to Kevin Lo for pointing out the problems and code suggestions.
2013-04-15 17:01:42 +00:00
if (rx_desc->status & E1000_RXD_STAT_IXSM)
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
return;
Corrections to the RX checksum code, make sure its disabled as well as enabled when necessary. And simplify the checksum routine itself, adding UDP bit to the test. Thanks to Kevin Lo for pointing out the problems and code suggestions.
2013-04-15 17:01:42 +00:00
if (rx_desc->errors & (E1000_RXD_ERR_TCPE | E1000_RXD_ERR_IPE))
return;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Corrections to the RX checksum code, make sure its disabled as well as enabled when necessary. And simplify the checksum routine itself, adding UDP bit to the test. Thanks to Kevin Lo for pointing out the problems and code suggestions.
2013-04-15 17:01:42 +00:00
/* IP Checksum Good? */
if (rx_desc->status & E1000_RXD_STAT_IPCS)
mp->m_pkthdr.csum_flags = (CSUM_IP_CHECKED | CSUM_IP_VALID);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Corrections to the RX checksum code, make sure its disabled as well as enabled when necessary. And simplify the checksum routine itself, adding UDP bit to the test. Thanks to Kevin Lo for pointing out the problems and code suggestions.
2013-04-15 17:01:42 +00:00
/* TCP or UDP checksum */
if (rx_desc->status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)) {
mp->m_pkthdr.csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
mp->m_pkthdr.csum_data = htons(0xffff);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
}
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
/*
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
* This routine is run via an vlan
* config EVENT
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
*/
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
static void
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
em_register_vlan(void *arg, if_t ifp, u16 vtag)
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
struct adapter *adapter = if_getsoftc(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 index, bit;
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
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if ((void*)adapter != 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;
if ((vtag == 0) || (vtag > 4095)) /* Invalid ID */
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 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
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
EM_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;
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +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;
/* Re-init to load the changes */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
em_init_locked(adapter);
EM_CORE_UNLOCK(adapter);
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
}
/*
* This routine is run via an vlan
* unconfig EVENT
*/
static void
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
em_unregister_vlan(void *arg, if_t ifp, u16 vtag)
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
{
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
struct adapter *adapter = if_getsoftc(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 index, bit;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (adapter != 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;
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
EM_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;
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +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;
/* Re-init to load the changes */
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
em_init_locked(adapter);
EM_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
em_setup_vlan_hw_support(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
u32 reg;
/*
** We get here thru init_locked, meaning
** a soft reset, this has already cleared
** the VFTA and other state, so if there
** have been no vlan's registered do nothing.
*/
if (adapter->num_vlans == 0)
return;
/*
** A soft reset zero's out the VFTA, so
** we need to repopulate it now.
*/
for (int i = 0; i < EM_VFTA_SIZE; i++)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
if (adapter->shadow_vfta[i] != 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
E1000_WRITE_REG_ARRAY(hw, E1000_VFTA,
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
i, adapter->shadow_vfta[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
reg = E1000_READ_REG(hw, E1000_CTRL);
reg |= E1000_CTRL_VME;
E1000_WRITE_REG(hw, E1000_CTRL, reg);
/* Enable the Filter Table */
reg = E1000_READ_REG(hw, E1000_RCTL);
reg &= ~E1000_RCTL_CFIEN;
reg |= E1000_RCTL_VFE;
E1000_WRITE_REG(hw, E1000_RCTL, reg);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
}
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_enable_intr(struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
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
u32 ims_mask = IMS_ENABLE_MASK;
Correct broken build.
2010-04-10 07:26:51 +00:00
if (hw->mac.type == e1000_82574) {
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(hw, EM_EIAC, EM_MSIX_MASK);
ims_mask |= EM_MSIX_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
}
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(hw, E1000_IMS, ims_mask);
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_disable_intr(struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
struct e1000_hw *hw = &adapter->hw;
A few more changes from yongari: - code flow in handler could let interrupt be reenabled when not wanted. - change where the RX lock is taken to improve performance. - adapter->msix is true for MSI systems also, it needs to explicitly test for 82574, good one :)
2010-04-09 23:15:37 +00:00
if (hw->mac.type == e1000_82574)
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(hw, EM_EIAC, 0);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff);
}
Corrected a workaround that should only be applied to one adapter. Workaround was causing device hangs when incorrectly applied to other adapters. PR: kern/66634
2005-01-01 19:57:23 +00:00
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/*
* 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
em_init_manageability(struct adapter *adapter)
{
/* A shared code workaround */
#define E1000_82542_MANC2H E1000_MANC2H
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 */
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
manc |= E1000_MANC_EN_MNG2HOST;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
#define E1000_MNG2HOST_PORT_623 (1 << 5)
#define E1000_MNG2HOST_PORT_664 (1 << 6)
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
manc2h |= E1000_MNG2HOST_PORT_623;
manc2h |= E1000_MNG2HOST_PORT_664;
E1000_WRITE_REG(&adapter->hw, E1000_MANC2H, manc2h);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc);
}
}
/*
* Give control back to hardware management
* controller if there is one.
*/
static void
em_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;
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
manc &= ~E1000_MANC_EN_MNG2HOST;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc);
}
}
/*
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
* em_get_hw_control sets the {CTRL_EXT|FWSM}:DRV_LOAD bit.
* For ASF and Pass Through versions of f/w this means
* that the driver is loaded. For AMT version type f/w
* this means that the network i/f is open.
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
*/
static void
em_get_hw_control(struct adapter *adapter)
{
u32 ctrl_ext, swsm;
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 (adapter->hw.mac.type == e1000_82573) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
swsm = E1000_READ_REG(&adapter->hw, E1000_SWSM);
E1000_WRITE_REG(&adapter->hw, E1000_SWSM,
swsm | E1000_SWSM_DRV_LOAD);
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;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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
/* else */
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);
return;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
}
/*
* em_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
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 ASF and Pass Through versions of f/w this means that
* the driver is no longer loaded. For AMT versions of the
* f/w this means that the network i/f is closed.
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
*/
static void
em_release_hw_control(struct adapter *adapter)
{
u32 ctrl_ext, swsm;
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 (!adapter->has_manage)
return;
if (adapter->hw.mac.type == e1000_82573) {
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
swsm = E1000_READ_REG(&adapter->hw, E1000_SWSM);
E1000_WRITE_REG(&adapter->hw, E1000_SWSM,
swsm & ~E1000_SWSM_DRV_LOAD);
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;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +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
/* else */
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);
return;
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +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
em_is_valid_ether_addr(u8 *addr)
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
{
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
char zero_addr[6] = { 0, 0, 0, 0, 0, 0 };
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
if ((addr[0] & 1) || (!bcmp(addr, zero_addr, ETHER_ADDR_LEN))) {
return (FALSE);
}
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (TRUE);
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +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
/*
** Parse the interface capabilities with regard
** to both system management and wake-on-lan for
** later use.
*/
static void
em_get_wakeup(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
u16 eeprom_data = 0, device_id, apme_mask;
adapter->has_manage = e1000_enable_mng_pass_thru(&adapter->hw);
apme_mask = EM_EEPROM_APME;
switch (adapter->hw.mac.type) {
case e1000_82573:
case e1000_82583:
adapter->has_amt = TRUE;
/* Falls thru */
case e1000_82571:
case e1000_82572:
case e1000_80003es2lan:
if (adapter->hw.bus.func == 1) {
e1000_read_nvm(&adapter->hw,
NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
break;
} else
e1000_read_nvm(&adapter->hw,
NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
break;
case e1000_ich8lan:
case e1000_ich9lan:
case e1000_ich10lan:
case e1000_pchlan:
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
case e1000_pch2lan:
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
apme_mask = E1000_WUC_APME;
adapter->has_amt = TRUE;
eeprom_data = E1000_READ_REG(&adapter->hw, E1000_WUC);
break;
default:
e1000_read_nvm(&adapter->hw,
NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
break;
}
if (eeprom_data & apme_mask)
adapter->wol = (E1000_WUFC_MAG | E1000_WUFC_MC);
/*
* We have the eeprom settings, now apply the special cases
* where the eeprom may be wrong or the board won't support
* wake on lan on a particular port
*/
device_id = pci_get_device(dev);
switch (device_id) {
case E1000_DEV_ID_82571EB_FIBER:
/* Wake events only supported on port A for dual fiber
* regardless of eeprom setting */
if (E1000_READ_REG(&adapter->hw, E1000_STATUS) &
E1000_STATUS_FUNC_1)
adapter->wol = 0;
break;
case E1000_DEV_ID_82571EB_QUAD_COPPER:
case E1000_DEV_ID_82571EB_QUAD_FIBER:
case E1000_DEV_ID_82571EB_QUAD_COPPER_LP:
/* if quad port adapter, disable WoL on all but port A */
if (global_quad_port_a != 0)
adapter->wol = 0;
/* Reset for multiple quad port adapters */
if (++global_quad_port_a == 4)
global_quad_port_a = 0;
break;
}
return;
}
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
/*
* 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
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
em_enable_wakeup(device_t 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 adapter *adapter = device_get_softc(dev);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if_t 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
u32 pmc, ctrl, ctrl_ext, rctl;
u16 status;
Do a sweep of the tree replacing calls to pci_find_extcap() with calls to pci_find_cap() instead.
2011-03-23 13:10:15 +00:00
if ((pci_find_cap(dev, PCIY_PMG, &pmc) != 0))
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
return;
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
/* Advertise the wakeup capability */
ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL);
ctrl |= (E1000_CTRL_SWDPIN2 | E1000_CTRL_SWDPIN3);
E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl);
E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN);
if ((adapter->hw.mac.type == e1000_ich8lan) ||
(adapter->hw.mac.type == e1000_pchlan) ||
(adapter->hw.mac.type == e1000_ich9lan) ||
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
(adapter->hw.mac.type == e1000_ich10lan))
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_suspend_workarounds_ich8lan(&adapter->hw);
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
/* Keep the laser running on Fiber adapters */
if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, ctrl_ext);
}
/*
** Determine type of Wakeup: note that wol
** is set with all bits on by default.
*/
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if ((if_getcapenable(ifp) & IFCAP_WOL_MAGIC) == 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
adapter->wol &= ~E1000_WUFC_MAG;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if ((if_getcapenable(ifp) & IFCAP_WOL_MCAST) == 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
adapter->wol &= ~E1000_WUFC_MC;
else {
rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
rctl |= E1000_RCTL_MPE;
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, rctl);
}
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
if ((adapter->hw.mac.type == e1000_pchlan) ||
(adapter->hw.mac.type == e1000_pch2lan)) {
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 (em_enable_phy_wakeup(adapter))
return;
} else {
E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN);
E1000_WRITE_REG(&adapter->hw, E1000_WUFC, adapter->wol);
}
if (adapter->hw.phy.type == e1000_phy_igp_3)
e1000_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
/* Request PME */
status = pci_read_config(dev, pmc + PCIR_POWER_STATUS, 2);
status &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getcapenable(ifp) & IFCAP_WOL)
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
status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
pci_write_config(dev, pmc + PCIR_POWER_STATUS, status, 2);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
return;
}
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
/*
** WOL in the newer chipset interfaces (pchlan)
** require thing to be copied into the phy
*/
static int
em_enable_phy_wakeup(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
u32 mreg, ret = 0;
u16 preg;
/* copy MAC RARs to PHY RARs */
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
e1000_copy_rx_addrs_to_phy_ich8lan(hw);
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
/* copy MAC MTA to PHY MTA */
for (int i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
mreg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
e1000_write_phy_reg(hw, BM_MTA(i), (u16)(mreg & 0xFFFF));
e1000_write_phy_reg(hw, BM_MTA(i) + 1,
(u16)((mreg >> 16) & 0xFFFF));
}
/* configure PHY Rx Control register */
e1000_read_phy_reg(&adapter->hw, BM_RCTL, &preg);
mreg = E1000_READ_REG(hw, E1000_RCTL);
if (mreg & E1000_RCTL_UPE)
preg |= BM_RCTL_UPE;
if (mreg & E1000_RCTL_MPE)
preg |= BM_RCTL_MPE;
preg &= ~(BM_RCTL_MO_MASK);
if (mreg & E1000_RCTL_MO_3)
preg |= (((mreg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
<< BM_RCTL_MO_SHIFT);
if (mreg & E1000_RCTL_BAM)
preg |= BM_RCTL_BAM;
if (mreg & E1000_RCTL_PMCF)
preg |= BM_RCTL_PMCF;
mreg = E1000_READ_REG(hw, E1000_CTRL);
if (mreg & E1000_CTRL_RFCE)
preg |= BM_RCTL_RFCE;
e1000_write_phy_reg(&adapter->hw, BM_RCTL, preg);
/* enable PHY wakeup in MAC register */
E1000_WRITE_REG(hw, E1000_WUC,
E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
E1000_WRITE_REG(hw, E1000_WUFC, adapter->wol);
/* configure and enable PHY wakeup in PHY registers */
e1000_write_phy_reg(&adapter->hw, BM_WUFC, adapter->wol);
e1000_write_phy_reg(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
/* activate PHY wakeup */
ret = hw->phy.ops.acquire(hw);
if (ret) {
printf("Could not acquire PHY\n");
return ret;
}
e1000_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
(BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
ret = e1000_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &preg);
if (ret) {
printf("Could not read PHY page 769\n");
goto out;
}
preg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
ret = e1000_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, preg);
if (ret)
printf("Could not set PHY Host Wakeup bit\n");
out:
hw->phy.ops.release(hw);
return ret;
}
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
em_led_func(void *arg, int onoff)
{
struct adapter *adapter = arg;
EM_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);
}
EM_CORE_UNLOCK(adapter);
}
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
/*
** Disable the L0S and L1 LINK states
*/
static void
em_disable_aspm(struct adapter *adapter)
{
int base, reg;
u16 link_cap,link_ctrl;
device_t dev = adapter->dev;
switch (adapter->hw.mac.type) {
case e1000_82573:
case e1000_82574:
case e1000_82583:
break;
default:
return;
}
Do a sweep of the tree replacing calls to pci_find_extcap() with calls to pci_find_cap() instead.
2011-03-23 13:10:15 +00:00
if (pci_find_cap(dev, PCIY_EXPRESS, &base) != 0)
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
return;
Align the PCI Express #defines with the style used for the PCI-X #defines. This also has the advantage that it makes the names more compact, iand also allows us to correct the non-uniform naming of the PCIM_LINK_* defines, making them all consistent amongst themselves. This is a mostly mechanical rename: s/PCIR_EXPRESS_/PCIER_/g s/PCIM_EXP_/PCIEM_/g s/PCIM_LINK_/PCIEM_LINK_/g When this is MFC'd, #defines will be added for the old names to assist out-of-tree drivers. Discussed with: jhb MFC after: 1 week
2012-09-18 22:04:59 +00:00
reg = base + PCIER_LINK_CAP;
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
link_cap = pci_read_config(dev, reg, 2);
Align the PCI Express #defines with the style used for the PCI-X #defines. This also has the advantage that it makes the names more compact, iand also allows us to correct the non-uniform naming of the PCIM_LINK_* defines, making them all consistent amongst themselves. This is a mostly mechanical rename: s/PCIR_EXPRESS_/PCIER_/g s/PCIM_EXP_/PCIEM_/g s/PCIM_LINK_/PCIEM_LINK_/g When this is MFC'd, #defines will be added for the old names to assist out-of-tree drivers. Discussed with: jhb MFC after: 1 week
2012-09-18 22:04:59 +00:00
if ((link_cap & PCIEM_LINK_CAP_ASPM) == 0)
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
return;
Align the PCI Express #defines with the style used for the PCI-X #defines. This also has the advantage that it makes the names more compact, iand also allows us to correct the non-uniform naming of the PCIM_LINK_* defines, making them all consistent amongst themselves. This is a mostly mechanical rename: s/PCIR_EXPRESS_/PCIER_/g s/PCIM_EXP_/PCIEM_/g s/PCIM_LINK_/PCIEM_LINK_/g When this is MFC'd, #defines will be added for the old names to assist out-of-tree drivers. Discussed with: jhb MFC after: 1 week
2012-09-18 22:04:59 +00:00
reg = base + PCIER_LINK_CTL;
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
link_ctrl = pci_read_config(dev, reg, 2);
Switch some PCI register reads from using magic numbers to using the names defined in pcireg.h MFC after: 1 week
2012-09-19 12:27:23 +00:00
link_ctrl &= ~PCIEM_LINK_CTL_ASPMC;
The purpose of this change is to add a routine to disable ASPM L0S and L1 LINK states on 82573, 82574, and 82583. The theory is that this is behind certain hangs being experienced by some customers. Also included a small optimization in the rxeof routine that was in my internal code. Change the PBA size for pchlan, it was incorrect. MFC after: 3 days
2010-11-24 22:24:07 +00:00
pci_write_config(dev, reg, link_ctrl, 2);
return;
}
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
/**********************************************************************
*
An attemp to make driver more readable and attaractive for further hacking: - Remove all spaces at eol. - Improve style(9) in most frequently edited functions. - In em_encap() push variables for 82544 workaround in the block where they are only used. - In em_get_buf() remove unused variable.
2006-01-20 11:38:25 +00:00
* Update the board statistics counters.
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
*
**********************************************************************/
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_update_stats_counters(struct adapter *adapter)
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
{
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
if(adapter->hw.phy.media_type == e1000_media_type_copper ||
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
(E1000_READ_REG(&adapter->hw, E1000_STATUS) & E1000_STATUS_LU)) {
adapter->stats.symerrs += E1000_READ_REG(&adapter->hw, E1000_SYMERRS);
adapter->stats.sec += E1000_READ_REG(&adapter->hw, E1000_SEC);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
}
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->stats.crcerrs += E1000_READ_REG(&adapter->hw, E1000_CRCERRS);
adapter->stats.mpc += E1000_READ_REG(&adapter->hw, E1000_MPC);
adapter->stats.scc += E1000_READ_REG(&adapter->hw, E1000_SCC);
adapter->stats.ecol += E1000_READ_REG(&adapter->hw, E1000_ECOL);
adapter->stats.mcc += E1000_READ_REG(&adapter->hw, E1000_MCC);
adapter->stats.latecol += E1000_READ_REG(&adapter->hw, E1000_LATECOL);
adapter->stats.colc += E1000_READ_REG(&adapter->hw, E1000_COLC);
adapter->stats.dc += E1000_READ_REG(&adapter->hw, E1000_DC);
adapter->stats.rlec += E1000_READ_REG(&adapter->hw, E1000_RLEC);
adapter->stats.xonrxc += E1000_READ_REG(&adapter->hw, E1000_XONRXC);
adapter->stats.xontxc += E1000_READ_REG(&adapter->hw, E1000_XONTXC);
Simplify hang detection by stealing the techniques used in ixl(4) and applying them to em(4). Rely on iterations through the local timer, and the tx queue state to determine if an actual hang has occurred. Any time a descriptor is used (packet sent), the tx queue is flagged as busy. Then when txeof runs, it either clears the flag when all is clean, or resets it to 1 if ANY are cleaned, if nothing is cleaned it increments the flag. Local timer simply checks to see if busy ever reaches MAX (10, which is compile time configurable), and then sets it as HUNG, at that point there is one more timer cycle in which to have any cleans, if not a watchdog reset will occur. Differential Revision: https://reviews.freebsd.org/D2019 Submitted by: jfv Reviewed by: hiren Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks
2015-06-02 18:28:41 +00:00
adapter->stats.xoffrxc += E1000_READ_REG(&adapter->hw, E1000_XOFFRXC);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->stats.xofftxc += E1000_READ_REG(&adapter->hw, E1000_XOFFTXC);
adapter->stats.fcruc += E1000_READ_REG(&adapter->hw, E1000_FCRUC);
adapter->stats.prc64 += E1000_READ_REG(&adapter->hw, E1000_PRC64);
adapter->stats.prc127 += E1000_READ_REG(&adapter->hw, E1000_PRC127);
adapter->stats.prc255 += E1000_READ_REG(&adapter->hw, E1000_PRC255);
adapter->stats.prc511 += E1000_READ_REG(&adapter->hw, E1000_PRC511);
adapter->stats.prc1023 += E1000_READ_REG(&adapter->hw, E1000_PRC1023);
adapter->stats.prc1522 += E1000_READ_REG(&adapter->hw, E1000_PRC1522);
adapter->stats.gprc += E1000_READ_REG(&adapter->hw, E1000_GPRC);
adapter->stats.bprc += E1000_READ_REG(&adapter->hw, E1000_BPRC);
adapter->stats.mprc += E1000_READ_REG(&adapter->hw, E1000_MPRC);
adapter->stats.gptc += E1000_READ_REG(&adapter->hw, E1000_GPTC);
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +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 */
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->stats.gorc += E1000_READ_REG(&adapter->hw, E1000_GORCL) +
((u64)E1000_READ_REG(&adapter->hw, E1000_GORCH) << 32);
adapter->stats.gotc += E1000_READ_REG(&adapter->hw, E1000_GOTCL) +
((u64)E1000_READ_REG(&adapter->hw, E1000_GOTCH) << 32);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->stats.rnbc += E1000_READ_REG(&adapter->hw, E1000_RNBC);
adapter->stats.ruc += E1000_READ_REG(&adapter->hw, E1000_RUC);
adapter->stats.rfc += E1000_READ_REG(&adapter->hw, E1000_RFC);
adapter->stats.roc += E1000_READ_REG(&adapter->hw, E1000_ROC);
adapter->stats.rjc += E1000_READ_REG(&adapter->hw, E1000_RJC);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
adapter->stats.tor += E1000_READ_REG(&adapter->hw, E1000_TORH);
adapter->stats.tot += E1000_READ_REG(&adapter->hw, E1000_TOTH);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
adapter->stats.tpr += E1000_READ_REG(&adapter->hw, E1000_TPR);
adapter->stats.tpt += E1000_READ_REG(&adapter->hw, E1000_TPT);
adapter->stats.ptc64 += E1000_READ_REG(&adapter->hw, E1000_PTC64);
adapter->stats.ptc127 += E1000_READ_REG(&adapter->hw, E1000_PTC127);
adapter->stats.ptc255 += E1000_READ_REG(&adapter->hw, E1000_PTC255);
adapter->stats.ptc511 += E1000_READ_REG(&adapter->hw, E1000_PTC511);
adapter->stats.ptc1023 += E1000_READ_REG(&adapter->hw, E1000_PTC1023);
adapter->stats.ptc1522 += E1000_READ_REG(&adapter->hw, E1000_PTC1522);
adapter->stats.mptc += E1000_READ_REG(&adapter->hw, E1000_MPTC);
adapter->stats.bptc += E1000_READ_REG(&adapter->hw, E1000_BPTC);
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
/* Interrupt Counts */
adapter->stats.iac += E1000_READ_REG(&adapter->hw, E1000_IAC);
adapter->stats.icrxptc += E1000_READ_REG(&adapter->hw, E1000_ICRXPTC);
adapter->stats.icrxatc += E1000_READ_REG(&adapter->hw, E1000_ICRXATC);
adapter->stats.ictxptc += E1000_READ_REG(&adapter->hw, E1000_ICTXPTC);
adapter->stats.ictxatc += E1000_READ_REG(&adapter->hw, E1000_ICTXATC);
adapter->stats.ictxqec += E1000_READ_REG(&adapter->hw, E1000_ICTXQEC);
adapter->stats.ictxqmtc += E1000_READ_REG(&adapter->hw, E1000_ICTXQMTC);
adapter->stats.icrxdmtc += E1000_READ_REG(&adapter->hw, E1000_ICRXDMTC);
adapter->stats.icrxoc += E1000_READ_REG(&adapter->hw, E1000_ICRXOC);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (adapter->hw.mac.type >= e1000_82543) {
adapter->stats.algnerrc +=
E1000_READ_REG(&adapter->hw, E1000_ALGNERRC);
adapter->stats.rxerrc +=
E1000_READ_REG(&adapter->hw, E1000_RXERRC);
adapter->stats.tncrs +=
E1000_READ_REG(&adapter->hw, E1000_TNCRS);
adapter->stats.cexterr +=
E1000_READ_REG(&adapter->hw, E1000_CEXTERR);
adapter->stats.tsctc +=
E1000_READ_REG(&adapter->hw, E1000_TSCTC);
adapter->stats.tsctfc +=
E1000_READ_REG(&adapter->hw, E1000_TSCTFC);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
}
- Use if_inc_counter() to increment various counters. - Do not ever set a counter to a value. For those counters that we don't increment, but return directly from hardware create cases in if_get_counter() method. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-09-18 15:56:14 +00:00
}
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
- Use if_inc_counter() to increment various counters. - Do not ever set a counter to a value. For those counters that we don't increment, but return directly from hardware create cases in if_get_counter() method. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-09-18 15:56:14 +00:00
static uint64_t
em_get_counter(if_t ifp, ift_counter cnt)
{
struct adapter *adapter;
Added support for 82545EM and 82546EB based adapters. Added Vlan support. MFC after: 1 week
2002-06-03 22:30:51 +00:00
- Use if_inc_counter() to increment various counters. - Do not ever set a counter to a value. For those counters that we don't increment, but return directly from hardware create cases in if_get_counter() method. Sponsored by: Netflix Sponsored by: Nginx, Inc.
2014-09-18 15:56:14 +00:00
adapter = if_getsoftc(ifp);
switch (cnt) {
case IFCOUNTER_COLLISIONS:
return (adapter->stats.colc);
case IFCOUNTER_IERRORS:
return (adapter->dropped_pkts + adapter->stats.rxerrc +
adapter->stats.crcerrs + adapter->stats.algnerrc +
adapter->stats.ruc + adapter->stats.roc +
adapter->stats.mpc + adapter->stats.cexterr);
case IFCOUNTER_OERRORS:
return (adapter->stats.ecol + adapter->stats.latecol +
adapter->watchdog_events);
default:
return (if_get_counter_default(ifp, cnt));
}
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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
em_sysctl_reg_handler(SYSCTL_HANDLER_ARGS)
{
struct adapter *adapter;
u_int val;
adapter = oidp->oid_arg1;
val = E1000_READ_REG(&adapter->hw, oidp->oid_arg2);
return (sysctl_handle_int(oidp, &val, 0, req));
}
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 20:57:41 +00:00
/*
* Add sysctl variables, one per statistic, to the system.
*/
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +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 20:57:41 +00:00
em_add_hw_stats(struct adapter *adapter)
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
device_t dev = adapter->dev;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +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 tx_ring *txr = adapter->tx_rings;
struct rx_ring *rxr = adapter->rx_rings;
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 20:57:41 +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);
struct e1000_hw_stats *stats = &adapter->stats;
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 sysctl_oid *stat_node, *queue_node, *int_node;
struct sysctl_oid_list *stat_list, *queue_list, *int_list;
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 20:57:41 +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
#define QUEUE_NAME_LEN 32
char namebuf[QUEUE_NAME_LEN];
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 20:57:41 +00:00
/* Driver Statistics */
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "link_irq",
CTLFLAG_RD, &adapter->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 20:57:41 +00:00
"Link MSIX IRQ Handled");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "mbuf_alloc_fail",
CTLFLAG_RD, &adapter->mbuf_alloc_failed,
"Std mbuf failed");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "cluster_alloc_fail",
CTLFLAG_RD, &adapter->mbuf_cluster_failed,
"Std mbuf cluster failed");
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");
SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "device_control",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_CTRL,
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
em_sysctl_reg_handler, "IU",
"Device Control Register");
SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_control",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_RCTL,
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
em_sysctl_reg_handler, "IU",
"Receiver Control Register");
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 20:57:41 +00:00
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");
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++, rxr++) {
snprintf(namebuf, QUEUE_NAME_LEN, "queue_tx_%d", i);
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
queue_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
CTLFLAG_RD, NULL, "TX Queue Name");
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
queue_list = SYSCTL_CHILDREN(queue_node);
SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_head",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | 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
em_sysctl_reg_handler, "IU",
"Transmit Descriptor Head");
SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_tail",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | 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
em_sysctl_reg_handler, "IU",
"Transmit Descriptor Tail");
SYSCTL_ADD_ULONG(ctx, queue_list, OID_AUTO, "tx_irq",
CTLFLAG_RD, &txr->tx_irq,
"Queue MSI-X Transmit Interrupts");
SYSCTL_ADD_ULONG(ctx, queue_list, OID_AUTO, "no_desc_avail",
CTLFLAG_RD, &txr->no_desc_avail,
"Queue No Descriptor Available");
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
snprintf(namebuf, QUEUE_NAME_LEN, "queue_rx_%d", i);
queue_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf,
CTLFLAG_RD, NULL, "RX Queue Name");
queue_list = SYSCTL_CHILDREN(queue_node);
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_PROC(ctx, queue_list, OID_AUTO, "rxd_head",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | 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
em_sysctl_reg_handler, "IU",
"Receive Descriptor Head");
SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "rxd_tail",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | 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
em_sysctl_reg_handler, "IU",
"Receive Descriptor Tail");
SYSCTL_ADD_ULONG(ctx, queue_list, OID_AUTO, "rx_irq",
CTLFLAG_RD, &rxr->rx_irq,
"Queue MSI-X Receive Interrupts");
}
/* 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 20:57:41 +00:00
stat_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "mac_stats",
CTLFLAG_RD, NULL, "Statistics");
stat_list = SYSCTL_CHILDREN(stat_node);
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(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 20:57:41 +00:00
CTLFLAG_RD, &stats->ecol,
"Excessive collisions");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "single_coll",
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
CTLFLAG_RD, &stats->scc,
"Single collisions");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "multiple_coll",
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
CTLFLAG_RD, &stats->mcc,
"Multiple collisions");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "late_coll",
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
CTLFLAG_RD, &stats->latecol,
"Late collisions");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "collision_count",
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
CTLFLAG_RD, &stats->colc,
"Collision Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "symbol_errors",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.symerrs,
"Symbol Errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "sequence_errors",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.sec,
"Sequence Errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "defer_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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.dc,
"Defer Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "missed_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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.mpc,
"Missed Packets");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_no_buff",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.rnbc,
"Receive No Buffers");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_undersize",
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
CTLFLAG_RD, &adapter->stats.ruc,
"Receive Undersize");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_fragmented",
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
CTLFLAG_RD, &adapter->stats.rfc,
"Fragmented Packets Received ");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_oversize",
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
CTLFLAG_RD, &adapter->stats.roc,
"Oversized Packets Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_jabber",
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
CTLFLAG_RD, &adapter->stats.rjc,
"Recevied Jabber");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_errs",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.rxerrc,
"Receive Errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "crc_errs",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.crcerrs,
"CRC errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "alignment_errs",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.algnerrc,
"Alignment Errors");
/* On 82575 these are collision counts */
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "coll_ext_errs",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.cexterr,
"Collision/Carrier extension errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xon_recvd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.xonrxc,
"XON Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xon_txd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.xontxc,
"XON Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xoff_recvd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.xoffrxc,
"XOFF Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xoff_txd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.xofftxc,
"XOFF Transmitted");
/* Packet Reception Stats */
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "total_pkts_recvd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.tpr,
"Total Packets Received ");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_pkts_recvd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.gprc,
"Good Packets Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_recvd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.bprc,
"Broadcast Packets Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_recvd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.mprc,
"Multicast Packets Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_64",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.prc64,
"64 byte frames received ");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_65_127",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.prc127,
"65-127 byte frames received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_128_255",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.prc255,
"128-255 byte frames received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_256_511",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.prc511,
"256-511 byte frames received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_512_1023",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.prc1023,
"512-1023 byte frames received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_1024_1522",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.prc1522,
"1023-1522 byte frames received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_octets_recvd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.gorc,
"Good Octets Received");
/* Packet Transmission Stats */
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_octets_txd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->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");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "total_pkts_txd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.tpt,
"Total Packets Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_pkts_txd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.gptc,
"Good Packets Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_txd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.bptc,
"Broadcast Packets Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_txd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.mptc,
"Multicast Packets Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_64",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.ptc64,
"64 byte frames transmitted ");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_65_127",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.ptc127,
"65-127 byte frames transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_128_255",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.ptc255,
"128-255 byte frames transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_256_511",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.ptc511,
"256-511 byte frames transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_512_1023",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.ptc1023,
"512-1023 byte frames transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_1024_1522",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.ptc1522,
"1024-1522 byte frames transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tso_txd",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.tsctc,
"TSO Contexts Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tso_ctx_fail",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.tsctfc,
"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);
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "asserts",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.iac,
"Interrupt Assertion Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "rx_pkt_timer",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.icrxptc,
"Interrupt Cause Rx Pkt Timer Expire Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "rx_abs_timer",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.icrxatc,
"Interrupt Cause Rx Abs Timer Expire Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "tx_pkt_timer",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.ictxptc,
"Interrupt Cause Tx Pkt Timer Expire Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "tx_abs_timer",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.ictxatc,
"Interrupt Cause Tx Abs Timer Expire Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "tx_queue_empty",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.ictxqec,
"Interrupt Cause Tx Queue Empty Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "tx_queue_min_thresh",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.ictxqmtc,
"Interrupt Cause Tx Queue Min Thresh Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "rx_desc_min_thresh",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.icrxdmtc,
"Interrupt Cause Rx Desc Min Thresh Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "rx_overrun",
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 20:57:41 +00:00
CTLFLAG_RD, &adapter->stats.icrxoc,
"Interrupt Cause Receiver Overrun Count");
This is the first commit of the Intel gigabit driver for PRO/1000 cards. Submitted by:Prafulla Deuskar Reviewed by: Paul Saab MFC after:1 week
2001-12-02 07:37:17 +00:00
}
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +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.
*
**********************************************************************/
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
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 20:57:41 +00:00
em_sysctl_nvm_info(SYSCTL_HANDLER_ARGS)
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +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
struct adapter *adapter = (struct adapter *)arg1;
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
int error;
int result;
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
result = -1;
error = sysctl_handle_int(oidp, &result, 0, req);
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
if (error || !req->newptr)
return (error);
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
/*
* This value will cause a hex dump of the
* first 32 16-bit words of the EEPROM to
* the screen.
*/
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 (result == 1)
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
em_print_nvm_info(adapter);
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (error);
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +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 20:57:41 +00:00
static void
em_print_nvm_info(struct adapter *adapter)
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +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 20:57:41 +00:00
u16 eeprom_data;
int i, j, row = 0;
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +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 20:57:41 +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);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +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 20:57:41 +00:00
printf("\n");
Add support for Quad port adapter Add sysctl's to display statistics/debug_info Set WAIT_FOR_AUTONEG_DEFAULT to zero by default Increment packet in/out statistics inline instead of every two seconds. MFC after: 3 days
2003-06-05 17:51:38 +00:00
}
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
static int
em_sysctl_int_delay(SYSCTL_HANDLER_ARGS)
{
struct em_int_delay_info *info;
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
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
u32 regval;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
int error, usecs, ticks;
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
info = (struct em_int_delay_info *)arg1;
usecs = info->value;
error = sysctl_handle_int(oidp, &usecs, 0, req);
if (error != 0 || req->newptr == NULL)
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (error);
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
if (usecs < 0 || usecs > EM_TICKS_TO_USECS(65535))
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (EINVAL);
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
info->value = usecs;
Merge in the new driver (6.5.0) of Intel. This has a new shared code infrastructure that is family specific and modular. There is also support for our latest gigabit nic, the 82575 that is MSI/X and multiqueue capable. The new shared code changes some interfaces to the core code but testing at Intel has been going on for months, it is fairly stable. I have attempted to be careful in retaining any fixes that CURRENT had and we did not, I apologize in advance if any thing gets clobbered, I'm sure I'll hear about it :) Approved by pdeuskar
2007-05-04 00:00:12 +00:00
ticks = EM_USECS_TO_TICKS(usecs);
if_lem.c: make sure that lem_rxeof() can drain the entire rx queue irrespective of the setting of lem_rx_process_limit, while giving a chance to the taskqueue scheduler to act after each chunk. This makes lem_rxeof similar to the one in if_em.c and if_igb.c . if_lem.c and if_em.c: add a sysctl to manually configure the 'itr' moderation register. Approved by: Jack Vogel
2013-05-09 17:07:30 +00:00
if (info->offset == E1000_ITR) /* units are 256ns here */
ticks *= 4;
Some more minor cleanups of em(4) driver: - Destroy mutex in case of attach failure. [1] - Lock properly em_watchdog(). [1] - Lock properly em_sysctl_int_delay(). [1] - Remove unused global adapter linked list. - Remove unused dma_size field from struct em_dma_alloc. - Do not touch interface statistics, that must be edited only by upper layers. [1] Submitted by: yongari [1]
2005-10-20 09:55:49 +00:00
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter = info->adapter;
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_LOCK(adapter);
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
regval = E1000_READ_OFFSET(&adapter->hw, info->offset);
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
regval = (regval & ~0xffff) | (ticks & 0xffff);
/* Handle a few special cases. */
switch (info->offset) {
case E1000_RDTR:
break;
case E1000_TIDV:
if (ticks == 0) {
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
adapter->txd_cmd &= ~E1000_TXD_CMD_IDE;
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
/* Don't write 0 into the TIDV register. */
regval++;
} 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
adapter->txd_cmd |= E1000_TXD_CMD_IDE;
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
break;
}
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
E1000_WRITE_OFFSET(&adapter->hw, info->offset, regval);
Driver version 6.7.3 - Bring HEAD up to the latest shared code - Fix TSO problem using limited MSS and forwarding - Dual lock implementation - New device support - For my ease, this code can compile in either 6.x or later - brings this driver in sync with the 6.3
2007-11-20 21:41:22 +00:00
EM_CORE_UNLOCK(adapter);
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
return (0);
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
}
static void
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
em_add_int_delay_sysctl(struct adapter *adapter, const char *name,
- Second style(9) megacleanup. - Rename "adapter" to "sc"/"softc", to be like other drivers. (-13 Kb less source code)
2006-02-15 08:39:50 +00:00
const char *description, struct em_int_delay_info *info,
int offset, int value)
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
{
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
info->adapter = adapter;
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
info->offset = offset;
info->value = value;
Revert back changes to made in rev 1.109 of if_em.c which were unnecessary. This makes it easier for us to get the changes into -current and to -stable quickly.
2006-08-03 19:05:04 +00:00
SYSCTL_ADD_PROC(device_get_sysctl_ctx(adapter->dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(adapter->dev)),
Add facilities for tuning the "em" driver's interrupt delays without recompiling the driver. See the comments near the top of "if_em.h" for descriptions of these delays. Four new loader tunables control the system-wide default values: hw.em.tx_int_delay hw.em.rx_int_delay hw.em.tx_abs_int_delay hw.em.rx_abs_int_delay The tunables are specified in microseconds. The valid range is 0-67108 usec., and 0 means that the timer is disabled. There are also four new sysctls (actually, a set of four for each "em" device in the system) to query and change the interrupt delays after the system is up: hw.em0.tx_int_delay hw.em0.rx_int_delay hw.em0.tx_abs_int_delay (not present for 82542/3/4 adapters) hw.em0.rx_abs_int_delay (not present for 82542/3/4 adapters) It seems to be OK to change these values even while the adapter is passing traffic. Approved by: Prafulla Deuskar <pdeuskar@FreeBSD.ORG> MFC after: 4 weeks
2003-08-01 17:33:59 +00:00
OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW,
info, 0, em_sysctl_int_delay, "I", description);
}
Backout bogus checkin to HEAD Approved by: scottl
2006-10-28 08:11:07 +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
em_set_sysctl_value(struct adapter *adapter, const char *name,
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +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)),
Fix multiple incorrect SYSCTL arguments in the kernel: - Wrong integer type was specified. - Wrong or missing "access" specifier. The "access" specifier sometimes included the SYSCTL type, which it should not, except for procedural SYSCTL nodes. - Logical OR where binary OR was expected. - Properly assert the "access" argument passed to all SYSCTL macros, using the CTASSERT macro. This applies to both static- and dynamically created SYSCTLs. - Properly assert the the data type for both static and dynamic SYSCTLs. In the case of static SYSCTLs we only assert that the data pointed to by the SYSCTL data pointer has the correct size, hence there is no easy way to assert types in the C language outside a C-function. - Rewrote some code which doesn't pass a constant "access" specifier when creating dynamic SYSCTL nodes, which is now a requirement. - Updated "EXAMPLES" section in SYSCTL manual page. MFC after: 3 days Sponsored by: Mellanox Technologies
2014-10-21 07:31:21 +00:00
OID_AUTO, name, CTLFLAG_RW, limit, value, description);
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +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
/*
** Set flow control using sysctl:
** Flow control values:
** 0 - off
** 1 - rx pause
** 2 - tx pause
** 3 - full
*/
static int
em_set_flowcntl(SYSCTL_HANDLER_ARGS)
{
int error;
static int input = 3; /* default is full */
struct adapter *adapter = (struct adapter *) arg1;
error = sysctl_handle_int(oidp, &input, 0, req);
if ((error) || (req->newptr == NULL))
return (error);
if (input == adapter->fc) /* no change? */
return (error);
switch (input) {
case e1000_fc_rx_pause:
case e1000_fc_tx_pause:
case e1000_fc_full:
case e1000_fc_none:
adapter->hw.fc.requested_mode = input;
adapter->fc = input;
break;
default:
/* Do nothing */
return (error);
}
adapter->hw.fc.current_mode = adapter->hw.fc.requested_mode;
e1000_force_mac_fc(&adapter->hw);
return (error);
}
Change the interface to the Energy Efficient Ethernet (EEE) setting in the igb and em driver. This was necessitated by a shared code change that I was given late in the game, a data type changed from bool to int, in the last update I dealt with it by a cast, but it was pointed out (thanks jhb) that there was a potential problem with this. John suggested this safer approach, and it is fine with me... MFC after:2 days (to catch the 9.1 update)
2012-07-07 20:21:05 +00:00
/*
** Manage Energy Efficient Ethernet:
** Control values:
** 0/1 - enabled/disabled
*/
static int
em_sysctl_eee(SYSCTL_HANDLER_ARGS)
{
struct adapter *adapter = (struct adapter *) arg1;
int error, value;
value = adapter->hw.dev_spec.ich8lan.eee_disable;
error = sysctl_handle_int(oidp, &value, 0, req);
if (error || req->newptr == NULL)
return (error);
EM_CORE_LOCK(adapter);
adapter->hw.dev_spec.ich8lan.eee_disable = (value != 0);
em_init_locked(adapter);
EM_CORE_UNLOCK(adapter);
return (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
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
static int
em_sysctl_debug_info(SYSCTL_HANDLER_ARGS)
{
struct adapter *adapter;
int error;
int result;
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
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
result = -1;
error = sysctl_handle_int(oidp, &result, 0, req);
if (error || !req->newptr)
return (error);
if (result == 1) {
adapter = (struct adapter *)arg1;
em_print_debug_info(adapter);
}
return (error);
}
/*
** This routine is meant to be fluid, add whatever is
** needed for debugging a problem. -jfv
*/
static void
em_print_debug_info(struct adapter *adapter)
{
device_t dev = adapter->dev;
struct tx_ring *txr = adapter->tx_rings;
struct rx_ring *rxr = adapter->rx_rings;
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getdrvflags(adapter->ifp) & IFF_DRV_RUNNING)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
printf("Interface is RUNNING ");
else
printf("Interface is NOT RUNNING\n");
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
Convert em(4) to use the driver API. Submitted by: Anuranjan Shukla <anshukla@juniper.net> Obtained from: Juniper Networks, Inc.
2014-06-02 18:52:03 +00:00
if (if_getdrvflags(adapter->ifp) & IFF_DRV_OACTIVE)
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
printf("and INACTIVE\n");
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
else
printf("and ACTIVE\n");
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++, rxr++) {
device_printf(dev, "TX Queue %d ------\n", i);
device_printf(dev, "hw tdh = %d, hw tdt = %d\n",
E1000_READ_REG(&adapter->hw, E1000_TDH(i)),
E1000_READ_REG(&adapter->hw, E1000_TDT(i)));
device_printf(dev, "Tx Queue Status = %d\n", txr->busy);
device_printf(dev, "TX descriptors avail = %d\n",
txr->tx_avail);
device_printf(dev, "Tx Descriptors avail failure = %ld\n",
txr->no_desc_avail);
device_printf(dev, "RX Queue %d ------\n", i);
device_printf(dev, "hw rdh = %d, hw rdt = %d\n",
E1000_READ_REG(&adapter->hw, E1000_RDH(i)),
E1000_READ_REG(&adapter->hw, E1000_RDT(i)));
device_printf(dev, "RX discarded packets = %ld\n",
rxr->rx_discarded);
device_printf(dev, "RX Next to Check = %d\n", rxr->next_to_check);
device_printf(dev, "RX Next to Refresh = %d\n", rxr->next_to_refresh);
}
Bug fix delta to the em driver: - Chasin down bogus watchdogs has led to an improved design to this handling, the hang decision takes place in the tx cleanup, with only a simple report check in local_timer. Our tests have shown no false watchdogs with this code. - VLAN fixes from jhb, the shadow vfta should be per interface, but as global it was not. Thanks John. - Bug fixes in the support for new PCH2 hardware. - Thanks for all the help and feedback on the driver, changes to lem with be coming shortly as well.
2010-10-26 00:07:58 +00:00
}
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
#ifdef EM_MULTIQUEUE
/*
* 82574 only:
* Write a new value to the EEPROM increasing the number of MSIX
* vectors from 3 to 5, for proper multiqueue support.
*/
static void
em_enable_vectors_82574(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
device_t dev = adapter->dev;
u16 edata;
e1000_read_nvm(hw, EM_NVM_PCIE_CTRL, 1, &edata);
printf("Current cap: %#06x\n", edata);
if (((edata & EM_NVM_MSIX_N_MASK) >> EM_NVM_MSIX_N_SHIFT) != 4) {
device_printf(dev, "Writing to eeprom: increasing "
"reported MSIX vectors from 3 to 5...\n");
edata &= ~(EM_NVM_MSIX_N_MASK);
edata |= 4 << EM_NVM_MSIX_N_SHIFT;
e1000_write_nvm(hw, EM_NVM_PCIE_CTRL, 1, &edata);
e1000_update_nvm_checksum(hw);
device_printf(dev, "Writing to eeprom: done\n");
}
}
#endif
#ifdef DDB
DB_COMMAND(em_reset_dev, em_ddb_reset_dev)
{
devclass_t dc;
int max_em;
dc = devclass_find("em");
max_em = devclass_get_maxunit(dc);
for (int index = 0; index < (max_em - 1); index++) {
device_t dev;
dev = devclass_get_device(dc, index);
if (device_get_driver(dev) == &em_driver) {
struct adapter *adapter = device_get_softc(dev);
Add an adapter CORE lock in the DDB hook em_dump_queue to avoid WITNESS panic in em_init_locked() while debugging. MFC after: 2 weeks Sponsored by: Limelight Networks
2015-07-16 16:32:57 +00:00
EM_CORE_LOCK(adapter);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
em_init_locked(adapter);
Add an adapter CORE lock in the DDB hook em_dump_queue to avoid WITNESS panic in em_init_locked() while debugging. MFC after: 2 weeks Sponsored by: Limelight Networks
2015-07-16 16:32:57 +00:00
EM_CORE_UNLOCK(adapter);
Change EM_MULTIQUEUE to a real kernconf entry and enable support for up to 2 rx/tx queues for the 82574. Program the 82574 to enable 5 msix vectors, assign 1 to each rx queue, 1 to each tx queue and 1 to the link handler. Inspired by DragonFlyBSD, enable some RSS logic for handling tx queue handling/processing. Move multiqueue handler functions so that they line up better in a diff review to if_igb.c Always enqueue tx work to be done in em_mq_start, if unable to acquire the TX lock, then this will be processed in the background later by the taskqueue. Remove mbuf argument from em_start_mq_locked() as the work is always enqueued. (stolen from igb) Setup TARC, TXDCTL and RXDCTL registers for better performance and stability in multiqueue and singlequeue implementations. Handle Intel errata 3 and generic multiqueue behavior with the initialization of TARC(0) and TARC(1) Bind interrupt threads to cpus in order. (stolen from igb) Add 2 new DDB functions, one to display the queue(s) and their settings and one to reset the adapter. Primarily used for debugging. In the multiqueue configuration, bump RXD and TXD ring size to max for the adapter (4096). Setup an RDTR of 64 and an RADV of 128 in multiqueue configuration to cut down on the number of interrupts. RADV was arbitrarily set to 2x RDTR and can be adjusted as needed. Cleanup the display in top a bit to make it clearer where the taskqueue threads are running and what they should be doing. Ensure that both queues are processed by em_local_timer() by writing them both to the IMS register to generate soft interrupts. Ensure that an soft interrupt is generated when em_msix_link() is run so that any races between assertion of the link/status interrupt and a rx/tx interrupt are handled. Document existing tuneables: hw.em.eee_setting, hw.em.msix, hw.em.smart_pwr_down, hw.em.sbp Document use of hw.em.num_queues and the new kernel option EM_MULTIQUEUE Thanks to Intel for their continued support of FreeBSD. Reviewed by: erj jfv hiren gnn wblock Obtained from: Intel Corporation MFC after: 2 weeks Relnotes: Yes Sponsored by: Limelight Networks Differential Revision: https://reviews.freebsd.org/D1994
2015-06-03 18:01:09 +00:00
}
}
}
DB_COMMAND(em_dump_queue, em_ddb_dump_queue)
{
devclass_t dc;
int max_em;
dc = devclass_find("em");
max_em = devclass_get_maxunit(dc);
for (int index = 0; index < (max_em - 1); index++) {
device_t dev;
dev = devclass_get_device(dc, index);
if (device_get_driver(dev) == &em_driver)
em_print_debug_info(device_get_softc(dev));
}
}
#endif
Reference in New Issue Copy Permalink