d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
b82a4c003c
freebsd-skq/sys/dev/ixgbe/ixgbe.c

4884 lines
135 KiB
C
Raw Normal View History

This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/******************************************************************************
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
Copyright (c) 2001-2010, Intel Corporation
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +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.
******************************************************************************/
/*$FreeBSD$*/
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
#ifdef HAVE_KERNEL_OPTION_HEADERS
#include "opt_device_polling.h"
#endif
A couple late breaking bugs that testing have turned up. - change include style so build in kernel try OR standalone work. - Limit HWCSUM - I was led to believe that it would work with RSS, but our testing had odd issues which suggests this is false. - A fatfinger error in the ioctl code made ifconfig up not work. Approved by: re
2007-07-12 19:04:11 +00:00
#include "ixgbe.h"
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
* Set this to one to display debug statistics
*********************************************************************/
int ixgbe_display_debug_stats = 0;
/*********************************************************************
* Driver version
*********************************************************************/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
char ixgbe_driver_version[] = "2.1.6";
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
* PCI Device ID Table
*
* Used by probe to select devices to load on
* Last field stores an index into ixgbe_strings
* Last entry must be all 0s
*
* { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index }
*********************************************************************/
static ixgbe_vendor_info_t ixgbe_vendor_info_array[] =
{
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598AF_DUAL_PORT, 0, 0, 0},
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598AF_SINGLE_PORT, 0, 0, 0},
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598EB_CX4, 0, 0, 0},
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598AT, 0, 0, 0},
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598AT2, 0, 0, 0},
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598, 0, 0, 0},
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598_DA_DUAL_PORT, 0, 0, 0},
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598_CX4_DUAL_PORT, 0, 0, 0},
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598EB_XF_LR, 0, 0, 0},
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM, 0, 0, 0},
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598EB_SFP_LOM, 0, 0, 0},
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_KX4, 0, 0, 0},
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_KX4_MEZZ, 0, 0, 0},
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_SFP, 0, 0, 0},
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_XAUI_LOM, 0, 0, 0},
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_CX4, 0, 0, 0},
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_T3_LOM, 0, 0, 0},
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
{IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_COMBO_BACKPLANE, 0, 0, 0},
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* required last entry */
{0, 0, 0, 0, 0}
};
/*********************************************************************
* Table of branding strings
*********************************************************************/
static char *ixgbe_strings[] = {
"Intel(R) PRO/10GbE PCI-Express Network Driver"
};
/*********************************************************************
* Function prototypes
*********************************************************************/
static int ixgbe_probe(device_t);
static int ixgbe_attach(device_t);
static int ixgbe_detach(device_t);
static int ixgbe_shutdown(device_t);
static void ixgbe_start(struct ifnet *);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
static void ixgbe_start_locked(struct tx_ring *, struct ifnet *);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
#if __FreeBSD_version >= 800000
static int ixgbe_mq_start(struct ifnet *, struct mbuf *);
static int ixgbe_mq_start_locked(struct ifnet *,
struct tx_ring *, struct mbuf *);
static void ixgbe_qflush(struct ifnet *);
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static int ixgbe_ioctl(struct ifnet *, u_long, caddr_t);
static void ixgbe_init(void *);
static void ixgbe_init_locked(struct adapter *);
static void ixgbe_stop(void *);
static void ixgbe_media_status(struct ifnet *, struct ifmediareq *);
static int ixgbe_media_change(struct ifnet *);
static void ixgbe_identify_hardware(struct adapter *);
static int ixgbe_allocate_pci_resources(struct adapter *);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
static int ixgbe_allocate_msix(struct adapter *);
static int ixgbe_allocate_legacy(struct adapter *);
static int ixgbe_allocate_queues(struct adapter *);
static int ixgbe_setup_msix(struct adapter *);
static void ixgbe_free_pci_resources(struct adapter *);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void ixgbe_local_timer(void *);
static void ixgbe_setup_interface(device_t, struct adapter *);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
static void ixgbe_config_link(struct adapter *);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static int ixgbe_allocate_transmit_buffers(struct tx_ring *);
static int ixgbe_setup_transmit_structures(struct adapter *);
static void ixgbe_setup_transmit_ring(struct tx_ring *);
static void ixgbe_initialize_transmit_units(struct adapter *);
static void ixgbe_free_transmit_structures(struct adapter *);
static void ixgbe_free_transmit_buffers(struct tx_ring *);
static int ixgbe_allocate_receive_buffers(struct rx_ring *);
static int ixgbe_setup_receive_structures(struct adapter *);
static int ixgbe_setup_receive_ring(struct rx_ring *);
static void ixgbe_initialize_receive_units(struct adapter *);
static void ixgbe_free_receive_structures(struct adapter *);
static void ixgbe_free_receive_buffers(struct rx_ring *);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
static void ixgbe_setup_hw_rsc(struct rx_ring *);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void ixgbe_enable_intr(struct adapter *);
static void ixgbe_disable_intr(struct adapter *);
static void ixgbe_update_stats_counters(struct adapter *);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
static bool ixgbe_txeof(struct tx_ring *);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
static bool ixgbe_rxeof(struct ix_queue *, int);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
static void ixgbe_rx_checksum(u32, struct mbuf *);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void ixgbe_set_promisc(struct adapter *);
static void ixgbe_disable_promisc(struct adapter *);
static void ixgbe_set_multi(struct adapter *);
static void ixgbe_print_hw_stats(struct adapter *);
static void ixgbe_print_debug_info(struct adapter *);
static void ixgbe_update_link_status(struct adapter *);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
static void ixgbe_refresh_mbufs(struct rx_ring *, int);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
static int ixgbe_xmit(struct tx_ring *, struct mbuf **);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static int ixgbe_sysctl_stats(SYSCTL_HANDLER_ARGS);
static int ixgbe_sysctl_debug(SYSCTL_HANDLER_ARGS);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
static int ixgbe_set_flowcntl(SYSCTL_HANDLER_ARGS);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static int ixgbe_dma_malloc(struct adapter *, bus_size_t,
struct ixgbe_dma_alloc *, int);
static void ixgbe_dma_free(struct adapter *, struct ixgbe_dma_alloc *);
static void ixgbe_add_rx_process_limit(struct adapter *, const char *,
const char *, int *, int);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
static bool ixgbe_tx_ctx_setup(struct tx_ring *, struct mbuf *);
static bool ixgbe_tso_setup(struct tx_ring *, struct mbuf *, u32 *);
static void ixgbe_set_ivar(struct adapter *, u8, u8, s8);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void ixgbe_configure_ivars(struct adapter *);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
static u8 * ixgbe_mc_array_itr(struct ixgbe_hw *, u8 **, u32 *);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
static void ixgbe_setup_vlan_hw_support(struct adapter *);
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
static void ixgbe_register_vlan(void *, struct ifnet *, u16);
static void ixgbe_unregister_vlan(void *, struct ifnet *, u16);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
static __inline void ixgbe_rx_discard(struct rx_ring *, int);
static __inline void ixgbe_rx_input(struct rx_ring *, struct ifnet *,
struct mbuf *, u32);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/* Support for pluggable optic modules */
static bool ixgbe_sfp_probe(struct adapter *);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Legacy (single vector interrupt handler */
static void ixgbe_legacy_irq(void *);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* The MSI/X Interrupt handlers */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
static void ixgbe_msix_que(void *);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
static void ixgbe_msix_link(void *);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/* Deferred interrupt tasklets */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
static void ixgbe_handle_que(void *, int);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
static void ixgbe_handle_link(void *, int);
static void ixgbe_handle_msf(void *, int);
static void ixgbe_handle_mod(void *, int);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
static void ixgbe_atr(struct tx_ring *, struct mbuf *);
static void ixgbe_reinit_fdir(void *, int);
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
* FreeBSD Device Interface Entry Points
*********************************************************************/
static device_method_t ixgbe_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, ixgbe_probe),
DEVMETHOD(device_attach, ixgbe_attach),
DEVMETHOD(device_detach, ixgbe_detach),
DEVMETHOD(device_shutdown, ixgbe_shutdown),
{0, 0}
};
static driver_t ixgbe_driver = {
"ix", ixgbe_methods, sizeof(struct adapter),
};
static devclass_t ixgbe_devclass;
DRIVER_MODULE(ixgbe, pci, ixgbe_driver, ixgbe_devclass, 0, 0);
MODULE_DEPEND(ixgbe, pci, 1, 1, 1);
MODULE_DEPEND(ixgbe, ether, 1, 1, 1);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
/*
** TUNEABLE PARAMETERS:
*/
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
** AIM: Adaptive Interrupt Moderation
** which means that the interrupt rate
** is varied over time based on the
** traffic for that interrupt vector
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
*/
static int ixgbe_enable_aim = TRUE;
TUNABLE_INT("hw.ixgbe.enable_aim", &ixgbe_enable_aim);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* How many packets rxeof tries to clean at a time */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
static int ixgbe_rx_process_limit = 128;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
TUNABLE_INT("hw.ixgbe.rx_process_limit", &ixgbe_rx_process_limit);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/* Flow control setting, default to full */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
static int ixgbe_flow_control = ixgbe_fc_full;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
TUNABLE_INT("hw.ixgbe.flow_control", &ixgbe_flow_control);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/*
** Smart speed setting, default to on
** this only works as a compile option
** right now as its during attach, set
** this to 'ixgbe_smart_speed_off' to
** disable.
*/
static int ixgbe_smart_speed = ixgbe_smart_speed_on;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/*
* MSIX should be the default for best performance,
* but this allows it to be forced off for testing.
*/
static int ixgbe_enable_msix = 1;
TUNABLE_INT("hw.ixgbe.enable_msix", &ixgbe_enable_msix);
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
/*
* Header split has seemed to be beneficial in
* all circumstances tested, so its on by default
* however this variable will allow it to be disabled
* for some debug purposes.
*/
static bool ixgbe_header_split = TRUE;
TUNABLE_INT("hw.ixgbe.hdr_split", &ixgbe_header_split);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
* Number of Queues, can be set to 0,
* it then autoconfigures based on the
* number of cpus. Each queue is a pair
* of RX and TX rings with a msix vector
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
*/
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
static int ixgbe_num_queues = 0;
TUNABLE_INT("hw.ixgbe.num_queues", &ixgbe_num_queues);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/*
** Number of TX descriptors per ring,
** setting higher than RX as this seems
** the better performing choice.
*/
static int ixgbe_txd = PERFORM_TXD;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
TUNABLE_INT("hw.ixgbe.txd", &ixgbe_txd);
/* Number of RX descriptors per ring */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
static int ixgbe_rxd = PERFORM_RXD;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
TUNABLE_INT("hw.ixgbe.rxd", &ixgbe_rxd);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Keep running tab on them for sanity check */
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
static int ixgbe_total_ports;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/*
** Shadow VFTA table, this is needed because
** the real filter table gets cleared during
** a soft reset and we need to repopulate it.
*/
static u32 ixgbe_shadow_vfta[IXGBE_VFTA_SIZE];
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/*
** The number of scatter-gather segments
** differs for 82598 and 82599, default to
** the former.
*/
static int ixgbe_num_segs = IXGBE_82598_SCATTER;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
/*
** For Flow Director: this is the
** number of TX packets we sample
** for the filter pool, this means
** every 20th packet will be probed.
**
** This feature can be disabled by
** setting this to 0.
*/
static int atr_sample_rate = 20;
/*
** Flow Director actually 'steals'
** part of the packet buffer as its
** filter pool, this variable controls
** how much it uses:
** 0 = 64K, 1 = 128K, 2 = 256K
*/
static int fdir_pballoc = 1;
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
* Device identification routine
*
* ixgbe_probe determines if the driver should be loaded on
* adapter based on PCI vendor/device id of the adapter.
*
* return 0 on success, positive on failure
*********************************************************************/
static int
ixgbe_probe(device_t dev)
{
ixgbe_vendor_info_t *ent;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
u16 pci_vendor_id = 0;
u16 pci_device_id = 0;
u16 pci_subvendor_id = 0;
u16 pci_subdevice_id = 0;
char adapter_name[256];
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
INIT_DEBUGOUT("ixgbe_probe: begin");
pci_vendor_id = pci_get_vendor(dev);
if (pci_vendor_id != IXGBE_INTEL_VENDOR_ID)
return (ENXIO);
pci_device_id = pci_get_device(dev);
pci_subvendor_id = pci_get_subvendor(dev);
pci_subdevice_id = pci_get_subdevice(dev);
ent = ixgbe_vendor_info_array;
while (ent->vendor_id != 0) {
if ((pci_vendor_id == ent->vendor_id) &&
(pci_device_id == ent->device_id) &&
((pci_subvendor_id == ent->subvendor_id) ||
(ent->subvendor_id == 0)) &&
((pci_subdevice_id == ent->subdevice_id) ||
(ent->subdevice_id == 0))) {
sprintf(adapter_name, "%s, Version - %s",
ixgbe_strings[ent->index],
ixgbe_driver_version);
device_set_desc_copy(dev, adapter_name);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
++ixgbe_total_ports;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (0);
}
ent++;
}
return (ENXIO);
}
/*********************************************************************
* Device initialization routine
*
* The attach entry point is called when the driver is being loaded.
* This routine identifies the type of hardware, allocates all resources
* and initializes the hardware.
*
* return 0 on success, positive on failure
*********************************************************************/
static int
ixgbe_attach(device_t dev)
{
struct adapter *adapter;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
struct ixgbe_hw *hw;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
int error = 0;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
u16 pci_device_id, csum;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
u32 ctrl_ext;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
INIT_DEBUGOUT("ixgbe_attach: begin");
/* Allocate, clear, and link in our adapter structure */
adapter = device_get_softc(dev);
adapter->dev = adapter->osdep.dev = dev;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
hw = &adapter->hw;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Core Lock Init*/
IXGBE_CORE_LOCK_INIT(adapter, device_get_nameunit(dev));
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/* Keep track of optics */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
pci_device_id = pci_get_device(dev);
switch (pci_device_id) {
case IXGBE_DEV_ID_82598_CX4_DUAL_PORT :
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
case IXGBE_DEV_ID_82598EB_CX4 :
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
adapter->optics = IFM_10G_CX4;
break;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
case IXGBE_DEV_ID_82598 :
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
case IXGBE_DEV_ID_82598AF_DUAL_PORT :
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
case IXGBE_DEV_ID_82598_DA_DUAL_PORT :
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
case IXGBE_DEV_ID_82598AF_SINGLE_PORT :
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
case IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM :
case IXGBE_DEV_ID_82598EB_SFP_LOM :
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
case IXGBE_DEV_ID_82598AT :
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
adapter->optics = IFM_10G_SR;
break;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
case IXGBE_DEV_ID_82598AT2 :
adapter->optics = IFM_10G_T;
break;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
case IXGBE_DEV_ID_82598EB_XF_LR :
adapter->optics = IFM_10G_LR;
break;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
case IXGBE_DEV_ID_82599_SFP :
adapter->optics = IFM_10G_SR;
ixgbe_num_segs = IXGBE_82599_SCATTER;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
break;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
case IXGBE_DEV_ID_82599_KX4 :
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
case IXGBE_DEV_ID_82599_KX4_MEZZ:
case IXGBE_DEV_ID_82599_CX4 :
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
adapter->optics = IFM_10G_CX4;
ixgbe_num_segs = IXGBE_82599_SCATTER;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
break;
case IXGBE_DEV_ID_82599_XAUI_LOM :
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
case IXGBE_DEV_ID_82599_COMBO_BACKPLANE :
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
ixgbe_num_segs = IXGBE_82599_SCATTER;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
break;
case IXGBE_DEV_ID_82599_T3_LOM:
ixgbe_num_segs = IXGBE_82599_SCATTER;
adapter->optics = IFM_10G_T;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
default:
break;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* SYSCTL APIs */
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
OID_AUTO, "stats", CTLTYPE_INT | CTLFLAG_RW,
adapter, 0, ixgbe_sysctl_stats, "I", "Statistics");
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
OID_AUTO, "debug", CTLTYPE_INT | CTLFLAG_RW,
adapter, 0, ixgbe_sysctl_debug, "I", "Debug Info");
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
OID_AUTO, "flow_control", CTLTYPE_INT | CTLFLAG_RW,
adapter, 0, ixgbe_set_flowcntl, "I", "Flow Control");
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
OID_AUTO, "enable_aim", CTLTYPE_INT|CTLFLAG_RW,
&ixgbe_enable_aim, 1, "Interrupt Moderation");
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Set up the timer callout */
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
callout_init_mtx(&adapter->timer, &adapter->core_mtx, 0);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Determine hardware revision */
ixgbe_identify_hardware(adapter);
/* Do base PCI setup - map BAR0 */
if (ixgbe_allocate_pci_resources(adapter)) {
device_printf(dev, "Allocation of PCI resources failed\n");
error = ENXIO;
goto err_out;
}
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
/* Do descriptor calc and sanity checks */
if (((ixgbe_txd * sizeof(union ixgbe_adv_tx_desc)) % DBA_ALIGN) != 0 ||
ixgbe_txd < MIN_TXD || ixgbe_txd > MAX_TXD) {
device_printf(dev, "TXD config issue, using default!\n");
adapter->num_tx_desc = DEFAULT_TXD;
} else
adapter->num_tx_desc = ixgbe_txd;
/*
** With many RX rings it is easy to exceed the
** system mbuf allocation. Tuning nmbclusters
** can alleviate this.
*/
Multiqueue RX is not correctly enabled on the new 82599 adapter, the SRRCTL register needs to be setup per queue. Approved by: re
2009-07-01 16:13:01 +00:00
if (nmbclusters > 0 ) {
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
int s;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
s = (ixgbe_rxd * adapter->num_queues) * ixgbe_total_ports;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
if (s > nmbclusters) {
device_printf(dev, "RX Descriptors exceed "
"system mbuf max, using default instead!\n");
ixgbe_rxd = DEFAULT_RXD;
}
}
if (((ixgbe_rxd * sizeof(union ixgbe_adv_rx_desc)) % DBA_ALIGN) != 0 ||
ixgbe_rxd < MIN_TXD || ixgbe_rxd > MAX_TXD) {
device_printf(dev, "RXD config issue, using default!\n");
adapter->num_rx_desc = DEFAULT_RXD;
} else
adapter->num_rx_desc = ixgbe_rxd;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Allocate our TX/RX Queues */
if (ixgbe_allocate_queues(adapter)) {
error = ENOMEM;
goto err_out;
}
/* Initialize the shared code */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
error = ixgbe_init_shared_code(hw);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
if (error == IXGBE_ERR_SFP_NOT_PRESENT) {
/*
** No optics in this port, set up
** so the timer routine will probe
** for later insertion.
*/
adapter->sfp_probe = TRUE;
error = 0;
} else if (error == IXGBE_ERR_SFP_NOT_SUPPORTED) {
device_printf(dev,"Unsupported SFP+ module detected!\n");
error = EIO;
goto err_late;
} else if (error) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
device_printf(dev,"Unable to initialize the shared code\n");
error = EIO;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
goto err_late;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Make sure we have a good EEPROM before we read from it */
if (ixgbe_validate_eeprom_checksum(&adapter->hw, &csum) < 0) {
device_printf(dev,"The EEPROM Checksum Is Not Valid\n");
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
error = EIO;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
goto err_late;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Pick up the smart speed setting */
if (hw->mac.type == ixgbe_mac_82599EB)
hw->phy.smart_speed = ixgbe_smart_speed;
/* Get Hardware Flow Control setting */
hw->fc.requested_mode = ixgbe_fc_full;
hw->fc.pause_time = IXGBE_FC_PAUSE;
hw->fc.low_water = IXGBE_FC_LO;
hw->fc.high_water = IXGBE_FC_HI;
hw->fc.send_xon = TRUE;
error = ixgbe_init_hw(hw);
if (error == IXGBE_ERR_EEPROM_VERSION) {
device_printf(dev, "This device is a pre-production adapter/"
"LOM. Please be aware there may be issues associated "
"with your hardware.\n If you are experiencing problems "
"please contact your Intel or hardware representative "
"who provided you with this hardware.\n");
} else if (error == IXGBE_ERR_SFP_NOT_SUPPORTED)
device_printf(dev,"Unsupported SFP+ Module\n");
if (error) {
error = EIO;
device_printf(dev,"Hardware Initialization Failure\n");
goto err_late;
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if ((adapter->msix > 1) && (ixgbe_enable_msix))
error = ixgbe_allocate_msix(adapter);
else
error = ixgbe_allocate_legacy(adapter);
if (error)
goto err_late;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Setup OS specific network interface */
ixgbe_setup_interface(dev, adapter);
/* Sysctl for limiting the amount of work done in the taskqueue */
ixgbe_add_rx_process_limit(adapter, "rx_processing_limit",
"max number of rx packets to process", &adapter->rx_process_limit,
ixgbe_rx_process_limit);
/* Initialize statistics */
ixgbe_update_stats_counters(adapter);
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +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
ixgbe_register_vlan, adapter, EVENTHANDLER_PRI_FIRST);
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +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
ixgbe_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Print PCIE bus type/speed/width info */
ixgbe_get_bus_info(hw);
device_printf(dev,"PCI Express Bus: Speed %s %s\n",
((hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0Gb/s":
(hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5Gb/s":"Unknown"),
(hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" :
(hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" :
(hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" :
("Unknown"));
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if ((hw->bus.width <= ixgbe_bus_width_pcie_x4) &&
(hw->bus.speed == ixgbe_bus_speed_2500)) {
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
device_printf(dev, "PCI-Express bandwidth available"
" for this card\n is not sufficient for"
" optimal performance.\n");
device_printf(dev, "For optimal performance a x8 "
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
"PCIE, or x4 PCIE 2 slot is required.\n");
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* let hardware know driver is loaded */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ctrl_ext |= IXGBE_CTRL_EXT_DRV_LOAD;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
INIT_DEBUGOUT("ixgbe_attach: end");
return (0);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
err_late:
ixgbe_free_transmit_structures(adapter);
ixgbe_free_receive_structures(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
err_out:
ixgbe_free_pci_resources(adapter);
return (error);
}
/*********************************************************************
* Device removal routine
*
* The detach entry point is called when the driver is being removed.
* This routine stops the adapter and deallocates all the resources
* that were allocated for driver operation.
*
* return 0 on success, positive on failure
*********************************************************************/
static int
ixgbe_detach(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ix_queue *que = adapter->queues;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
u32 ctrl_ext;
INIT_DEBUGOUT("ixgbe_detach: begin");
/* Make sure VLANS are not using driver */
if (adapter->ifp->if_vlantrunk != NULL) {
device_printf(dev,"Vlan in use, detach first\n");
return (EBUSY);
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_LOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_stop(adapter);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_UNLOCK(adapter);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
for (int i = 0; i < adapter->num_queues; i++, que++) {
if (que->tq) {
taskqueue_drain(que->tq, &que->que_task);
taskqueue_free(que->tq);
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
}
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/* Drain the Link queue */
if (adapter->tq) {
taskqueue_drain(adapter->tq, &adapter->link_task);
taskqueue_drain(adapter->tq, &adapter->mod_task);
taskqueue_drain(adapter->tq, &adapter->msf_task);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
taskqueue_drain(adapter->tq, &adapter->fdir_task);
#endif
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
taskqueue_free(adapter->tq);
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/* let hardware know driver is unloading */
ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
ctrl_ext &= ~IXGBE_CTRL_EXT_DRV_LOAD;
IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT, ctrl_ext);
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +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);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ether_ifdetach(adapter->ifp);
callout_drain(&adapter->timer);
ixgbe_free_pci_resources(adapter);
bus_generic_detach(dev);
if_free(adapter->ifp);
ixgbe_free_transmit_structures(adapter);
ixgbe_free_receive_structures(adapter);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_LOCK_DESTROY(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (0);
}
/*********************************************************************
*
* Shutdown entry point
*
**********************************************************************/
static int
ixgbe_shutdown(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_LOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_stop(adapter);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_UNLOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (0);
}
/*********************************************************************
* Transmit entry point
*
* ixgbe_start is called by the stack to initiate a transmit.
* The driver will remain in this routine as long as there are
* packets to transmit and transmit resources are available.
* In case resources are not available stack is notified and
* the packet is requeued.
**********************************************************************/
static void
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
ixgbe_start_locked(struct tx_ring *txr, struct ifnet * ifp)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
struct mbuf *m_head;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct adapter *adapter = txr->adapter;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_TX_LOCK_ASSERT(txr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) !=
IFF_DRV_RUNNING)
return;
if (!adapter->link_active)
return;
while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
if (m_head == NULL)
break;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if (ixgbe_xmit(txr, &m_head)) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (m_head == NULL)
break;
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
break;
}
/* Send a copy of the frame to the BPF listener */
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
ETHER_BPF_MTAP(ifp, m_head);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Set watchdog on */
txr->watchdog_check = TRUE;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
return;
}
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/*
* Legacy TX start - called by the stack, this
* always uses the first tx ring, and should
* not be used with multiqueue tx enabled.
*/
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void
ixgbe_start(struct ifnet *ifp)
{
struct adapter *adapter = ifp->if_softc;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct tx_ring *txr = adapter->tx_rings;
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
IXGBE_TX_LOCK(txr);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
ixgbe_start_locked(txr, ifp);
IXGBE_TX_UNLOCK(txr);
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
#if __FreeBSD_version >= 800000
/*
** Multiqueue Transmit driver
**
*/
static int
ixgbe_mq_start(struct ifnet *ifp, struct mbuf *m)
{
struct adapter *adapter = ifp->if_softc;
struct tx_ring *txr;
int i = 0, err = 0;
/* Which queue to use */
if ((m->m_flags & M_FLOWID) != 0)
i = m->m_pkthdr.flowid % adapter->num_queues;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
else /* use the cpu we're on */
i = curcpu % adapter->num_queues;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
txr = &adapter->tx_rings[i];
if (IXGBE_TX_TRYLOCK(txr)) {
err = ixgbe_mq_start_locked(ifp, txr, m);
IXGBE_TX_UNLOCK(txr);
} else
err = drbr_enqueue(ifp, txr->br, m);
return (err);
}
static int
ixgbe_mq_start_locked(struct ifnet *ifp, struct tx_ring *txr, struct mbuf *m)
{
struct adapter *adapter = txr->adapter;
struct mbuf *next;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
int enqueued, err = 0;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
IFF_DRV_RUNNING || adapter->link_active == 0) {
if (m != NULL)
err = drbr_enqueue(ifp, txr->br, m);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
return (err);
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
enqueued = 0;
if (m == NULL)
next = drbr_dequeue(ifp, txr->br);
else
next = m;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/* Process the queue */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
while (next != NULL) {
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if ((err = ixgbe_xmit(txr, &next)) != 0) {
if (next != NULL)
err = drbr_enqueue(ifp, txr->br, next);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
break;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
enqueued++;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
drbr_stats_update(ifp, next->m_pkthdr.len, next->m_flags);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Send a copy of the frame to the BPF listener */
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
ETHER_BPF_MTAP(ifp, next);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
break;
if (txr->tx_avail <= IXGBE_TX_OP_THRESHOLD) {
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
break;
}
next = drbr_dequeue(ifp, txr->br);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (enqueued > 0)
txr->watchdog_check = TRUE;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
return (err);
}
/*
** Flush all ring buffers
*/
static void
ixgbe_qflush(struct ifnet *ifp)
{
struct adapter *adapter = ifp->if_softc;
struct tx_ring *txr = adapter->tx_rings;
struct mbuf *m;
for (int i = 0; i < adapter->num_queues; i++, txr++) {
IXGBE_TX_LOCK(txr);
while ((m = buf_ring_dequeue_sc(txr->br)) != NULL)
m_freem(m);
IXGBE_TX_UNLOCK(txr);
}
if_qflush(ifp);
}
#endif /* __FreeBSD_version >= 800000 */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
* Ioctl entry point
*
* ixgbe_ioctl is called when the user wants to configure the
* interface.
*
* return 0 on success, positive on failure
**********************************************************************/
static int
ixgbe_ioctl(struct ifnet * ifp, u_long command, caddr_t data)
{
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct adapter *adapter = ifp->if_softc;
struct ifreq *ifr = (struct ifreq *) data;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
int error = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
switch (command) {
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
case SIOCSIFMTU:
IOCTL_DEBUGOUT("ioctl: SIOCSIFMTU (Set Interface MTU)");
if (ifr->ifr_mtu > IXGBE_MAX_FRAME_SIZE - ETHER_HDR_LEN) {
error = EINVAL;
} else {
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_LOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ifp->if_mtu = ifr->ifr_mtu;
adapter->max_frame_size =
ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
ixgbe_init_locked(adapter);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_UNLOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
break;
case SIOCSIFFLAGS:
IOCTL_DEBUGOUT("ioctl: SIOCSIFFLAGS (Set Interface Flags)");
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_LOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (ifp->if_flags & IFF_UP) {
A couple late breaking bugs that testing have turned up. - change include style so build in kernel try OR standalone work. - Limit HWCSUM - I was led to believe that it would work with RSS, but our testing had odd issues which suggests this is false. - A fatfinger error in the ioctl code made ifconfig up not work. Approved by: re
2007-07-12 19:04:11 +00:00
if ((ifp->if_drv_flags & IFF_DRV_RUNNING)) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if ((ifp->if_flags ^ adapter->if_flags) &
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
(IFF_PROMISC | IFF_ALLMULTI)) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_disable_promisc(adapter);
ixgbe_set_promisc(adapter);
}
} else
ixgbe_init_locked(adapter);
} else
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
ixgbe_stop(adapter);
adapter->if_flags = ifp->if_flags;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_UNLOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
IOCTL_DEBUGOUT("ioctl: SIOC(ADD|DEL)MULTI");
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_LOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_disable_intr(adapter);
ixgbe_set_multi(adapter);
ixgbe_enable_intr(adapter);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_UNLOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
break;
case SIOCSIFMEDIA:
case SIOCGIFMEDIA:
IOCTL_DEBUGOUT("ioctl: SIOCxIFMEDIA (Get/Set Interface Media)");
error = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
break;
case SIOCSIFCAP:
{
int mask = ifr->ifr_reqcap ^ ifp->if_capenable;
IOCTL_DEBUGOUT("ioctl: SIOCSIFCAP (Set Capabilities)");
if (mask & IFCAP_HWCSUM)
ifp->if_capenable ^= IFCAP_HWCSUM;
if (mask & IFCAP_TSO4)
ifp->if_capenable ^= IFCAP_TSO4;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (mask & IFCAP_LRO)
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
ifp->if_capenable ^= IFCAP_LRO;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (mask & IFCAP_VLAN_HWTAGGING)
ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
ixgbe_init(adapter);
VLAN_CAPABILITIES(ifp);
break;
}
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
default:
IOCTL_DEBUGOUT1("ioctl: UNKNOWN (0x%X)\n", (int)command);
error = ether_ioctl(ifp, command, data);
break;
}
return (error);
}
/*********************************************************************
* Init entry point
*
* This routine is used in two ways. It is used by the stack as
* init entry point in network interface structure. It is also used
* by the driver as a hw/sw initialization routine to get to a
* consistent state.
*
* return 0 on success, positive on failure
**********************************************************************/
#define IXGBE_MHADD_MFS_SHIFT 16
static void
ixgbe_init_locked(struct adapter *adapter)
{
struct ifnet *ifp = adapter->ifp;
device_t dev = adapter->dev;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ixgbe_hw *hw = &adapter->hw;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
u32 k, txdctl, mhadd, gpie;
u32 rxdctl, rxctrl;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
int err;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
mtx_assert(&adapter->core_mtx, MA_OWNED);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
INIT_DEBUGOUT("ixgbe_init: begin");
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_reset_hw(hw);
hw->adapter_stopped = FALSE;
ixgbe_stop_adapter(hw);
callout_stop(&adapter->timer);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* reprogram the RAR[0] in case user changed it. */
ixgbe_set_rar(hw, 0, adapter->hw.mac.addr, 0, IXGBE_RAH_AV);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Get the latest mac address, User can use a LAA */
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
bcopy(IF_LLADDR(adapter->ifp), hw->mac.addr,
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
IXGBE_ETH_LENGTH_OF_ADDRESS);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
ixgbe_set_rar(hw, 0, hw->mac.addr, 0, 1);
hw->addr_ctrl.rar_used_count = 1;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Prepare transmit descriptors and buffers */
if (ixgbe_setup_transmit_structures(adapter)) {
device_printf(dev,"Could not setup transmit structures\n");
ixgbe_stop(adapter);
return;
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_init_hw(hw);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_initialize_transmit_units(adapter);
/* Setup Multicast table */
ixgbe_set_multi(adapter);
/*
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
** Determine the correct mbuf pool
** for doing jumbo/headersplit
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
*/
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
if (ifp->if_mtu > ETHERMTU)
adapter->rx_mbuf_sz = MJUMPAGESIZE;
else
adapter->rx_mbuf_sz = MCLBYTES;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Prepare receive descriptors and buffers */
if (ixgbe_setup_receive_structures(adapter)) {
device_printf(dev,"Could not setup receive structures\n");
ixgbe_stop(adapter);
return;
}
/* Configure RX settings */
ixgbe_initialize_receive_units(adapter);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
gpie = IXGBE_READ_REG(&adapter->hw, IXGBE_GPIE);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (hw->mac.type == ixgbe_mac_82599EB) {
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
gpie |= IXGBE_SDP1_GPIEN;
gpie |= IXGBE_SDP2_GPIEN;
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Enable Fan Failure Interrupt */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
if (hw->device_id == IXGBE_DEV_ID_82598AT)
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
gpie |= IXGBE_SDP1_GPIEN;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if (adapter->msix > 1) {
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Enable Enhanced MSIX mode */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
gpie |= IXGBE_GPIE_MSIX_MODE;
gpie |= IXGBE_GPIE_EIAME | IXGBE_GPIE_PBA_SUPPORT |
IXGBE_GPIE_OCD;
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Set the various hardware offload abilities */
ifp->if_hwassist = 0;
if (ifp->if_capenable & IFCAP_TSO4)
ifp->if_hwassist |= CSUM_TSO;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if (ifp->if_capenable & IFCAP_TXCSUM)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ifp->if_hwassist = (CSUM_TCP | CSUM_UDP);
/* Set MTU size */
if (ifp->if_mtu > ETHERMTU) {
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
mhadd &= ~IXGBE_MHADD_MFS_MASK;
mhadd |= adapter->max_frame_size << IXGBE_MHADD_MFS_SHIFT;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
/* Now enable all the queues */
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (int i = 0; i < adapter->num_queues; i++) {
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
txdctl |= IXGBE_TXDCTL_ENABLE;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Set WTHRESH to 8, burst writeback */
txdctl |= (8 << 16);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(i), txdctl);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (int i = 0; i < adapter->num_queues; i++) {
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
if (hw->mac.type == ixgbe_mac_82598EB) {
/*
** PTHRESH = 21
** HTHRESH = 4
** WTHRESH = 8
*/
rxdctl &= ~0x3FFFFF;
rxdctl |= 0x080420;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
rxdctl |= IXGBE_RXDCTL_ENABLE;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(i), rxdctl);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
for (k = 0; k < 10; k++) {
if (IXGBE_READ_REG(hw, IXGBE_RXDCTL(i)) &
IXGBE_RXDCTL_ENABLE)
break;
else
msec_delay(1);
}
wmb();
IXGBE_WRITE_REG(hw, IXGBE_RDT(i), adapter->num_rx_desc - 1);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/* Set up VLAN offloads and filter */
ixgbe_setup_vlan_hw_support(adapter);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/* Enable Receive engine */
rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (hw->mac.type == ixgbe_mac_82598EB)
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rxctrl |= IXGBE_RXCTRL_DMBYPS;
rxctrl |= IXGBE_RXCTRL_RXEN;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_enable_rx_dma(hw, rxctrl);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
callout_reset(&adapter->timer, hz, ixgbe_local_timer, adapter);
/* Set up MSI/X routing */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
if (ixgbe_enable_msix)
ixgbe_configure_ivars(adapter);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
else { /* Simple settings for Legacy/MSI */
ixgbe_set_ivar(adapter, 0, 0, 0);
ixgbe_set_ivar(adapter, 0, 0, 1);
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
/* Init Flow director */
if (adapter->hw.mac.type == ixgbe_mac_82599EB)
ixgbe_init_fdir_signature_82599(&adapter->hw, fdir_pballoc);
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/*
** Check on any SFP devices that
** need to be kick-started
*/
err = hw->phy.ops.identify(hw);
if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
device_printf(dev,
"Unsupported SFP+ module type was detected.\n");
ixgbe_detach(dev);
return;
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Config/Enable Link */
ixgbe_config_link(adapter);
/* And now turn on interrupts */
ixgbe_enable_intr(adapter);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Now inform the stack we're ready */
ifp->if_drv_flags |= IFF_DRV_RUNNING;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
return;
}
static void
ixgbe_init(void *arg)
{
struct adapter *adapter = arg;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_LOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_init_locked(adapter);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_UNLOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/*
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
**
** MSIX Interrupt Handlers and Tasklets
**
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
*/
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
static inline void
ixgbe_enable_queue(struct adapter *adapter, u32 vector)
{
struct ixgbe_hw *hw = &adapter->hw;
u64 queue = (u64)(1 << vector);
u32 mask;
if (hw->mac.type == ixgbe_mac_82598EB) {
mask = (IXGBE_EIMS_RTX_QUEUE & queue);
IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
} else {
mask = (queue & 0xFFFFFFFF);
if (mask)
IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
mask = (queue >> 32);
if (mask)
IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
}
}
static inline void
ixgbe_disable_queue(struct adapter *adapter, u32 vector)
{
struct ixgbe_hw *hw = &adapter->hw;
u64 queue = (u64)(1 << vector);
u32 mask;
if (hw->mac.type == ixgbe_mac_82598EB) {
mask = (IXGBE_EIMS_RTX_QUEUE & queue);
IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
} else {
mask = (queue & 0xFFFFFFFF);
if (mask)
IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
mask = (queue >> 32);
if (mask)
IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
}
}
static inline void
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
ixgbe_rearm_queues(struct adapter *adapter, u64 queues)
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
{
u32 mask;
if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
mask = (IXGBE_EIMS_RTX_QUEUE & queues);
IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
} else {
mask = (queues & 0xFFFFFFFF);
IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
mask = (queues >> 32);
IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
}
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
static void
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_handle_que(void *context, int pending)
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
{
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ix_queue *que = context;
struct adapter *adapter = que->adapter;
struct tx_ring *txr = que->txr;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct ifnet *ifp = adapter->ifp;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
u32 loop = MAX_LOOP;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
bool more_rx, more_tx;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_TX_LOCK(txr);
do {
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
more_rx = ixgbe_rxeof(que, adapter->rx_process_limit);
more_tx = ixgbe_txeof(txr);
} while (loop-- && (more_rx || more_tx));
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
#if __FreeBSD_version >= 800000
if (!drbr_empty(ifp, txr->br))
ixgbe_mq_start_locked(ifp, txr, NULL);
#else
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
ixgbe_start_locked(txr, ifp);
#endif
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_TX_UNLOCK(txr);
/* Reenable this interrupt */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_enable_queue(adapter, que->msix);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
*
* Legacy Interrupt Service routine
*
**********************************************************************/
static void
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
ixgbe_legacy_irq(void *arg)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ix_queue *que = arg;
struct adapter *adapter = que->adapter;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
struct ixgbe_hw *hw = &adapter->hw;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct tx_ring *txr = adapter->tx_rings;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
bool more_tx, more_rx;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
u32 reg_eicr, loop = MAX_LOOP;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
reg_eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
++que->irqs;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
if (reg_eicr == 0) {
ixgbe_enable_intr(adapter);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
return;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
more_rx = ixgbe_rxeof(que, adapter->rx_process_limit);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
IXGBE_TX_LOCK(txr);
do {
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
more_tx = ixgbe_txeof(txr);
} while (loop-- && more_tx);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
IXGBE_TX_UNLOCK(txr);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (more_rx || more_tx)
taskqueue_enqueue(que->tq, &que->que_task);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Check for fan failure */
if ((hw->phy.media_type == ixgbe_media_type_copper) &&
(reg_eicr & IXGBE_EICR_GPI_SDP1)) {
device_printf(adapter->dev, "\nCRITICAL: FAN FAILURE!! "
"REPLACE IMMEDIATELY!!\n");
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EICR_GPI_SDP1);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
/* Link status change */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (reg_eicr & IXGBE_EICR_LSC)
taskqueue_enqueue(adapter->tq, &adapter->link_task);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
ixgbe_enable_intr(adapter);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
return;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
/*********************************************************************
*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
* MSI Queue Interrupt Service routine
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
*
**********************************************************************/
void
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_msix_que(void *arg)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ix_queue *que = arg;
struct adapter *adapter = que->adapter;
struct tx_ring *txr = que->txr;
struct rx_ring *rxr = que->rxr;
bool more_tx, more_rx;
u32 newitr = 0;
ixgbe_disable_queue(adapter, que->msix);
++que->irqs;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
more_rx = ixgbe_rxeof(que, adapter->rx_process_limit);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
IXGBE_TX_LOCK(txr);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
more_tx = ixgbe_txeof(txr);
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
IXGBE_TX_UNLOCK(txr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
more_rx = ixgbe_rxeof(que, adapter->rx_process_limit);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Do AIM now? */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (ixgbe_enable_aim == FALSE)
goto no_calc;
/*
** Do Adaptive Interrupt Moderation:
** - Write out last calculated setting
** - Calculate based on average size over
** the last interval.
*/
if (que->eitr_setting)
IXGBE_WRITE_REG(&adapter->hw,
IXGBE_EITR(que->msix), que->eitr_setting);
que->eitr_setting = 0;
/* Idle, do nothing */
if ((txr->bytes == 0) && (rxr->bytes == 0))
goto no_calc;
if ((txr->bytes) && (txr->packets))
newitr = txr->bytes/txr->packets;
if ((rxr->bytes) && (rxr->packets))
newitr = max(newitr,
(rxr->bytes / rxr->packets));
newitr += 24; /* account for hardware frame, crc */
/* set an upper boundary */
newitr = min(newitr, 3000);
/* Be nice to the mid range */
if ((newitr > 300) && (newitr < 1200))
newitr = (newitr / 3);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
else
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
newitr = (newitr / 2);
if (adapter->hw.mac.type == ixgbe_mac_82598EB)
newitr |= newitr << 16;
else
newitr |= IXGBE_EITR_CNT_WDIS;
/* save for next interrupt */
que->eitr_setting = newitr;
/* Reset state */
txr->bytes = 0;
txr->packets = 0;
rxr->bytes = 0;
rxr->packets = 0;
no_calc:
if (more_tx || more_rx)
taskqueue_enqueue(que->tq, &que->que_task);
else /* Reenable this interrupt */
ixgbe_enable_queue(adapter, que->msix);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
return;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
static void
ixgbe_msix_link(void *arg)
{
struct adapter *adapter = arg;
struct ixgbe_hw *hw = &adapter->hw;
u32 reg_eicr;
++adapter->link_irq;
/* First get the cause */
reg_eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/* Clear interrupt with write */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
IXGBE_WRITE_REG(hw, IXGBE_EICR, reg_eicr);
/* Link status change */
if (reg_eicr & IXGBE_EICR_LSC)
taskqueue_enqueue(adapter->tq, &adapter->link_task);
if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
if (reg_eicr & IXGBE_EICR_FLOW_DIR) {
/* This is probably overkill :) */
if (!atomic_cmpset_int(&adapter->fdir_reinit, 0, 1))
return;
/* Clear the interrupt */
IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_FLOW_DIR);
/* Turn off the interface */
adapter->ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
taskqueue_enqueue(adapter->tq, &adapter->fdir_task);
} else
#endif
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
if (reg_eicr & IXGBE_EICR_ECC) {
device_printf(adapter->dev, "\nCRITICAL: ECC ERROR!! "
"Please Reboot!!\n");
IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
} else
if (reg_eicr & IXGBE_EICR_GPI_SDP1) {
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/* Clear the interrupt */
IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
taskqueue_enqueue(adapter->tq, &adapter->msf_task);
} else if (reg_eicr & IXGBE_EICR_GPI_SDP2) {
/* Clear the interrupt */
IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2);
taskqueue_enqueue(adapter->tq, &adapter->mod_task);
}
}
/* Check for fan failure */
if ((hw->device_id == IXGBE_DEV_ID_82598AT) &&
(reg_eicr & IXGBE_EICR_GPI_SDP1)) {
device_printf(adapter->dev, "\nCRITICAL: FAN FAILURE!! "
"REPLACE IMMEDIATELY!!\n");
IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
}
IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
return;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
*
* Media Ioctl callback
*
* This routine is called whenever the user queries the status of
* the interface using ifconfig.
*
**********************************************************************/
static void
ixgbe_media_status(struct ifnet * ifp, struct ifmediareq * ifmr)
{
struct adapter *adapter = ifp->if_softc;
INIT_DEBUGOUT("ixgbe_media_status: begin");
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_CORE_LOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_update_link_status(adapter);
ifmr->ifm_status = IFM_AVALID;
ifmr->ifm_active = IFM_ETHER;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if (!adapter->link_active) {
IXGBE_CORE_UNLOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ifmr->ifm_status |= IFM_ACTIVE;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
switch (adapter->link_speed) {
case IXGBE_LINK_SPEED_1GB_FULL:
ifmr->ifm_active |= IFM_1000_T | IFM_FDX;
break;
case IXGBE_LINK_SPEED_10GB_FULL:
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
ifmr->ifm_active |= adapter->optics | IFM_FDX;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
break;
}
IXGBE_CORE_UNLOCK(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
/*********************************************************************
*
* Media Ioctl callback
*
* This routine is called when the user changes speed/duplex using
* media/mediopt option with ifconfig.
*
**********************************************************************/
static int
ixgbe_media_change(struct ifnet * ifp)
{
struct adapter *adapter = ifp->if_softc;
struct ifmedia *ifm = &adapter->media;
INIT_DEBUGOUT("ixgbe_media_change: begin");
if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
return (EINVAL);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
switch (IFM_SUBTYPE(ifm->ifm_media)) {
case IFM_AUTO:
adapter->hw.phy.autoneg_advertised =
IXGBE_LINK_SPEED_1GB_FULL | IXGBE_LINK_SPEED_10GB_FULL;
break;
default:
device_printf(adapter->dev, "Only auto media type\n");
return (EINVAL);
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (0);
}
/*********************************************************************
*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
* This routine maps the mbufs to tx descriptors, allowing the
* TX engine to transmit the packets.
* - return 0 on success, positive on failure
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
*
**********************************************************************/
static int
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
ixgbe_xmit(struct tx_ring *txr, struct mbuf **m_headp)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct adapter *adapter = txr->adapter;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
u32 olinfo_status = 0, cmd_type_len;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
u32 paylen = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
int i, j, error, nsegs;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
int first, last = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
struct mbuf *m_head;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
bus_dma_segment_t segs[ixgbe_num_segs];
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bus_dmamap_t map;
struct ixgbe_tx_buf *txbuf, *txbuf_mapped;
union ixgbe_adv_tx_desc *txd = NULL;
m_head = *m_headp;
/* Basic descriptor defines */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
cmd_type_len = (IXGBE_ADVTXD_DTYP_DATA |
IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (m_head->m_flags & M_VLANTAG)
cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
/*
* Important to capture the first descriptor
* used because it will contain the index of
* the one we tell the hardware to report back
*/
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
first = txr->next_avail_desc;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
txbuf = &txr->tx_buffers[first];
txbuf_mapped = txbuf;
map = txbuf->map;
/*
* Map the packet for DMA.
*/
error = bus_dmamap_load_mbuf_sg(txr->txtag, map,
*m_headp, segs, &nsegs, BUS_DMA_NOWAIT);
if (error == EFBIG) {
struct mbuf *m;
m = m_defrag(*m_headp, M_DONTWAIT);
if (m == NULL) {
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
adapter->mbuf_defrag_failed++;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
m_freem(*m_headp);
*m_headp = NULL;
return (ENOBUFS);
}
*m_headp = m;
/* Try it again */
error = bus_dmamap_load_mbuf_sg(txr->txtag, map,
*m_headp, segs, &nsegs, BUS_DMA_NOWAIT);
if (error == ENOMEM) {
adapter->no_tx_dma_setup++;
return (error);
} else if (error != 0) {
adapter->no_tx_dma_setup++;
m_freem(*m_headp);
*m_headp = NULL;
return (error);
}
} else if (error == ENOMEM) {
adapter->no_tx_dma_setup++;
return (error);
} else if (error != 0) {
adapter->no_tx_dma_setup++;
m_freem(*m_headp);
*m_headp = NULL;
return (error);
}
/* Make certain there are enough descriptors */
if (nsegs > txr->tx_avail - 2) {
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
txr->no_desc_avail++;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
error = ENOBUFS;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
goto xmit_fail;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
m_head = *m_headp;
/*
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
** Set up the appropriate offload context
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
** this becomes the first descriptor of
** a packet.
*/
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
if (ixgbe_tso_setup(txr, m_head, &paylen)) {
cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
olinfo_status |= IXGBE_TXD_POPTS_IXSM << 8;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
olinfo_status |= IXGBE_TXD_POPTS_TXSM << 8;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
olinfo_status |= paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
++adapter->tso_tx;
} else
return (ENXIO);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
} else if (ixgbe_tx_ctx_setup(txr, m_head))
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
olinfo_status |= IXGBE_TXD_POPTS_TXSM << 8;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
#ifdef IXGBE_IEEE1588
/* This is changing soon to an mtag detection */
if (we detect this mbuf has a TSTAMP mtag)
cmd_type_len |= IXGBE_ADVTXD_MAC_TSTAMP;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
#endif
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
/* Do the flow director magic */
if ((txr->atr_sample) && (!adapter->fdir_reinit)) {
++txr->atr_count;
if (txr->atr_count >= atr_sample_rate) {
ixgbe_atr(txr, m_head);
txr->atr_count = 0;
}
}
#endif
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/* Record payload length */
if (paylen == 0)
olinfo_status |= m_head->m_pkthdr.len <<
IXGBE_ADVTXD_PAYLEN_SHIFT;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
i = txr->next_avail_desc;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
for (j = 0; j < nsegs; j++) {
bus_size_t seglen;
bus_addr_t segaddr;
txbuf = &txr->tx_buffers[i];
txd = &txr->tx_base[i];
seglen = segs[j].ds_len;
segaddr = htole64(segs[j].ds_addr);
txd->read.buffer_addr = segaddr;
txd->read.cmd_type_len = htole32(txr->txd_cmd |
cmd_type_len |seglen);
txd->read.olinfo_status = htole32(olinfo_status);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
last = i; /* descriptor that will get completion IRQ */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (++i == adapter->num_tx_desc)
i = 0;
txbuf->m_head = NULL;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
txbuf->eop_index = -1;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
txd->read.cmd_type_len |=
htole32(IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
txr->tx_avail -= nsegs;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
txr->next_avail_desc = i;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
txbuf->m_head = m_head;
txbuf->map = map;
bus_dmamap_sync(txr->txtag, map, BUS_DMASYNC_PREWRITE);
/* Set the index of the descriptor that will be marked done */
txbuf = &txr->tx_buffers[first];
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
txbuf->eop_index = last;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
/*
* Advance the Transmit Descriptor Tail (Tdt), this tells the
* hardware that this frame is available to transmit.
*/
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
++txr->total_packets;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
txr->watchdog_time = ticks;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
IXGBE_WRITE_REG(&adapter->hw, IXGBE_TDT(txr->me), i);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Do a clean if descriptors are low */
if (txr->tx_avail <= IXGBE_TX_CLEANUP_THRESHOLD)
ixgbe_txeof(txr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (0);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
xmit_fail:
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bus_dmamap_unload(txr->txtag, txbuf->map);
return (error);
}
static void
ixgbe_set_promisc(struct adapter *adapter)
{
u_int32_t reg_rctl;
struct ifnet *ifp = adapter->ifp;
reg_rctl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
if (ifp->if_flags & IFF_PROMISC) {
reg_rctl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, reg_rctl);
} else if (ifp->if_flags & IFF_ALLMULTI) {
reg_rctl |= IXGBE_FCTRL_MPE;
reg_rctl &= ~IXGBE_FCTRL_UPE;
IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, reg_rctl);
}
return;
}
static void
ixgbe_disable_promisc(struct adapter * adapter)
{
u_int32_t reg_rctl;
reg_rctl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
reg_rctl &= (~IXGBE_FCTRL_UPE);
reg_rctl &= (~IXGBE_FCTRL_MPE);
IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, reg_rctl);
return;
}
/*********************************************************************
* Multicast Update
*
* This routine is called whenever multicast address list is updated.
*
**********************************************************************/
#define IXGBE_RAR_ENTRIES 16
static void
ixgbe_set_multi(struct adapter *adapter)
{
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
u32 fctrl;
u8 mta[MAX_NUM_MULTICAST_ADDRESSES * IXGBE_ETH_LENGTH_OF_ADDRESS];
u8 *update_ptr;
struct ifmultiaddr *ifma;
int mcnt = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
struct ifnet *ifp = adapter->ifp;
IOCTL_DEBUGOUT("ixgbe_set_multi: begin");
fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
if (ifp->if_flags & IFF_PROMISC)
fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
else if (ifp->if_flags & IFF_ALLMULTI) {
fctrl |= IXGBE_FCTRL_MPE;
fctrl &= ~IXGBE_FCTRL_UPE;
} else
fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
#if __FreeBSD_version < 800000
IF_ADDR_LOCK(ifp);
#else
Use if_maddr_rlock()/if_maddr_runlock() rather than IF_ADDR_LOCK()/ IF_ADDR_UNLOCK() across network device drivers when accessing the per-interface multicast address list, if_multiaddrs. This will allow us to change the locking strategy without affecting our driver programming interface or binary interface. For two wireless drivers, remove unnecessary locking, since they don't actually access the multicast address list. Approved by: re (kib) MFC after: 6 weeks
2009-06-26 11:45:06 +00:00
if_maddr_rlock(ifp);
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
bcopy(LLADDR((struct sockaddr_dl *) ifma->ifma_addr),
&mta[mcnt * IXGBE_ETH_LENGTH_OF_ADDRESS],
IXGBE_ETH_LENGTH_OF_ADDRESS);
mcnt++;
}
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
#if __FreeBSD_version < 800000
IF_ADDR_UNLOCK(ifp);
#else
Use if_maddr_rlock()/if_maddr_runlock() rather than IF_ADDR_LOCK()/ IF_ADDR_UNLOCK() across network device drivers when accessing the per-interface multicast address list, if_multiaddrs. This will allow us to change the locking strategy without affecting our driver programming interface or binary interface. For two wireless drivers, remove unnecessary locking, since they don't actually access the multicast address list. Approved by: re (kib) MFC after: 6 weeks
2009-06-26 11:45:06 +00:00
if_maddr_runlock(ifp);
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
update_ptr = mta;
ixgbe_update_mc_addr_list(&adapter->hw,
update_ptr, mcnt, ixgbe_mc_array_itr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/*
* This is an iterator function now needed by the multicast
* shared code. It simply feeds the shared code routine the
* addresses in the array of ixgbe_set_multi() one by one.
*/
static u8 *
ixgbe_mc_array_itr(struct ixgbe_hw *hw, u8 **update_ptr, u32 *vmdq)
{
u8 *addr = *update_ptr;
u8 *newptr;
*vmdq = 0;
newptr = addr + IXGBE_ETH_LENGTH_OF_ADDRESS;
*update_ptr = newptr;
return addr;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
* Timer routine
*
* This routine checks for link status,updates statistics,
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
* and runs the watchdog check.
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
*
**********************************************************************/
static void
ixgbe_local_timer(void *arg)
{
struct adapter *adapter = arg;
struct ifnet *ifp = adapter->ifp;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
device_t dev = adapter->dev;
struct tx_ring *txr = adapter->tx_rings;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
mtx_assert(&adapter->core_mtx, MA_OWNED);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/* Check for pluggable optics */
if (adapter->sfp_probe)
if (!ixgbe_sfp_probe(adapter))
goto out; /* Nothing to do */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_update_link_status(adapter);
ixgbe_update_stats_counters(adapter);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Debug display */
if (ixgbe_display_debug_stats && ifp->if_drv_flags & IFF_DRV_RUNNING)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_print_hw_stats(adapter);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/*
* If the interface has been paused
* then don't do the watchdog check
*/
if (IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)
goto out;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
** Check for time since any descriptor was cleaned
*/
for (int i = 0; i < adapter->num_queues; i++, txr++) {
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
IXGBE_TX_LOCK(txr);
if (txr->watchdog_check == FALSE) {
IXGBE_TX_UNLOCK(txr);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
continue;
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if ((ticks - txr->watchdog_time) > IXGBE_WATCHDOG)
goto hung;
IXGBE_TX_UNLOCK(txr);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
out:
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_rearm_queues(adapter, adapter->que_mask);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
callout_reset(&adapter->timer, hz, ixgbe_local_timer, adapter);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
return;
hung:
device_printf(adapter->dev, "Watchdog timeout -- resetting\n");
device_printf(dev,"Queue(%d) tdh = %d, hw tdt = %d\n", txr->me,
IXGBE_READ_REG(&adapter->hw, IXGBE_TDH(txr->me)),
IXGBE_READ_REG(&adapter->hw, IXGBE_TDT(txr->me)));
device_printf(dev,"TX(%d) desc avail = %d,"
"Next TX to Clean = %d\n",
txr->me, txr->tx_avail, txr->next_to_clean);
adapter->ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
adapter->watchdog_events++;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
IXGBE_TX_UNLOCK(txr);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
ixgbe_init_locked(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/*
** Note: this routine updates the OS on the link state
** the real check of the hardware only happens with
** a link interrupt.
*/
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void
ixgbe_update_link_status(struct adapter *adapter)
{
struct ifnet *ifp = adapter->ifp;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct tx_ring *txr = adapter->tx_rings;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
device_t dev = adapter->dev;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
if (adapter->link_up){
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (adapter->link_active == FALSE) {
if (bootverbose)
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_printf(dev,"Link is up %d Gbps %s \n",
((adapter->link_speed == 128)? 10:1),
"Full Duplex");
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
adapter->link_active = TRUE;
if_link_state_change(ifp, LINK_STATE_UP);
}
} else { /* Link down */
if (adapter->link_active == TRUE) {
if (bootverbose)
device_printf(dev,"Link is Down\n");
if_link_state_change(ifp, LINK_STATE_DOWN);
adapter->link_active = FALSE;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (int i = 0; i < adapter->num_queues;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
i++, txr++)
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
txr->watchdog_check = FALSE;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
}
return;
}
/*********************************************************************
*
* This routine disables all traffic on the adapter by issuing a
* global reset on the MAC and deallocates TX/RX buffers.
*
**********************************************************************/
static void
ixgbe_stop(void *arg)
{
struct ifnet *ifp;
struct adapter *adapter = arg;
ifp = adapter->ifp;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
mtx_assert(&adapter->core_mtx, MA_OWNED);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
INIT_DEBUGOUT("ixgbe_stop: begin\n");
ixgbe_disable_intr(adapter);
/* Tell the stack that the interface is no longer active */
ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
ixgbe_reset_hw(&adapter->hw);
adapter->hw.adapter_stopped = FALSE;
ixgbe_stop_adapter(&adapter->hw);
callout_stop(&adapter->timer);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
/* reprogram the RAR[0] in case user changed it. */
ixgbe_set_rar(&adapter->hw, 0, adapter->hw.mac.addr, 0, IXGBE_RAH_AV);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
/*********************************************************************
*
* Determine hardware revision.
*
**********************************************************************/
static void
ixgbe_identify_hardware(struct adapter *adapter)
{
device_t dev = adapter->dev;
/* Save off the information about this board */
adapter->hw.vendor_id = pci_get_vendor(dev);
adapter->hw.device_id = pci_get_device(dev);
adapter->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1);
adapter->hw.subsystem_vendor_id =
pci_read_config(dev, PCIR_SUBVEND_0, 2);
adapter->hw.subsystem_device_id =
pci_read_config(dev, PCIR_SUBDEV_0, 2);
return;
}
/*********************************************************************
*
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
* Setup the Legacy or MSI Interrupt handler
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
*
**********************************************************************/
static int
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
ixgbe_allocate_legacy(struct adapter *adapter)
{
device_t dev = adapter->dev;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ix_queue *que = adapter->queues;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
int error, rid = 0;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/* MSI RID at 1 */
if (adapter->msix == 1)
rid = 1;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* We allocate a single interrupt resource */
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
adapter->res = bus_alloc_resource_any(dev,
SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE);
if (adapter->res == NULL) {
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_printf(dev, "Unable to allocate bus resource: "
"interrupt\n");
return (ENXIO);
}
/*
* Try allocating a fast interrupt and the associated deferred
* processing contexts.
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
TASK_INIT(&que->que_task, 0, ixgbe_handle_que, que);
que->tq = taskqueue_create_fast("ixgbe_que", M_NOWAIT,
taskqueue_thread_enqueue, &que->tq);
taskqueue_start_threads(&que->tq, 1, PI_NET, "%s ixq",
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_get_nameunit(adapter->dev));
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/* Tasklets for Link, SFP and Multispeed Fiber */
TASK_INIT(&adapter->link_task, 0, ixgbe_handle_link, adapter);
TASK_INIT(&adapter->mod_task, 0, ixgbe_handle_mod, adapter);
TASK_INIT(&adapter->msf_task, 0, ixgbe_handle_msf, adapter);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
TASK_INIT(&adapter->fdir_task, 0, ixgbe_reinit_fdir, adapter);
#endif
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
adapter->tq = taskqueue_create_fast("ixgbe_link", M_NOWAIT,
taskqueue_thread_enqueue, &adapter->tq);
taskqueue_start_threads(&adapter->tq, 1, PI_NET, "%s linkq",
device_get_nameunit(adapter->dev));
if ((error = bus_setup_intr(dev, adapter->res,
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
INTR_TYPE_NET | INTR_MPSAFE, NULL, ixgbe_legacy_irq,
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
que, &adapter->tag)) != 0) {
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_printf(dev, "Failed to register fast interrupt "
"handler: %d\n", error);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
taskqueue_free(que->tq);
taskqueue_free(adapter->tq);
que->tq = NULL;
adapter->tq = NULL;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
return (error);
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* For simplicity in the handlers */
adapter->que_mask = IXGBE_EIMS_ENABLE_MASK;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
return (0);
}
/*********************************************************************
*
* Setup MSIX Interrupt resources and handlers
*
**********************************************************************/
static int
ixgbe_allocate_msix(struct adapter *adapter)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
device_t dev = adapter->dev;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ix_queue *que = adapter->queues;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
int error, rid, vector = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
for (int i = 0; i < adapter->num_queues; i++, vector++, que++) {
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
rid = vector + 1;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
que->res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid,
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
RF_SHAREABLE | RF_ACTIVE);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (que->res == NULL) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
device_printf(dev,"Unable to allocate"
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
" bus resource: que interrupt [%d]\n", vector);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (ENXIO);
}
/* Set the handler function */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
error = bus_setup_intr(dev, que->res,
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
INTR_TYPE_NET | INTR_MPSAFE, NULL,
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_msix_que, que, &que->tag);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (error) {
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
que->res = NULL;
device_printf(dev, "Failed to register QUE handler");
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (error);
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
que->msix = vector;
adapter->que_mask |= (u64)(1 << que->msix);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/*
** Bind the msix vector, and thus the
** ring to the corresponding cpu.
*/
Decided to limit the interrupt bind to multiqueue config as done in igb.
2009-06-25 18:40:27 +00:00
if (adapter->num_queues > 1)
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
bus_bind_intr(dev, que->res, i);
Change intr_bind to bus_bind_intr, thanks to John Baldwin for pointing out this simplification.
2009-06-25 17:16:26 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
TASK_INIT(&que->que_task, 0, ixgbe_handle_que, que);
que->tq = taskqueue_create_fast("ixgbe_que", M_NOWAIT,
taskqueue_thread_enqueue, &que->tq);
taskqueue_start_threads(&que->tq, 1, PI_NET, "%s que",
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
device_get_nameunit(adapter->dev));
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* and Link */
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
rid = vector + 1;
adapter->res = bus_alloc_resource_any(dev,
SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE);
if (!adapter->res) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
device_printf(dev,"Unable to allocate"
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
" bus resource: Link interrupt [%d]\n", rid);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (ENXIO);
}
/* Set the link handler function */
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
error = bus_setup_intr(dev, adapter->res,
INTR_TYPE_NET | INTR_MPSAFE, NULL,
ixgbe_msix_link, adapter, &adapter->tag);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (error) {
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
adapter->res = NULL;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
device_printf(dev, "Failed to register LINK handler");
return (error);
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
adapter->linkvec = vector;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/* Tasklets for Link, SFP and Multispeed Fiber */
TASK_INIT(&adapter->link_task, 0, ixgbe_handle_link, adapter);
TASK_INIT(&adapter->mod_task, 0, ixgbe_handle_mod, adapter);
TASK_INIT(&adapter->msf_task, 0, ixgbe_handle_msf, adapter);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
TASK_INIT(&adapter->fdir_task, 0, ixgbe_reinit_fdir, adapter);
#endif
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
adapter->tq = taskqueue_create_fast("ixgbe_link", M_NOWAIT,
taskqueue_thread_enqueue, &adapter->tq);
taskqueue_start_threads(&adapter->tq, 1, PI_NET, "%s linkq",
device_get_nameunit(adapter->dev));
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (0);
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/*
* Setup Either MSI/X or MSI
*/
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static int
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
ixgbe_setup_msix(struct adapter *adapter)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_t dev = adapter->dev;
int rid, want, queues, msgs;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/* Override by tuneable */
if (ixgbe_enable_msix == 0)
goto msi;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* First try MSI/X */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rid = PCIR_BAR(MSIX_82598_BAR);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
adapter->msix_mem = bus_alloc_resource_any(dev,
SYS_RES_MEMORY, &rid, RF_ACTIVE);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
if (!adapter->msix_mem) {
rid += 4; /* 82599 maps in higher BAR */
adapter->msix_mem = bus_alloc_resource_any(dev,
SYS_RES_MEMORY, &rid, RF_ACTIVE);
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if (!adapter->msix_mem) {
/* May not be enabled */
device_printf(adapter->dev,
"Unable to map MSIX table \n");
goto msi;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
msgs = pci_msix_count(dev);
if (msgs == 0) { /* system has msix disabled */
bus_release_resource(dev, SYS_RES_MEMORY,
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rid, adapter->msix_mem);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
adapter->msix_mem = NULL;
goto msi;
}
/* Figure out a reasonable auto config value */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
queues = (mp_ncpus > (msgs-1)) ? (msgs-1) : mp_ncpus;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if (ixgbe_num_queues == 0)
ixgbe_num_queues = queues;
/*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
** Want one vector (RX/TX pair) per queue
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
** plus an additional for Link.
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
want = ixgbe_num_queues + 1;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if (msgs >= want)
msgs = want;
else {
device_printf(adapter->dev,
"MSIX Configuration Problem, "
"%d vectors but %d queues wanted!\n",
msgs, want);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
return (0); /* Will go to Legacy setup */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if ((msgs) && pci_alloc_msix(dev, &msgs) == 0) {
device_printf(adapter->dev,
"Using MSIX interrupts with %d vectors\n", msgs);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
adapter->num_queues = ixgbe_num_queues;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
return (msgs);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
msi:
msgs = pci_msi_count(dev);
if (msgs == 1 && pci_alloc_msi(dev, &msgs) == 0)
device_printf(adapter->dev,"Using MSI interrupt\n");
return (msgs);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static int
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
ixgbe_allocate_pci_resources(struct adapter *adapter)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
int rid;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
device_t dev = adapter->dev;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
rid = PCIR_BAR(0);
adapter->pci_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
&rid, RF_ACTIVE);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if (!(adapter->pci_mem)) {
device_printf(dev,"Unable to allocate bus resource: memory\n");
return (ENXIO);
}
adapter->osdep.mem_bus_space_tag =
rman_get_bustag(adapter->pci_mem);
adapter->osdep.mem_bus_space_handle =
rman_get_bushandle(adapter->pci_mem);
adapter->hw.hw_addr = (u8 *) &adapter->osdep.mem_bus_space_handle;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Legacy defaults */
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
adapter->num_queues = 1;
adapter->hw.back = &adapter->osdep;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/*
** Now setup MSI or MSI/X, should
** return us the number of supported
** vectors. (Will be 1 for MSI)
*/
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
adapter->msix = ixgbe_setup_msix(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (0);
}
static void
ixgbe_free_pci_resources(struct adapter * adapter)
{
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ix_queue *que = adapter->queues;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
device_t dev = adapter->dev;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
int rid, memrid;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
if (adapter->hw.mac.type == ixgbe_mac_82598EB)
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
memrid = PCIR_BAR(MSIX_82598_BAR);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
else
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
memrid = PCIR_BAR(MSIX_82599_BAR);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
** There is a slight possibility of a failure mode
** in attach that will result in entering this function
** before interrupt resources have been initialized, and
** in that case we do not want to execute the loops below
** We can detect this reliably by the state of the adapter
** res pointer.
*/
if (adapter->res == NULL)
goto mem;
/*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
** Release all msix queue resources:
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
for (int i = 0; i < adapter->num_queues; i++, que++) {
rid = que->msix + 1;
if (que->tag != NULL) {
bus_teardown_intr(dev, que->res, que->tag);
que->tag = NULL;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (que->res != NULL)
bus_release_resource(dev, SYS_RES_IRQ, rid, que->res);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/* Clean the Legacy or Link interrupt last */
if (adapter->linkvec) /* we are doing MSIX */
rid = adapter->linkvec + 1;
else
(adapter->msix != 0) ? (rid = 1):(rid = 0);
if (adapter->tag != NULL) {
bus_teardown_intr(dev, adapter->res, adapter->tag);
adapter->tag = NULL;
}
if (adapter->res != NULL)
bus_release_resource(dev, SYS_RES_IRQ, rid, adapter->res);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
mem:
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if (adapter->msix)
pci_release_msi(dev);
if (adapter->msix_mem != NULL)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bus_release_resource(dev, SYS_RES_MEMORY,
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
memrid, adapter->msix_mem);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if (adapter->pci_mem != NULL)
bus_release_resource(dev, SYS_RES_MEMORY,
PCIR_BAR(0), adapter->pci_mem);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
/*********************************************************************
*
* Setup networking device structure and register an interface.
*
**********************************************************************/
static void
ixgbe_setup_interface(device_t dev, struct adapter *adapter)
{
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct ixgbe_hw *hw = &adapter->hw;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
struct ifnet *ifp;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
INIT_DEBUGOUT("ixgbe_setup_interface: begin");
ifp = adapter->ifp = if_alloc(IFT_ETHER);
if (ifp == NULL)
panic("%s: can not if_alloc()\n", device_get_nameunit(dev));
if_initname(ifp, device_get_name(dev), device_get_unit(dev));
ifp->if_mtu = ETHERMTU;
ifp->if_baudrate = 1000000000;
ifp->if_init = ixgbe_init;
ifp->if_softc = adapter;
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_ioctl = ixgbe_ioctl;
ifp->if_start = ixgbe_start;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
#if __FreeBSD_version >= 800000
ifp->if_transmit = ixgbe_mq_start;
ifp->if_qflush = ixgbe_qflush;
#endif
ifp->if_snd.ifq_maxlen = adapter->num_tx_desc - 2;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ether_ifattach(ifp, adapter->hw.mac.addr);
adapter->max_frame_size =
ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
/*
* Tell the upper layer(s) we support long frames.
*/
ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_TSO4 | IFCAP_VLAN_HWCSUM;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_LRO;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ifp->if_capenable = ifp->if_capabilities;
/*
* Specify the media types supported by this adapter and register
* callbacks to update media and link information
*/
ifmedia_init(&adapter->media, IFM_IMASK, ixgbe_media_change,
ixgbe_media_status);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
ifmedia_add(&adapter->media, IFM_ETHER | adapter->optics |
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IFM_FDX, 0, NULL);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
if (hw->device_id == IXGBE_DEV_ID_82598AT) {
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
ifmedia_add(&adapter->media,
IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
ifmedia_add(&adapter->media,
IFM_ETHER | IFM_1000_T, 0, NULL);
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
return;
}
static void
ixgbe_config_link(struct adapter *adapter)
{
struct ixgbe_hw *hw = &adapter->hw;
u32 autoneg, err = 0;
bool sfp, negotiate;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
sfp = ixgbe_is_sfp(hw);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (sfp) {
if (hw->phy.multispeed_fiber) {
hw->mac.ops.setup_sfp(hw);
taskqueue_enqueue(adapter->tq, &adapter->msf_task);
} else
taskqueue_enqueue(adapter->tq, &adapter->mod_task);
} else {
if (hw->mac.ops.check_link)
err = ixgbe_check_link(hw, &autoneg,
&adapter->link_up, FALSE);
if (err)
goto out;
autoneg = hw->phy.autoneg_advertised;
if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
err = hw->mac.ops.get_link_capabilities(hw,
&autoneg, &negotiate);
if (err)
goto out;
if (hw->mac.ops.setup_link)
err = hw->mac.ops.setup_link(hw, autoneg,
negotiate, adapter->link_up);
}
out:
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
/********************************************************************
* Manage DMA'able memory.
*******************************************************************/
static void
ixgbe_dmamap_cb(void *arg, bus_dma_segment_t * segs, int nseg, int error)
{
if (error)
return;
*(bus_addr_t *) arg = segs->ds_addr;
return;
}
static int
ixgbe_dma_malloc(struct adapter *adapter, bus_size_t size,
struct ixgbe_dma_alloc *dma, int mapflags)
{
device_t dev = adapter->dev;
int r;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
r = bus_dma_tag_create(bus_get_dma_tag(adapter->dev), /* parent */
DBA_ALIGN, 0, /* alignment, bounds */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
size, /* maxsize */
1, /* nsegments */
size, /* maxsegsize */
BUS_DMA_ALLOCNOW, /* flags */
NULL, /* lockfunc */
NULL, /* lockfuncarg */
&dma->dma_tag);
if (r != 0) {
device_printf(dev,"ixgbe_dma_malloc: bus_dma_tag_create failed; "
"error %u\n", r);
goto fail_0;
}
r = bus_dmamem_alloc(dma->dma_tag, (void **)&dma->dma_vaddr,
BUS_DMA_NOWAIT, &dma->dma_map);
if (r != 0) {
device_printf(dev,"ixgbe_dma_malloc: bus_dmamem_alloc failed; "
"error %u\n", r);
goto fail_1;
}
r = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr,
size,
ixgbe_dmamap_cb,
&dma->dma_paddr,
mapflags | BUS_DMA_NOWAIT);
if (r != 0) {
device_printf(dev,"ixgbe_dma_malloc: bus_dmamap_load failed; "
"error %u\n", r);
goto fail_2;
}
dma->dma_size = size;
return (0);
fail_2:
bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
fail_1:
bus_dma_tag_destroy(dma->dma_tag);
fail_0:
dma->dma_map = NULL;
dma->dma_tag = NULL;
return (r);
}
static void
ixgbe_dma_free(struct adapter *adapter, struct ixgbe_dma_alloc *dma)
{
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
bus_dmamap_sync(dma->dma_tag, dma->dma_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bus_dmamap_unload(dma->dma_tag, dma->dma_map);
bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
bus_dma_tag_destroy(dma->dma_tag);
}
/*********************************************************************
*
* Allocate memory for the transmit and receive rings, and then
* the descriptors associated with each, called only once at attach.
*
**********************************************************************/
static int
ixgbe_allocate_queues(struct adapter *adapter)
{
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
device_t dev = adapter->dev;
struct ix_queue *que;
struct tx_ring *txr;
struct rx_ring *rxr;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
int rsize, tsize, error = IXGBE_SUCCESS;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
int txconf = 0, rxconf = 0;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* First allocate the top level queue structs */
if (!(adapter->queues =
(struct ix_queue *) malloc(sizeof(struct ix_queue) *
adapter->num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) {
device_printf(dev, "Unable to allocate queue memory\n");
error = ENOMEM;
goto fail;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* First allocate the TX ring struct memory */
if (!(adapter->tx_rings =
(struct tx_ring *) malloc(sizeof(struct tx_ring) *
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
adapter->num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
device_printf(dev, "Unable to allocate TX ring memory\n");
error = ENOMEM;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
goto tx_fail;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
/* Next allocate the RX */
if (!(adapter->rx_rings =
(struct rx_ring *) malloc(sizeof(struct rx_ring) *
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
adapter->num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
device_printf(dev, "Unable to allocate RX ring memory\n");
error = ENOMEM;
goto rx_fail;
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* For the ring itself */
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
tsize = roundup2(adapter->num_tx_desc *
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
sizeof(union ixgbe_adv_tx_desc), DBA_ALIGN);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*
* Now set up the TX queues, txconf is needed to handle the
* possibility that things fail midcourse and we need to
* undo memory gracefully
*/
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (int i = 0; i < adapter->num_queues; i++, txconf++) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Set up some basics */
txr = &adapter->tx_rings[i];
txr->adapter = adapter;
txr->me = i;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Initialize the TX side lock */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
snprintf(txr->mtx_name, sizeof(txr->mtx_name), "%s:tx(%d)",
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_get_nameunit(dev), txr->me);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
mtx_init(&txr->tx_mtx, txr->mtx_name, NULL, MTX_DEF);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
if (ixgbe_dma_malloc(adapter, tsize,
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
&txr->txdma, BUS_DMA_NOWAIT)) {
device_printf(dev,
"Unable to allocate TX Descriptor memory\n");
error = ENOMEM;
goto err_tx_desc;
}
txr->tx_base = (union ixgbe_adv_tx_desc *)txr->txdma.dma_vaddr;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
bzero((void *)txr->tx_base, tsize);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Now allocate transmit buffers for the ring */
if (ixgbe_allocate_transmit_buffers(txr)) {
device_printf(dev,
"Critical Failure setting up transmit buffers\n");
error = ENOMEM;
goto err_tx_desc;
}
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
#if __FreeBSD_version >= 800000
/* Allocate a buf ring */
txr->br = buf_ring_alloc(IXGBE_BR_SIZE, M_DEVBUF,
M_WAITOK, &txr->tx_mtx);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (txr->br == NULL) {
device_printf(dev,
"Critical Failure setting up buf ring\n");
error = ENOMEM;
goto err_tx_desc;
}
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
/*
* Next the RX queues...
*/
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
rsize = roundup2(adapter->num_rx_desc *
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
sizeof(union ixgbe_adv_rx_desc), DBA_ALIGN);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (int i = 0; i < adapter->num_queues; i++, rxconf++) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
rxr = &adapter->rx_rings[i];
/* Set up some basics */
rxr->adapter = adapter;
rxr->me = i;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/* Initialize the RX side lock */
snprintf(rxr->mtx_name, sizeof(rxr->mtx_name), "%s:rx(%d)",
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_get_nameunit(dev), rxr->me);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
mtx_init(&rxr->rx_mtx, rxr->mtx_name, NULL, MTX_DEF);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (ixgbe_dma_malloc(adapter, rsize,
&rxr->rxdma, BUS_DMA_NOWAIT)) {
device_printf(dev,
"Unable to allocate RxDescriptor memory\n");
error = ENOMEM;
goto err_rx_desc;
}
rxr->rx_base = (union ixgbe_adv_rx_desc *)rxr->rxdma.dma_vaddr;
bzero((void *)rxr->rx_base, rsize);
/* Allocate receive buffers for the ring*/
if (ixgbe_allocate_receive_buffers(rxr)) {
device_printf(dev,
"Critical Failure setting up receive buffers\n");
error = ENOMEM;
goto err_rx_desc;
}
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/*
** Finally set up the queue holding structs
*/
for (int i = 0; i < adapter->num_queues; i++) {
que = &adapter->queues[i];
que->adapter = adapter;
que->txr = &adapter->tx_rings[i];
que->rxr = &adapter->rx_rings[i];
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (0);
err_rx_desc:
for (rxr = adapter->rx_rings; rxconf > 0; rxr++, rxconf--)
ixgbe_dma_free(adapter, &rxr->rxdma);
err_tx_desc:
for (txr = adapter->tx_rings; txconf > 0; txr++, txconf--)
ixgbe_dma_free(adapter, &txr->txdma);
free(adapter->rx_rings, M_DEVBUF);
rx_fail:
free(adapter->tx_rings, M_DEVBUF);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
tx_fail:
free(adapter->queues, M_DEVBUF);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
fail:
return (error);
}
/*********************************************************************
*
* Allocate memory for tx_buffer structures. The tx_buffer stores all
* the information needed to transmit a packet on the wire. This is
* called only once at attach, setup is done every reset.
*
**********************************************************************/
static int
ixgbe_allocate_transmit_buffers(struct tx_ring *txr)
{
struct adapter *adapter = txr->adapter;
device_t dev = adapter->dev;
struct ixgbe_tx_buf *txbuf;
int error, i;
/*
* Setup DMA descriptor areas.
*/
if ((error = bus_dma_tag_create(NULL, /* parent */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
1, 0, /* alignment, bounds */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
IXGBE_TSO_SIZE, /* maxsize */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
ixgbe_num_segs, /* nsegments */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
PAGE_SIZE, /* maxsegsize */
0, /* flags */
NULL, /* lockfunc */
NULL, /* lockfuncarg */
&txr->txtag))) {
device_printf(dev,"Unable to allocate TX DMA tag\n");
goto fail;
}
if (!(txr->tx_buffers =
(struct ixgbe_tx_buf *) malloc(sizeof(struct ixgbe_tx_buf) *
adapter->num_tx_desc, M_DEVBUF, M_NOWAIT | M_ZERO))) {
device_printf(dev, "Unable to allocate tx_buffer memory\n");
error = ENOMEM;
goto fail;
}
/* Create the descriptor buffer dma maps */
txbuf = txr->tx_buffers;
for (i = 0; i < adapter->num_tx_desc; i++, txbuf++) {
error = bus_dmamap_create(txr->txtag, 0, &txbuf->map);
if (error != 0) {
device_printf(dev, "Unable to create TX DMA map\n");
goto fail;
}
}
return 0;
fail:
/* We free all, it handles case where we are in the middle */
ixgbe_free_transmit_structures(adapter);
return (error);
}
/*********************************************************************
*
* Initialize a transmit ring.
*
**********************************************************************/
static void
ixgbe_setup_transmit_ring(struct tx_ring *txr)
{
struct adapter *adapter = txr->adapter;
struct ixgbe_tx_buf *txbuf;
int i;
/* Clear the old ring contents */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
IXGBE_TX_LOCK(txr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bzero((void *)txr->tx_base,
(sizeof(union ixgbe_adv_tx_desc)) * adapter->num_tx_desc);
/* Reset indices */
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
txr->next_avail_desc = 0;
txr->next_to_clean = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* 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;
}
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/* Clear the EOP index */
txbuf->eop_index = -1;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
/* Set the rate at which we sample packets */
if (adapter->hw.mac.type == ixgbe_mac_82599EB)
txr->atr_sample = atr_sample_rate;
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Set number of descriptors available */
txr->tx_avail = adapter->num_tx_desc;
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
IXGBE_TX_UNLOCK(txr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
/*********************************************************************
*
* Initialize all transmit rings.
*
**********************************************************************/
static int
ixgbe_setup_transmit_structures(struct adapter *adapter)
{
struct tx_ring *txr = adapter->tx_rings;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_setup_transmit_ring(txr);
return (0);
}
/*********************************************************************
*
* Enable transmit unit.
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
*
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
**********************************************************************/
static void
ixgbe_initialize_transmit_units(struct adapter *adapter)
{
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct tx_ring *txr = adapter->tx_rings;
struct ixgbe_hw *hw = &adapter->hw;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Setup the Base and Length of the Tx Descriptor Ring */
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++) {
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
u64 tdba = txr->txdma.dma_paddr;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
u32 txctrl;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_WRITE_REG(hw, IXGBE_TDBAL(i),
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
(tdba & 0x00000000ffffffffULL));
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_WRITE_REG(hw, IXGBE_TDBAH(i), (tdba >> 32));
IXGBE_WRITE_REG(hw, IXGBE_TDLEN(i),
adapter->num_tx_desc * sizeof(struct ixgbe_legacy_tx_desc));
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Setup the HW Tx Head and Tail descriptor pointers */
IXGBE_WRITE_REG(hw, IXGBE_TDH(i), 0);
IXGBE_WRITE_REG(hw, IXGBE_TDT(i), 0);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Setup Transmit Descriptor Cmd Settings */
txr->txd_cmd = IXGBE_TXD_CMD_IFCS;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
txr->watchdog_check = FALSE;
/* Disable Head Writeback */
switch (hw->mac.type) {
case ixgbe_mac_82598EB:
txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(i));
break;
case ixgbe_mac_82599EB:
default:
txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(i));
break;
}
txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
switch (hw->mac.type) {
case ixgbe_mac_82598EB:
IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(i), txctrl);
break;
case ixgbe_mac_82599EB:
default:
IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(i), txctrl);
break;
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
if (hw->mac.type == ixgbe_mac_82599EB) {
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
u32 dmatxctl, rttdcs;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
dmatxctl |= IXGBE_DMATXCTL_TE;
IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Disable arbiter to set MTQC */
rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
rttdcs |= IXGBE_RTTDCS_ARBDIS;
IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
IXGBE_WRITE_REG(hw, IXGBE_MTQC, IXGBE_MTQC_64Q_1PB);
rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
/*********************************************************************
*
* Free all transmit rings.
*
**********************************************************************/
static void
ixgbe_free_transmit_structures(struct adapter *adapter)
{
struct tx_ring *txr = adapter->tx_rings;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (int i = 0; i < adapter->num_queues; i++, txr++) {
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_TX_LOCK(txr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_free_transmit_buffers(txr);
ixgbe_dma_free(adapter, &txr->txdma);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_TX_UNLOCK(txr);
IXGBE_TX_LOCK_DESTROY(txr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
free(adapter->tx_rings, M_DEVBUF);
}
/*********************************************************************
*
* Free transmit ring related data structures.
*
**********************************************************************/
static void
ixgbe_free_transmit_buffers(struct tx_ring *txr)
{
struct adapter *adapter = txr->adapter;
struct ixgbe_tx_buf *tx_buffer;
int i;
INIT_DEBUGOUT("free_transmit_ring: begin");
if (txr->tx_buffers == NULL)
return;
tx_buffer = txr->tx_buffers;
for (i = 0; i < adapter->num_tx_desc; i++, tx_buffer++) {
if (tx_buffer->m_head != NULL) {
bus_dmamap_sync(txr->txtag, tx_buffer->map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(txr->txtag,
tx_buffer->map);
m_freem(tx_buffer->m_head);
tx_buffer->m_head = NULL;
if (tx_buffer->map != NULL) {
bus_dmamap_destroy(txr->txtag,
tx_buffer->map);
tx_buffer->map = NULL;
}
} else if (tx_buffer->map != NULL) {
bus_dmamap_unload(txr->txtag,
tx_buffer->map);
bus_dmamap_destroy(txr->txtag,
tx_buffer->map);
tx_buffer->map = NULL;
}
}
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
#if __FreeBSD_version >= 800000
This delta fixes two bugs: - When a vlan event occurs a check was not made that the event was actually for the interface, thus resulting in a panic. All three drivers have this vulnerability. Add a check for this condition. - Secondly, there was a duplicate buf_ring free in the em driver resulting in a panic on unload. Remove. Approved by: re
2009-07-24 16:57:49 +00:00
if (txr->br != NULL)
buf_ring_free(txr->br, M_DEVBUF);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +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;
}
/*********************************************************************
*
* Advanced Context Descriptor setup for VLAN or CSUM
*
**********************************************************************/
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
static boolean_t
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
ixgbe_tx_ctx_setup(struct tx_ring *txr, struct mbuf *mp)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
struct adapter *adapter = txr->adapter;
struct ixgbe_adv_tx_context_desc *TXD;
struct ixgbe_tx_buf *tx_buffer;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
struct ether_vlan_header *eh;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
struct ip *ip;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
struct ip6_hdr *ip6;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
int ehdrlen, ip_hlen = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
u16 etype;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
u8 ipproto = 0;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
bool offload = TRUE;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
int ctxd = txr->next_avail_desc;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
u16 vtag = 0;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if ((mp->m_pkthdr.csum_flags & CSUM_OFFLOAD) == 0)
offload = FALSE;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
tx_buffer = &txr->tx_buffers[ctxd];
TXD = (struct ixgbe_adv_tx_context_desc *) &txr->tx_base[ctxd];
/*
** In advanced descriptors the vlan tag must
** be placed into the descriptor itself.
*/
if (mp->m_flags & M_VLANTAG) {
vtag = htole16(mp->m_pkthdr.ether_vtag);
vlan_macip_lens |= (vtag << IXGBE_ADVTXD_VLAN_SHIFT);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
} else if (offload == FALSE)
return FALSE;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*
* Determine where frame payload starts.
* Jump over vlan headers if already present,
* helpful for QinQ too.
*/
eh = mtod(mp, struct ether_vlan_header *);
if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
etype = ntohs(eh->evl_proto);
ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
} else {
etype = ntohs(eh->evl_encap_proto);
ehdrlen = ETHER_HDR_LEN;
}
/* Set the ether header length */
vlan_macip_lens |= ehdrlen << IXGBE_ADVTXD_MACLEN_SHIFT;
switch (etype) {
case ETHERTYPE_IP:
ip = (struct ip *)(mp->m_data + ehdrlen);
ip_hlen = ip->ip_hl << 2;
if (mp->m_len < ehdrlen + ip_hlen)
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
return (FALSE);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ipproto = ip->ip_p;
type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
break;
case ETHERTYPE_IPV6:
ip6 = (struct ip6_hdr *)(mp->m_data + ehdrlen);
ip_hlen = sizeof(struct ip6_hdr);
if (mp->m_len < ehdrlen + ip_hlen)
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
return (FALSE);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ipproto = ip6->ip6_nxt;
type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV6;
break;
default:
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
offload = FALSE;
break;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
vlan_macip_lens |= ip_hlen;
type_tucmd_mlhl |= IXGBE_ADVTXD_DCMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
switch (ipproto) {
case IPPROTO_TCP:
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if (mp->m_pkthdr.csum_flags & CSUM_TCP)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
break;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
case IPPROTO_UDP:
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if (mp->m_pkthdr.csum_flags & CSUM_UDP)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_UDP;
break;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
default:
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
offload = FALSE;
break;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
/* Now copy bits into descriptor */
TXD->vlan_macip_lens |= htole32(vlan_macip_lens);
TXD->type_tucmd_mlhl |= htole32(type_tucmd_mlhl);
TXD->seqnum_seed = htole32(0);
TXD->mss_l4len_idx = htole32(0);
tx_buffer->m_head = NULL;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
tx_buffer->eop_index = -1;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* We've consumed the first desc, adjust counters */
if (++ctxd == adapter->num_tx_desc)
ctxd = 0;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
txr->next_avail_desc = ctxd;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
--txr->tx_avail;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
return (offload);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
/**********************************************************************
*
* Setup work for hardware segmentation offload (TSO) on
* adapters using advanced tx descriptors
*
**********************************************************************/
static boolean_t
ixgbe_tso_setup(struct tx_ring *txr, struct mbuf *mp, u32 *paylen)
{
struct adapter *adapter = txr->adapter;
struct ixgbe_adv_tx_context_desc *TXD;
struct ixgbe_tx_buf *tx_buffer;
u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
u32 mss_l4len_idx = 0;
u16 vtag = 0;
int ctxd, ehdrlen, hdrlen, ip_hlen, tcp_hlen;
struct ether_vlan_header *eh;
struct ip *ip;
struct tcphdr *th;
/*
* Determine where frame payload starts.
* Jump over vlan headers if already present
*/
eh = mtod(mp, struct ether_vlan_header *);
if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
else
ehdrlen = ETHER_HDR_LEN;
/* Ensure we have at least the IP+TCP header in the first mbuf. */
if (mp->m_len < ehdrlen + sizeof(struct ip) + sizeof(struct tcphdr))
return FALSE;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
ctxd = txr->next_avail_desc;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
tx_buffer = &txr->tx_buffers[ctxd];
TXD = (struct ixgbe_adv_tx_context_desc *) &txr->tx_base[ctxd];
ip = (struct ip *)(mp->m_data + ehdrlen);
if (ip->ip_p != IPPROTO_TCP)
return FALSE; /* 0 */
ip->ip_sum = 0;
ip_hlen = ip->ip_hl << 2;
th = (struct tcphdr *)((caddr_t)ip + ip_hlen);
th->th_sum = in_pseudo(ip->ip_src.s_addr,
ip->ip_dst.s_addr, htons(IPPROTO_TCP));
tcp_hlen = th->th_off << 2;
hdrlen = ehdrlen + ip_hlen + tcp_hlen;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* This is used in the transmit desc in encap */
*paylen = mp->m_pkthdr.len - hdrlen;
/* VLAN MACLEN IPLEN */
if (mp->m_flags & M_VLANTAG) {
vtag = htole16(mp->m_pkthdr.ether_vtag);
vlan_macip_lens |= (vtag << IXGBE_ADVTXD_VLAN_SHIFT);
}
vlan_macip_lens |= ehdrlen << IXGBE_ADVTXD_MACLEN_SHIFT;
vlan_macip_lens |= ip_hlen;
TXD->vlan_macip_lens |= htole32(vlan_macip_lens);
/* ADV DTYPE TUCMD */
type_tucmd_mlhl |= IXGBE_ADVTXD_DCMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
TXD->type_tucmd_mlhl |= htole32(type_tucmd_mlhl);
/* MSS L4LEN IDX */
mss_l4len_idx |= (mp->m_pkthdr.tso_segsz << IXGBE_ADVTXD_MSS_SHIFT);
mss_l4len_idx |= (tcp_hlen << IXGBE_ADVTXD_L4LEN_SHIFT);
TXD->mss_l4len_idx = htole32(mss_l4len_idx);
TXD->seqnum_seed = htole32(0);
tx_buffer->m_head = NULL;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
tx_buffer->eop_index = -1;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (++ctxd == adapter->num_tx_desc)
ctxd = 0;
txr->tx_avail--;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
txr->next_avail_desc = ctxd;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return TRUE;
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
/*
** This routine parses packet headers so that Flow
** Director can make a hashed filter table entry
** allowing traffic flows to be identified and kept
** on the same cpu. This would be a performance
** hit, but we only do it at IXGBE_FDIR_RATE of
** packets.
*/
static void
ixgbe_atr(struct tx_ring *txr, struct mbuf *mp)
{
struct adapter *adapter = txr->adapter;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ix_queue *que;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
struct ixgbe_atr_input atr_input;
struct ip *ip;
struct tcphdr *th;
struct udphdr *uh;
struct ether_vlan_header *eh;
int ehdrlen, ip_hlen;
u16 etype, vlan_id, src_port, dst_port, flex_bytes;
u32 src_ipv4_addr, dst_ipv4_addr;
u8 l4type = 0, ipproto = 0;
eh = mtod(mp, struct ether_vlan_header *);
if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
else
ehdrlen = ETHER_HDR_LEN;
etype = ntohs(eh->evl_proto);
/* Only handling IPv4 */
if (etype != ETHERTYPE_IP)
return;
ip = (struct ip *)(mp->m_data + ehdrlen);
ipproto = ip->ip_p;
ip_hlen = ip->ip_hl << 2;
src_port = dst_port = 0;
/* check if we're UDP or TCP */
switch (ipproto) {
case IPPROTO_TCP:
th = (struct tcphdr *)((caddr_t)ip + ip_hlen);
src_port = th->th_sport;
dst_port = th->th_dport;
l4type |= IXGBE_ATR_L4TYPE_TCP;
break;
case IPPROTO_UDP:
uh = (struct udphdr *)((caddr_t)ip + ip_hlen);
src_port = uh->uh_sport;
dst_port = uh->uh_dport;
l4type |= IXGBE_ATR_L4TYPE_UDP;
break;
default:
return;
}
memset(&atr_input, 0, sizeof(struct ixgbe_atr_input));
vlan_id = htole16(mp->m_pkthdr.ether_vtag);
src_ipv4_addr = ip->ip_src.s_addr;
dst_ipv4_addr = ip->ip_dst.s_addr;
flex_bytes = etype;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
que = &adapter->queues[txr->me];
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
ixgbe_atr_set_vlan_id_82599(&atr_input, vlan_id);
ixgbe_atr_set_src_port_82599(&atr_input, dst_port);
ixgbe_atr_set_dst_port_82599(&atr_input, src_port);
ixgbe_atr_set_flex_byte_82599(&atr_input, flex_bytes);
ixgbe_atr_set_l4type_82599(&atr_input, l4type);
/* src and dst are inverted, think how the receiver sees them */
ixgbe_atr_set_src_ipv4_82599(&atr_input, dst_ipv4_addr);
ixgbe_atr_set_dst_ipv4_82599(&atr_input, src_ipv4_addr);
/* This assumes the Rx queue and Tx queue are bound to the same CPU */
ixgbe_fdir_add_signature_filter_82599(&adapter->hw,
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
&atr_input, que->msix);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
}
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/**********************************************************************
*
* Examine each tx_buffer in the used queue. If the hardware is done
* processing the packet then free associated resources. The
* tx_buffer is put back on the free queue.
*
**********************************************************************/
static boolean_t
ixgbe_txeof(struct tx_ring *txr)
{
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
struct adapter *adapter = txr->adapter;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
struct ifnet *ifp = adapter->ifp;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
u32 first, last, done;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
struct ixgbe_tx_buf *tx_buffer;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
struct ixgbe_legacy_tx_desc *tx_desc, *eop_desc;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
mtx_assert(&txr->tx_mtx, MA_OWNED);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (txr->tx_avail == adapter->num_tx_desc)
return FALSE;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
first = txr->next_to_clean;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
tx_buffer = &txr->tx_buffers[first];
/* For cleanup we just use legacy struct */
tx_desc = (struct ixgbe_legacy_tx_desc *)&txr->tx_base[first];
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
last = tx_buffer->eop_index;
if (last == -1)
return FALSE;
eop_desc = (struct ixgbe_legacy_tx_desc *)&txr->tx_base[last];
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/*
** Get the index of the first descriptor
** BEYOND the EOP and call that 'done'.
** I do this so the comparison in the
** inner while loop below can be simple
*/
if (++last == adapter->num_tx_desc) last = 0;
done = last;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
BUS_DMASYNC_POSTREAD);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/*
** Only the EOP descriptor of a packet now has the DD
** bit set, this is what we look for...
*/
while (eop_desc->upper.fields.status & IXGBE_TXD_STAT_DD) {
/* We clean the range of the packet */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
while (first != done) {
tx_desc->upper.data = 0;
tx_desc->lower.data = 0;
tx_desc->buffer_addr = 0;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
++txr->tx_avail;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (tx_buffer->m_head) {
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
txr->bytes +=
tx_buffer->m_head->m_pkthdr.len;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bus_dmamap_sync(txr->txtag,
tx_buffer->map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(txr->txtag,
tx_buffer->map);
m_freem(tx_buffer->m_head);
tx_buffer->m_head = NULL;
tx_buffer->map = NULL;
}
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
tx_buffer->eop_index = -1;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
txr->watchdog_time = ticks;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (++first == adapter->num_tx_desc)
first = 0;
tx_buffer = &txr->tx_buffers[first];
tx_desc =
(struct ixgbe_legacy_tx_desc *)&txr->tx_base[first];
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
++txr->packets;
++ifp->if_opackets;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* See if there is more work now */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
last = tx_buffer->eop_index;
if (last != -1) {
eop_desc =
(struct ixgbe_legacy_tx_desc *)&txr->tx_base[last];
/* Get next done point */
if (++last == adapter->num_tx_desc) last = 0;
done = last;
} else
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
break;
}
bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
txr->next_to_clean = first;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*
* If we have enough room, clear IFF_DRV_OACTIVE to tell the stack that
* it is OK to send packets. If there are no pending descriptors,
* clear the timeout. Otherwise, if some descriptors have been freed,
* restart the timeout.
*/
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (txr->tx_avail > IXGBE_TX_CLEANUP_THRESHOLD) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (txr->tx_avail == adapter->num_tx_desc) {
txr->watchdog_check = FALSE;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
return FALSE;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
return TRUE;
}
/*********************************************************************
*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
* Refresh mbuf buffers for RX descriptor rings
* - now keeps its own state so discards due to resource
* exhaustion are unnecessary, if an mbuf cannot be obtained
* it just returns, keeping its placeholder, thus it can simply
* be recalled to try again.
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
*
**********************************************************************/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
static void
ixgbe_refresh_mbufs(struct rx_ring *rxr, int limit)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
struct adapter *adapter = rxr->adapter;
bus_dma_segment_t seg[2];
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
struct ixgbe_rx_buf *rxbuf;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
struct mbuf *mh, *mp;
bus_dmamap_t map;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
int i, nsegs, error, cleaned;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
i = rxr->next_to_refresh;
cleaned = -1; /* Signify no completions */
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
while (i != limit) {
rxbuf = &rxr->rx_buffers[i];
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (rxbuf->m_head == NULL) {
mh = m_gethdr(M_DONTWAIT, MT_DATA);
if (mh == NULL)
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
goto update;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
} else /* reuse */
mh = rxbuf->m_head;
mh->m_len = MHLEN;
mh->m_flags |= M_PKTHDR;
if (rxbuf->m_pack == NULL) {
mp = m_getjcl(M_DONTWAIT, MT_DATA,
M_PKTHDR, adapter->rx_mbuf_sz);
if (mp == NULL)
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
goto update;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
mp->m_len = adapter->rx_mbuf_sz;
mp->m_flags &= ~M_PKTHDR;
} else { /* reusing */
mp = rxbuf->m_pack;
mp->m_len = adapter->rx_mbuf_sz;
mp->m_flags &= ~M_PKTHDR;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
mp->m_next = NULL;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/*
** Need to create a chain for the following
** dmamap call at this point.
*/
mh->m_next = mp;
mh->m_pkthdr.len = mh->m_len + mp->m_len;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Get the memory mapping */
error = bus_dmamap_load_mbuf_sg(rxr->rxtag,
rxr->spare_map, mh, seg, &nsegs, BUS_DMA_NOWAIT);
if (error != 0) {
printf("GET BUF: dmamap load failure - %d\n", error);
m_free(mh);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
goto update;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Unload old mapping and update buffer struct */
if (rxbuf->m_head != NULL)
bus_dmamap_unload(rxr->rxtag, rxbuf->map);
map = rxbuf->map;
rxbuf->map = rxr->spare_map;
rxr->spare_map = map;
rxbuf->m_head = mh;
rxbuf->m_pack = mp;
bus_dmamap_sync(rxr->rxtag,
rxbuf->map, BUS_DMASYNC_PREREAD);
/* Update descriptor */
rxr->rx_base[i].read.hdr_addr = htole64(seg[0].ds_addr);
rxr->rx_base[i].read.pkt_addr = htole64(seg[1].ds_addr);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
cleaned = i;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Calculate next index */
if (++i == adapter->num_rx_desc)
i = 0;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* This is the work marker for refresh */
rxr->next_to_refresh = i;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
update:
if (cleaned != -1) /* If we refreshed some, bump tail */
IXGBE_WRITE_REG(&adapter->hw, IXGBE_RDT(rxr->me), cleaned);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
return;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
*
* Allocate memory for rx_buffer structures. Since we use one
* rx_buffer per received packet, the maximum number of rx_buffer's
* that we'll need is equal to the number of receive descriptors
* that we've allocated.
*
**********************************************************************/
static int
ixgbe_allocate_receive_buffers(struct rx_ring *rxr)
{
struct adapter *adapter = rxr->adapter;
device_t dev = adapter->dev;
struct ixgbe_rx_buf *rxbuf;
int i, bsize, error;
bsize = sizeof(struct ixgbe_rx_buf) * adapter->num_rx_desc;
if (!(rxr->rx_buffers =
(struct ixgbe_rx_buf *) malloc(bsize,
M_DEVBUF, M_NOWAIT | M_ZERO))) {
device_printf(dev, "Unable to allocate rx_buffer memory\n");
error = ENOMEM;
goto fail;
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/*
** The tag is made to accomodate the largest buffer size
** with packet split (hence the two segments, even though
** it may not always use this.
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if ((error = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
1, 0, /* alignment, bounds */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
MJUM16BYTES, /* maxsize */
2, /* nsegments */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
MJUMPAGESIZE, /* maxsegsize */
0, /* flags */
NULL, /* lockfunc */
NULL, /* lockfuncarg */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
&rxr->rxtag))) {
device_printf(dev, "Unable to create RX DMA tag\n");
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
goto fail;
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/* Create the spare map (used by getbuf) */
error = bus_dmamap_create(rxr->rxtag, BUS_DMA_NOWAIT,
&rxr->spare_map);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (error) {
device_printf(dev, "%s: bus_dmamap_create failed: %d\n",
__func__, error);
goto fail;
}
for (i = 0; i < adapter->num_rx_desc; i++, rxbuf++) {
rxbuf = &rxr->rx_buffers[i];
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
error = bus_dmamap_create(rxr->rxtag,
BUS_DMA_NOWAIT, &rxbuf->map);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (error) {
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
device_printf(dev, "Unable to create RX DMA map\n");
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
goto fail;
}
}
return (0);
fail:
/* Frees all, but can handle partial completion */
ixgbe_free_receive_structures(adapter);
return (error);
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/*
** Used to detect a descriptor that has
** been merged by Hardware RSC.
*/
static inline u32
ixgbe_rsc_count(union ixgbe_adv_rx_desc *rx)
{
return (le32toh(rx->wb.lower.lo_dword.data) &
IXGBE_RXDADV_RSCCNT_MASK) >> IXGBE_RXDADV_RSCCNT_SHIFT;
}
/*********************************************************************
*
* Initialize Hardware RSC (LRO) feature on 82599
* for an RX ring, this is toggled by the LRO capability
* even though it is transparent to the stack.
*
**********************************************************************/
static void
ixgbe_setup_hw_rsc(struct rx_ring *rxr)
{
struct adapter *adapter = rxr->adapter;
struct ixgbe_hw *hw = &adapter->hw;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
u32 rscctrl, rdrxctl;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
rdrxctl |= IXGBE_RDRXCTL_RSCACKC;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(rxr->me));
rscctrl |= IXGBE_RSCCTL_RSCEN;
/*
** Limit the total number of descriptors that
** can be combined, so it does not exceed 64K
*/
if (adapter->rx_mbuf_sz == MCLBYTES)
rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
else /* using 4K clusters */
rscctrl |= IXGBE_RSCCTL_MAXDESC_8;
IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(rxr->me), rscctrl);
/* Enable TCP header recognition */
IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0),
(IXGBE_READ_REG(hw, IXGBE_PSRTYPE(0)) |
IXGBE_PSRTYPE_TCPHDR));
/* Disable RSC for ACK packets */
IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
(IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
rxr->hw_rsc = TRUE;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
*
* Initialize a receive ring and its buffers.
*
**********************************************************************/
static int
ixgbe_setup_receive_ring(struct rx_ring *rxr)
{
struct adapter *adapter;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
struct ifnet *ifp;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_t dev;
struct ixgbe_rx_buf *rxbuf;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
bus_dma_segment_t seg[2];
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct lro_ctrl *lro = &rxr->lro;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
int rsize, nsegs, error;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
adapter = rxr->adapter;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
ifp = adapter->ifp;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
dev = adapter->dev;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Clear the ring contents */
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
IXGBE_RX_LOCK(rxr);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rsize = roundup2(adapter->num_rx_desc *
sizeof(union ixgbe_adv_rx_desc), DBA_ALIGN);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bzero((void *)rxr->rx_base, rsize);
/*
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
** Free current RX buffer structs and their mbufs
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
*/
for (int i = 0; i < adapter->num_rx_desc; i++) {
rxbuf = &rxr->rx_buffers[i];
if (rxbuf->m_head != NULL) {
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
bus_dmamap_sync(rxr->rxtag, rxbuf->map,
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
BUS_DMASYNC_POSTREAD);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
bus_dmamap_unload(rxr->rxtag, rxbuf->map);
if (rxbuf->m_head) {
rxbuf->m_head->m_next = rxbuf->m_pack;
m_freem(rxbuf->m_head);
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
rxbuf->m_head = NULL;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rxbuf->m_pack = NULL;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* Now replenish the mbufs */
for (int j = 0; j != adapter->num_rx_desc; ++j) {
rxbuf = &rxr->rx_buffers[j];
rxbuf->m_head = m_gethdr(M_DONTWAIT, MT_DATA);
if (rxbuf->m_head == NULL)
panic("RX ring hdr initialization failed!\n");
rxbuf->m_head->m_len = MHLEN;
rxbuf->m_head->m_flags |= M_PKTHDR;
rxbuf->m_head->m_pkthdr.len = rxbuf->m_head->m_len;
rxbuf->m_pack = m_getjcl(M_DONTWAIT, MT_DATA,
M_PKTHDR, adapter->rx_mbuf_sz);
if (rxbuf->m_pack == NULL)
panic("RX ring pkt initialization failed!\n");
rxbuf->m_pack->m_len = adapter->rx_mbuf_sz;
rxbuf->m_head->m_next = rxbuf->m_pack;
rxbuf->m_head->m_pkthdr.len += rxbuf->m_pack->m_len;
/* Get the memory mapping */
error = bus_dmamap_load_mbuf_sg(rxr->rxtag,
rxbuf->map, rxbuf->m_head, seg,
&nsegs, BUS_DMA_NOWAIT);
if (error != 0)
panic("RX ring dma initialization failed!\n");
bus_dmamap_sync(rxr->rxtag,
rxbuf->map, BUS_DMASYNC_PREREAD);
/* Update descriptor */
rxr->rx_base[j].read.hdr_addr = htole64(seg[0].ds_addr);
rxr->rx_base[j].read.pkt_addr = htole64(seg[1].ds_addr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Setup our descriptor indices */
rxr->next_to_check = 0;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
rxr->next_to_refresh = 0;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
rxr->lro_enabled = FALSE;
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
/* Use header split if configured */
if (ixgbe_header_split)
rxr->hdr_split = TRUE;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/*
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
** Now set up the LRO interface:
** 82598 uses software LRO, the
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
** 82599 additionally uses a
** hardware assist.
**
** Disable RSC when RXCSUM is off
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if ((adapter->hw.mac.type == ixgbe_mac_82599EB) &&
(ifp->if_capenable & IFCAP_RXCSUM))
ixgbe_setup_hw_rsc(rxr);
else if (ifp->if_capenable & IFCAP_LRO) {
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
int err = tcp_lro_init(lro);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (err)
panic("LRO Initialization failed!\n");
INIT_DEBUGOUT("RX Soft LRO Initialized\n");
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
rxr->lro_enabled = TRUE;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
lro->ifp = adapter->ifp;
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
IXGBE_RX_UNLOCK(rxr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return (0);
}
/*********************************************************************
*
* Initialize all receive rings.
*
**********************************************************************/
static int
ixgbe_setup_receive_structures(struct adapter *adapter)
{
struct rx_ring *rxr = adapter->rx_rings;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
int j;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (j = 0; j < adapter->num_queues; j++, rxr++)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (ixgbe_setup_receive_ring(rxr))
goto fail;
return (0);
fail:
/*
* Free RX buffers allocated so far, we will only handle
* the rings that completed, the failing case will have
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
* cleaned up for itself. 'j' failed, so its the terminus.
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
*/
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
for (int i = 0; i < j; ++i) {
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rxr = &adapter->rx_rings[i];
for (int n = 0; n < adapter->num_rx_desc; n++) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
struct ixgbe_rx_buf *rxbuf;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rxbuf = &rxr->rx_buffers[n];
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (rxbuf->m_head != NULL) {
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
bus_dmamap_sync(rxr->rxtag, rxbuf->map,
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
BUS_DMASYNC_POSTREAD);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
bus_dmamap_unload(rxr->rxtag, rxbuf->map);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
m_freem(rxbuf->m_head);
rxbuf->m_head = NULL;
}
}
}
return (ENOBUFS);
}
/*********************************************************************
*
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
* Setup receive registers and features.
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
*
**********************************************************************/
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
#define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void
ixgbe_initialize_receive_units(struct adapter *adapter)
{
struct rx_ring *rxr = adapter->rx_rings;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
struct ixgbe_hw *hw = &adapter->hw;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
struct ifnet *ifp = adapter->ifp;
Multiqueue RX is not correctly enabled on the new 82599 adapter, the SRRCTL register needs to be setup per queue. Approved by: re
2009-07-01 16:13:01 +00:00
u32 bufsz, rxctrl, fctrl, srrctl, rxcsum;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
u32 reta, mrqc = 0, hlreg, random[10];
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*
* Make sure receives are disabled while
* setting up the descriptor ring
*/
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
IXGBE_WRITE_REG(hw, IXGBE_RXCTRL,
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
rxctrl & ~IXGBE_RXCTRL_RXEN);
/* Enable broadcasts */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
fctrl |= IXGBE_FCTRL_BAM;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
fctrl |= IXGBE_FCTRL_DPF;
fctrl |= IXGBE_FCTRL_PMCF;
IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/* Set for Jumbo Frames? */
Multiqueue RX is not correctly enabled on the new 82599 adapter, the SRRCTL register needs to be setup per queue. Approved by: re
2009-07-01 16:13:01 +00:00
hlreg = IXGBE_READ_REG(hw, IXGBE_HLREG0);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
if (ifp->if_mtu > ETHERMTU) {
hlreg |= IXGBE_HLREG0_JUMBOEN;
Multiqueue RX is not correctly enabled on the new 82599 adapter, the SRRCTL register needs to be setup per queue. Approved by: re
2009-07-01 16:13:01 +00:00
bufsz = 4096 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
} else {
hlreg &= ~IXGBE_HLREG0_JUMBOEN;
Multiqueue RX is not correctly enabled on the new 82599 adapter, the SRRCTL register needs to be setup per queue. Approved by: re
2009-07-01 16:13:01 +00:00
bufsz = 2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
}
IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (int i = 0; i < adapter->num_queues; i++, rxr++) {
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
u64 rdba = rxr->rxdma.dma_paddr;
Multiqueue RX is not correctly enabled on the new 82599 adapter, the SRRCTL register needs to be setup per queue. Approved by: re
2009-07-01 16:13:01 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Setup the Base and Length of the Rx Descriptor Ring */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_WRITE_REG(hw, IXGBE_RDBAL(i),
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
(rdba & 0x00000000ffffffffULL));
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_WRITE_REG(hw, IXGBE_RDBAH(i), (rdba >> 32));
IXGBE_WRITE_REG(hw, IXGBE_RDLEN(i),
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
adapter->num_rx_desc * sizeof(union ixgbe_adv_rx_desc));
Multiqueue RX is not correctly enabled on the new 82599 adapter, the SRRCTL register needs to be setup per queue. Approved by: re
2009-07-01 16:13:01 +00:00
/* Set up the SRRCTL register */
srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
srrctl |= bufsz;
if (rxr->hdr_split) {
/* Use a standard mbuf for the header */
srrctl |= ((IXGBE_RX_HDR <<
IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT)
& IXGBE_SRRCTL_BSIZEHDR_MASK);
srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
/* PSRTYPE must be initialized in 82599 */
u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
IXGBE_PSRTYPE_UDPHDR |
IXGBE_PSRTYPE_IPV4HDR |
IXGBE_PSRTYPE_IPV6HDR;
IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype);
}
} else
srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(i), srrctl);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Setup the HW Rx Head and Tail Descriptor Pointers */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_WRITE_REG(hw, IXGBE_RDH(i), 0);
IXGBE_WRITE_REG(hw, IXGBE_RDT(i), 0);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/* Setup RSS */
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if (adapter->num_queues > 1) {
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
int i, j;
reta = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* set up random bits */
arc4rand(&random, sizeof(random), 0);
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Set up the redirection table */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
for (i = 0, j = 0; i < 128; i++, j++) {
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if (j == adapter->num_queues) j = 0;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
reta = (reta << 8) | (j * 0x11);
if ((i & 3) == 3)
IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Now fill our hash function seeds */
for (int i = 0; i < 10; i++)
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), random[i]);
/* Perform hash on these packet types */
Multiqueue RX is not correctly enabled on the new 82599 adapter, the SRRCTL register needs to be setup per queue. Approved by: re
2009-07-01 16:13:01 +00:00
mrqc = IXGBE_MRQC_RSSEN
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
| IXGBE_MRQC_RSS_FIELD_IPV4
| IXGBE_MRQC_RSS_FIELD_IPV4_TCP
| IXGBE_MRQC_RSS_FIELD_IPV4_UDP
| IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP
| IXGBE_MRQC_RSS_FIELD_IPV6_EX
| IXGBE_MRQC_RSS_FIELD_IPV6
| IXGBE_MRQC_RSS_FIELD_IPV6_TCP
| IXGBE_MRQC_RSS_FIELD_IPV6_UDP
| IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* RSS and RX IPP Checksum are mutually exclusive */
rxcsum |= IXGBE_RXCSUM_PCSD;
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if (ifp->if_capenable & IFCAP_RXCSUM)
rxcsum |= IXGBE_RXCSUM_PCSD;
if (!(rxcsum & IXGBE_RXCSUM_PCSD))
rxcsum |= IXGBE_RXCSUM_IPPCSE;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
/*********************************************************************
*
* Free all receive rings.
*
**********************************************************************/
static void
ixgbe_free_receive_structures(struct adapter *adapter)
{
struct rx_ring *rxr = adapter->rx_rings;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
for (int i = 0; i < adapter->num_queues; i++, rxr++) {
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct lro_ctrl *lro = &rxr->lro;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
ixgbe_free_receive_buffers(rxr);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Free LRO memory */
tcp_lro_free(lro);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* Free the ring memory as well */
ixgbe_dma_free(adapter, &rxr->rxdma);
}
free(adapter->rx_rings, M_DEVBUF);
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
*
* Free receive ring data structures
*
**********************************************************************/
void
ixgbe_free_receive_buffers(struct rx_ring *rxr)
{
struct adapter *adapter = NULL;
struct ixgbe_rx_buf *rxbuf = NULL;
INIT_DEBUGOUT("free_receive_buffers: begin");
adapter = rxr->adapter;
if (rxr->rx_buffers != NULL) {
rxbuf = &rxr->rx_buffers[0];
for (int i = 0; i < adapter->num_rx_desc; i++) {
if (rxbuf->map != NULL) {
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
bus_dmamap_sync(rxr->rxtag, rxbuf->map,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(rxr->rxtag, rxbuf->map);
bus_dmamap_destroy(rxr->rxtag, rxbuf->map);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
if (rxbuf->m_head != NULL) {
m_freem(rxbuf->m_head);
}
rxbuf->m_head = NULL;
++rxbuf;
}
}
if (rxr->rx_buffers != NULL) {
free(rxr->rx_buffers, M_DEVBUF);
rxr->rx_buffers = NULL;
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
if (rxr->rxtag != NULL) {
bus_dma_tag_destroy(rxr->rxtag);
rxr->rxtag = NULL;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
return;
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
static __inline void
ixgbe_rx_input(struct rx_ring *rxr, struct ifnet *ifp, struct mbuf *m, u32 ptype)
{
/*
* ATM LRO is only for IPv4/TCP packets and TCP checksum of the packet
* should be computed by hardware. Also it should not have VLAN tag in
* ethernet header.
*/
if (rxr->lro_enabled &&
(ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 &&
(ptype & IXGBE_RXDADV_PKTTYPE_ETQF) == 0 &&
(ptype & (IXGBE_RXDADV_PKTTYPE_IPV4 | IXGBE_RXDADV_PKTTYPE_TCP)) ==
(IXGBE_RXDADV_PKTTYPE_IPV4 | IXGBE_RXDADV_PKTTYPE_TCP) &&
(m->m_pkthdr.csum_flags & (CSUM_DATA_VALID | CSUM_PSEUDO_HDR)) ==
(CSUM_DATA_VALID | CSUM_PSEUDO_HDR)) {
/*
* Send to the stack if:
** - LRO not enabled, or
** - no LRO resources, or
** - lro enqueue fails
*/
if (rxr->lro.lro_cnt != 0)
if (tcp_lro_rx(&rxr->lro, m, 0) == 0)
return;
}
(*ifp->if_input)(ifp, m);
}
static __inline void
ixgbe_rx_discard(struct rx_ring *rxr, int i)
{
struct adapter *adapter = rxr->adapter;
struct ixgbe_rx_buf *rbuf;
struct mbuf *mh, *mp;
rbuf = &rxr->rx_buffers[i];
if (rbuf->fmp != NULL) /* Partial chain ? */
m_freem(rbuf->fmp);
mh = rbuf->m_head;
mp = rbuf->m_pack;
/* Reuse loaded DMA map and just update mbuf chain */
mh->m_len = MHLEN;
mh->m_flags |= M_PKTHDR;
mh->m_next = NULL;
mp->m_len = mp->m_pkthdr.len = adapter->rx_mbuf_sz;
mp->m_data = mp->m_ext.ext_buf;
mp->m_next = NULL;
return;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
*
* This routine executes in interrupt context. It replenishes
* the mbufs in the descriptor and sends data which has been
* dma'ed into host memory to upper layer.
*
* We loop at most count times if count is > 0, or until done if
* count < 0.
*
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
* Return TRUE for more work, FALSE for all clean.
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
*********************************************************************/
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
static bool
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_rxeof(struct ix_queue *que, int count)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct adapter *adapter = que->adapter;
struct rx_ring *rxr = que->rxr;
struct ifnet *ifp = adapter->ifp;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct lro_ctrl *lro = &rxr->lro;
struct lro_entry *queued;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
int i, nextp, processed = 0;
u32 staterr = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
union ixgbe_adv_rx_desc *cur;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ixgbe_rx_buf *rbuf, *nbuf;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_RX_LOCK(rxr);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
for (i = rxr->next_to_check; count != 0;) {
struct mbuf *sendmp, *mh, *mp;
u32 rsc, ptype;
u16 hlen, plen, hdr, vtag;
bool eop;
/* Sync the ring. */
bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
cur = &rxr->rx_base[i];
staterr = le32toh(cur->wb.upper.status_error);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if ((staterr & IXGBE_RXD_STAT_DD) == 0)
break;
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
break;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
count--;
sendmp = NULL;
nbuf = NULL;
rsc = 0;
cur->wb.upper.status_error = 0;
rbuf = &rxr->rx_buffers[i];
mh = rbuf->m_head;
mp = rbuf->m_pack;
plen = le16toh(cur->wb.upper.length);
ptype = le32toh(cur->wb.lower.lo_dword.data) &
IXGBE_RXDADV_PKTTYPE_MASK;
hdr = le16toh(cur->wb.lower.lo_dword.hs_rss.hdr_info);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
vtag = le16toh(cur->wb.upper.vlan);
eop = ((staterr & IXGBE_RXD_STAT_EOP) != 0);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Make sure all parts of a bad packet are discarded */
if (((staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) != 0) ||
(rxr->discard)) {
ifp->if_ierrors++;
rxr->rx_discarded++;
if (!eop)
rxr->discard = TRUE;
else
rxr->discard = FALSE;
ixgbe_rx_discard(rxr, i);
goto next_desc;
}
/*
** On 82599 which supports a hardware
** LRO (called HW RSC), packets need
** not be fragmented across sequential
** descriptors, rather the next descriptor
** is indicated in bits of the descriptor.
** This also means that we might proceses
** more than one packet at a time, something
** that has never been true before, it
** required eliminating global chain pointers
** in favor of what we are doing here. -jfv
*/
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (!eop) {
/*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
** Figure out the next descriptor
** of this frame.
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
*/
if (rxr->hw_rsc == TRUE) {
rsc = ixgbe_rsc_count(cur);
rxr->rsc_num += (rsc - 1);
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (rsc) { /* Get hardware index */
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
nextp = ((staterr &
IXGBE_RXDADV_NEXTP_MASK) >>
IXGBE_RXDADV_NEXTP_SHIFT);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
} else { /* Just sequential */
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
nextp = i + 1;
if (nextp == adapter->num_rx_desc)
nextp = 0;
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
nbuf = &rxr->rx_buffers[nextp];
prefetch(nbuf);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
** The header mbuf is ONLY used when header
** split is enabled, otherwise we get normal
** behavior, ie, both header and payload
** are DMA'd into the payload buffer.
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
**
** Rather than using the fmp/lmp global pointers
** we now keep the head of a packet chain in the
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
** buffer struct and pass this along from one
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
** descriptor to the next, until we get EOP.
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (rxr->hdr_split && (rbuf->fmp == NULL)) {
/* This must be an initial descriptor */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
hlen = (hdr & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
IXGBE_RXDADV_HDRBUFLEN_SHIFT;
if (hlen > IXGBE_RX_HDR)
hlen = IXGBE_RX_HDR;
mh->m_len = hlen;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
mh->m_flags |= M_PKTHDR;
mh->m_next = NULL;
mh->m_pkthdr.len = mh->m_len;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Null buf pointer so it is refreshed */
rbuf->m_head = NULL;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
** Check the payload length, this
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
** could be zero if its a small
** packet.
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if (plen > 0) {
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
mp->m_len = plen;
mp->m_next = NULL;
mp->m_flags &= ~M_PKTHDR;
mh->m_next = mp;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
mh->m_pkthdr.len += mp->m_len;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Null buf pointer so it is refreshed */
rbuf->m_pack = NULL;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rxr->rx_split_packets++;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/*
** Now create the forward
** chain so when complete
** we wont have to.
*/
if (eop == 0) {
/* stash the chain head */
nbuf->fmp = mh;
/* Make forward chain */
if (plen)
mp->m_next = nbuf->m_pack;
else
mh->m_next = nbuf->m_pack;
} else {
/* Singlet, prepare to send */
sendmp = mh;
if (staterr & IXGBE_RXD_STAT_VP) {
sendmp->m_pkthdr.ether_vtag = vtag;
sendmp->m_flags |= M_VLANTAG;
}
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
} else {
/*
** Either no header split, or a
** secondary piece of a fragmented
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
** split packet.
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
mp->m_len = plen;
/*
** See if there is a stored head
** that determines what we are
*/
sendmp = rbuf->fmp;
rbuf->m_pack = rbuf->fmp = NULL;
if (sendmp != NULL) /* secondary frag */
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
sendmp->m_pkthdr.len += mp->m_len;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
else {
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* first desc of a non-ps chain */
sendmp = mp;
sendmp->m_flags |= M_PKTHDR;
sendmp->m_pkthdr.len = mp->m_len;
if (staterr & IXGBE_RXD_STAT_VP) {
sendmp->m_pkthdr.ether_vtag = vtag;
sendmp->m_flags |= M_VLANTAG;
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
}
/* Pass the head pointer on */
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (eop == 0) {
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
nbuf->fmp = sendmp;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
sendmp = NULL;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
mp->m_next = nbuf->m_pack;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
++processed;
/* Sending this frame? */
if (eop) {
sendmp->m_pkthdr.rcvif = ifp;
ifp->if_ipackets++;
rxr->rx_packets++;
/* capture data for AIM */
rxr->bytes += sendmp->m_pkthdr.len;
rxr->rx_bytes += sendmp->m_pkthdr.len;
if ((ifp->if_capenable & IFCAP_RXCSUM) != 0)
ixgbe_rx_checksum(staterr, sendmp);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
#if __FreeBSD_version >= 800000
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
sendmp->m_pkthdr.flowid = que->msix;
sendmp->m_flags |= M_FLOWID;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
#endif
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
next_desc:
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Advance our pointers to the next descriptor. */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (++i == adapter->num_rx_desc)
i = 0;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Now send to the stack or do LRO */
if (sendmp != NULL)
ixgbe_rx_input(rxr, ifp, sendmp, ptype);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Every 8 descriptors we go to refresh mbufs */
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (processed == 8) {
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_refresh_mbufs(rxr, i);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
processed = 0;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
/* Refresh any remaining buf structs */
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (processed != 0) {
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
ixgbe_refresh_mbufs(rxr, i);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
processed = 0;
}
rxr->next_to_check = i;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/*
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
* Flush any outstanding LRO work
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
SLIST_REMOVE_HEAD(&lro->lro_active, next);
tcp_lro_flush(lro, queued);
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_RX_UNLOCK(rxr);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/*
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
** We still have cleaning to do?
** Schedule another interrupt if so.
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
if ((staterr & IXGBE_RXD_STAT_DD) != 0) {
ixgbe_rearm_queues(adapter, (u64)(1 << que->msix));
return (TRUE);
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
return (FALSE);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*********************************************************************
*
* Verify that the hardware indicated that the checksum is valid.
* Inform the stack about the status of checksum so that stack
* doesn't spend time verifying the checksum.
*
*********************************************************************/
static void
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
ixgbe_rx_checksum(u32 staterr, struct mbuf * mp)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
u16 status = (u16) staterr;
u8 errors = (u8) (staterr >> 24);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
if (status & IXGBE_RXD_STAT_IPCS) {
if (!(errors & IXGBE_RXD_ERR_IPE)) {
/* IP Checksum Good */
mp->m_pkthdr.csum_flags = CSUM_IP_CHECKED;
mp->m_pkthdr.csum_flags |= CSUM_IP_VALID;
} else
mp->m_pkthdr.csum_flags = 0;
}
if (status & IXGBE_RXD_STAT_L4CS) {
if (!(errors & IXGBE_RXD_ERR_TCPE)) {
mp->m_pkthdr.csum_flags |=
(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
mp->m_pkthdr.csum_data = htons(0xffff);
}
}
return;
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
/*
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
** This routine is run via an vlan config EVENT,
** it enables us to use the HW Filter table since
** we can get the vlan id. This just creates the
** entry in the soft version of the VFTA, init will
** repopulate the real table.
*/
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void
Improvement on the last change, this gives a precise way to tell the one and only interface that a vlan event is for. Thanks to John Baldwin for the patch. Approved by: re
2009-07-24 21:35:52 +00:00
ixgbe_register_vlan(void *arg, struct ifnet *ifp, u16 vtag)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
struct adapter *adapter = ifp->if_softc;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
u16 index, bit;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
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
if (ifp->if_softc != arg) /* Not our event */
This delta fixes two bugs: - When a vlan event occurs a check was not made that the event was actually for the interface, thus resulting in a panic. All three drivers have this vulnerability. Add a check for this condition. - Secondly, there was a duplicate buf_ring free in the em driver resulting in a panic on unload. Remove. Approved by: re
2009-07-24 16:57:49 +00:00
return;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if ((vtag == 0) || (vtag > 4095)) /* Invalid */
return;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
index = (vtag >> 5) & 0x7F;
bit = vtag & 0x1F;
ixgbe_shadow_vfta[index] |= (1 << bit);
++adapter->num_vlans;
/* Re-init to load the changes */
ixgbe_init(adapter);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
/*
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
** This routine is run via an vlan
** unconfig EVENT, remove our entry
** in the soft vfta.
*/
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
static void
Improvement on the last change, this gives a precise way to tell the one and only interface that a vlan event is for. Thanks to John Baldwin for the patch. Approved by: re
2009-07-24 21:35:52 +00:00
ixgbe_unregister_vlan(void *arg, struct ifnet *ifp, u16 vtag)
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
{
struct adapter *adapter = ifp->if_softc;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
u16 index, bit;
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
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
if (ifp->if_softc != arg)
This delta fixes two bugs: - When a vlan event occurs a check was not made that the event was actually for the interface, thus resulting in a panic. All three drivers have this vulnerability. Add a check for this condition. - Secondly, there was a duplicate buf_ring free in the em driver resulting in a panic on unload. Remove. Approved by: re
2009-07-24 16:57:49 +00:00
return;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if ((vtag == 0) || (vtag > 4095)) /* Invalid */
return;
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
index = (vtag >> 5) & 0x7F;
bit = vtag & 0x1F;
ixgbe_shadow_vfta[index] &= ~(1 << bit);
--adapter->num_vlans;
/* Re-init to load the changes */
ixgbe_init(adapter);
}
static void
ixgbe_setup_vlan_hw_support(struct adapter *adapter)
{
struct ixgbe_hw *hw = &adapter->hw;
u32 ctrl;
/*
** 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 < IXGBE_VFTA_SIZE; i++)
if (ixgbe_shadow_vfta[i] != 0)
IXGBE_WRITE_REG(hw, IXGBE_VFTA(i),
ixgbe_shadow_vfta[i]);
/* Enable the Filter Table */
ctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
ctrl &= ~IXGBE_VLNCTRL_CFIEN;
ctrl |= IXGBE_VLNCTRL_VFE;
if (hw->mac.type == ixgbe_mac_82598EB)
ctrl |= IXGBE_VLNCTRL_VME;
IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, ctrl);
/* On 82599 the VLAN enable is per/queue in RXDCTL */
if (hw->mac.type == ixgbe_mac_82599EB)
for (int i = 0; i < adapter->num_queues; i++) {
ctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
ctrl |= IXGBE_RXDCTL_VME;
IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(i), ctrl);
}
This updates the ixgbe driver to Intel internal version 1.4.7 Shared code changes, core driver fixes, vlan event/filter support Also Kip Macy's fix to allow any number of queues, thanks Kip!
2008-07-30 18:15:18 +00:00
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void
ixgbe_enable_intr(struct adapter *adapter)
{
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
struct ixgbe_hw *hw = &adapter->hw;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ix_queue *que = adapter->queues;
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Enable Fan Failure detection */
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
if (hw->device_id == IXGBE_DEV_ID_82598AT)
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
mask |= IXGBE_EIMS_GPI_SDP1;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/* 82599 specific interrupts */
if (adapter->hw.mac.type == ixgbe_mac_82599EB) {
mask |= IXGBE_EIMS_ECC;
mask |= IXGBE_EIMS_GPI_SDP1;
mask |= IXGBE_EIMS_GPI_SDP2;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#ifdef IXGBE_FDIR
mask |= IXGBE_EIMS_FLOW_DIR;
#endif
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* With RSS we use auto clear */
if (adapter->msix_mem) {
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
mask = IXGBE_EIMS_ENABLE_MASK;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Dont autoclear Link */
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
mask &= ~IXGBE_EIMS_OTHER;
mask &= ~IXGBE_EIMS_LSC;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
IXGBE_WRITE_REG(hw, IXGBE_EIAC, mask);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
/*
** Now enable all queues, this is done seperately to
** allow for handling the extended (beyond 32) MSIX
** vectors that can be used by 82599
*/
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
for (int i = 0; i < adapter->num_queues; i++, que++)
ixgbe_enable_queue(adapter, que->msix);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_WRITE_FLUSH(hw);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
static void
ixgbe_disable_intr(struct adapter *adapter)
{
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
if (adapter->msix_mem)
IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, 0);
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
} else {
IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
IXGBE_WRITE_FLUSH(&adapter->hw);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
u16
ixgbe_read_pci_cfg(struct ixgbe_hw *hw, u32 reg)
{
u16 value;
value = pci_read_config(((struct ixgbe_osdep *)hw->back)->dev,
reg, 2);
return (value);
}
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
void
ixgbe_write_pci_cfg(struct ixgbe_hw *hw, u32 reg, u16 value)
{
pci_write_config(((struct ixgbe_osdep *)hw->back)->dev,
reg, value, 2);
return;
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
/*
** Setup the correct IVAR register for a particular MSIX interrupt
** (yes this is all very magic and confusing :)
** - entry is the register array entry
** - vector is the MSIX vector for this queue
** - type is RX/TX/MISC
*/
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void
Update for the Intel 10G driver, this adds support for newest hardware, adds multiqueue tx interface, infrastructure cleanup to allow up to 32 MSIX vectors on newer Nehalem systems. Bug fixes, etc.
2009-06-24 18:27:07 +00:00
ixgbe_set_ivar(struct adapter *adapter, u8 entry, u8 vector, s8 type)
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
{
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
struct ixgbe_hw *hw = &adapter->hw;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
u32 ivar, index;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
vector |= IXGBE_IVAR_ALLOC_VAL;
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
switch (hw->mac.type) {
case ixgbe_mac_82598EB:
if (type == -1)
entry = IXGBE_IVAR_OTHER_CAUSES_INDEX;
else
entry += (type * 64);
index = (entry >> 2) & 0x1F;
ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
ivar &= ~(0xFF << (8 * (entry & 0x3)));
ivar |= (vector << (8 * (entry & 0x3)));
IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR(index), ivar);
break;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
case ixgbe_mac_82599EB:
if (type == -1) { /* MISC IVAR */
index = (entry & 1) * 8;
ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC);
ivar &= ~(0xFF << index);
ivar |= (vector << index);
IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar);
} else { /* RX/TX IVARS */
index = (16 * (entry & 1)) + (8 * type);
ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(entry >> 1));
ivar &= ~(0xFF << index);
ivar |= (vector << index);
IXGBE_WRITE_REG(hw, IXGBE_IVAR(entry >> 1), ivar);
}
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
default:
break;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
static void
ixgbe_configure_ivars(struct adapter *adapter)
{
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ix_queue *que = adapter->queues;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
for (int i = 0; i < adapter->num_queues; i++, que++) {
/* First the RX queue entry */
ixgbe_set_ivar(adapter, i, que->msix, 0);
/* ... and the TX */
ixgbe_set_ivar(adapter, i, que->msix, 1);
/* Set an Initial EITR value */
IXGBE_WRITE_REG(&adapter->hw,
IXGBE_EITR(que->msix), IXGBE_LOW_LATENCY);
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/* For the Link interrupt */
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
ixgbe_set_ivar(adapter, 1, adapter->linkvec, -1);
}
/*
** ixgbe_sfp_probe - called in the local timer to
** determine if a port had optics inserted.
*/
static bool ixgbe_sfp_probe(struct adapter *adapter)
{
struct ixgbe_hw *hw = &adapter->hw;
device_t dev = adapter->dev;
bool result = FALSE;
if ((hw->phy.type == ixgbe_phy_nl) &&
(hw->phy.sfp_type == ixgbe_sfp_type_not_present)) {
s32 ret = hw->phy.ops.identify_sfp(hw);
if (ret)
goto out;
ret = hw->phy.ops.reset(hw);
if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) {
device_printf(dev,"Unsupported SFP+ module detected!");
printf(" Reload driver with supported module.\n");
adapter->sfp_probe = FALSE;
goto out;
} else
device_printf(dev,"SFP+ module detected!\n");
/* We now have supported optics */
adapter->sfp_probe = FALSE;
result = TRUE;
}
out:
return (result);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
}
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
/*
** Tasklet handler for MSIX Link interrupts
** - do outside interrupt since it might sleep
*/
static void
ixgbe_handle_link(void *context, int pending)
{
struct adapter *adapter = context;
ixgbe_check_link(&adapter->hw,
&adapter->link_speed, &adapter->link_up, 0);
ixgbe_update_link_status(adapter);
}
/*
** Tasklet for handling SFP module interrupts
*/
static void
ixgbe_handle_mod(void *context, int pending)
{
struct adapter *adapter = context;
struct ixgbe_hw *hw = &adapter->hw;
device_t dev = adapter->dev;
u32 err;
err = hw->phy.ops.identify_sfp(hw);
if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
device_printf(dev,
"Unsupported SFP+ module type was detected.\n");
return;
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
err = hw->mac.ops.setup_sfp(hw);
if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
device_printf(dev,
"Setup failure - unsupported SFP+ module type.\n");
return;
}
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
taskqueue_enqueue(adapter->tq, &adapter->msf_task);
return;
}
/*
** Tasklet for handling MSF (multispeed fiber) interrupts
*/
static void
ixgbe_handle_msf(void *context, int pending)
{
struct adapter *adapter = context;
struct ixgbe_hw *hw = &adapter->hw;
u32 autoneg;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
bool negotiate;
autoneg = hw->phy.autoneg_advertised;
if ((!autoneg) && (hw->mac.ops.get_link_capabilities))
hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiate);
if (hw->mac.ops.setup_link)
hw->mac.ops.setup_link(hw, autoneg, negotiate, TRUE);
#if 0
ixgbe_check_link(&adapter->hw, &speed, &adapter->link_up, 0);
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
ixgbe_update_link_status(adapter);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#endif
return;
}
#ifdef IXGBE_FDIR
/*
** Tasklet for reinitializing the Flow Director filter table
*/
static void
ixgbe_reinit_fdir(void *context, int pending)
{
struct adapter *adapter = context;
Remove extraneous semicolons, no functional changes. Submitted by: Marc Balmer <marc@msys.ch> MFC after: 1 week
2010-01-07 21:01:37 +00:00
struct ifnet *ifp = adapter->ifp;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
if (adapter->fdir_reinit != 1) /* Shouldn't happen */
return;
ixgbe_reinit_fdir_tables_82599(&adapter->hw);
adapter->fdir_reinit = 0;
/* Restart the interface */
ifp->if_drv_flags |= IFF_DRV_RUNNING;
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
return;
}
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
#endif
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/**********************************************************************
*
* Update the board statistics counters.
*
**********************************************************************/
static void
ixgbe_update_stats_counters(struct adapter *adapter)
{
Remove extraneous semicolons, no functional changes. Submitted by: Marc Balmer <marc@msys.ch> MFC after: 1 week
2010-01-07 21:01:37 +00:00
struct ifnet *ifp = adapter->ifp;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
struct ixgbe_hw *hw = &adapter->hw;
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
u32 missed_rx = 0, bprc, lxon, lxoff, total;
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
u64 total_missed_rx = 0;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
for (int i = 0; i < 8; i++) {
Stats missed packet handling was still not quite right, thanks to Dmitrij Tejblum for the correction, need a variable with scope only within the for loop for all queues. MFC: 3 days
2009-09-11 00:00:23 +00:00
u32 mp;
mp = IXGBE_READ_REG(hw, IXGBE_MPC(i));
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
/* missed_rx tallies misses for the gprc workaround */
Stats missed packet handling was still not quite right, thanks to Dmitrij Tejblum for the correction, need a variable with scope only within the for loop for all queues. MFC: 3 days
2009-09-11 00:00:23 +00:00
missed_rx += mp;
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
/* global total per queue */
Stats missed packet handling was still not quite right, thanks to Dmitrij Tejblum for the correction, need a variable with scope only within the for loop for all queues. MFC: 3 days
2009-09-11 00:00:23 +00:00
adapter->stats.mpc[i] += mp;
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
/* Running comprehensive total for stats display */
total_missed_rx += adapter->stats.mpc[i];
if (hw->mac.type == ixgbe_mac_82598EB)
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
/* Hardware workaround, gprc counts missed packets */
adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
adapter->stats.gprc -= missed_rx;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
if (hw->mac.type == ixgbe_mac_82599EB) {
adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
IXGBE_READ_REG(hw, IXGBE_GORCH); /* clears register */
adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
IXGBE_READ_REG(hw, IXGBE_GOTCH); /* clears register */
adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORL);
IXGBE_READ_REG(hw, IXGBE_TORH); /* clears register */
adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
} else {
adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
/* 82598 only has a counter in the high register */
adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
}
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
/*
* Workaround: mprc hardware is incorrectly counting
* broadcasts, so for now we subtract those.
*/
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
adapter->stats.bprc += bprc;
adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
adapter->stats.mprc -= bprc;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
adapter->stats.lxontxc += lxon;
lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
adapter->stats.lxofftxc += lxoff;
total = lxon + lxoff;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
adapter->stats.gptc -= total;
adapter->stats.mptc -= total;
adapter->stats.ptc64 -= total;
adapter->stats.gotc -= total * ETHER_MIN_LEN;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
/* Fill out the OS statistics structure */
ifp->if_ipackets = adapter->stats.gprc;
ifp->if_opackets = adapter->stats.gptc;
ifp->if_ibytes = adapter->stats.gorc;
ifp->if_obytes = adapter->stats.gotc;
ifp->if_imcasts = adapter->stats.mprc;
ifp->if_collisions = 0;
/* Rx Errors */
A couple of changes: first, make header split on by default and remove from its dependency on LRO, my tests have shown that its always beneficial, even when doing bridging. Second, fix up a few problems in the statistics code, the adapter dependencies had gotten lost so some code that should only run on 82599 was always running, this resulted in bogus flow control numbers on 82598.
2009-08-24 20:41:51 +00:00
ifp->if_ierrors = total_missed_rx + adapter->stats.crcerrs +
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
adapter->stats.rlec;
}
/**********************************************************************
*
* This routine is called only when ixgbe_display_debug_stats is enabled.
* This routine provides a way to take a look at important statistics
* maintained by the driver and hardware.
*
**********************************************************************/
static void
ixgbe_print_hw_stats(struct adapter * adapter)
{
device_t dev = adapter->dev;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"Std Mbuf Failed = %lu\n",
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
adapter->mbuf_defrag_failed);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"Missed Packets = %llu\n",
(long long)adapter->stats.mpc[0]);
device_printf(dev,"Receive length errors = %llu\n",
((long long)adapter->stats.roc +
(long long)adapter->stats.ruc));
device_printf(dev,"Crc errors = %llu\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
(long long)adapter->stats.crcerrs);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"Driver dropped packets = %lu\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
adapter->dropped_pkts);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_printf(dev, "watchdog timeouts = %ld\n",
adapter->watchdog_events);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"XON Rcvd = %llu\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
(long long)adapter->stats.lxonrxc);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"XON Xmtd = %llu\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
(long long)adapter->stats.lxontxc);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"XOFF Rcvd = %llu\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
(long long)adapter->stats.lxoffrxc);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"XOFF Xmtd = %llu\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
(long long)adapter->stats.lxofftxc);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"Total Packets Rcvd = %llu\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
(long long)adapter->stats.tpr);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"Good Packets Rcvd = %llu\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
(long long)adapter->stats.gprc);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"Good Packets Xmtd = %llu\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
(long long)adapter->stats.gptc);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"TSO Transmissions = %lu\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
adapter->tso_tx);
return;
}
/**********************************************************************
*
* This routine is called only when em_display_debug_stats is enabled.
* This routine provides a way to take a look at important statistics
* maintained by the driver and hardware.
*
**********************************************************************/
static void
ixgbe_print_debug_info(struct adapter *adapter)
{
device_t dev = adapter->dev;
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
struct ixgbe_hw *hw = &adapter->hw;
struct ix_queue *que = adapter->queues;
struct rx_ring *rxr;
struct tx_ring *txr;
struct lro_ctrl *lro;
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
device_printf(dev,"Error Byte Count = %u \n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
IXGBE_READ_REG(hw, IXGBE_ERRBC));
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
for (int i = 0; i < adapter->num_queues; i++, que++) {
txr = que->txr;
rxr = que->rxr;
lro = &rxr->lro;
device_printf(dev,"QUE(%d) IRQs Handled: %lu\n",
que->msix, (long)que->irqs);
device_printf(dev,"RX[%d]: rdh = %d, hw rdt = %d\n",
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
i, IXGBE_READ_REG(hw, IXGBE_RDH(i)),
IXGBE_READ_REG(hw, IXGBE_RDT(i)));
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
device_printf(dev,"TX[%d] tdh = %d, hw tdt = %d\n", i,
IXGBE_READ_REG(hw, IXGBE_TDH(i)),
IXGBE_READ_REG(hw, IXGBE_TDT(i)));
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
device_printf(dev,"RX(%d) Packets Received: %lld\n",
rxr->me, (long long)rxr->rx_packets);
device_printf(dev,"RX(%d) Split RX Packets: %lld\n",
rxr->me, (long long)rxr->rx_split_packets);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_printf(dev,"RX(%d) Bytes Received: %lu\n",
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
rxr->me, (long)rxr->rx_bytes);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_printf(dev,"RX(%d) LRO Queued= %d\n",
rxr->me, lro->lro_queued);
device_printf(dev,"RX(%d) LRO Flushed= %d\n",
rxr->me, lro->lro_flushed);
Update driver to Intel version 2.0.7: This adds new feature support for the 82599, a hardware assist to LRO, doing this required a large revamp to the RX cleanup code because the descriptor ring may not be processed out of order, this necessitated the elimination of global pointers. Additionally, the RX routine now does not refresh mbufs on every descriptor, rather it will do a range, and then update the hardware pointer at that time. These are performance oriented changes. The TX side now has a cleaner simpler watchdog algorithm as well, in TX cleanup a read of ticks is stored, that can then be compared in local_timer to determine if there is a hang. Various other cleanups along the way, thanks to all who have provided input and testing.
2009-12-07 21:30:54 +00:00
device_printf(dev,"RX(%d) HW LRO Merges= %lu\n",
rxr->me, (long)rxr->rsc_num);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_printf(dev,"TX(%d) Packets Sent: %lu\n",
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
txr->me, (long)txr->total_packets);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
device_printf(dev,"TX(%d) NO Desc Avail: %lu\n",
Update the driver to Intel version 2.1.6 - add some new hardware support for 82599 - Big change to interrupt architecture, it now uses a queue which contains an RX/TX pair as the recipient of the interrupt. This will reduce overall system interrupts/msix usage. - Improved RX mbuf handling: the old get_buf routine is no longer synchronized with rxeof, this allows the elimination of packet discards due to mbuf allocation failure. - Much simplified and improved AIM code, it now happens in the queue interrupt context and takes into account both the traffic on the RX AND TX side. - variety of small tweaks, like ring size, that have been seen as performance improvements. - Thanks to those that provided feedback or suggested changes, I hope I've caught all of them.
2010-03-27 00:21:40 +00:00
txr->me, (long)txr->no_desc_avail);
This is driver version 1.4.4 of the Intel ixgbe driver. -It has new hardware support -It uses a new method of TX cleanup called Head Write Back -It includes the provisional generic TCP LRO feature contributed by Myricom and made general purpose by me. This should move into the stack upon approval but for this driver drop its in here. -Also bug fixes and etc... MFC in a week if no serious issues arise.
2008-05-16 18:46:30 +00:00
}
device_printf(dev,"Link IRQ Handled: %lu\n",
(long)adapter->link_irq);
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
return;
}
static int
ixgbe_sysctl_stats(SYSCTL_HANDLER_ARGS)
{
int error;
int result;
struct adapter *adapter;
result = -1;
error = sysctl_handle_int(oidp, &result, 0, req);
if (error || !req->newptr)
return (error);
if (result == 1) {
adapter = (struct adapter *) arg1;
ixgbe_print_hw_stats(adapter);
}
return error;
}
static int
ixgbe_sysctl_debug(SYSCTL_HANDLER_ARGS)
{
int error, result;
struct adapter *adapter;
result = -1;
error = sysctl_handle_int(oidp, &result, 0, req);
if (error || !req->newptr)
return (error);
if (result == 1) {
adapter = (struct adapter *) arg1;
ixgbe_print_debug_info(adapter);
}
return error;
}
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
/*
** Set flow control using sysctl:
** Flow control values:
** 0 - off
** 1 - rx pause
** 2 - tx pause
** 3 - full
*/
static int
ixgbe_set_flowcntl(SYSCTL_HANDLER_ARGS)
{
int error;
struct adapter *adapter;
error = sysctl_handle_int(oidp, &ixgbe_flow_control, 0, req);
if (error)
return (error);
adapter = (struct adapter *) arg1;
switch (ixgbe_flow_control) {
case ixgbe_fc_rx_pause:
case ixgbe_fc_tx_pause:
case ixgbe_fc_full:
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
adapter->hw.fc.requested_mode = ixgbe_flow_control;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
break;
case ixgbe_fc_none:
default:
Updated ixgbe driver - version 1.6.2 -This version has header split, and as a result a number of aspects of the code have been improved/simplified. - Interrupt handling refined for performance - Many small bugs fixed along the way MFC after: ASAP - in time for 7.1
2008-11-26 23:41:18 +00:00
adapter->hw.fc.requested_mode = ixgbe_fc_none;
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
}
Update the ixgbe driver to 1.7.4, this includes support for the new 82599 adapter family, adds header split, and many small fixes. The driver should now be added to the GENERIC kernel. MFC after: 2 weeks
2009-04-10 00:22:48 +00:00
ixgbe_fc_enable(&adapter->hw, 0);
This is an update to the new Intel 10G 82598 driver. The first drop was Beta, this code is expected to be the release version. Note that this driver code will build in either 6.2 or 7. If you use the code in 6.2 you will not get TSO or MSI/X support but it will function in a legacy mode. Approved by: re
2007-09-04 02:31:35 +00:00
return error;
}
Add the actual source too :) Approved by: re
2007-07-11 23:03:16 +00:00
static void
ixgbe_add_rx_process_limit(struct adapter *adapter, const char *name,
const char *description, int *limit, int value)
{
*limit = value;
SYSCTL_ADD_INT(device_get_sysctl_ctx(adapter->dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(adapter->dev)),
OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW, limit, value, description);
}
Reference in New Issue Copy Permalink