d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
3f929d8cdd
freebsd-dev/sys/dev/e1000/if_lem.c

4648 lines
136 KiB
C
Raw Normal View History

Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/******************************************************************************
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
Copyright (c) 2001-2012, Intel Corporation
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +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.
******************************************************************************/
/*$FreeBSD$*/
#ifdef HAVE_KERNEL_OPTION_HEADERS
#include "opt_device_polling.h"
#include "opt_inet.h"
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
#include "opt_inet6.h"
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
#endif
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/endian.h>
#include <sys/kernel.h>
#include <sys/kthread.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/module.h>
#include <sys/rman.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>
#include <sys/eventhandler.h>
#include <machine/bus.h>
#include <machine/resource.h>
#include <net/bpf.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <net/if_dl.h>
#include <net/if_media.h>
#include <net/if_types.h>
#include <net/if_vlan_var.h>
#include <netinet/in_systm.h>
#include <netinet/in.h>
#include <netinet/if_ether.h>
#include <netinet/ip.h>
#include <netinet/ip6.h>
#include <netinet/tcp.h>
#include <netinet/udp.h>
#include <machine/in_cksum.h>
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
#include <dev/led/led.h>
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
#include <dev/pci/pcivar.h>
#include <dev/pci/pcireg.h>
#include "e1000_api.h"
#include "if_lem.h"
/*********************************************************************
* Legacy Em Driver version:
*********************************************************************/
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
char lem_driver_version[] = "1.0.5";
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/*********************************************************************
* PCI Device ID Table
*
* Used by probe to select devices to load on
* Last field stores an index into e1000_strings
* Last entry must be all 0s
*
* { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index }
*********************************************************************/
static em_vendor_info_t lem_vendor_info_array[] =
{
/* Intel(R) PRO/1000 Network Connection */
{ 0x8086, E1000_DEV_ID_82540EM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82540EM_LOM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82540EP, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82540EP_LOM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82540EP_LP, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82541EI, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82541ER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82541ER_LOM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82541EI_MOBILE, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82541GI, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82541GI_LF, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82541GI_MOBILE, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82542, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82543GC_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82543GC_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82544EI_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82544EI_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82544GC_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82544GC_LOM, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82545EM_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82545EM_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82545GM_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82545GM_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82545GM_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82546EB_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82546EB_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82546EB_QUAD_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82546GB_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82546GB_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82546GB_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82546GB_PCIE, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82546GB_QUAD_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3,
PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82547EI, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82547EI_MOBILE, PCI_ANY_ID, PCI_ANY_ID, 0},
{ 0x8086, E1000_DEV_ID_82547GI, PCI_ANY_ID, PCI_ANY_ID, 0},
/* required last entry */
{ 0, 0, 0, 0, 0}
};
/*********************************************************************
* Table of branding strings for all supported NICs.
*********************************************************************/
static char *lem_strings[] = {
"Intel(R) PRO/1000 Legacy Network Connection"
};
/*********************************************************************
* Function prototypes
*********************************************************************/
static int lem_probe(device_t);
static int lem_attach(device_t);
static int lem_detach(device_t);
static int lem_shutdown(device_t);
static int lem_suspend(device_t);
static int lem_resume(device_t);
static void lem_start(struct ifnet *);
static void lem_start_locked(struct ifnet *ifp);
static int lem_ioctl(struct ifnet *, u_long, caddr_t);
static void lem_init(void *);
static void lem_init_locked(struct adapter *);
static void lem_stop(void *);
static void lem_media_status(struct ifnet *, struct ifmediareq *);
static int lem_media_change(struct ifnet *);
static void lem_identify_hardware(struct adapter *);
static int lem_allocate_pci_resources(struct adapter *);
static int lem_allocate_irq(struct adapter *adapter);
static void lem_free_pci_resources(struct adapter *);
static void lem_local_timer(void *);
static int lem_hardware_init(struct adapter *);
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
static int lem_setup_interface(device_t, struct adapter *);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
static void lem_setup_transmit_structures(struct adapter *);
static void lem_initialize_transmit_unit(struct adapter *);
static int lem_setup_receive_structures(struct adapter *);
static void lem_initialize_receive_unit(struct adapter *);
static void lem_enable_intr(struct adapter *);
static void lem_disable_intr(struct adapter *);
static void lem_free_transmit_structures(struct adapter *);
static void lem_free_receive_structures(struct adapter *);
static void lem_update_stats_counters(struct adapter *);
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
static void lem_add_hw_stats(struct adapter *adapter);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
static void lem_txeof(struct adapter *);
static void lem_tx_purge(struct adapter *);
static int lem_allocate_receive_structures(struct adapter *);
static int lem_allocate_transmit_structures(struct adapter *);
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
static bool lem_rxeof(struct adapter *, int, int *);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
#ifndef __NO_STRICT_ALIGNMENT
static int lem_fixup_rx(struct adapter *);
#endif
static void lem_receive_checksum(struct adapter *, struct e1000_rx_desc *,
struct mbuf *);
static void lem_transmit_checksum_setup(struct adapter *, struct mbuf *,
u32 *, u32 *);
static void lem_set_promisc(struct adapter *);
static void lem_disable_promisc(struct adapter *);
static void lem_set_multi(struct adapter *);
static void lem_update_link_status(struct adapter *);
static int lem_get_buf(struct adapter *, int);
static void lem_register_vlan(void *, struct ifnet *, u16);
static void lem_unregister_vlan(void *, struct ifnet *, u16);
static void lem_setup_vlan_hw_support(struct adapter *);
static int lem_xmit(struct adapter *, struct mbuf **);
static void lem_smartspeed(struct adapter *);
static int lem_82547_fifo_workaround(struct adapter *, int);
static void lem_82547_update_fifo_head(struct adapter *, int);
static int lem_82547_tx_fifo_reset(struct adapter *);
static void lem_82547_move_tail(void *);
static int lem_dma_malloc(struct adapter *, bus_size_t,
struct em_dma_alloc *, int);
static void lem_dma_free(struct adapter *, struct em_dma_alloc *);
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
static int lem_sysctl_nvm_info(SYSCTL_HANDLER_ARGS);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
static void lem_print_nvm_info(struct adapter *);
static int lem_is_valid_ether_addr(u8 *);
static u32 lem_fill_descriptors (bus_addr_t address, u32 length,
PDESC_ARRAY desc_array);
static int lem_sysctl_int_delay(SYSCTL_HANDLER_ARGS);
static void lem_add_int_delay_sysctl(struct adapter *, const char *,
const char *, struct em_int_delay_info *, int, int);
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
static void lem_set_flow_cntrl(struct adapter *, const char *,
const char *, int *, int);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Management and WOL Support */
static void lem_init_manageability(struct adapter *);
static void lem_release_manageability(struct adapter *);
static void lem_get_hw_control(struct adapter *);
static void lem_release_hw_control(struct adapter *);
static void lem_get_wakeup(device_t);
static void lem_enable_wakeup(device_t);
static int lem_enable_phy_wakeup(struct adapter *);
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
static void lem_led_func(void *, int);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
static void lem_intr(void *);
static int lem_irq_fast(void *);
static void lem_handle_rxtx(void *context, int pending);
static void lem_handle_link(void *context, int pending);
static void lem_add_rx_process_limit(struct adapter *, const char *,
const char *, int *, int);
#ifdef DEVICE_POLLING
Fix poll handler declaration.
2010-03-31 16:42:22 +00:00
static poll_handler_t lem_poll;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
#endif /* POLLING */
/*********************************************************************
* FreeBSD Device Interface Entry Points
*********************************************************************/
static device_method_t lem_methods[] = {
/* Device interface */
DEVMETHOD(device_probe, lem_probe),
DEVMETHOD(device_attach, lem_attach),
DEVMETHOD(device_detach, lem_detach),
DEVMETHOD(device_shutdown, lem_shutdown),
DEVMETHOD(device_suspend, lem_suspend),
DEVMETHOD(device_resume, lem_resume),
Use DEVMETHOD_END macro defined in sys/bus.h instead of {0, 0} sentinel on device_method_t arrays Reviewed by: cognet Approved by: cognet
2013-01-30 18:01:20 +00:00
DEVMETHOD_END
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
};
static driver_t lem_driver = {
"em", lem_methods, sizeof(struct adapter),
};
extern devclass_t em_devclass;
DRIVER_MODULE(lem, pci, lem_driver, em_devclass, 0, 0);
MODULE_DEPEND(lem, pci, 1, 1, 1);
MODULE_DEPEND(lem, ether, 1, 1, 1);
/*********************************************************************
* Tunable default values.
*********************************************************************/
#define EM_TICKS_TO_USECS(ticks) ((1024 * (ticks) + 500) / 1000)
#define EM_USECS_TO_TICKS(usecs) ((1000 * (usecs) + 512) / 1024)
static int lem_tx_int_delay_dflt = EM_TICKS_TO_USECS(EM_TIDV);
static int lem_rx_int_delay_dflt = EM_TICKS_TO_USECS(EM_RDTR);
static int lem_tx_abs_int_delay_dflt = EM_TICKS_TO_USECS(EM_TADV);
static int lem_rx_abs_int_delay_dflt = EM_TICKS_TO_USECS(EM_RADV);
static int lem_rxd = EM_DEFAULT_RXD;
static int lem_txd = EM_DEFAULT_TXD;
static int lem_smart_pwr_down = FALSE;
/* Controls whether promiscuous also shows bad packets */
static int lem_debug_sbp = FALSE;
TUNABLE_INT("hw.em.tx_int_delay", &lem_tx_int_delay_dflt);
TUNABLE_INT("hw.em.rx_int_delay", &lem_rx_int_delay_dflt);
TUNABLE_INT("hw.em.tx_abs_int_delay", &lem_tx_abs_int_delay_dflt);
TUNABLE_INT("hw.em.rx_abs_int_delay", &lem_rx_abs_int_delay_dflt);
TUNABLE_INT("hw.em.rxd", &lem_rxd);
TUNABLE_INT("hw.em.txd", &lem_txd);
TUNABLE_INT("hw.em.smart_pwr_down", &lem_smart_pwr_down);
TUNABLE_INT("hw.em.sbp", &lem_debug_sbp);
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
/* Interrupt style - default to fast */
static int lem_use_legacy_irq = 0;
TUNABLE_INT("hw.em.use_legacy_irq", &lem_use_legacy_irq);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* How many packets rxeof tries to clean at a time */
static int lem_rx_process_limit = 100;
TUNABLE_INT("hw.em.rx_process_limit", &lem_rx_process_limit);
/* Flow control setting - default to FULL */
static int lem_fc_setting = e1000_fc_full;
TUNABLE_INT("hw.em.fc_setting", &lem_fc_setting);
/* Global used in WOL setup with multiport cards */
static int global_quad_port_a = 0;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP /* see ixgbe.c for details */
#include <dev/netmap/if_lem_netmap.h>
#endif /* DEV_NETMAP */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/*********************************************************************
* Device identification routine
*
* em_probe determines if the driver should be loaded on
* adapter based on PCI vendor/device id of the adapter.
*
* return BUS_PROBE_DEFAULT on success, positive on failure
*********************************************************************/
static int
lem_probe(device_t dev)
{
char adapter_name[60];
u16 pci_vendor_id = 0;
u16 pci_device_id = 0;
u16 pci_subvendor_id = 0;
u16 pci_subdevice_id = 0;
em_vendor_info_t *ent;
INIT_DEBUGOUT("em_probe: begin");
pci_vendor_id = pci_get_vendor(dev);
if (pci_vendor_id != EM_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 = lem_vendor_info_array;
while (ent->vendor_id != 0) {
if ((pci_vendor_id == ent->vendor_id) &&
(pci_device_id == ent->device_id) &&
((pci_subvendor_id == ent->subvendor_id) ||
(ent->subvendor_id == PCI_ANY_ID)) &&
((pci_subdevice_id == ent->subdevice_id) ||
(ent->subdevice_id == PCI_ANY_ID))) {
sprintf(adapter_name, "%s %s",
lem_strings[ent->index],
lem_driver_version);
device_set_desc_copy(dev, adapter_name);
return (BUS_PROBE_DEFAULT);
}
ent++;
}
return (ENXIO);
}
/*********************************************************************
* Device initialization routine
*
* The attach entry point is called when the driver is being loaded.
* This routine identifies the type of hardware, allocates all resources
* and initializes the hardware.
*
* return 0 on success, positive on failure
*********************************************************************/
static int
lem_attach(device_t dev)
{
struct adapter *adapter;
int tsize, rsize;
int error = 0;
INIT_DEBUGOUT("lem_attach: begin");
adapter = device_get_softc(dev);
adapter->dev = adapter->osdep.dev = dev;
EM_CORE_LOCK_INIT(adapter, device_get_nameunit(dev));
EM_TX_LOCK_INIT(adapter, device_get_nameunit(dev));
EM_RX_LOCK_INIT(adapter, device_get_nameunit(dev));
/* SYSCTL stuff */
SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
OID_AUTO, "nvm", CTLTYPE_INT|CTLFLAG_RW, adapter, 0,
lem_sysctl_nvm_info, "I", "NVM Information");
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
callout_init_mtx(&adapter->timer, &adapter->core_mtx, 0);
callout_init_mtx(&adapter->tx_fifo_timer, &adapter->tx_mtx, 0);
/* Determine hardware and mac info */
lem_identify_hardware(adapter);
/* Setup PCI resources */
if (lem_allocate_pci_resources(adapter)) {
device_printf(dev, "Allocation of PCI resources failed\n");
error = ENXIO;
goto err_pci;
}
/* Do Shared Code initialization */
if (e1000_setup_init_funcs(&adapter->hw, TRUE)) {
device_printf(dev, "Setup of Shared code failed\n");
error = ENXIO;
goto err_pci;
}
e1000_get_bus_info(&adapter->hw);
/* Set up some sysctls for the tunable interrupt delays */
lem_add_int_delay_sysctl(adapter, "rx_int_delay",
"receive interrupt delay in usecs", &adapter->rx_int_delay,
E1000_REGISTER(&adapter->hw, E1000_RDTR), lem_rx_int_delay_dflt);
lem_add_int_delay_sysctl(adapter, "tx_int_delay",
"transmit interrupt delay in usecs", &adapter->tx_int_delay,
E1000_REGISTER(&adapter->hw, E1000_TIDV), lem_tx_int_delay_dflt);
if (adapter->hw.mac.type >= e1000_82540) {
lem_add_int_delay_sysctl(adapter, "rx_abs_int_delay",
"receive interrupt delay limit in usecs",
&adapter->rx_abs_int_delay,
E1000_REGISTER(&adapter->hw, E1000_RADV),
lem_rx_abs_int_delay_dflt);
lem_add_int_delay_sysctl(adapter, "tx_abs_int_delay",
"transmit interrupt delay limit in usecs",
&adapter->tx_abs_int_delay,
E1000_REGISTER(&adapter->hw, E1000_TADV),
lem_tx_abs_int_delay_dflt);
}
/* Sysctls for limiting the amount of work done in the taskqueue */
lem_add_rx_process_limit(adapter, "rx_processing_limit",
"max number of rx packets to process", &adapter->rx_process_limit,
lem_rx_process_limit);
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
/* Sysctl for setting the interface flow control */
lem_set_flow_cntrl(adapter, "flow_control",
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
"flow control setting",
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
&adapter->fc_setting, lem_fc_setting);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/*
* Validate number of transmit and receive descriptors. It
* must not exceed hardware maximum, and must be multiple
* of E1000_DBA_ALIGN.
*/
if (((lem_txd * sizeof(struct e1000_tx_desc)) % EM_DBA_ALIGN) != 0 ||
(adapter->hw.mac.type >= e1000_82544 && lem_txd > EM_MAX_TXD) ||
(adapter->hw.mac.type < e1000_82544 && lem_txd > EM_MAX_TXD_82543) ||
(lem_txd < EM_MIN_TXD)) {
device_printf(dev, "Using %d TX descriptors instead of %d!\n",
EM_DEFAULT_TXD, lem_txd);
adapter->num_tx_desc = EM_DEFAULT_TXD;
} else
adapter->num_tx_desc = lem_txd;
if (((lem_rxd * sizeof(struct e1000_rx_desc)) % EM_DBA_ALIGN) != 0 ||
(adapter->hw.mac.type >= e1000_82544 && lem_rxd > EM_MAX_RXD) ||
(adapter->hw.mac.type < e1000_82544 && lem_rxd > EM_MAX_RXD_82543) ||
(lem_rxd < EM_MIN_RXD)) {
device_printf(dev, "Using %d RX descriptors instead of %d!\n",
EM_DEFAULT_RXD, lem_rxd);
adapter->num_rx_desc = EM_DEFAULT_RXD;
} else
adapter->num_rx_desc = lem_rxd;
adapter->hw.mac.autoneg = DO_AUTO_NEG;
adapter->hw.phy.autoneg_wait_to_complete = FALSE;
adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
adapter->rx_buffer_len = 2048;
e1000_init_script_state_82541(&adapter->hw, TRUE);
e1000_set_tbi_compatibility_82543(&adapter->hw, TRUE);
/* Copper options */
if (adapter->hw.phy.media_type == e1000_media_type_copper) {
adapter->hw.phy.mdix = AUTO_ALL_MODES;
adapter->hw.phy.disable_polarity_correction = FALSE;
adapter->hw.phy.ms_type = EM_MASTER_SLAVE;
}
/*
* Set the frame limits assuming
* standard ethernet sized frames.
*/
adapter->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHERNET_FCS_SIZE;
adapter->min_frame_size = ETH_ZLEN + ETHERNET_FCS_SIZE;
/*
* This controls when hardware reports transmit completion
* status.
*/
adapter->hw.mac.report_tx_early = 1;
tsize = roundup2(adapter->num_tx_desc * sizeof(struct e1000_tx_desc),
EM_DBA_ALIGN);
/* Allocate Transmit Descriptor ring */
if (lem_dma_malloc(adapter, tsize, &adapter->txdma, BUS_DMA_NOWAIT)) {
device_printf(dev, "Unable to allocate tx_desc memory\n");
error = ENOMEM;
goto err_tx_desc;
}
adapter->tx_desc_base =
(struct e1000_tx_desc *)adapter->txdma.dma_vaddr;
rsize = roundup2(adapter->num_rx_desc * sizeof(struct e1000_rx_desc),
EM_DBA_ALIGN);
/* Allocate Receive Descriptor ring */
if (lem_dma_malloc(adapter, rsize, &adapter->rxdma, BUS_DMA_NOWAIT)) {
device_printf(dev, "Unable to allocate rx_desc memory\n");
error = ENOMEM;
goto err_rx_desc;
}
adapter->rx_desc_base =
(struct e1000_rx_desc *)adapter->rxdma.dma_vaddr;
Do not allocate multicast array memory in multicast filter configuration function. For failed memory allocations, em(4)/lem(4) called panic(9) which is not acceptable on production box. igb(4)/ixgb(4)/ix(4) allocated the required memory in stack which consumed 768 bytes of stack memory which looks too big. To address these issues, allocate multicast array memory in device attach time and make multicast configuration success under any conditions. This change also removes the excessive use of memory in stack. Reviewed by: jfv
2010-08-28 00:34:22 +00:00
/* Allocate multicast array memory. */
adapter->mta = malloc(sizeof(u8) * ETH_ADDR_LEN *
MAX_NUM_MULTICAST_ADDRESSES, M_DEVBUF, M_NOWAIT);
if (adapter->mta == NULL) {
device_printf(dev, "Can not allocate multicast setup array\n");
error = ENOMEM;
goto err_hw_init;
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/*
** Start from a known state, this is
** important in reading the nvm and
** mac from that.
*/
e1000_reset_hw(&adapter->hw);
/* Make sure we have a good EEPROM before we read from it */
if (e1000_validate_nvm_checksum(&adapter->hw) < 0) {
/*
** Some PCI-E parts fail the first check due to
** the link being in sleep state, call it again,
** if it fails a second time its a real issue.
*/
if (e1000_validate_nvm_checksum(&adapter->hw) < 0) {
device_printf(dev,
"The EEPROM Checksum Is Not Valid\n");
error = EIO;
goto err_hw_init;
}
}
/* Copy the permanent MAC address out of the EEPROM */
if (e1000_read_mac_addr(&adapter->hw) < 0) {
device_printf(dev, "EEPROM read error while reading MAC"
" address\n");
error = EIO;
goto err_hw_init;
}
if (!lem_is_valid_ether_addr(adapter->hw.mac.addr)) {
device_printf(dev, "Invalid MAC address\n");
error = EIO;
goto err_hw_init;
}
/* Initialize the hardware */
if (lem_hardware_init(adapter)) {
device_printf(dev, "Unable to initialize the hardware\n");
error = EIO;
goto err_hw_init;
}
/* Allocate transmit descriptors and buffers */
if (lem_allocate_transmit_structures(adapter)) {
device_printf(dev, "Could not setup transmit structures\n");
error = ENOMEM;
goto err_tx_struct;
}
/* Allocate receive descriptors and buffers */
if (lem_allocate_receive_structures(adapter)) {
device_printf(dev, "Could not setup receive structures\n");
error = ENOMEM;
goto err_rx_struct;
}
/*
** Do interrupt configuration
*/
error = lem_allocate_irq(adapter);
if (error)
goto err_rx_struct;
/*
* Get Wake-on-Lan and Management info for later use
*/
lem_get_wakeup(dev);
/* Setup OS specific network interface */
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
if (lem_setup_interface(dev, adapter) != 0)
goto err_rx_struct;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Initialize statistics */
lem_update_stats_counters(adapter);
adapter->hw.mac.get_link_status = 1;
lem_update_link_status(adapter);
/* Indicate SOL/IDER usage */
if (e1000_check_reset_block(&adapter->hw))
device_printf(dev,
"PHY reset is blocked due to SOL/IDER session.\n");
/* Do we need workaround for 82544 PCI-X adapter? */
if (adapter->hw.bus.type == e1000_bus_type_pcix &&
adapter->hw.mac.type == e1000_82544)
adapter->pcix_82544 = TRUE;
else
adapter->pcix_82544 = FALSE;
/* Register for VLAN events */
adapter->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
lem_register_vlan, adapter, EVENTHANDLER_PRI_FIRST);
adapter->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
lem_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST);
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
lem_add_hw_stats(adapter);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Non-AMT based hardware can now take control from firmware */
if (adapter->has_manage && !adapter->has_amt)
lem_get_hw_control(adapter);
/* Tell the stack that the interface is not active */
adapter->ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
adapter->led_dev = led_create(lem_led_func, adapter,
device_get_nameunit(dev));
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
lem_netmap_attach(adapter);
#endif /* DEV_NETMAP */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
INIT_DEBUGOUT("lem_attach: end");
return (0);
err_rx_struct:
lem_free_transmit_structures(adapter);
err_tx_struct:
err_hw_init:
lem_release_hw_control(adapter);
lem_dma_free(adapter, &adapter->rxdma);
err_rx_desc:
lem_dma_free(adapter, &adapter->txdma);
err_tx_desc:
err_pci:
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
if (adapter->ifp != NULL)
if_free(adapter->ifp);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
lem_free_pci_resources(adapter);
Do not allocate multicast array memory in multicast filter configuration function. For failed memory allocations, em(4)/lem(4) called panic(9) which is not acceptable on production box. igb(4)/ixgb(4)/ix(4) allocated the required memory in stack which consumed 768 bytes of stack memory which looks too big. To address these issues, allocate multicast array memory in device attach time and make multicast configuration success under any conditions. This change also removes the excessive use of memory in stack. Reviewed by: jfv
2010-08-28 00:34:22 +00:00
free(adapter->mta, M_DEVBUF);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
EM_TX_LOCK_DESTROY(adapter);
EM_RX_LOCK_DESTROY(adapter);
EM_CORE_LOCK_DESTROY(adapter);
return (error);
}
/*********************************************************************
* Device removal routine
*
* The detach entry point is called when the driver is being removed.
* This routine stops the adapter and deallocates all the resources
* that were allocated for driver operation.
*
* return 0 on success, positive on failure
*********************************************************************/
static int
lem_detach(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
struct ifnet *ifp = adapter->ifp;
INIT_DEBUGOUT("em_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);
}
#ifdef DEVICE_POLLING
if (ifp->if_capenable & IFCAP_POLLING)
ether_poll_deregister(ifp);
#endif
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
if (adapter->led_dev != NULL)
led_destroy(adapter->led_dev);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
EM_CORE_LOCK(adapter);
EM_TX_LOCK(adapter);
adapter->in_detach = 1;
lem_stop(adapter);
e1000_phy_hw_reset(&adapter->hw);
lem_release_manageability(adapter);
EM_TX_UNLOCK(adapter);
EM_CORE_UNLOCK(adapter);
/* 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);
ether_ifdetach(adapter->ifp);
callout_drain(&adapter->timer);
callout_drain(&adapter->tx_fifo_timer);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
netmap_detach(ifp);
#endif /* DEV_NETMAP */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
lem_free_pci_resources(adapter);
bus_generic_detach(dev);
if_free(ifp);
lem_free_transmit_structures(adapter);
lem_free_receive_structures(adapter);
/* Free Transmit Descriptor ring */
if (adapter->tx_desc_base) {
lem_dma_free(adapter, &adapter->txdma);
adapter->tx_desc_base = NULL;
}
/* Free Receive Descriptor ring */
if (adapter->rx_desc_base) {
lem_dma_free(adapter, &adapter->rxdma);
adapter->rx_desc_base = NULL;
}
lem_release_hw_control(adapter);
Do not allocate multicast array memory in multicast filter configuration function. For failed memory allocations, em(4)/lem(4) called panic(9) which is not acceptable on production box. igb(4)/ixgb(4)/ix(4) allocated the required memory in stack which consumed 768 bytes of stack memory which looks too big. To address these issues, allocate multicast array memory in device attach time and make multicast configuration success under any conditions. This change also removes the excessive use of memory in stack. Reviewed by: jfv
2010-08-28 00:34:22 +00:00
free(adapter->mta, M_DEVBUF);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
EM_TX_LOCK_DESTROY(adapter);
EM_RX_LOCK_DESTROY(adapter);
EM_CORE_LOCK_DESTROY(adapter);
return (0);
}
/*********************************************************************
*
* Shutdown entry point
*
**********************************************************************/
static int
lem_shutdown(device_t dev)
{
return lem_suspend(dev);
}
/*
* Suspend/resume device methods.
*/
static int
lem_suspend(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
EM_CORE_LOCK(adapter);
lem_release_manageability(adapter);
lem_release_hw_control(adapter);
lem_enable_wakeup(dev);
EM_CORE_UNLOCK(adapter);
return bus_generic_suspend(dev);
}
static int
lem_resume(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
struct ifnet *ifp = adapter->ifp;
EM_CORE_LOCK(adapter);
lem_init_locked(adapter);
lem_init_manageability(adapter);
EM_CORE_UNLOCK(adapter);
lem_start(ifp);
return bus_generic_resume(dev);
}
static void
lem_start_locked(struct ifnet *ifp)
{
struct adapter *adapter = ifp->if_softc;
struct mbuf *m_head;
EM_TX_LOCK_ASSERT(adapter);
if ((ifp->if_drv_flags & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) !=
IFF_DRV_RUNNING)
return;
if (!adapter->link_active)
return;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
/*
* Force a cleanup if number of TX descriptors
* available hits the threshold
*/
if (adapter->num_tx_desc_avail <= EM_TX_CLEANUP_THRESHOLD) {
lem_txeof(adapter);
/* Now do we at least have a minimal? */
if (adapter->num_tx_desc_avail <= EM_TX_OP_THRESHOLD) {
adapter->no_tx_desc_avail1++;
return;
}
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
if (m_head == NULL)
break;
/*
* Encapsulation can modify our pointer, and or make it
* NULL on failure. In that event, we can't requeue.
*/
if (lem_xmit(adapter, &m_head)) {
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 */
ETHER_BPF_MTAP(ifp, m_head);
/* Set timeout in case hardware has problems transmitting. */
adapter->watchdog_check = TRUE;
Remove multiqueue stack related stuff form lem, it is unneeded for legacy hardware. Also remove some TSO related cruft. Add some watchdog_time setting that was missing, thanks to Mikolaj Golub for pointing that out.
2010-04-14 18:29:01 +00:00
adapter->watchdog_time = ticks;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
if (adapter->num_tx_desc_avail <= EM_TX_OP_THRESHOLD)
ifp->if_drv_flags |= IFF_DRV_OACTIVE;
return;
}
static void
lem_start(struct ifnet *ifp)
{
struct adapter *adapter = ifp->if_softc;
EM_TX_LOCK(adapter);
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
lem_start_locked(ifp);
EM_TX_UNLOCK(adapter);
}
/*********************************************************************
* Ioctl entry point
*
* em_ioctl is called when the user wants to configure the
* interface.
*
* return 0 on success, positive on failure
**********************************************************************/
static int
lem_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
{
struct adapter *adapter = ifp->if_softc;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
struct ifreq *ifr = (struct ifreq *)data;
#if defined(INET) || defined(INET6)
struct ifaddr *ifa = (struct ifaddr *)data;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
#endif
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
bool avoid_reset = FALSE;
int error = 0;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (adapter->in_detach)
return (error);
switch (command) {
case SIOCSIFADDR:
#ifdef INET
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
if (ifa->ifa_addr->sa_family == AF_INET)
avoid_reset = TRUE;
#endif
#ifdef INET6
if (ifa->ifa_addr->sa_family == AF_INET6)
avoid_reset = TRUE;
#endif
/*
** Calling init results in link renegotiation,
** so we avoid doing it when possible.
*/
if (avoid_reset) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
ifp->if_flags |= IFF_UP;
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
lem_init(adapter);
#ifdef INET
if (!(ifp->if_flags & IFF_NOARP))
arp_ifinit(ifp, ifa);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
#endif
Part 2 of 2 New deltas for the 1G drivers. There have still been intermittent problems with apparent TX hangs for some customers. These have been problematic to reproduce but I believe these changes will address them. Testing on a number of fronts have been positive. EM: there is an important 'chicken bit' fix for 82574 in the shared code this is supported in the core here. - The TX path has been tightened up to improve performance. In particular UDP with jumbo frames was having problems, and the changes here have improved that. - OACTIVE has been used more carefully on the theory that some hangs may be due to a problem in this interaction - Problems with the RX init code, the "lazy" allocation and ring initialization has been found to cause problems in some newer client systems, and as it really is not that big a win (its not in a hot path) it seems best to remove it. - HWTSO was broken when VLAN HWTAGGING or HWFILTER is used, I found this was due to an error in setting up the descriptors in em_xmit. IGB: - TX is also improved here. With multiqueue I realized its very important to handle OACTIVE only under the CORE lock so there are no races between the queues. - Flow Control handling was broken in a couple ways, I have changed and I hope improved that in this delta. - UDP also had a problem in the TX path here, it was change to improve that. - On some hardware, with the driver static, a weird stray interrupt seems to sometimes fire and cause a panic in the RX mbuf refresh code. This is addressed by setting interrupts late in the init path, and also to set all interrupts bits off at the start of that.
2011-12-10 07:08:52 +00:00
} else
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
error = ether_ioctl(ifp, command, data);
break;
case SIOCSIFMTU:
{
int max_frame_size;
IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFMTU (Set Interface MTU)");
EM_CORE_LOCK(adapter);
switch (adapter->hw.mac.type) {
case e1000_82542:
max_frame_size = ETHER_MAX_LEN;
break;
default:
max_frame_size = MAX_JUMBO_FRAME_SIZE;
}
if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN -
ETHER_CRC_LEN) {
EM_CORE_UNLOCK(adapter);
error = EINVAL;
break;
}
ifp->if_mtu = ifr->ifr_mtu;
adapter->max_frame_size =
ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
lem_init_locked(adapter);
EM_CORE_UNLOCK(adapter);
break;
}
case SIOCSIFFLAGS:
IOCTL_DEBUGOUT("ioctl rcv'd:\
SIOCSIFFLAGS (Set Interface Flags)");
EM_CORE_LOCK(adapter);
if (ifp->if_flags & IFF_UP) {
if ((ifp->if_drv_flags & IFF_DRV_RUNNING)) {
if ((ifp->if_flags ^ adapter->if_flags) &
(IFF_PROMISC | IFF_ALLMULTI)) {
lem_disable_promisc(adapter);
lem_set_promisc(adapter);
}
} else
lem_init_locked(adapter);
} else
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
EM_TX_LOCK(adapter);
lem_stop(adapter);
EM_TX_UNLOCK(adapter);
}
adapter->if_flags = ifp->if_flags;
EM_CORE_UNLOCK(adapter);
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
IOCTL_DEBUGOUT("ioctl rcv'd: SIOC(ADD|DEL)MULTI");
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
EM_CORE_LOCK(adapter);
lem_disable_intr(adapter);
lem_set_multi(adapter);
if (adapter->hw.mac.type == e1000_82542 &&
adapter->hw.revision_id == E1000_REVISION_2) {
lem_initialize_receive_unit(adapter);
}
#ifdef DEVICE_POLLING
if (!(ifp->if_capenable & IFCAP_POLLING))
#endif
lem_enable_intr(adapter);
EM_CORE_UNLOCK(adapter);
}
break;
case SIOCSIFMEDIA:
/* Check SOL/IDER usage */
EM_CORE_LOCK(adapter);
if (e1000_check_reset_block(&adapter->hw)) {
EM_CORE_UNLOCK(adapter);
device_printf(adapter->dev, "Media change is"
" blocked due to SOL/IDER session.\n");
break;
}
EM_CORE_UNLOCK(adapter);
case SIOCGIFMEDIA:
IOCTL_DEBUGOUT("ioctl rcv'd: \
SIOCxIFMEDIA (Get/Set Interface Media)");
error = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
break;
case SIOCSIFCAP:
{
int mask, reinit;
IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFCAP (Set Capabilities)");
reinit = 0;
mask = ifr->ifr_reqcap ^ ifp->if_capenable;
#ifdef DEVICE_POLLING
if (mask & IFCAP_POLLING) {
if (ifr->ifr_reqcap & IFCAP_POLLING) {
error = ether_poll_register(lem_poll, ifp);
if (error)
return (error);
EM_CORE_LOCK(adapter);
lem_disable_intr(adapter);
ifp->if_capenable |= IFCAP_POLLING;
EM_CORE_UNLOCK(adapter);
} else {
error = ether_poll_deregister(ifp);
/* Enable interrupt even in error case */
EM_CORE_LOCK(adapter);
lem_enable_intr(adapter);
ifp->if_capenable &= ~IFCAP_POLLING;
EM_CORE_UNLOCK(adapter);
}
}
#endif
if (mask & IFCAP_HWCSUM) {
ifp->if_capenable ^= IFCAP_HWCSUM;
reinit = 1;
}
if (mask & IFCAP_VLAN_HWTAGGING) {
ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
reinit = 1;
}
if ((mask & IFCAP_WOL) &&
(ifp->if_capabilities & IFCAP_WOL) != 0) {
if (mask & IFCAP_WOL_MCAST)
ifp->if_capenable ^= IFCAP_WOL_MCAST;
if (mask & IFCAP_WOL_MAGIC)
ifp->if_capenable ^= IFCAP_WOL_MAGIC;
}
if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING))
lem_init(adapter);
VLAN_CAPABILITIES(ifp);
break;
}
default:
error = ether_ioctl(ifp, command, data);
break;
}
return (error);
}
/*********************************************************************
* Init entry point
*
* This routine is used in two ways. It is used by the stack as
* init entry point in network interface structure. It is also used
* by the driver as a hw/sw initialization routine to get to a
* consistent state.
*
* return 0 on success, positive on failure
**********************************************************************/
static void
lem_init_locked(struct adapter *adapter)
{
struct ifnet *ifp = adapter->ifp;
device_t dev = adapter->dev;
u32 pba;
INIT_DEBUGOUT("lem_init: begin");
EM_CORE_LOCK_ASSERT(adapter);
EM_TX_LOCK(adapter);
lem_stop(adapter);
EM_TX_UNLOCK(adapter);
/*
* Packet Buffer Allocation (PBA)
* Writing PBA sets the receive portion of the buffer
* the remainder is used for the transmit buffer.
*
* Devices before the 82547 had a Packet Buffer of 64K.
* Default allocation: PBA=48K for Rx, leaving 16K for Tx.
* After the 82547 the buffer was reduced to 40K.
* Default allocation: PBA=30K for Rx, leaving 10K for Tx.
* Note: default does not leave enough room for Jumbo Frame >10k.
*/
switch (adapter->hw.mac.type) {
case e1000_82547:
case e1000_82547_rev_2: /* 82547: Total Packet Buffer is 40K */
if (adapter->max_frame_size > 8192)
pba = E1000_PBA_22K; /* 22K for Rx, 18K for Tx */
else
pba = E1000_PBA_30K; /* 30K for Rx, 10K for Tx */
adapter->tx_fifo_head = 0;
adapter->tx_head_addr = pba << EM_TX_HEAD_ADDR_SHIFT;
adapter->tx_fifo_size =
(E1000_PBA_40K - pba) << EM_PBA_BYTES_SHIFT;
break;
default:
/* Devices before 82547 had a Packet Buffer of 64K. */
if (adapter->max_frame_size > 8192)
pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */
else
pba = E1000_PBA_48K; /* 48K for Rx, 16K for Tx */
}
INIT_DEBUGOUT1("lem_init: pba=%dK",pba);
E1000_WRITE_REG(&adapter->hw, E1000_PBA, pba);
/* Get the latest mac address, User can use a LAA */
bcopy(IF_LLADDR(adapter->ifp), adapter->hw.mac.addr,
ETHER_ADDR_LEN);
/* Put the address into the Receive Address Array */
e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
/* Initialize the hardware */
if (lem_hardware_init(adapter)) {
device_printf(dev, "Unable to initialize the hardware\n");
return;
}
lem_update_link_status(adapter);
/* Setup VLAN support, basic and offload if available */
E1000_WRITE_REG(&adapter->hw, E1000_VET, ETHERTYPE_VLAN);
/* Set hardware offload abilities */
ifp->if_hwassist = 0;
if (adapter->hw.mac.type >= e1000_82543) {
if (ifp->if_capenable & IFCAP_TXCSUM)
ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP);
}
/* Configure for OS presence */
lem_init_manageability(adapter);
/* Prepare transmit descriptors and buffers */
lem_setup_transmit_structures(adapter);
lem_initialize_transmit_unit(adapter);
/* Setup Multicast table */
lem_set_multi(adapter);
/* Prepare receive descriptors and buffers */
if (lem_setup_receive_structures(adapter)) {
device_printf(dev, "Could not setup receive structures\n");
EM_TX_LOCK(adapter);
lem_stop(adapter);
EM_TX_UNLOCK(adapter);
return;
}
lem_initialize_receive_unit(adapter);
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
/* Use real VLAN Filter support? */
if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
/* Use real VLAN Filter support */
lem_setup_vlan_hw_support(adapter);
else {
u32 ctrl;
ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL);
ctrl |= E1000_CTRL_VME;
E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl);
}
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Don't lose promiscuous settings */
lem_set_promisc(adapter);
ifp->if_drv_flags |= IFF_DRV_RUNNING;
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
callout_reset(&adapter->timer, hz, lem_local_timer, adapter);
e1000_clear_hw_cntrs_base_generic(&adapter->hw);
#ifdef DEVICE_POLLING
/*
* Only enable interrupts if we are not polling, make sure
* they are off otherwise.
*/
if (ifp->if_capenable & IFCAP_POLLING)
lem_disable_intr(adapter);
else
#endif /* DEVICE_POLLING */
lem_enable_intr(adapter);
/* AMT based hardware can now take control from firmware */
if (adapter->has_manage && adapter->has_amt)
lem_get_hw_control(adapter);
}
static void
lem_init(void *arg)
{
struct adapter *adapter = arg;
EM_CORE_LOCK(adapter);
lem_init_locked(adapter);
EM_CORE_UNLOCK(adapter);
}
#ifdef DEVICE_POLLING
/*********************************************************************
*
* Legacy polling routine
*
*********************************************************************/
static int
lem_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
{
struct adapter *adapter = ifp->if_softc;
u32 reg_icr, rx_done = 0;
EM_CORE_LOCK(adapter);
if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
EM_CORE_UNLOCK(adapter);
return (rx_done);
}
if (cmd == POLL_AND_CHECK_STATUS) {
reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
callout_stop(&adapter->timer);
adapter->hw.mac.get_link_status = 1;
lem_update_link_status(adapter);
callout_reset(&adapter->timer, hz,
lem_local_timer, adapter);
}
}
EM_CORE_UNLOCK(adapter);
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
lem_rxeof(adapter, count, &rx_done);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
EM_TX_LOCK(adapter);
lem_txeof(adapter);
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
lem_start_locked(ifp);
EM_TX_UNLOCK(adapter);
return (rx_done);
}
#endif /* DEVICE_POLLING */
/*********************************************************************
*
* Legacy Interrupt Service routine
*
*********************************************************************/
static void
lem_intr(void *arg)
{
struct adapter *adapter = arg;
struct ifnet *ifp = adapter->ifp;
u32 reg_icr;
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
if ((ifp->if_capenable & IFCAP_POLLING) ||
((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0))
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
return;
EM_CORE_LOCK(adapter);
reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
if (reg_icr & E1000_ICR_RXO)
adapter->rx_overruns++;
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
if ((reg_icr == 0xffffffff) || (reg_icr == 0)) {
EM_CORE_UNLOCK(adapter);
return;
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
callout_stop(&adapter->timer);
adapter->hw.mac.get_link_status = 1;
lem_update_link_status(adapter);
/* Deal with TX cruft when link lost */
lem_tx_purge(adapter);
callout_reset(&adapter->timer, hz,
lem_local_timer, adapter);
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
EM_CORE_UNLOCK(adapter);
return;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
EM_CORE_UNLOCK(adapter);
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
lem_rxeof(adapter, -1, NULL);
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
EM_TX_LOCK(adapter);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
lem_txeof(adapter);
if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
lem_start_locked(ifp);
EM_TX_UNLOCK(adapter);
return;
}
static void
lem_handle_link(void *context, int pending)
{
struct adapter *adapter = context;
struct ifnet *ifp = adapter->ifp;
if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
return;
EM_CORE_LOCK(adapter);
callout_stop(&adapter->timer);
lem_update_link_status(adapter);
/* Deal with TX cruft when link lost */
lem_tx_purge(adapter);
callout_reset(&adapter->timer, hz, lem_local_timer, adapter);
EM_CORE_UNLOCK(adapter);
}
/* Combined RX/TX handler, used by Legacy and MSI */
static void
lem_handle_rxtx(void *context, int pending)
{
struct adapter *adapter = context;
struct ifnet *ifp = adapter->ifp;
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
lem_rxeof(adapter, adapter->rx_process_limit, NULL);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
EM_TX_LOCK(adapter);
lem_txeof(adapter);
if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
lem_start_locked(ifp);
EM_TX_UNLOCK(adapter);
}
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
lem_enable_intr(adapter);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
/*********************************************************************
*
* Fast Legacy/MSI Combined Interrupt Service routine
*
*********************************************************************/
static int
lem_irq_fast(void *arg)
{
struct adapter *adapter = arg;
struct ifnet *ifp;
u32 reg_icr;
ifp = adapter->ifp;
reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR);
/* Hot eject? */
if (reg_icr == 0xffffffff)
return FILTER_STRAY;
/* Definitely not our interrupt. */
if (reg_icr == 0x0)
return FILTER_STRAY;
/*
* Mask interrupts until the taskqueue is finished running. This is
* cheap, just assume that it is needed. This also works around the
* MSI message reordering errata on certain systems.
*/
lem_disable_intr(adapter);
taskqueue_enqueue(adapter->tq, &adapter->rxtx_task);
/* Link status change */
if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
adapter->hw.mac.get_link_status = 1;
taskqueue_enqueue(taskqueue_fast, &adapter->link_task);
}
if (reg_icr & E1000_ICR_RXO)
adapter->rx_overruns++;
return FILTER_HANDLED;
}
/*********************************************************************
*
* Media Ioctl callback
*
* This routine is called whenever the user queries the status of
* the interface using ifconfig.
*
**********************************************************************/
static void
lem_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
{
struct adapter *adapter = ifp->if_softc;
u_char fiber_type = IFM_1000_SX;
INIT_DEBUGOUT("lem_media_status: begin");
EM_CORE_LOCK(adapter);
lem_update_link_status(adapter);
ifmr->ifm_status = IFM_AVALID;
ifmr->ifm_active = IFM_ETHER;
if (!adapter->link_active) {
EM_CORE_UNLOCK(adapter);
return;
}
ifmr->ifm_status |= IFM_ACTIVE;
if ((adapter->hw.phy.media_type == e1000_media_type_fiber) ||
(adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) {
if (adapter->hw.mac.type == e1000_82545)
fiber_type = IFM_1000_LX;
ifmr->ifm_active |= fiber_type | IFM_FDX;
} else {
switch (adapter->link_speed) {
case 10:
ifmr->ifm_active |= IFM_10_T;
break;
case 100:
ifmr->ifm_active |= IFM_100_TX;
break;
case 1000:
ifmr->ifm_active |= IFM_1000_T;
break;
}
if (adapter->link_duplex == FULL_DUPLEX)
ifmr->ifm_active |= IFM_FDX;
else
ifmr->ifm_active |= IFM_HDX;
}
EM_CORE_UNLOCK(adapter);
}
/*********************************************************************
*
* Media Ioctl callback
*
* This routine is called when the user changes speed/duplex using
* media/mediopt option with ifconfig.
*
**********************************************************************/
static int
lem_media_change(struct ifnet *ifp)
{
struct adapter *adapter = ifp->if_softc;
struct ifmedia *ifm = &adapter->media;
INIT_DEBUGOUT("lem_media_change: begin");
if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
return (EINVAL);
EM_CORE_LOCK(adapter);
switch (IFM_SUBTYPE(ifm->ifm_media)) {
case IFM_AUTO:
adapter->hw.mac.autoneg = DO_AUTO_NEG;
adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
break;
case IFM_1000_LX:
case IFM_1000_SX:
case IFM_1000_T:
adapter->hw.mac.autoneg = DO_AUTO_NEG;
adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
break;
case IFM_100_TX:
adapter->hw.mac.autoneg = FALSE;
adapter->hw.phy.autoneg_advertised = 0;
if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL;
else
adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF;
break;
case IFM_10_T:
adapter->hw.mac.autoneg = FALSE;
adapter->hw.phy.autoneg_advertised = 0;
if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL;
else
adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF;
break;
default:
device_printf(adapter->dev, "Unsupported media type\n");
}
lem_init_locked(adapter);
EM_CORE_UNLOCK(adapter);
return (0);
}
/*********************************************************************
*
* This routine maps the mbufs to tx descriptors.
*
* return 0 on success, positive on failure
**********************************************************************/
static int
lem_xmit(struct adapter *adapter, struct mbuf **m_headp)
{
bus_dma_segment_t segs[EM_MAX_SCATTER];
bus_dmamap_t map;
struct em_buffer *tx_buffer, *tx_buffer_mapped;
struct e1000_tx_desc *ctxd = NULL;
struct mbuf *m_head;
u32 txd_upper, txd_lower, txd_used, txd_saved;
int error, nsegs, i, j, first, last = 0;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
m_head = *m_headp;
txd_upper = txd_lower = txd_used = txd_saved = 0;
Fix for a panic when TX checksum offload is done and a packet has only a header in the first mbuf, the checksum code will dereference a pointer into the non-existing IP header. Do a check for the size and pullup if needed. Thanks to Michael Tuexen for this fix. MFC: asap - should be in 8.1 IMHO
2010-07-12 21:47:30 +00:00
/*
** When doing checksum offload, it is critical to
** make sure the first mbuf has more than header,
** because that routine expects data to be present.
*/
if ((m_head->m_pkthdr.csum_flags & CSUM_OFFLOAD) &&
(m_head->m_len < ETHER_HDR_LEN + sizeof(struct ip))) {
m_head = m_pullup(m_head, ETHER_HDR_LEN + sizeof(struct ip));
*m_headp = m_head;
if (m_head == NULL)
return (ENOBUFS);
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/*
* Map the packet for DMA
*
* Capture the first descriptor index,
* this descriptor will have the index
* of the EOP which is the only one that
* now gets a DONE bit writeback.
*/
first = adapter->next_avail_tx_desc;
tx_buffer = &adapter->tx_buffer_area[first];
tx_buffer_mapped = tx_buffer;
map = tx_buffer->map;
error = bus_dmamap_load_mbuf_sg(adapter->txtag, map,
*m_headp, segs, &nsegs, BUS_DMA_NOWAIT);
/*
* There are two types of errors we can (try) to handle:
* - EFBIG means the mbuf chain was too long and bus_dma ran
* out of segments. Defragment the mbuf chain and try again.
* - ENOMEM means bus_dma could not obtain enough bounce buffers
* at this point in time. Defer sending and try again later.
* All other errors, in particular EINVAL, are fatal and prevent the
* mbuf chain from ever going through. Drop it and report error.
*/
if (error == EFBIG) {
struct mbuf *m;
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_defrag(*m_headp, M_NOWAIT);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (m == NULL) {
adapter->mbuf_alloc_failed++;
m_freem(*m_headp);
*m_headp = NULL;
return (ENOBUFS);
}
*m_headp = m;
/* Try it again */
error = bus_dmamap_load_mbuf_sg(adapter->txtag, map,
*m_headp, segs, &nsegs, BUS_DMA_NOWAIT);
if (error) {
adapter->no_tx_dma_setup++;
m_freem(*m_headp);
*m_headp = NULL;
return (error);
}
} else if (error != 0) {
adapter->no_tx_dma_setup++;
return (error);
}
if (nsegs > (adapter->num_tx_desc_avail - 2)) {
adapter->no_tx_desc_avail2++;
bus_dmamap_unload(adapter->txtag, map);
return (ENOBUFS);
}
m_head = *m_headp;
/* Do hardware assists */
if (m_head->m_pkthdr.csum_flags & CSUM_OFFLOAD)
lem_transmit_checksum_setup(adapter, m_head,
&txd_upper, &txd_lower);
i = adapter->next_avail_tx_desc;
if (adapter->pcix_82544)
txd_saved = i;
/* Set up our transmit descriptors */
for (j = 0; j < nsegs; j++) {
bus_size_t seg_len;
bus_addr_t seg_addr;
/* If adapter is 82544 and on PCIX bus */
if(adapter->pcix_82544) {
DESC_ARRAY desc_array;
u32 array_elements, counter;
/*
* Check the Address and Length combination and
* split the data accordingly
*/
array_elements = lem_fill_descriptors(segs[j].ds_addr,
segs[j].ds_len, &desc_array);
for (counter = 0; counter < array_elements; counter++) {
if (txd_used == adapter->num_tx_desc_avail) {
adapter->next_avail_tx_desc = txd_saved;
adapter->no_tx_desc_avail2++;
bus_dmamap_unload(adapter->txtag, map);
return (ENOBUFS);
}
tx_buffer = &adapter->tx_buffer_area[i];
ctxd = &adapter->tx_desc_base[i];
ctxd->buffer_addr = htole64(
desc_array.descriptor[counter].address);
ctxd->lower.data = htole32(
(adapter->txd_cmd | txd_lower | (u16)
desc_array.descriptor[counter].length));
ctxd->upper.data =
htole32((txd_upper));
last = i;
if (++i == adapter->num_tx_desc)
i = 0;
tx_buffer->m_head = NULL;
tx_buffer->next_eop = -1;
txd_used++;
}
} else {
tx_buffer = &adapter->tx_buffer_area[i];
ctxd = &adapter->tx_desc_base[i];
seg_addr = segs[j].ds_addr;
seg_len = segs[j].ds_len;
ctxd->buffer_addr = htole64(seg_addr);
ctxd->lower.data = htole32(
adapter->txd_cmd | txd_lower | seg_len);
ctxd->upper.data =
htole32(txd_upper);
last = i;
if (++i == adapter->num_tx_desc)
i = 0;
tx_buffer->m_head = NULL;
tx_buffer->next_eop = -1;
}
}
adapter->next_avail_tx_desc = i;
if (adapter->pcix_82544)
adapter->num_tx_desc_avail -= txd_used;
else
adapter->num_tx_desc_avail -= nsegs;
if (m_head->m_flags & M_VLANTAG) {
/* Set the vlan id. */
ctxd->upper.fields.special =
htole16(m_head->m_pkthdr.ether_vtag);
/* Tell hardware to add tag */
ctxd->lower.data |= htole32(E1000_TXD_CMD_VLE);
}
tx_buffer->m_head = m_head;
tx_buffer_mapped->map = tx_buffer->map;
tx_buffer->map = map;
bus_dmamap_sync(adapter->txtag, map, BUS_DMASYNC_PREWRITE);
/*
* Last Descriptor of Packet
* needs End Of Packet (EOP)
* and Report Status (RS)
*/
ctxd->lower.data |=
htole32(E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS);
/*
* Keep track in the first buffer which
* descriptor will be written back
*/
tx_buffer = &adapter->tx_buffer_area[first];
tx_buffer->next_eop = last;
Update code from Intel: - Sync shared code with Intel internal - New client chipset support added - em driver - fixes to 82574, limit queues to 1 but use MSIX - em driver - large changes in TX checksum offload and tso code, thanks to yongari. - some small changes for watchdog issues. - igb driver - local timer watchdog code was missing locking this and a couple other watchdog related fixes. - bug in rx discard found by Andrew Boyer, check for null pointer MFC: a week
2010-09-28 00:13:15 +00:00
adapter->watchdog_time = ticks;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/*
* Advance the Transmit Descriptor Tail (TDT), this tells the E1000
* that this frame is available to transmit.
*/
bus_dmamap_sync(adapter->txdma.dma_tag, adapter->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
if (adapter->hw.mac.type == e1000_82547 &&
adapter->link_duplex == HALF_DUPLEX)
lem_82547_move_tail(adapter);
else {
E1000_WRITE_REG(&adapter->hw, E1000_TDT(0), i);
if (adapter->hw.mac.type == e1000_82547)
lem_82547_update_fifo_head(adapter,
m_head->m_pkthdr.len);
}
return (0);
}
/*********************************************************************
*
* 82547 workaround to avoid controller hang in half-duplex environment.
* The workaround is to avoid queuing a large packet that would span
* the internal Tx FIFO ring boundary. We need to reset the FIFO pointers
* in this case. We do that only when FIFO is quiescent.
*
**********************************************************************/
static void
lem_82547_move_tail(void *arg)
{
struct adapter *adapter = arg;
struct e1000_tx_desc *tx_desc;
u16 hw_tdt, sw_tdt, length = 0;
bool eop = 0;
EM_TX_LOCK_ASSERT(adapter);
hw_tdt = E1000_READ_REG(&adapter->hw, E1000_TDT(0));
sw_tdt = adapter->next_avail_tx_desc;
while (hw_tdt != sw_tdt) {
tx_desc = &adapter->tx_desc_base[hw_tdt];
length += tx_desc->lower.flags.length;
eop = tx_desc->lower.data & E1000_TXD_CMD_EOP;
if (++hw_tdt == adapter->num_tx_desc)
hw_tdt = 0;
if (eop) {
if (lem_82547_fifo_workaround(adapter, length)) {
adapter->tx_fifo_wrk_cnt++;
callout_reset(&adapter->tx_fifo_timer, 1,
lem_82547_move_tail, adapter);
break;
}
E1000_WRITE_REG(&adapter->hw, E1000_TDT(0), hw_tdt);
lem_82547_update_fifo_head(adapter, length);
length = 0;
}
}
}
static int
lem_82547_fifo_workaround(struct adapter *adapter, int len)
{
int fifo_space, fifo_pkt_len;
fifo_pkt_len = roundup2(len + EM_FIFO_HDR, EM_FIFO_HDR);
if (adapter->link_duplex == HALF_DUPLEX) {
fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
if (fifo_pkt_len >= (EM_82547_PKT_THRESH + fifo_space)) {
if (lem_82547_tx_fifo_reset(adapter))
return (0);
else
return (1);
}
}
return (0);
}
static void
lem_82547_update_fifo_head(struct adapter *adapter, int len)
{
int fifo_pkt_len = roundup2(len + EM_FIFO_HDR, EM_FIFO_HDR);
/* tx_fifo_head is always 16 byte aligned */
adapter->tx_fifo_head += fifo_pkt_len;
if (adapter->tx_fifo_head >= adapter->tx_fifo_size) {
adapter->tx_fifo_head -= adapter->tx_fifo_size;
}
}
static int
lem_82547_tx_fifo_reset(struct adapter *adapter)
{
u32 tctl;
if ((E1000_READ_REG(&adapter->hw, E1000_TDT(0)) ==
E1000_READ_REG(&adapter->hw, E1000_TDH(0))) &&
(E1000_READ_REG(&adapter->hw, E1000_TDFT) ==
E1000_READ_REG(&adapter->hw, E1000_TDFH)) &&
(E1000_READ_REG(&adapter->hw, E1000_TDFTS) ==
E1000_READ_REG(&adapter->hw, E1000_TDFHS)) &&
(E1000_READ_REG(&adapter->hw, E1000_TDFPC) == 0)) {
/* Disable TX unit */
tctl = E1000_READ_REG(&adapter->hw, E1000_TCTL);
E1000_WRITE_REG(&adapter->hw, E1000_TCTL,
tctl & ~E1000_TCTL_EN);
/* Reset FIFO pointers */
E1000_WRITE_REG(&adapter->hw, E1000_TDFT,
adapter->tx_head_addr);
E1000_WRITE_REG(&adapter->hw, E1000_TDFH,
adapter->tx_head_addr);
E1000_WRITE_REG(&adapter->hw, E1000_TDFTS,
adapter->tx_head_addr);
E1000_WRITE_REG(&adapter->hw, E1000_TDFHS,
adapter->tx_head_addr);
/* Re-enable TX unit */
E1000_WRITE_REG(&adapter->hw, E1000_TCTL, tctl);
E1000_WRITE_FLUSH(&adapter->hw);
adapter->tx_fifo_head = 0;
adapter->tx_fifo_reset_cnt++;
return (TRUE);
}
else {
return (FALSE);
}
}
static void
lem_set_promisc(struct adapter *adapter)
{
struct ifnet *ifp = adapter->ifp;
u32 reg_rctl;
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
if (ifp->if_flags & IFF_PROMISC) {
reg_rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
/* Turn this on if you want to see bad packets */
if (lem_debug_sbp)
reg_rctl |= E1000_RCTL_SBP;
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
} else if (ifp->if_flags & IFF_ALLMULTI) {
reg_rctl |= E1000_RCTL_MPE;
reg_rctl &= ~E1000_RCTL_UPE;
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
}
}
static void
lem_disable_promisc(struct adapter *adapter)
{
u32 reg_rctl;
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
reg_rctl &= (~E1000_RCTL_UPE);
reg_rctl &= (~E1000_RCTL_MPE);
reg_rctl &= (~E1000_RCTL_SBP);
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
}
/*********************************************************************
* Multicast Update
*
* This routine is called whenever multicast address list is updated.
*
**********************************************************************/
static void
lem_set_multi(struct adapter *adapter)
{
struct ifnet *ifp = adapter->ifp;
struct ifmultiaddr *ifma;
u32 reg_rctl = 0;
u8 *mta; /* Multicast array memory */
int mcnt = 0;
IOCTL_DEBUGOUT("lem_set_multi: begin");
Do not allocate multicast array memory in multicast filter configuration function. For failed memory allocations, em(4)/lem(4) called panic(9) which is not acceptable on production box. igb(4)/ixgb(4)/ix(4) allocated the required memory in stack which consumed 768 bytes of stack memory which looks too big. To address these issues, allocate multicast array memory in device attach time and make multicast configuration success under any conditions. This change also removes the excessive use of memory in stack. Reviewed by: jfv
2010-08-28 00:34:22 +00:00
mta = adapter->mta;
bzero(mta, sizeof(u8) * ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (adapter->hw.mac.type == e1000_82542 &&
adapter->hw.revision_id == E1000_REVISION_2) {
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
e1000_pci_clear_mwi(&adapter->hw);
reg_rctl |= E1000_RCTL_RST;
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
msec_delay(5);
}
#if __FreeBSD_version < 800000
IF_ADDR_LOCK(ifp);
#else
if_maddr_rlock(ifp);
#endif
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
if (ifma->ifma_addr->sa_family != AF_LINK)
continue;
if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
break;
bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
&mta[mcnt * ETH_ADDR_LEN], ETH_ADDR_LEN);
mcnt++;
}
#if __FreeBSD_version < 800000
IF_ADDR_UNLOCK(ifp);
#else
if_maddr_runlock(ifp);
#endif
if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
reg_rctl |= E1000_RCTL_MPE;
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
} else
e1000_update_mc_addr_list(&adapter->hw, mta, mcnt);
if (adapter->hw.mac.type == e1000_82542 &&
adapter->hw.revision_id == E1000_REVISION_2) {
reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
reg_rctl &= ~E1000_RCTL_RST;
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl);
msec_delay(5);
if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
e1000_pci_set_mwi(&adapter->hw);
}
}
/*********************************************************************
* Timer routine
*
* This routine checks for link status and updates statistics.
*
**********************************************************************/
static void
lem_local_timer(void *arg)
{
struct adapter *adapter = arg;
EM_CORE_LOCK_ASSERT(adapter);
lem_update_link_status(adapter);
lem_update_stats_counters(adapter);
lem_smartspeed(adapter);
/*
* We check the watchdog: the time since
* the last TX descriptor was cleaned.
* This implies a functional TX engine.
*/
if ((adapter->watchdog_check == TRUE) &&
(ticks - adapter->watchdog_time > EM_WATCHDOG))
goto hung;
callout_reset(&adapter->timer, hz, lem_local_timer, adapter);
return;
hung:
device_printf(adapter->dev, "Watchdog timeout -- resetting\n");
adapter->ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
adapter->watchdog_events++;
lem_init_locked(adapter);
}
static void
lem_update_link_status(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
struct ifnet *ifp = adapter->ifp;
device_t dev = adapter->dev;
u32 link_check = 0;
/* Get the cached link value or read phy for real */
switch (hw->phy.media_type) {
case e1000_media_type_copper:
if (hw->mac.get_link_status) {
/* Do the work to read phy */
e1000_check_for_link(hw);
link_check = !hw->mac.get_link_status;
if (link_check) /* ESB2 fix */
e1000_cfg_on_link_up(hw);
} else
link_check = TRUE;
break;
case e1000_media_type_fiber:
e1000_check_for_link(hw);
link_check = (E1000_READ_REG(hw, E1000_STATUS) &
E1000_STATUS_LU);
break;
case e1000_media_type_internal_serdes:
e1000_check_for_link(hw);
link_check = adapter->hw.mac.serdes_has_link;
break;
default:
case e1000_media_type_unknown:
break;
}
/* Now check for a transition */
if (link_check && (adapter->link_active == 0)) {
e1000_get_speed_and_duplex(hw, &adapter->link_speed,
&adapter->link_duplex);
if (bootverbose)
device_printf(dev, "Link is up %d Mbps %s\n",
adapter->link_speed,
((adapter->link_duplex == FULL_DUPLEX) ?
"Full Duplex" : "Half Duplex"));
adapter->link_active = 1;
adapter->smartspeed = 0;
ifp->if_baudrate = adapter->link_speed * 1000000;
if_link_state_change(ifp, LINK_STATE_UP);
} else if (!link_check && (adapter->link_active == 1)) {
ifp->if_baudrate = adapter->link_speed = 0;
adapter->link_duplex = 0;
if (bootverbose)
device_printf(dev, "Link is Down\n");
adapter->link_active = 0;
/* Link down, disable watchdog */
adapter->watchdog_check = FALSE;
if_link_state_change(ifp, LINK_STATE_DOWN);
}
}
/*********************************************************************
*
* This routine disables all traffic on the adapter by issuing a
* global reset on the MAC and deallocates TX/RX buffers.
*
* This routine should always be called with BOTH the CORE
* and TX locks.
**********************************************************************/
static void
lem_stop(void *arg)
{
struct adapter *adapter = arg;
struct ifnet *ifp = adapter->ifp;
EM_CORE_LOCK_ASSERT(adapter);
EM_TX_LOCK_ASSERT(adapter);
INIT_DEBUGOUT("lem_stop: begin");
lem_disable_intr(adapter);
callout_stop(&adapter->timer);
callout_stop(&adapter->tx_fifo_timer);
/* Tell the stack that the interface is no longer active */
ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
e1000_reset_hw(&adapter->hw);
if (adapter->hw.mac.type >= e1000_82544)
E1000_WRITE_REG(&adapter->hw, E1000_WUC, 0);
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
e1000_led_off(&adapter->hw);
e1000_cleanup_led(&adapter->hw);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
/*********************************************************************
*
* Determine hardware revision.
*
**********************************************************************/
static void
lem_identify_hardware(struct adapter *adapter)
{
device_t dev = adapter->dev;
/* Make sure our PCI config space has the necessary stuff set */
adapter->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
if (!((adapter->hw.bus.pci_cmd_word & PCIM_CMD_BUSMASTEREN) &&
(adapter->hw.bus.pci_cmd_word & PCIM_CMD_MEMEN))) {
device_printf(dev, "Memory Access and/or Bus Master bits "
"were not set!\n");
adapter->hw.bus.pci_cmd_word |=
(PCIM_CMD_BUSMASTEREN | PCIM_CMD_MEMEN);
pci_write_config(dev, PCIR_COMMAND,
adapter->hw.bus.pci_cmd_word, 2);
}
/* Save off the information about this board */
adapter->hw.vendor_id = pci_get_vendor(dev);
adapter->hw.device_id = pci_get_device(dev);
adapter->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1);
adapter->hw.subsystem_vendor_id =
pci_read_config(dev, PCIR_SUBVEND_0, 2);
adapter->hw.subsystem_device_id =
pci_read_config(dev, PCIR_SUBDEV_0, 2);
/* Do Shared Code Init and Setup */
if (e1000_set_mac_type(&adapter->hw)) {
device_printf(dev, "Setup init failure\n");
return;
}
}
static int
lem_allocate_pci_resources(struct adapter *adapter)
{
device_t dev = adapter->dev;
int val, rid, error = E1000_SUCCESS;
rid = PCIR_BAR(0);
adapter->memory = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
&rid, RF_ACTIVE);
if (adapter->memory == NULL) {
device_printf(dev, "Unable to allocate bus resource: memory\n");
return (ENXIO);
}
adapter->osdep.mem_bus_space_tag =
rman_get_bustag(adapter->memory);
adapter->osdep.mem_bus_space_handle =
rman_get_bushandle(adapter->memory);
adapter->hw.hw_addr = (u8 *)&adapter->osdep.mem_bus_space_handle;
/* Only older adapters use IO mapping */
if (adapter->hw.mac.type > e1000_82543) {
/* Figure our where our IO BAR is ? */
for (rid = PCIR_BAR(0); rid < PCIR_CIS;) {
val = pci_read_config(dev, rid, 4);
if (EM_BAR_TYPE(val) == EM_BAR_TYPE_IO) {
adapter->io_rid = rid;
break;
}
rid += 4;
/* check for 64bit BAR */
if (EM_BAR_MEM_TYPE(val) == EM_BAR_MEM_TYPE_64BIT)
rid += 4;
}
if (rid >= PCIR_CIS) {
device_printf(dev, "Unable to locate IO BAR\n");
return (ENXIO);
}
adapter->ioport = bus_alloc_resource_any(dev,
SYS_RES_IOPORT, &adapter->io_rid, RF_ACTIVE);
if (adapter->ioport == NULL) {
device_printf(dev, "Unable to allocate bus resource: "
"ioport\n");
return (ENXIO);
}
adapter->hw.io_base = 0;
adapter->osdep.io_bus_space_tag =
rman_get_bustag(adapter->ioport);
adapter->osdep.io_bus_space_handle =
rman_get_bushandle(adapter->ioport);
}
adapter->hw.back = &adapter->osdep;
return (error);
}
/*********************************************************************
*
* Setup the Legacy or MSI Interrupt handler
*
**********************************************************************/
int
lem_allocate_irq(struct adapter *adapter)
{
device_t dev = adapter->dev;
int error, rid = 0;
/* Manually turn off all interrupts */
E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff);
/* We allocate a single interrupt resource */
adapter->res[0] = bus_alloc_resource_any(dev,
SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE);
if (adapter->res[0] == NULL) {
device_printf(dev, "Unable to allocate bus resource: "
"interrupt\n");
return (ENXIO);
}
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
/* Do Legacy setup? */
if (lem_use_legacy_irq) {
if ((error = bus_setup_intr(dev, adapter->res[0],
INTR_TYPE_NET | INTR_MPSAFE, NULL, lem_intr, adapter,
&adapter->tag[0])) != 0) {
device_printf(dev,
"Failed to register interrupt handler");
return (error);
}
return (0);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
/*
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
* Use a Fast interrupt and the associated
* deferred processing contexts.
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
*/
TASK_INIT(&adapter->rxtx_task, 0, lem_handle_rxtx, adapter);
TASK_INIT(&adapter->link_task, 0, lem_handle_link, adapter);
adapter->tq = taskqueue_create_fast("lem_taskq", M_NOWAIT,
taskqueue_thread_enqueue, &adapter->tq);
taskqueue_start_threads(&adapter->tq, 1, PI_NET, "%s taskq",
device_get_nameunit(adapter->dev));
if ((error = bus_setup_intr(dev, adapter->res[0],
INTR_TYPE_NET, lem_irq_fast, NULL, adapter,
&adapter->tag[0])) != 0) {
device_printf(dev, "Failed to register fast interrupt "
"handler: %d\n", error);
taskqueue_free(adapter->tq);
adapter->tq = NULL;
return (error);
}
return (0);
}
static void
lem_free_pci_resources(struct adapter *adapter)
{
device_t dev = adapter->dev;
if (adapter->tag[0] != NULL) {
bus_teardown_intr(dev, adapter->res[0],
adapter->tag[0]);
adapter->tag[0] = NULL;
}
if (adapter->res[0] != NULL) {
bus_release_resource(dev, SYS_RES_IRQ,
0, adapter->res[0]);
}
if (adapter->memory != NULL)
bus_release_resource(dev, SYS_RES_MEMORY,
PCIR_BAR(0), adapter->memory);
if (adapter->ioport != NULL)
bus_release_resource(dev, SYS_RES_IOPORT,
adapter->io_rid, adapter->ioport);
}
/*********************************************************************
*
* Initialize the hardware to a configuration
* as specified by the adapter structure.
*
**********************************************************************/
static int
lem_hardware_init(struct adapter *adapter)
{
device_t dev = adapter->dev;
u16 rx_buffer_size;
INIT_DEBUGOUT("lem_hardware_init: begin");
/* Issue a global reset */
e1000_reset_hw(&adapter->hw);
/* When hardware is reset, fifo_head is also reset */
adapter->tx_fifo_head = 0;
/*
* These parameters control the automatic generation (Tx) and
* response (Rx) to Ethernet PAUSE frames.
* - High water mark should allow for at least two frames to be
* received after sending an XOFF.
* - Low water mark works best when it is very near the high water mark.
* This allows the receiver to restart by sending XON when it has
* drained a bit. Here we use an arbitary value of 1500 which will
* restart after one full frame is pulled from the buffer. There
* could be several smaller frames in the buffer and if so they will
* not trigger the XON until their total number reduces the buffer
* by 1500.
* - The pause time is fairly large at 1000 x 512ns = 512 usec.
*/
rx_buffer_size = ((E1000_READ_REG(&adapter->hw, E1000_PBA) &
0xffff) << 10 );
adapter->hw.fc.high_water = rx_buffer_size -
roundup2(adapter->max_frame_size, 1024);
adapter->hw.fc.low_water = adapter->hw.fc.high_water - 1500;
adapter->hw.fc.pause_time = EM_FC_PAUSE_TIME;
adapter->hw.fc.send_xon = TRUE;
/* Set Flow control, use the tunable location if sane */
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
if ((lem_fc_setting >= 0) && (lem_fc_setting < 4))
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
adapter->hw.fc.requested_mode = lem_fc_setting;
else
adapter->hw.fc.requested_mode = e1000_fc_none;
if (e1000_init_hw(&adapter->hw) < 0) {
device_printf(dev, "Hardware Initialization Failed\n");
return (EIO);
}
e1000_check_for_link(&adapter->hw);
return (0);
}
/*********************************************************************
*
* Setup networking device structure and register an interface.
*
**********************************************************************/
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
static int
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
lem_setup_interface(device_t dev, struct adapter *adapter)
{
struct ifnet *ifp;
INIT_DEBUGOUT("lem_setup_interface: begin");
ifp = adapter->ifp = if_alloc(IFT_ETHER);
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
if (ifp == NULL) {
device_printf(dev, "can not allocate ifnet structure\n");
return (-1);
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if_initname(ifp, device_get_name(dev), device_get_unit(dev));
ifp->if_init = lem_init;
ifp->if_softc = adapter;
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_ioctl = lem_ioctl;
ifp->if_start = lem_start;
IFQ_SET_MAXLEN(&ifp->if_snd, adapter->num_tx_desc - 1);
ifp->if_snd.ifq_drv_maxlen = adapter->num_tx_desc - 1;
IFQ_SET_READY(&ifp->if_snd);
ether_ifattach(ifp, adapter->hw.mac.addr);
ifp->if_capabilities = ifp->if_capenable = 0;
if (adapter->hw.mac.type >= e1000_82543) {
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM;
ifp->if_capenable |= IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
/*
* Tell the upper layer(s) we support long frames.
*/
ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
ifp->if_capenable |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
/*
** Dont turn this on by default, if vlans are
** created on another pseudo device (eg. lagg)
** then vlan events are not passed thru, breaking
** operation, but with HW FILTER off it works. If
** using vlans directly on the em driver you can
** enable this and get full hardware tag filtering.
*/
ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
#ifdef DEVICE_POLLING
ifp->if_capabilities |= IFCAP_POLLING;
#endif
Small changes preparing for MFC, need to conditionalize the buf_ring_free call, and lem is missing the WOL change put into em.
2010-05-14 22:18:34 +00:00
/* Enable only WOL MAGIC by default */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (adapter->wol) {
ifp->if_capabilities |= IFCAP_WOL;
Small changes preparing for MFC, need to conditionalize the buf_ring_free call, and lem is missing the WOL change put into em.
2010-05-14 22:18:34 +00:00
ifp->if_capenable |= IFCAP_WOL_MAGIC;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
/*
* Specify the media types supported by this adapter and register
* callbacks to update media and link information
*/
ifmedia_init(&adapter->media, IFM_IMASK,
lem_media_change, lem_media_status);
if ((adapter->hw.phy.media_type == e1000_media_type_fiber) ||
(adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) {
u_char fiber_type = IFM_1000_SX; /* default type */
if (adapter->hw.mac.type == e1000_82545)
fiber_type = IFM_1000_LX;
ifmedia_add(&adapter->media, IFM_ETHER | fiber_type | IFM_FDX,
0, NULL);
ifmedia_add(&adapter->media, IFM_ETHER | fiber_type, 0, NULL);
} else {
ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T, 0, NULL);
ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T | IFM_FDX,
0, NULL);
ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX,
0, NULL);
ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
0, NULL);
if (adapter->hw.phy.type != e1000_phy_ife) {
ifmedia_add(&adapter->media,
IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
ifmedia_add(&adapter->media,
IFM_ETHER | IFM_1000_T, 0, NULL);
}
}
ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
Do not call voluntary panic(9) in case of if_alloc() failure. Reviewed by: jfv
2010-08-28 00:09:19 +00:00
return (0);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
/*********************************************************************
*
* Workaround for SmartSpeed on 82541 and 82547 controllers
*
**********************************************************************/
static void
lem_smartspeed(struct adapter *adapter)
{
u16 phy_tmp;
if (adapter->link_active || (adapter->hw.phy.type != e1000_phy_igp) ||
adapter->hw.mac.autoneg == 0 ||
(adapter->hw.phy.autoneg_advertised & ADVERTISE_1000_FULL) == 0)
return;
if (adapter->smartspeed == 0) {
/* If Master/Slave config fault is asserted twice,
* we assume back-to-back */
e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_tmp);
if (!(phy_tmp & SR_1000T_MS_CONFIG_FAULT))
return;
e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_tmp);
if (phy_tmp & SR_1000T_MS_CONFIG_FAULT) {
e1000_read_phy_reg(&adapter->hw,
PHY_1000T_CTRL, &phy_tmp);
if(phy_tmp & CR_1000T_MS_ENABLE) {
phy_tmp &= ~CR_1000T_MS_ENABLE;
e1000_write_phy_reg(&adapter->hw,
PHY_1000T_CTRL, phy_tmp);
adapter->smartspeed++;
if(adapter->hw.mac.autoneg &&
!e1000_copper_link_autoneg(&adapter->hw) &&
!e1000_read_phy_reg(&adapter->hw,
PHY_CONTROL, &phy_tmp)) {
phy_tmp |= (MII_CR_AUTO_NEG_EN |
MII_CR_RESTART_AUTO_NEG);
e1000_write_phy_reg(&adapter->hw,
PHY_CONTROL, phy_tmp);
}
}
}
return;
} else if(adapter->smartspeed == EM_SMARTSPEED_DOWNSHIFT) {
/* If still no link, perhaps using 2/3 pair cable */
e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_tmp);
phy_tmp |= CR_1000T_MS_ENABLE;
e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_tmp);
if(adapter->hw.mac.autoneg &&
!e1000_copper_link_autoneg(&adapter->hw) &&
!e1000_read_phy_reg(&adapter->hw, PHY_CONTROL, &phy_tmp)) {
phy_tmp |= (MII_CR_AUTO_NEG_EN |
MII_CR_RESTART_AUTO_NEG);
e1000_write_phy_reg(&adapter->hw, PHY_CONTROL, phy_tmp);
}
}
/* Restart process after EM_SMARTSPEED_MAX iterations */
if(adapter->smartspeed++ == EM_SMARTSPEED_MAX)
adapter->smartspeed = 0;
}
/*
* Manage DMA'able memory.
*/
static void
lem_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
{
if (error)
return;
*(bus_addr_t *) arg = segs[0].ds_addr;
}
static int
lem_dma_malloc(struct adapter *adapter, bus_size_t size,
struct em_dma_alloc *dma, int mapflags)
{
int error;
error = bus_dma_tag_create(bus_get_dma_tag(adapter->dev), /* parent */
EM_DBA_ALIGN, 0, /* alignment, bounds */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
size, /* maxsize */
1, /* nsegments */
size, /* maxsegsize */
0, /* flags */
NULL, /* lockfunc */
NULL, /* lockarg */
&dma->dma_tag);
if (error) {
device_printf(adapter->dev,
"%s: bus_dma_tag_create failed: %d\n",
__func__, error);
goto fail_0;
}
error = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr,
BUS_DMA_NOWAIT | BUS_DMA_COHERENT, &dma->dma_map);
if (error) {
device_printf(adapter->dev,
"%s: bus_dmamem_alloc(%ju) failed: %d\n",
__func__, (uintmax_t)size, error);
goto fail_2;
}
dma->dma_paddr = 0;
error = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr,
size, lem_dmamap_cb, &dma->dma_paddr, mapflags | BUS_DMA_NOWAIT);
if (error || dma->dma_paddr == 0) {
device_printf(adapter->dev,
"%s: bus_dmamap_load failed: %d\n",
__func__, error);
goto fail_3;
}
return (0);
fail_3:
bus_dmamap_unload(dma->dma_tag, dma->dma_map);
fail_2:
bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
bus_dma_tag_destroy(dma->dma_tag);
fail_0:
dma->dma_map = NULL;
dma->dma_tag = NULL;
return (error);
}
static void
lem_dma_free(struct adapter *adapter, struct em_dma_alloc *dma)
{
if (dma->dma_tag == NULL)
return;
if (dma->dma_map != NULL) {
bus_dmamap_sync(dma->dma_tag, dma->dma_map,
BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(dma->dma_tag, dma->dma_map);
bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
dma->dma_map = NULL;
}
bus_dma_tag_destroy(dma->dma_tag);
dma->dma_tag = NULL;
}
/*********************************************************************
*
* Allocate memory for tx_buffer structures. The tx_buffer stores all
* the information needed to transmit a packet on the wire.
*
**********************************************************************/
static int
lem_allocate_transmit_structures(struct adapter *adapter)
{
device_t dev = adapter->dev;
struct em_buffer *tx_buffer;
int error;
/*
* Create DMA tags for tx descriptors
*/
if ((error = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
1, 0, /* alignment, bounds */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
MCLBYTES * EM_MAX_SCATTER, /* maxsize */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
EM_MAX_SCATTER, /* nsegments */
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
MCLBYTES, /* maxsegsize */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
0, /* flags */
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
NULL, /* lockfunc */
NULL, /* lockarg */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
&adapter->txtag)) != 0) {
device_printf(dev, "Unable to allocate TX DMA tag\n");
goto fail;
}
adapter->tx_buffer_area = malloc(sizeof(struct em_buffer) *
adapter->num_tx_desc, M_DEVBUF, M_NOWAIT | M_ZERO);
if (adapter->tx_buffer_area == NULL) {
device_printf(dev, "Unable to allocate tx_buffer memory\n");
error = ENOMEM;
goto fail;
}
/* Create the descriptor buffer dma maps */
for (int i = 0; i < adapter->num_tx_desc; i++) {
tx_buffer = &adapter->tx_buffer_area[i];
error = bus_dmamap_create(adapter->txtag, 0, &tx_buffer->map);
if (error != 0) {
device_printf(dev, "Unable to create TX DMA map\n");
goto fail;
}
tx_buffer->next_eop = -1;
}
return (0);
fail:
lem_free_transmit_structures(adapter);
return (error);
}
/*********************************************************************
*
* (Re)Initialize transmit structures.
*
**********************************************************************/
static void
lem_setup_transmit_structures(struct adapter *adapter)
{
struct em_buffer *tx_buffer;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
/* we are already locked */
struct netmap_adapter *na = NA(adapter->ifp);
struct netmap_slot *slot = netmap_reset(na, NR_TX, 0, 0);
#endif /* DEV_NETMAP */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Clear the old ring contents */
bzero(adapter->tx_desc_base,
(sizeof(struct e1000_tx_desc)) * adapter->num_tx_desc);
/* Free any existing TX buffers */
for (int i = 0; i < adapter->num_tx_desc; i++, tx_buffer++) {
tx_buffer = &adapter->tx_buffer_area[i];
bus_dmamap_sync(adapter->txtag, tx_buffer->map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(adapter->txtag, tx_buffer->map);
m_freem(tx_buffer->m_head);
tx_buffer->m_head = NULL;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
if (slot) {
(This commit only touches code within the DEV_NETMAP blocks) Introduce some functions to map NIC ring indexes into netmap ring indexes and vice versa. This way we can implement the bound checks only in one place (and hopefully in a correct way). On passing, make the code and comments more uniform across the various drivers.
2012-02-15 23:13:29 +00:00
/* the i-th NIC entry goes to slot si */
A bunch of netmap fixes: USERSPACE: 1. add support for devices with different number of rx and tx queues; 2. add better support for zero-copy operation, adding an extra field to the netmap ring to indicate how many buffers we have already processed but not yet released (with help from Eddie Kohler); 3. The two changes above unfortunately require an API change, so while at it add a version field and some spares to the ioctl() argument to help detect mismatches. 4. update the manual page for the two changes above; 5. update sample applications in tools/tools/netmap KERNEL: 1. simplify the internal structures moving the global wait queues to the 'struct netmap_adapter'; 2. simplify the functions that map kring<->nic ring indexes 3. normalize device-specific code, helps mainteinance; 4. start exploring the impact of micro-optimizations (prefetch etc.) in the ixgbe driver. Use 'legacy' descriptors on the tx ring and prefetch slots gives about 20% speedup at 900 MHz. Another 7-10% would come from removing the explict calls to bus_dmamap* in the core (they are effectively NOPs in this case, but it takes expensive load of the per-buffer dma maps to figure out that they are all NULL. Rx performance not investigated. I am postponing the MFC so i can import a few more improvements before merging.
2012-02-27 19:05:01 +00:00
int si = netmap_idx_n2k(&na->tx_rings[0], i);
small code cleanup in preparation for future modifications in the memory allocator used by netmap. No functional change, two small bug fixes: - in if_re.c add a missing bus_dmamap_sync() - in netmap.c comment out a spurious free() in an error handling block
2012-01-10 19:57:23 +00:00
uint64_t paddr;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
void *addr;
small code cleanup in preparation for future modifications in the memory allocator used by netmap. No functional change, two small bug fixes: - in if_re.c add a missing bus_dmamap_sync() - in netmap.c comment out a spurious free() in an error handling block
2012-01-10 19:57:23 +00:00
addr = PNMB(slot + si, &paddr);
adapter->tx_desc_base[si].buffer_addr = htole64(paddr);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
/* reload the map for netmap mode */
small code cleanup in preparation for future modifications in the memory allocator used by netmap. No functional change, two small bug fixes: - in if_re.c add a missing bus_dmamap_sync() - in netmap.c comment out a spurious free() in an error handling block
2012-01-10 19:57:23 +00:00
netmap_load_map(adapter->txtag, tx_buffer->map, addr);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
}
#endif /* DEV_NETMAP */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
tx_buffer->next_eop = -1;
}
/* Reset state */
Clear transmit checksum offload context state upon lem(4) interface initialization. Prior to this change packets may be transmitted with an incorrect checksum. Em(4) already has an equivalent change in r213234. Obtained From: Sandvine MFC After: 1 week Approved by: re (bz)
2011-09-17 13:48:09 +00:00
adapter->last_hw_offload = 0;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
adapter->next_avail_tx_desc = 0;
adapter->next_tx_to_clean = 0;
adapter->num_tx_desc_avail = adapter->num_tx_desc;
bus_dmamap_sync(adapter->txdma.dma_tag, adapter->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
return;
}
/*********************************************************************
*
* Enable transmit unit.
*
**********************************************************************/
static void
lem_initialize_transmit_unit(struct adapter *adapter)
{
u32 tctl, tipg = 0;
u64 bus_addr;
INIT_DEBUGOUT("lem_initialize_transmit_unit: begin");
/* Setup the Base and Length of the Tx Descriptor Ring */
bus_addr = adapter->txdma.dma_paddr;
E1000_WRITE_REG(&adapter->hw, E1000_TDLEN(0),
adapter->num_tx_desc * sizeof(struct e1000_tx_desc));
E1000_WRITE_REG(&adapter->hw, E1000_TDBAH(0),
(u32)(bus_addr >> 32));
E1000_WRITE_REG(&adapter->hw, E1000_TDBAL(0),
(u32)bus_addr);
/* Setup the HW Tx Head and Tail descriptor pointers */
E1000_WRITE_REG(&adapter->hw, E1000_TDT(0), 0);
E1000_WRITE_REG(&adapter->hw, E1000_TDH(0), 0);
HW_DEBUGOUT2("Base = %x, Length = %x\n",
E1000_READ_REG(&adapter->hw, E1000_TDBAL(0)),
E1000_READ_REG(&adapter->hw, E1000_TDLEN(0)));
/* Set the default values for the Tx Inter Packet Gap timer */
switch (adapter->hw.mac.type) {
case e1000_82542:
tipg = DEFAULT_82542_TIPG_IPGT;
tipg |= DEFAULT_82542_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
tipg |= DEFAULT_82542_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
break;
default:
if ((adapter->hw.phy.media_type == e1000_media_type_fiber) ||
(adapter->hw.phy.media_type ==
e1000_media_type_internal_serdes))
tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
else
tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
tipg |= DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
tipg |= DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
}
E1000_WRITE_REG(&adapter->hw, E1000_TIPG, tipg);
E1000_WRITE_REG(&adapter->hw, E1000_TIDV, adapter->tx_int_delay.value);
if(adapter->hw.mac.type >= e1000_82540)
E1000_WRITE_REG(&adapter->hw, E1000_TADV,
adapter->tx_abs_int_delay.value);
/* Program the Transmit Control Register */
tctl = E1000_READ_REG(&adapter->hw, E1000_TCTL);
tctl &= ~E1000_TCTL_CT;
tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN |
(E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT));
/* This write will effectively turn on the transmit unit. */
E1000_WRITE_REG(&adapter->hw, E1000_TCTL, tctl);
/* Setup Transmit Descriptor Base Settings */
adapter->txd_cmd = E1000_TXD_CMD_IFCS;
if (adapter->tx_int_delay.value > 0)
adapter->txd_cmd |= E1000_TXD_CMD_IDE;
}
/*********************************************************************
*
* Free all transmit related data structures.
*
**********************************************************************/
static void
lem_free_transmit_structures(struct adapter *adapter)
{
struct em_buffer *tx_buffer;
INIT_DEBUGOUT("free_transmit_structures: begin");
if (adapter->tx_buffer_area != NULL) {
for (int i = 0; i < adapter->num_tx_desc; i++) {
tx_buffer = &adapter->tx_buffer_area[i];
if (tx_buffer->m_head != NULL) {
bus_dmamap_sync(adapter->txtag, tx_buffer->map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(adapter->txtag,
tx_buffer->map);
m_freem(tx_buffer->m_head);
tx_buffer->m_head = NULL;
} else if (tx_buffer->map != NULL)
bus_dmamap_unload(adapter->txtag,
tx_buffer->map);
if (tx_buffer->map != NULL) {
bus_dmamap_destroy(adapter->txtag,
tx_buffer->map);
tx_buffer->map = NULL;
}
}
}
if (adapter->tx_buffer_area != NULL) {
free(adapter->tx_buffer_area, M_DEVBUF);
adapter->tx_buffer_area = NULL;
}
if (adapter->txtag != NULL) {
bus_dma_tag_destroy(adapter->txtag);
adapter->txtag = NULL;
}
#if __FreeBSD_version >= 800000
if (adapter->br != NULL)
buf_ring_free(adapter->br, M_DEVBUF);
#endif
}
/*********************************************************************
*
* The offload context needs to be set when we transfer the first
* packet of a particular protocol (TCP/UDP). This routine has been
* enhanced to deal with inserted VLAN headers, and IPV6 (not complete)
*
* Added back the old method of keeping the current context type
* and not setting if unnecessary, as this is reported to be a
* big performance win. -jfv
**********************************************************************/
static void
lem_transmit_checksum_setup(struct adapter *adapter, struct mbuf *mp,
u32 *txd_upper, u32 *txd_lower)
{
struct e1000_context_desc *TXD = NULL;
struct em_buffer *tx_buffer;
struct ether_vlan_header *eh;
struct ip *ip = NULL;
struct ip6_hdr *ip6;
int curr_txd, ehdrlen;
u32 cmd, hdr_len, ip_hlen;
u16 etype;
u8 ipproto;
cmd = hdr_len = ipproto = 0;
Fix for a panic when TX checksum offload is done and a packet has only a header in the first mbuf, the checksum code will dereference a pointer into the non-existing IP header. Do a check for the size and pullup if needed. Thanks to Michael Tuexen for this fix. MFC: asap - should be in 8.1 IMHO
2010-07-12 21:47:30 +00:00
*txd_upper = *txd_lower = 0;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
curr_txd = adapter->next_avail_tx_desc;
/*
* 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;
}
/*
* We only support TCP/UDP for IPv4 and IPv6 for the moment.
* TODO: Support SCTP too when it hits the tree.
*/
switch (etype) {
case ETHERTYPE_IP:
ip = (struct ip *)(mp->m_data + ehdrlen);
ip_hlen = ip->ip_hl << 2;
/* Setup of IP header checksum. */
if (mp->m_pkthdr.csum_flags & CSUM_IP) {
/*
* Start offset for header checksum calculation.
* End offset for header checksum calculation.
* Offset of place to put the checksum.
*/
TXD = (struct e1000_context_desc *)
&adapter->tx_desc_base[curr_txd];
TXD->lower_setup.ip_fields.ipcss = ehdrlen;
TXD->lower_setup.ip_fields.ipcse =
htole16(ehdrlen + ip_hlen);
TXD->lower_setup.ip_fields.ipcso =
ehdrlen + offsetof(struct ip, ip_sum);
cmd |= E1000_TXD_CMD_IP;
*txd_upper |= E1000_TXD_POPTS_IXSM << 8;
}
hdr_len = ehdrlen + ip_hlen;
ipproto = ip->ip_p;
break;
case ETHERTYPE_IPV6:
ip6 = (struct ip6_hdr *)(mp->m_data + ehdrlen);
ip_hlen = sizeof(struct ip6_hdr); /* XXX: No header stacking. */
/* IPv6 doesn't have a header checksum. */
hdr_len = ehdrlen + ip_hlen;
ipproto = ip6->ip6_nxt;
break;
Fix for a panic when TX checksum offload is done and a packet has only a header in the first mbuf, the checksum code will dereference a pointer into the non-existing IP header. Do a check for the size and pullup if needed. Thanks to Michael Tuexen for this fix. MFC: asap - should be in 8.1 IMHO
2010-07-12 21:47:30 +00:00
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
default:
return;
}
switch (ipproto) {
case IPPROTO_TCP:
if (mp->m_pkthdr.csum_flags & CSUM_TCP) {
*txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
*txd_upper |= E1000_TXD_POPTS_TXSM << 8;
/* no need for context if already set */
if (adapter->last_hw_offload == CSUM_TCP)
return;
adapter->last_hw_offload = CSUM_TCP;
/*
* Start offset for payload checksum calculation.
* End offset for payload checksum calculation.
* Offset of place to put the checksum.
*/
TXD = (struct e1000_context_desc *)
&adapter->tx_desc_base[curr_txd];
TXD->upper_setup.tcp_fields.tucss = hdr_len;
TXD->upper_setup.tcp_fields.tucse = htole16(0);
TXD->upper_setup.tcp_fields.tucso =
hdr_len + offsetof(struct tcphdr, th_sum);
cmd |= E1000_TXD_CMD_TCP;
}
break;
case IPPROTO_UDP:
{
if (mp->m_pkthdr.csum_flags & CSUM_UDP) {
*txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
*txd_upper |= E1000_TXD_POPTS_TXSM << 8;
/* no need for context if already set */
if (adapter->last_hw_offload == CSUM_UDP)
return;
adapter->last_hw_offload = CSUM_UDP;
/*
* Start offset for header checksum calculation.
* End offset for header checksum calculation.
* Offset of place to put the checksum.
*/
TXD = (struct e1000_context_desc *)
&adapter->tx_desc_base[curr_txd];
TXD->upper_setup.tcp_fields.tucss = hdr_len;
TXD->upper_setup.tcp_fields.tucse = htole16(0);
TXD->upper_setup.tcp_fields.tucso =
hdr_len + offsetof(struct udphdr, uh_sum);
}
/* Fall Thru */
}
default:
break;
}
Fix for a panic when TX checksum offload is done and a packet has only a header in the first mbuf, the checksum code will dereference a pointer into the non-existing IP header. Do a check for the size and pullup if needed. Thanks to Michael Tuexen for this fix. MFC: asap - should be in 8.1 IMHO
2010-07-12 21:47:30 +00:00
if (TXD == NULL)
return;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
TXD->tcp_seg_setup.data = htole32(0);
TXD->cmd_and_length =
htole32(adapter->txd_cmd | E1000_TXD_CMD_DEXT | cmd);
tx_buffer = &adapter->tx_buffer_area[curr_txd];
tx_buffer->m_head = NULL;
tx_buffer->next_eop = -1;
if (++curr_txd == adapter->num_tx_desc)
curr_txd = 0;
adapter->num_tx_desc_avail--;
adapter->next_avail_tx_desc = curr_txd;
}
/**********************************************************************
*
* 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 void
lem_txeof(struct adapter *adapter)
{
int first, last, done, num_avail;
struct em_buffer *tx_buffer;
struct e1000_tx_desc *tx_desc, *eop_desc;
struct ifnet *ifp = adapter->ifp;
EM_TX_LOCK_ASSERT(adapter);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
if (ifp->if_capenable & IFCAP_NETMAP) {
selwakeuppri(&NA(ifp)->tx_rings[0].si, PI_NET);
return;
}
#endif /* DEV_NETMAP */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (adapter->num_tx_desc_avail == adapter->num_tx_desc)
return;
num_avail = adapter->num_tx_desc_avail;
first = adapter->next_tx_to_clean;
tx_desc = &adapter->tx_desc_base[first];
tx_buffer = &adapter->tx_buffer_area[first];
last = tx_buffer->next_eop;
eop_desc = &adapter->tx_desc_base[last];
/*
* What this does is get the index of the
* first descriptor AFTER the EOP of the
* first packet, that way we can do the
* simple comparison on the inner while loop.
*/
if (++last == adapter->num_tx_desc)
last = 0;
done = last;
bus_dmamap_sync(adapter->txdma.dma_tag, adapter->txdma.dma_map,
BUS_DMASYNC_POSTREAD);
while (eop_desc->upper.fields.status & E1000_TXD_STAT_DD) {
/* We clean the range of the packet */
while (first != done) {
tx_desc->upper.data = 0;
tx_desc->lower.data = 0;
tx_desc->buffer_addr = 0;
++num_avail;
if (tx_buffer->m_head) {
ifp->if_opackets++;
bus_dmamap_sync(adapter->txtag,
tx_buffer->map,
BUS_DMASYNC_POSTWRITE);
bus_dmamap_unload(adapter->txtag,
tx_buffer->map);
m_freem(tx_buffer->m_head);
tx_buffer->m_head = NULL;
}
tx_buffer->next_eop = -1;
adapter->watchdog_time = ticks;
if (++first == adapter->num_tx_desc)
first = 0;
tx_buffer = &adapter->tx_buffer_area[first];
tx_desc = &adapter->tx_desc_base[first];
}
/* See if we can continue to the next packet */
last = tx_buffer->next_eop;
if (last != -1) {
eop_desc = &adapter->tx_desc_base[last];
/* Get new done point */
if (++last == adapter->num_tx_desc) last = 0;
done = last;
} else
break;
}
bus_dmamap_sync(adapter->txdma.dma_tag, adapter->txdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
adapter->next_tx_to_clean = first;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
adapter->num_tx_desc_avail = num_avail;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/*
* If 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 watchdog.
*/
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
if (adapter->num_tx_desc_avail > EM_TX_CLEANUP_THRESHOLD) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
if (adapter->num_tx_desc_avail == adapter->num_tx_desc) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
adapter->watchdog_check = FALSE;
return;
}
}
}
/*********************************************************************
*
* When Link is lost sometimes there is work still in the TX ring
* which may result in a watchdog, rather than allow that we do an
* attempted cleanup and then reinit here. Note that this has been
* seens mostly with fiber adapters.
*
**********************************************************************/
static void
lem_tx_purge(struct adapter *adapter)
{
if ((!adapter->link_active) && (adapter->watchdog_check)) {
EM_TX_LOCK(adapter);
lem_txeof(adapter);
EM_TX_UNLOCK(adapter);
if (adapter->watchdog_check) /* Still outstanding? */
lem_init_locked(adapter);
}
}
/*********************************************************************
*
* Get a buffer from system mbuf buffer pool.
*
**********************************************************************/
static int
lem_get_buf(struct adapter *adapter, int i)
{
struct mbuf *m;
bus_dma_segment_t segs[1];
bus_dmamap_t map;
struct em_buffer *rx_buffer;
int error, nsegs;
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (m == NULL) {
adapter->mbuf_cluster_failed++;
return (ENOBUFS);
}
m->m_len = m->m_pkthdr.len = MCLBYTES;
if (adapter->max_frame_size <= (MCLBYTES - ETHER_ALIGN))
m_adj(m, ETHER_ALIGN);
/*
* Using memory from the mbuf cluster pool, invoke the
* bus_dma machinery to arrange the memory mapping.
*/
error = bus_dmamap_load_mbuf_sg(adapter->rxtag,
adapter->rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT);
if (error != 0) {
m_free(m);
return (error);
}
/* If nsegs is wrong then the stack is corrupt. */
KASSERT(nsegs == 1, ("Too many segments returned!"));
rx_buffer = &adapter->rx_buffer_area[i];
if (rx_buffer->m_head != NULL)
bus_dmamap_unload(adapter->rxtag, rx_buffer->map);
map = rx_buffer->map;
rx_buffer->map = adapter->rx_sparemap;
adapter->rx_sparemap = map;
bus_dmamap_sync(adapter->rxtag, rx_buffer->map, BUS_DMASYNC_PREREAD);
rx_buffer->m_head = m;
adapter->rx_desc_base[i].buffer_addr = htole64(segs[0].ds_addr);
return (0);
}
/*********************************************************************
*
* 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
lem_allocate_receive_structures(struct adapter *adapter)
{
device_t dev = adapter->dev;
struct em_buffer *rx_buffer;
int i, error;
adapter->rx_buffer_area = malloc(sizeof(struct em_buffer) *
adapter->num_rx_desc, M_DEVBUF, M_NOWAIT | M_ZERO);
if (adapter->rx_buffer_area == NULL) {
device_printf(dev, "Unable to allocate rx_buffer memory\n");
return (ENOMEM);
}
error = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
1, 0, /* alignment, bounds */
BUS_SPACE_MAXADDR, /* lowaddr */
BUS_SPACE_MAXADDR, /* highaddr */
NULL, NULL, /* filter, filterarg */
MCLBYTES, /* maxsize */
1, /* nsegments */
MCLBYTES, /* maxsegsize */
0, /* flags */
NULL, /* lockfunc */
NULL, /* lockarg */
&adapter->rxtag);
if (error) {
device_printf(dev, "%s: bus_dma_tag_create failed %d\n",
__func__, error);
goto fail;
}
/* Create the spare map (used by getbuf) */
error = bus_dmamap_create(adapter->rxtag, BUS_DMA_NOWAIT,
&adapter->rx_sparemap);
if (error) {
device_printf(dev, "%s: bus_dmamap_create failed: %d\n",
__func__, error);
goto fail;
}
rx_buffer = adapter->rx_buffer_area;
for (i = 0; i < adapter->num_rx_desc; i++, rx_buffer++) {
error = bus_dmamap_create(adapter->rxtag, BUS_DMA_NOWAIT,
&rx_buffer->map);
if (error) {
device_printf(dev, "%s: bus_dmamap_create failed: %d\n",
__func__, error);
goto fail;
}
}
return (0);
fail:
lem_free_receive_structures(adapter);
return (error);
}
/*********************************************************************
*
* (Re)initialize receive structures.
*
**********************************************************************/
static int
lem_setup_receive_structures(struct adapter *adapter)
{
struct em_buffer *rx_buffer;
int i, error;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
/* we are already under lock */
struct netmap_adapter *na = NA(adapter->ifp);
struct netmap_slot *slot = netmap_reset(na, NR_RX, 0, 0);
#endif
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/* Reset descriptor ring */
bzero(adapter->rx_desc_base,
(sizeof(struct e1000_rx_desc)) * adapter->num_rx_desc);
/* Free current RX buffers. */
rx_buffer = adapter->rx_buffer_area;
for (i = 0; i < adapter->num_rx_desc; i++, rx_buffer++) {
if (rx_buffer->m_head != NULL) {
bus_dmamap_sync(adapter->rxtag, rx_buffer->map,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(adapter->rxtag, rx_buffer->map);
m_freem(rx_buffer->m_head);
rx_buffer->m_head = NULL;
}
}
/* Allocate new ones. */
for (i = 0; i < adapter->num_rx_desc; i++) {
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
if (slot) {
(This commit only touches code within the DEV_NETMAP blocks) Introduce some functions to map NIC ring indexes into netmap ring indexes and vice versa. This way we can implement the bound checks only in one place (and hopefully in a correct way). On passing, make the code and comments more uniform across the various drivers.
2012-02-15 23:13:29 +00:00
/* the i-th NIC entry goes to slot si */
A bunch of netmap fixes: USERSPACE: 1. add support for devices with different number of rx and tx queues; 2. add better support for zero-copy operation, adding an extra field to the netmap ring to indicate how many buffers we have already processed but not yet released (with help from Eddie Kohler); 3. The two changes above unfortunately require an API change, so while at it add a version field and some spares to the ioctl() argument to help detect mismatches. 4. update the manual page for the two changes above; 5. update sample applications in tools/tools/netmap KERNEL: 1. simplify the internal structures moving the global wait queues to the 'struct netmap_adapter'; 2. simplify the functions that map kring<->nic ring indexes 3. normalize device-specific code, helps mainteinance; 4. start exploring the impact of micro-optimizations (prefetch etc.) in the ixgbe driver. Use 'legacy' descriptors on the tx ring and prefetch slots gives about 20% speedup at 900 MHz. Another 7-10% would come from removing the explict calls to bus_dmamap* in the core (they are effectively NOPs in this case, but it takes expensive load of the per-buffer dma maps to figure out that they are all NULL. Rx performance not investigated. I am postponing the MFC so i can import a few more improvements before merging.
2012-02-27 19:05:01 +00:00
int si = netmap_idx_n2k(&na->rx_rings[0], i);
small code cleanup in preparation for future modifications in the memory allocator used by netmap. No functional change, two small bug fixes: - in if_re.c add a missing bus_dmamap_sync() - in netmap.c comment out a spurious free() in an error handling block
2012-01-10 19:57:23 +00:00
uint64_t paddr;
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
void *addr;
small code cleanup in preparation for future modifications in the memory allocator used by netmap. No functional change, two small bug fixes: - in if_re.c add a missing bus_dmamap_sync() - in netmap.c comment out a spurious free() in an error handling block
2012-01-10 19:57:23 +00:00
addr = PNMB(slot + si, &paddr);
netmap_load_map(adapter->rxtag, rx_buffer->map, addr);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
/* Update descriptor */
small code cleanup in preparation for future modifications in the memory allocator used by netmap. No functional change, two small bug fixes: - in if_re.c add a missing bus_dmamap_sync() - in netmap.c comment out a spurious free() in an error handling block
2012-01-10 19:57:23 +00:00
adapter->rx_desc_base[i].buffer_addr = htole64(paddr);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
continue;
}
#endif /* DEV_NETMAP */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
error = lem_get_buf(adapter, i);
if (error)
return (error);
}
/* Setup our descriptor pointers */
adapter->next_rx_desc_to_check = 0;
bus_dmamap_sync(adapter->rxdma.dma_tag, adapter->rxdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
return (0);
}
/*********************************************************************
*
* Enable receive unit.
*
**********************************************************************/
#define MAX_INTS_PER_SEC 8000
#define DEFAULT_ITR 1000000000/(MAX_INTS_PER_SEC * 256)
static void
lem_initialize_receive_unit(struct adapter *adapter)
{
struct ifnet *ifp = adapter->ifp;
u64 bus_addr;
u32 rctl, rxcsum;
INIT_DEBUGOUT("lem_initialize_receive_unit: begin");
/*
* Make sure receives are disabled while setting
* up the descriptor ring
*/
rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, rctl & ~E1000_RCTL_EN);
if (adapter->hw.mac.type >= e1000_82540) {
E1000_WRITE_REG(&adapter->hw, E1000_RADV,
adapter->rx_abs_int_delay.value);
/*
* Set the interrupt throttling rate. Value is calculated
* as DEFAULT_ITR = 1/(MAX_INTS_PER_SEC * 256ns)
*/
E1000_WRITE_REG(&adapter->hw, E1000_ITR, DEFAULT_ITR);
}
/* Setup the Base and Length of the Rx Descriptor Ring */
bus_addr = adapter->rxdma.dma_paddr;
E1000_WRITE_REG(&adapter->hw, E1000_RDLEN(0),
adapter->num_rx_desc * sizeof(struct e1000_rx_desc));
E1000_WRITE_REG(&adapter->hw, E1000_RDBAH(0),
(u32)(bus_addr >> 32));
E1000_WRITE_REG(&adapter->hw, E1000_RDBAL(0),
(u32)bus_addr);
/* Setup the Receive Control Register */
rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
E1000_RCTL_RDMTS_HALF |
(adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
/* Make sure VLAN Filters are off */
rctl &= ~E1000_RCTL_VFE;
if (e1000_tbi_sbp_enabled_82543(&adapter->hw))
rctl |= E1000_RCTL_SBP;
else
rctl &= ~E1000_RCTL_SBP;
switch (adapter->rx_buffer_len) {
default:
case 2048:
rctl |= E1000_RCTL_SZ_2048;
break;
case 4096:
rctl |= E1000_RCTL_SZ_4096 |
E1000_RCTL_BSEX | E1000_RCTL_LPE;
break;
case 8192:
rctl |= E1000_RCTL_SZ_8192 |
E1000_RCTL_BSEX | E1000_RCTL_LPE;
break;
case 16384:
rctl |= E1000_RCTL_SZ_16384 |
E1000_RCTL_BSEX | E1000_RCTL_LPE;
break;
}
if (ifp->if_mtu > ETHERMTU)
rctl |= E1000_RCTL_LPE;
else
rctl &= ~E1000_RCTL_LPE;
/* Enable 82543 Receive Checksum Offload for TCP and UDP */
if ((adapter->hw.mac.type >= e1000_82543) &&
(ifp->if_capenable & IFCAP_RXCSUM)) {
rxcsum = E1000_READ_REG(&adapter->hw, E1000_RXCSUM);
rxcsum |= (E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL);
E1000_WRITE_REG(&adapter->hw, E1000_RXCSUM, rxcsum);
}
/* Enable Receives */
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, rctl);
/*
* Setup the HW Rx Head and
* Tail Descriptor Pointers
*/
E1000_WRITE_REG(&adapter->hw, E1000_RDH(0), 0);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
/* preserve buffers already made available to clients */
if (ifp->if_capenable & IFCAP_NETMAP) {
struct netmap_adapter *na = NA(adapter->ifp);
struct netmap_kring *kring = &na->rx_rings[0];
int t = na->num_rx_desc - 1 - kring->nr_hwavail;
if (t >= na->num_rx_desc)
t -= na->num_rx_desc;
E1000_WRITE_REG(&adapter->hw, E1000_RDT(0), t);
} else
#endif /* DEV_NETMAP */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
E1000_WRITE_REG(&adapter->hw, E1000_RDT(0), adapter->num_rx_desc - 1);
return;
}
/*********************************************************************
*
* Free receive related data structures.
*
**********************************************************************/
static void
lem_free_receive_structures(struct adapter *adapter)
{
struct em_buffer *rx_buffer;
int i;
INIT_DEBUGOUT("free_receive_structures: begin");
if (adapter->rx_sparemap) {
bus_dmamap_destroy(adapter->rxtag, adapter->rx_sparemap);
adapter->rx_sparemap = NULL;
}
/* Cleanup any existing buffers */
if (adapter->rx_buffer_area != NULL) {
rx_buffer = adapter->rx_buffer_area;
for (i = 0; i < adapter->num_rx_desc; i++, rx_buffer++) {
if (rx_buffer->m_head != NULL) {
bus_dmamap_sync(adapter->rxtag, rx_buffer->map,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(adapter->rxtag,
rx_buffer->map);
m_freem(rx_buffer->m_head);
rx_buffer->m_head = NULL;
} else if (rx_buffer->map != NULL)
bus_dmamap_unload(adapter->rxtag,
rx_buffer->map);
if (rx_buffer->map != NULL) {
bus_dmamap_destroy(adapter->rxtag,
rx_buffer->map);
rx_buffer->map = NULL;
}
}
}
if (adapter->rx_buffer_area != NULL) {
free(adapter->rx_buffer_area, M_DEVBUF);
adapter->rx_buffer_area = NULL;
}
if (adapter->rxtag != NULL) {
bus_dma_tag_destroy(adapter->rxtag);
adapter->rxtag = NULL;
}
}
/*********************************************************************
*
* 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.
*
* For polling we also now return the number of cleaned packets
*********************************************************************/
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
static bool
lem_rxeof(struct adapter *adapter, int count, int *done)
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
{
remove duplicate semicolons where possible. Approved by: cperciva MFC after: 1 week
2012-10-22 03:00:37 +00:00
struct ifnet *ifp = adapter->ifp;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
struct mbuf *mp;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
u8 status = 0, accept_frame = 0, eop = 0;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
u16 len, desc_len, prev_len_adj;
int i, rx_sent = 0;
struct e1000_rx_desc *current_desc;
EM_RX_LOCK(adapter);
i = adapter->next_rx_desc_to_check;
current_desc = &adapter->rx_desc_base[i];
bus_dmamap_sync(adapter->rxdma.dma_tag, adapter->rxdma.dma_map,
BUS_DMASYNC_POSTREAD);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
#ifdef DEV_NETMAP
if (ifp->if_capenable & IFCAP_NETMAP) {
A bunch of netmap fixes: USERSPACE: 1. add support for devices with different number of rx and tx queues; 2. add better support for zero-copy operation, adding an extra field to the netmap ring to indicate how many buffers we have already processed but not yet released (with help from Eddie Kohler); 3. The two changes above unfortunately require an API change, so while at it add a version field and some spares to the ioctl() argument to help detect mismatches. 4. update the manual page for the two changes above; 5. update sample applications in tools/tools/netmap KERNEL: 1. simplify the internal structures moving the global wait queues to the 'struct netmap_adapter'; 2. simplify the functions that map kring<->nic ring indexes 3. normalize device-specific code, helps mainteinance; 4. start exploring the impact of micro-optimizations (prefetch etc.) in the ixgbe driver. Use 'legacy' descriptors on the tx ring and prefetch slots gives about 20% speedup at 900 MHz. Another 7-10% would come from removing the explict calls to bus_dmamap* in the core (they are effectively NOPs in this case, but it takes expensive load of the per-buffer dma maps to figure out that they are all NULL. Rx performance not investigated. I am postponing the MFC so i can import a few more improvements before merging.
2012-02-27 19:05:01 +00:00
struct netmap_adapter *na = NA(ifp);
na->rx_rings[0].nr_kflags |= NKR_PENDINTR;
selwakeuppri(&na->rx_rings[0].si, PI_NET);
add netmap support for "em", "lem", "igb" and "re". On my hardware, "em" in netmap mode does about 1.388 Mpps on one card (on an Asus motherboard), and 1.1 Mpps on another card (PCIe bus). Both seem to be NIC-limited, because i have the same rate even with the CPU running at 150 MHz. On the "re" driver the tx throughput is around 420-450 Kpps on various (8111C and the like) chipsets. On the Rx side performance seems much better, and i can receive the full load generated by the "em" cards. "igb" is untested as i don't have the hardware.
2011-12-05 15:33:13 +00:00
EM_RX_UNLOCK(adapter);
return (0);
}
#endif /* DEV_NETMAP */
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (!((current_desc->status) & E1000_RXD_STAT_DD)) {
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
if (done != NULL)
*done = rx_sent;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
EM_RX_UNLOCK(adapter);
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
return (FALSE);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
while (count != 0 && ifp->if_drv_flags & IFF_DRV_RUNNING) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
struct mbuf *m = NULL;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
status = current_desc->status;
if ((status & E1000_RXD_STAT_DD) == 0)
break;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
mp = adapter->rx_buffer_area[i].m_head;
/*
* Can't defer bus_dmamap_sync(9) because TBI_ACCEPT
* needs to access the last received byte in the mbuf.
*/
bus_dmamap_sync(adapter->rxtag, adapter->rx_buffer_area[i].map,
BUS_DMASYNC_POSTREAD);
accept_frame = 1;
prev_len_adj = 0;
desc_len = le16toh(current_desc->length);
if (status & E1000_RXD_STAT_EOP) {
count--;
eop = 1;
if (desc_len < ETHER_CRC_LEN) {
len = 0;
prev_len_adj = ETHER_CRC_LEN - desc_len;
} else
len = desc_len - ETHER_CRC_LEN;
} else {
eop = 0;
len = desc_len;
}
if (current_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
u8 last_byte;
u32 pkt_len = desc_len;
if (adapter->fmp != NULL)
pkt_len += adapter->fmp->m_pkthdr.len;
last_byte = *(mtod(mp, caddr_t) + desc_len - 1);
if (TBI_ACCEPT(&adapter->hw, status,
current_desc->errors, pkt_len, last_byte,
adapter->min_frame_size, adapter->max_frame_size)) {
e1000_tbi_adjust_stats_82543(&adapter->hw,
&adapter->stats, pkt_len,
adapter->hw.mac.addr,
adapter->max_frame_size);
if (len > 0)
len--;
} else
accept_frame = 0;
}
if (accept_frame) {
if (lem_get_buf(adapter, i) != 0) {
ifp->if_iqdrops++;
goto discard;
}
/* Assign correct length to the current fragment */
mp->m_len = len;
if (adapter->fmp == NULL) {
mp->m_pkthdr.len = len;
adapter->fmp = mp; /* Store the first mbuf */
adapter->lmp = mp;
} else {
/* Chain mbuf's together */
mp->m_flags &= ~M_PKTHDR;
/*
* Adjust length of previous mbuf in chain if
* we received less than 4 bytes in the last
* descriptor.
*/
if (prev_len_adj > 0) {
adapter->lmp->m_len -= prev_len_adj;
adapter->fmp->m_pkthdr.len -=
prev_len_adj;
}
adapter->lmp->m_next = mp;
adapter->lmp = adapter->lmp->m_next;
adapter->fmp->m_pkthdr.len += len;
}
if (eop) {
adapter->fmp->m_pkthdr.rcvif = ifp;
ifp->if_ipackets++;
lem_receive_checksum(adapter, current_desc,
adapter->fmp);
#ifndef __NO_STRICT_ALIGNMENT
if (adapter->max_frame_size >
(MCLBYTES - ETHER_ALIGN) &&
lem_fixup_rx(adapter) != 0)
goto skip;
#endif
if (status & E1000_RXD_STAT_VP) {
adapter->fmp->m_pkthdr.ether_vtag =
When extracting the VLAN tag from if_em and if_lem receive descriptor rings, copy the whole VLAN tag, not just the VLAN ID. This fixes a problem in which VLAN priority information was dropped when using offloaded VLAN processing with these drivers. Discussed with: jfv, rrs Sponsored by: ADARA Networks, Inc. MFC after: 3 days
2012-01-05 17:30:15 +00:00
le16toh(current_desc->special);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
adapter->fmp->m_flags |= M_VLANTAG;
}
#ifndef __NO_STRICT_ALIGNMENT
skip:
#endif
m = adapter->fmp;
adapter->fmp = NULL;
adapter->lmp = NULL;
}
} else {
This patch fixes a nit in the em, lem, and igb driver statistics. Increment adapter->dropped_pkts instead of if_ierrors because if_ierrors is overwritten by hw stats collection. Submitted by: Andrew Boyer <aboyer@averesystems.com> Reviewed by: Jack F Vogel <jfv@freebsd.org> MFC after: 2 weeks
2012-09-23 22:53:39 +00:00
adapter->dropped_pkts++;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
discard:
/* Reuse loaded DMA map and just update mbuf chain */
mp = adapter->rx_buffer_area[i].m_head;
mp->m_len = mp->m_pkthdr.len = MCLBYTES;
mp->m_data = mp->m_ext.ext_buf;
mp->m_next = NULL;
if (adapter->max_frame_size <=
(MCLBYTES - ETHER_ALIGN))
m_adj(mp, ETHER_ALIGN);
if (adapter->fmp != NULL) {
m_freem(adapter->fmp);
adapter->fmp = NULL;
adapter->lmp = NULL;
}
m = NULL;
}
/* Zero out the receive descriptors status. */
current_desc->status = 0;
bus_dmamap_sync(adapter->rxdma.dma_tag, adapter->rxdma.dma_map,
BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
/* Advance our pointers to the next descriptor. */
if (++i == adapter->num_rx_desc)
i = 0;
/* Call into the stack */
if (m != NULL) {
adapter->next_rx_desc_to_check = i;
EM_RX_UNLOCK(adapter);
(*ifp->if_input)(ifp, m);
EM_RX_LOCK(adapter);
rx_sent++;
i = adapter->next_rx_desc_to_check;
}
current_desc = &adapter->rx_desc_base[i];
}
adapter->next_rx_desc_to_check = i;
/* Advance the E1000's Receive Queue #0 "Tail Pointer". */
if (--i < 0)
i = adapter->num_rx_desc - 1;
E1000_WRITE_REG(&adapter->hw, E1000_RDT(0), i);
Changes from John Baldwin adding to last commit, change rxeof api for poll friendliness, and eliminate unnecessary link tasklet use. Thanks John!
2010-06-16 16:37:36 +00:00
if (done != NULL)
*done = rx_sent;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
EM_RX_UNLOCK(adapter);
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
return ((status & E1000_RXD_STAT_DD) ? TRUE : FALSE);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
#ifndef __NO_STRICT_ALIGNMENT
/*
* When jumbo frames are enabled we should realign entire payload on
* architecures with strict alignment. This is serious design mistake of 8254x
* as it nullifies DMA operations. 8254x just allows RX buffer size to be
* 2048/4096/8192/16384. What we really want is 2048 - ETHER_ALIGN to align its
* payload. On architecures without strict alignment restrictions 8254x still
* performs unaligned memory access which would reduce the performance too.
* To avoid copying over an entire frame to align, we allocate a new mbuf and
* copy ethernet header to the new mbuf. The new mbuf is prepended into the
* existing mbuf chain.
*
* Be aware, best performance of the 8254x is achived only when jumbo frame is
* not used at all on architectures with strict alignment.
*/
static int
lem_fixup_rx(struct adapter *adapter)
{
struct mbuf *m, *n;
int error;
error = 0;
m = adapter->fmp;
if (m->m_len <= (MCLBYTES - ETHER_HDR_LEN)) {
bcopy(m->m_data, m->m_data + ETHER_HDR_LEN, m->m_len);
m->m_data += ETHER_HDR_LEN;
} else {
Mechanically substitute flags from historic mbuf allocator with malloc(9) flags in sys/dev.
2012-12-04 09:32:43 +00:00
MGETHDR(n, M_NOWAIT, MT_DATA);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
if (n != NULL) {
bcopy(m->m_data, n->m_data, ETHER_HDR_LEN);
m->m_data += ETHER_HDR_LEN;
m->m_len -= ETHER_HDR_LEN;
n->m_len = ETHER_HDR_LEN;
M_MOVE_PKTHDR(n, m);
n->m_next = m;
adapter->fmp = n;
} else {
adapter->dropped_pkts++;
m_freem(adapter->fmp);
adapter->fmp = NULL;
error = ENOMEM;
}
}
return (error);
}
#endif
/*********************************************************************
*
* 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
lem_receive_checksum(struct adapter *adapter,
struct e1000_rx_desc *rx_desc, struct mbuf *mp)
{
/* 82543 or newer only */
if ((adapter->hw.mac.type < e1000_82543) ||
/* Ignore Checksum bit is set */
(rx_desc->status & E1000_RXD_STAT_IXSM)) {
mp->m_pkthdr.csum_flags = 0;
return;
}
if (rx_desc->status & E1000_RXD_STAT_IPCS) {
/* Did it pass? */
if (!(rx_desc->errors & E1000_RXD_ERR_IPE)) {
/* IP Checksum Good */
mp->m_pkthdr.csum_flags = CSUM_IP_CHECKED;
mp->m_pkthdr.csum_flags |= CSUM_IP_VALID;
} else {
mp->m_pkthdr.csum_flags = 0;
}
}
if (rx_desc->status & E1000_RXD_STAT_TCPCS) {
/* Did it pass? */
if (!(rx_desc->errors & E1000_RXD_ERR_TCPE)) {
mp->m_pkthdr.csum_flags |=
(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
mp->m_pkthdr.csum_data = htons(0xffff);
}
}
}
/*
* This routine is run via an vlan
* config EVENT
*/
static void
lem_register_vlan(void *arg, struct ifnet *ifp, u16 vtag)
{
struct adapter *adapter = ifp->if_softc;
u32 index, bit;
if (ifp->if_softc != arg) /* Not our event */
return;
if ((vtag == 0) || (vtag > 4095)) /* Invalid ID */
return;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
EM_CORE_LOCK(adapter);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
index = (vtag >> 5) & 0x7F;
bit = vtag & 0x1F;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
adapter->shadow_vfta[index] |= (1 << bit);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
++adapter->num_vlans;
/* Re-init to load the changes */
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
lem_init_locked(adapter);
EM_CORE_UNLOCK(adapter);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
/*
* This routine is run via an vlan
* unconfig EVENT
*/
static void
lem_unregister_vlan(void *arg, struct ifnet *ifp, u16 vtag)
{
struct adapter *adapter = ifp->if_softc;
u32 index, bit;
if (ifp->if_softc != arg)
return;
if ((vtag == 0) || (vtag > 4095)) /* Invalid */
return;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
EM_CORE_LOCK(adapter);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
index = (vtag >> 5) & 0x7F;
bit = vtag & 0x1F;
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
adapter->shadow_vfta[index] &= ~(1 << bit);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
--adapter->num_vlans;
/* Re-init to load the changes */
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
lem_init_locked(adapter);
EM_CORE_UNLOCK(adapter);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
static void
lem_setup_vlan_hw_support(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
u32 reg;
/*
** We get here thru init_locked, meaning
** a soft reset, this has already cleared
** the VFTA and other state, so if there
** have been no vlan's registered do nothing.
*/
if (adapter->num_vlans == 0)
return;
/*
** A soft reset zero's out the VFTA, so
** we need to repopulate it now.
*/
for (int i = 0; i < EM_VFTA_SIZE; i++)
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
if (adapter->shadow_vfta[i] != 0)
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
E1000_WRITE_REG_ARRAY(hw, E1000_VFTA,
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
i, adapter->shadow_vfta[i]);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
reg = E1000_READ_REG(hw, E1000_CTRL);
reg |= E1000_CTRL_VME;
E1000_WRITE_REG(hw, E1000_CTRL, reg);
/* Enable the Filter Table */
reg = E1000_READ_REG(hw, E1000_RCTL);
reg &= ~E1000_RCTL_CFIEN;
reg |= E1000_RCTL_VFE;
E1000_WRITE_REG(hw, E1000_RCTL, reg);
/* Update the frame size */
E1000_WRITE_REG(&adapter->hw, E1000_RLPML,
adapter->max_frame_size + VLAN_TAG_SIZE);
}
static void
lem_enable_intr(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
u32 ims_mask = IMS_ENABLE_MASK;
E1000_WRITE_REG(hw, E1000_IMS, ims_mask);
}
static void
lem_disable_intr(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
Clean up some unused leftover code from em Make IRQ style a tuneable Fix lock handling in the interrupt handler MFC after:3 days
2012-07-31 18:44:10 +00:00
E1000_WRITE_REG(hw, E1000_IMC, 0xffffffff);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
/*
* Bit of a misnomer, what this really means is
* to enable OS management of the system... aka
* to disable special hardware management features
*/
static void
lem_init_manageability(struct adapter *adapter)
{
/* A shared code workaround */
if (adapter->has_manage) {
int manc = E1000_READ_REG(&adapter->hw, E1000_MANC);
/* disable hardware interception of ARP */
manc &= ~(E1000_MANC_ARP_EN);
E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc);
}
}
/*
* Give control back to hardware management
* controller if there is one.
*/
static void
lem_release_manageability(struct adapter *adapter)
{
if (adapter->has_manage) {
int manc = E1000_READ_REG(&adapter->hw, E1000_MANC);
/* re-enable hardware interception of ARP */
manc |= E1000_MANC_ARP_EN;
E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc);
}
}
/*
* lem_get_hw_control sets the {CTRL_EXT|FWSM}:DRV_LOAD bit.
* For ASF and Pass Through versions of f/w this means
* that the driver is loaded. For AMT version type f/w
* this means that the network i/f is open.
*/
static void
lem_get_hw_control(struct adapter *adapter)
{
u32 ctrl_ext;
ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT,
ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
return;
}
/*
* lem_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
* For ASF and Pass Through versions of f/w this means that
* the driver is no longer loaded. For AMT versions of the
* f/w this means that the network i/f is closed.
*/
static void
lem_release_hw_control(struct adapter *adapter)
{
u32 ctrl_ext;
if (!adapter->has_manage)
return;
ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT,
ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
return;
}
static int
lem_is_valid_ether_addr(u8 *addr)
{
char zero_addr[6] = { 0, 0, 0, 0, 0, 0 };
if ((addr[0] & 1) || (!bcmp(addr, zero_addr, ETHER_ADDR_LEN))) {
return (FALSE);
}
return (TRUE);
}
/*
** Parse the interface capabilities with regard
** to both system management and wake-on-lan for
** later use.
*/
static void
lem_get_wakeup(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
u16 eeprom_data = 0, device_id, apme_mask;
adapter->has_manage = e1000_enable_mng_pass_thru(&adapter->hw);
apme_mask = EM_EEPROM_APME;
switch (adapter->hw.mac.type) {
case e1000_82542:
case e1000_82543:
break;
case e1000_82544:
e1000_read_nvm(&adapter->hw,
NVM_INIT_CONTROL2_REG, 1, &eeprom_data);
apme_mask = EM_82544_APME;
break;
case e1000_82546:
case e1000_82546_rev_3:
if (adapter->hw.bus.func == 1) {
e1000_read_nvm(&adapter->hw,
NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
break;
} else
e1000_read_nvm(&adapter->hw,
NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
break;
default:
e1000_read_nvm(&adapter->hw,
NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
break;
}
if (eeprom_data & apme_mask)
adapter->wol = (E1000_WUFC_MAG | E1000_WUFC_MC);
/*
* We have the eeprom settings, now apply the special cases
* where the eeprom may be wrong or the board won't support
* wake on lan on a particular port
*/
device_id = pci_get_device(dev);
switch (device_id) {
case E1000_DEV_ID_82546GB_PCIE:
adapter->wol = 0;
break;
case E1000_DEV_ID_82546EB_FIBER:
case E1000_DEV_ID_82546GB_FIBER:
/* Wake events only supported on port A for dual fiber
* regardless of eeprom setting */
if (E1000_READ_REG(&adapter->hw, E1000_STATUS) &
E1000_STATUS_FUNC_1)
adapter->wol = 0;
break;
case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
/* if quad port adapter, disable WoL on all but port A */
if (global_quad_port_a != 0)
adapter->wol = 0;
/* Reset for multiple quad port adapters */
if (++global_quad_port_a == 4)
global_quad_port_a = 0;
break;
}
return;
}
/*
* Enable PCI Wake On Lan capability
*/
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
static void
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
lem_enable_wakeup(device_t dev)
{
struct adapter *adapter = device_get_softc(dev);
struct ifnet *ifp = adapter->ifp;
u32 pmc, ctrl, ctrl_ext, rctl;
u16 status;
Do a sweep of the tree replacing calls to pci_find_extcap() with calls to pci_find_cap() instead.
2011-03-23 13:10:15 +00:00
if ((pci_find_cap(dev, PCIY_PMG, &pmc) != 0))
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
return;
/* Advertise the wakeup capability */
ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL);
ctrl |= (E1000_CTRL_SWDPIN2 | E1000_CTRL_SWDPIN3);
E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl);
E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN);
/* Keep the laser running on Fiber adapters */
if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT);
ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, ctrl_ext);
}
/*
** Determine type of Wakeup: note that wol
** is set with all bits on by default.
*/
if ((ifp->if_capenable & IFCAP_WOL_MAGIC) == 0)
adapter->wol &= ~E1000_WUFC_MAG;
if ((ifp->if_capenable & IFCAP_WOL_MCAST) == 0)
adapter->wol &= ~E1000_WUFC_MC;
else {
rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL);
rctl |= E1000_RCTL_MPE;
E1000_WRITE_REG(&adapter->hw, E1000_RCTL, rctl);
}
if (adapter->hw.mac.type == e1000_pchlan) {
if (lem_enable_phy_wakeup(adapter))
return;
} else {
E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN);
E1000_WRITE_REG(&adapter->hw, E1000_WUFC, adapter->wol);
}
/* Request PME */
status = pci_read_config(dev, pmc + PCIR_POWER_STATUS, 2);
status &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
if (ifp->if_capenable & IFCAP_WOL)
status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
pci_write_config(dev, pmc + PCIR_POWER_STATUS, status, 2);
return;
}
/*
** WOL in the newer chipset interfaces (pchlan)
** require thing to be copied into the phy
*/
static int
lem_enable_phy_wakeup(struct adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
u32 mreg, ret = 0;
u16 preg;
/* copy MAC RARs to PHY RARs */
for (int i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
mreg = E1000_READ_REG(hw, E1000_RAL(i));
e1000_write_phy_reg(hw, BM_RAR_L(i), (u16)(mreg & 0xFFFF));
e1000_write_phy_reg(hw, BM_RAR_M(i),
(u16)((mreg >> 16) & 0xFFFF));
mreg = E1000_READ_REG(hw, E1000_RAH(i));
e1000_write_phy_reg(hw, BM_RAR_H(i), (u16)(mreg & 0xFFFF));
e1000_write_phy_reg(hw, BM_RAR_CTRL(i),
(u16)((mreg >> 16) & 0xFFFF));
}
/* copy MAC MTA to PHY MTA */
for (int i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
mreg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
e1000_write_phy_reg(hw, BM_MTA(i), (u16)(mreg & 0xFFFF));
e1000_write_phy_reg(hw, BM_MTA(i) + 1,
(u16)((mreg >> 16) & 0xFFFF));
}
/* configure PHY Rx Control register */
e1000_read_phy_reg(&adapter->hw, BM_RCTL, &preg);
mreg = E1000_READ_REG(hw, E1000_RCTL);
if (mreg & E1000_RCTL_UPE)
preg |= BM_RCTL_UPE;
if (mreg & E1000_RCTL_MPE)
preg |= BM_RCTL_MPE;
preg &= ~(BM_RCTL_MO_MASK);
if (mreg & E1000_RCTL_MO_3)
preg |= (((mreg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
<< BM_RCTL_MO_SHIFT);
if (mreg & E1000_RCTL_BAM)
preg |= BM_RCTL_BAM;
if (mreg & E1000_RCTL_PMCF)
preg |= BM_RCTL_PMCF;
mreg = E1000_READ_REG(hw, E1000_CTRL);
if (mreg & E1000_CTRL_RFCE)
preg |= BM_RCTL_RFCE;
e1000_write_phy_reg(&adapter->hw, BM_RCTL, preg);
/* enable PHY wakeup in MAC register */
E1000_WRITE_REG(hw, E1000_WUC,
E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
E1000_WRITE_REG(hw, E1000_WUFC, adapter->wol);
/* configure and enable PHY wakeup in PHY registers */
e1000_write_phy_reg(&adapter->hw, BM_WUFC, adapter->wol);
e1000_write_phy_reg(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
/* activate PHY wakeup */
ret = hw->phy.ops.acquire(hw);
if (ret) {
printf("Could not acquire PHY\n");
return ret;
}
e1000_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
(BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
ret = e1000_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &preg);
if (ret) {
printf("Could not read PHY page 769\n");
goto out;
}
preg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
ret = e1000_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, preg);
if (ret)
printf("Could not set PHY Host Wakeup bit\n");
out:
hw->phy.ops.release(hw);
return ret;
}
Hook the identification LEDs of igb(4), lem(4) and em(4) devices up with led(4) so they can be lit or f.e. made blink via `echo f2 > /dev/led/em0` for localization purposes. Approved by: jfv MFC afer: 1 week (after r205869)
2010-03-31 20:43:24 +00:00
static void
lem_led_func(void *arg, int onoff)
{
struct adapter *adapter = arg;
EM_CORE_LOCK(adapter);
if (onoff) {
e1000_setup_led(&adapter->hw);
e1000_led_on(&adapter->hw);
} else {
e1000_led_off(&adapter->hw);
e1000_cleanup_led(&adapter->hw);
}
EM_CORE_UNLOCK(adapter);
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
/*********************************************************************
* 82544 Coexistence issue workaround.
* There are 2 issues.
* 1. Transmit Hang issue.
* To detect this issue, following equation can be used...
* SIZE[3:0] + ADDR[2:0] = SUM[3:0].
* If SUM[3:0] is in between 1 to 4, we will have this issue.
*
* 2. DAC issue.
* To detect this issue, following equation can be used...
* SIZE[3:0] + ADDR[2:0] = SUM[3:0].
* If SUM[3:0] is in between 9 to c, we will have this issue.
*
*
* WORKAROUND:
* Make sure we do not have ending address
* as 1,2,3,4(Hang) or 9,a,b,c (DAC)
*
*************************************************************************/
static u32
lem_fill_descriptors (bus_addr_t address, u32 length,
PDESC_ARRAY desc_array)
{
u32 safe_terminator;
/* Since issue is sensitive to length and address.*/
/* Let us first check the address...*/
if (length <= 4) {
desc_array->descriptor[0].address = address;
desc_array->descriptor[0].length = length;
desc_array->elements = 1;
return (desc_array->elements);
}
safe_terminator = (u32)((((u32)address & 0x7) +
(length & 0xF)) & 0xF);
/* if it does not fall between 0x1 to 0x4 and 0x9 to 0xC then return */
if (safe_terminator == 0 ||
(safe_terminator > 4 &&
safe_terminator < 9) ||
(safe_terminator > 0xC &&
safe_terminator <= 0xF)) {
desc_array->descriptor[0].address = address;
desc_array->descriptor[0].length = length;
desc_array->elements = 1;
return (desc_array->elements);
}
desc_array->descriptor[0].address = address;
desc_array->descriptor[0].length = length - 4;
desc_array->descriptor[1].address = address + (length - 4);
desc_array->descriptor[1].length = 4;
desc_array->elements = 2;
return (desc_array->elements);
}
/**********************************************************************
*
* Update the board statistics counters.
*
**********************************************************************/
static void
lem_update_stats_counters(struct adapter *adapter)
{
struct ifnet *ifp;
if(adapter->hw.phy.media_type == e1000_media_type_copper ||
(E1000_READ_REG(&adapter->hw, E1000_STATUS) & E1000_STATUS_LU)) {
adapter->stats.symerrs += E1000_READ_REG(&adapter->hw, E1000_SYMERRS);
adapter->stats.sec += E1000_READ_REG(&adapter->hw, E1000_SEC);
}
adapter->stats.crcerrs += E1000_READ_REG(&adapter->hw, E1000_CRCERRS);
adapter->stats.mpc += E1000_READ_REG(&adapter->hw, E1000_MPC);
adapter->stats.scc += E1000_READ_REG(&adapter->hw, E1000_SCC);
adapter->stats.ecol += E1000_READ_REG(&adapter->hw, E1000_ECOL);
adapter->stats.mcc += E1000_READ_REG(&adapter->hw, E1000_MCC);
adapter->stats.latecol += E1000_READ_REG(&adapter->hw, E1000_LATECOL);
adapter->stats.colc += E1000_READ_REG(&adapter->hw, E1000_COLC);
adapter->stats.dc += E1000_READ_REG(&adapter->hw, E1000_DC);
adapter->stats.rlec += E1000_READ_REG(&adapter->hw, E1000_RLEC);
adapter->stats.xonrxc += E1000_READ_REG(&adapter->hw, E1000_XONRXC);
adapter->stats.xontxc += E1000_READ_REG(&adapter->hw, E1000_XONTXC);
adapter->stats.xoffrxc += E1000_READ_REG(&adapter->hw, E1000_XOFFRXC);
adapter->stats.xofftxc += E1000_READ_REG(&adapter->hw, E1000_XOFFTXC);
adapter->stats.fcruc += E1000_READ_REG(&adapter->hw, E1000_FCRUC);
adapter->stats.prc64 += E1000_READ_REG(&adapter->hw, E1000_PRC64);
adapter->stats.prc127 += E1000_READ_REG(&adapter->hw, E1000_PRC127);
adapter->stats.prc255 += E1000_READ_REG(&adapter->hw, E1000_PRC255);
adapter->stats.prc511 += E1000_READ_REG(&adapter->hw, E1000_PRC511);
adapter->stats.prc1023 += E1000_READ_REG(&adapter->hw, E1000_PRC1023);
adapter->stats.prc1522 += E1000_READ_REG(&adapter->hw, E1000_PRC1522);
adapter->stats.gprc += E1000_READ_REG(&adapter->hw, E1000_GPRC);
adapter->stats.bprc += E1000_READ_REG(&adapter->hw, E1000_BPRC);
adapter->stats.mprc += E1000_READ_REG(&adapter->hw, E1000_MPRC);
adapter->stats.gptc += E1000_READ_REG(&adapter->hw, E1000_GPTC);
/* For the 64-bit byte counters the low dword must be read first. */
/* Both registers clear on the read of the high dword */
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
adapter->stats.gorc += E1000_READ_REG(&adapter->hw, E1000_GORCL) +
((u64)E1000_READ_REG(&adapter->hw, E1000_GORCH) << 32);
adapter->stats.gotc += E1000_READ_REG(&adapter->hw, E1000_GOTCL) +
((u64)E1000_READ_REG(&adapter->hw, E1000_GOTCH) << 32);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
adapter->stats.rnbc += E1000_READ_REG(&adapter->hw, E1000_RNBC);
adapter->stats.ruc += E1000_READ_REG(&adapter->hw, E1000_RUC);
adapter->stats.rfc += E1000_READ_REG(&adapter->hw, E1000_RFC);
adapter->stats.roc += E1000_READ_REG(&adapter->hw, E1000_ROC);
adapter->stats.rjc += E1000_READ_REG(&adapter->hw, E1000_RJC);
adapter->stats.tor += E1000_READ_REG(&adapter->hw, E1000_TORH);
adapter->stats.tot += E1000_READ_REG(&adapter->hw, E1000_TOTH);
adapter->stats.tpr += E1000_READ_REG(&adapter->hw, E1000_TPR);
adapter->stats.tpt += E1000_READ_REG(&adapter->hw, E1000_TPT);
adapter->stats.ptc64 += E1000_READ_REG(&adapter->hw, E1000_PTC64);
adapter->stats.ptc127 += E1000_READ_REG(&adapter->hw, E1000_PTC127);
adapter->stats.ptc255 += E1000_READ_REG(&adapter->hw, E1000_PTC255);
adapter->stats.ptc511 += E1000_READ_REG(&adapter->hw, E1000_PTC511);
adapter->stats.ptc1023 += E1000_READ_REG(&adapter->hw, E1000_PTC1023);
adapter->stats.ptc1522 += E1000_READ_REG(&adapter->hw, E1000_PTC1522);
adapter->stats.mptc += E1000_READ_REG(&adapter->hw, E1000_MPTC);
adapter->stats.bptc += E1000_READ_REG(&adapter->hw, E1000_BPTC);
if (adapter->hw.mac.type >= e1000_82543) {
adapter->stats.algnerrc +=
E1000_READ_REG(&adapter->hw, E1000_ALGNERRC);
adapter->stats.rxerrc +=
E1000_READ_REG(&adapter->hw, E1000_RXERRC);
adapter->stats.tncrs +=
E1000_READ_REG(&adapter->hw, E1000_TNCRS);
adapter->stats.cexterr +=
E1000_READ_REG(&adapter->hw, E1000_CEXTERR);
adapter->stats.tsctc +=
E1000_READ_REG(&adapter->hw, E1000_TSCTC);
adapter->stats.tsctfc +=
E1000_READ_REG(&adapter->hw, E1000_TSCTFC);
}
ifp = adapter->ifp;
ifp->if_collisions = adapter->stats.colc;
/* Rx Errors */
ifp->if_ierrors = adapter->dropped_pkts + adapter->stats.rxerrc +
adapter->stats.crcerrs + adapter->stats.algnerrc +
adapter->stats.ruc + adapter->stats.roc +
adapter->stats.mpc + adapter->stats.cexterr;
/* Tx Errors */
ifp->if_oerrors = adapter->stats.ecol +
adapter->stats.latecol + adapter->watchdog_events;
}
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
/* Export a single 32-bit register via a read-only sysctl. */
static int
lem_sysctl_reg_handler(SYSCTL_HANDLER_ARGS)
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
{
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
struct adapter *adapter;
u_int val;
adapter = oidp->oid_arg1;
val = E1000_READ_REG(&adapter->hw, oidp->oid_arg2);
return (sysctl_handle_int(oidp, &val, 0, req));
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
/*
* Add sysctl variables, one per statistic, to the system.
*/
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
static void
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
lem_add_hw_stats(struct adapter *adapter)
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
{
device_t dev = adapter->dev;
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(dev);
struct sysctl_oid *tree = device_get_sysctl_tree(dev);
struct sysctl_oid_list *child = SYSCTL_CHILDREN(tree);
struct e1000_hw_stats *stats = &adapter->stats;
struct sysctl_oid *stat_node;
struct sysctl_oid_list *stat_list;
/* Driver Statistics */
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "mbuf_alloc_fail",
CTLFLAG_RD, &adapter->mbuf_alloc_failed,
"Std mbuf failed");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "cluster_alloc_fail",
CTLFLAG_RD, &adapter->mbuf_cluster_failed,
"Std mbuf cluster failed");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "dropped",
CTLFLAG_RD, &adapter->dropped_pkts,
"Driver dropped packets");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_dma_fail",
CTLFLAG_RD, &adapter->no_tx_dma_setup,
"Driver tx dma failure in xmit");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_desc_fail1",
CTLFLAG_RD, &adapter->no_tx_desc_avail1,
"Not enough tx descriptors failure in xmit");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_desc_fail2",
CTLFLAG_RD, &adapter->no_tx_desc_avail2,
"Not enough tx descriptors failure in xmit");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_overruns",
CTLFLAG_RD, &adapter->rx_overruns,
"RX overruns");
SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "watchdog_timeouts",
CTLFLAG_RD, &adapter->watchdog_events,
"Watchdog timeouts");
SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "device_control",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_CTRL,
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
lem_sysctl_reg_handler, "IU",
"Device Control Register");
SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_control",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_RCTL,
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
lem_sysctl_reg_handler, "IU",
"Receiver Control Register");
SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "fc_high_water",
CTLFLAG_RD, &adapter->hw.fc.high_water, 0,
"Flow Control High Watermark");
SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "fc_low_water",
CTLFLAG_RD, &adapter->hw.fc.low_water, 0,
"Flow Control Low Watermark");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "fifo_workaround",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->tx_fifo_wrk_cnt,
"TX FIFO workaround events");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "fifo_reset",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->tx_fifo_reset_cnt,
"TX FIFO resets");
SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txd_head",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_TDH(0),
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
lem_sysctl_reg_handler, "IU",
"Transmit Descriptor Head");
SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txd_tail",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_TDT(0),
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
lem_sysctl_reg_handler, "IU",
"Transmit Descriptor Tail");
SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rxd_head",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_RDH(0),
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
lem_sysctl_reg_handler, "IU",
"Receive Descriptor Head");
SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rxd_tail",
Specify a CTLTYPE_FOO so that a future sysctl(8) change does not need to rely on the format string.
2011-01-18 21:14:23 +00:00
CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_RDT(0),
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
lem_sysctl_reg_handler, "IU",
"Receive Descriptor Tail");
/* MAC stats get their own sub node */
stat_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "mac_stats",
CTLFLAG_RD, NULL, "Statistics");
stat_list = SYSCTL_CHILDREN(stat_node);
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "excess_coll",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &stats->ecol,
"Excessive collisions");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "single_coll",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &stats->scc,
"Single collisions");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "multiple_coll",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &stats->mcc,
"Multiple collisions");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "late_coll",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &stats->latecol,
"Late collisions");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "collision_count",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &stats->colc,
"Collision Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "symbol_errors",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.symerrs,
"Symbol Errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "sequence_errors",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.sec,
"Sequence Errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "defer_count",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.dc,
"Defer Count");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "missed_packets",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.mpc,
"Missed Packets");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_no_buff",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.rnbc,
"Receive No Buffers");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_undersize",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.ruc,
"Receive Undersize");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_fragmented",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.rfc,
"Fragmented Packets Received ");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_oversize",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.roc,
"Oversized Packets Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_jabber",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.rjc,
"Recevied Jabber");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_errs",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.rxerrc,
"Receive Errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "crc_errs",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.crcerrs,
"CRC errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "alignment_errs",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.algnerrc,
"Alignment Errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "coll_ext_errs",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.cexterr,
"Collision/Carrier extension errors");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xon_recvd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.xonrxc,
"XON Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xon_txd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.xontxc,
"XON Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xoff_recvd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.xoffrxc,
"XOFF Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xoff_txd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.xofftxc,
"XOFF Transmitted");
/* Packet Reception Stats */
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "total_pkts_recvd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.tpr,
"Total Packets Received ");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_pkts_recvd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.gprc,
"Good Packets Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_recvd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.bprc,
"Broadcast Packets Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_recvd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.mprc,
"Multicast Packets Received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_64",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.prc64,
"64 byte frames received ");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_65_127",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.prc127,
"65-127 byte frames received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_128_255",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.prc255,
"128-255 byte frames received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_256_511",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.prc511,
"256-511 byte frames received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_512_1023",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.prc1023,
"512-1023 byte frames received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_1024_1522",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.prc1522,
"1023-1522 byte frames received");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_octets_recvd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.gorc,
"Good Octets Received");
/* Packet Transmission Stats */
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_octets_txd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.gotc,
"Good Octets Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "total_pkts_txd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.tpt,
"Total Packets Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_pkts_txd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.gptc,
"Good Packets Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_txd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.bptc,
"Broadcast Packets Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_txd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.mptc,
"Multicast Packets Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_64",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.ptc64,
"64 byte frames transmitted ");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_65_127",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.ptc127,
"65-127 byte frames transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_128_255",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.ptc255,
"128-255 byte frames transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_256_511",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.ptc511,
"256-511 byte frames transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_512_1023",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.ptc1023,
"512-1023 byte frames transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_1024_1522",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.ptc1522,
"1024-1522 byte frames transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tso_txd",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.tsctc,
"TSO Contexts Transmitted");
sysctl(9) cleanup checkpoint: amd64 GENERIC builds cleanly. Commit the Intel drivers.
2011-01-12 19:53:23 +00:00
SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tso_ctx_fail",
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
CTLFLAG_RD, &adapter->stats.tsctfc,
"TSO Contexts Failed");
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
/**********************************************************************
*
* This routine provides a way to dump out the adapter eeprom,
* often a useful debug/service tool. This only dumps the first
* 32 words, stuff that matters is in that extent.
*
**********************************************************************/
static int
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
lem_sysctl_nvm_info(SYSCTL_HANDLER_ARGS)
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
{
struct adapter *adapter;
int error;
int result;
result = -1;
error = sysctl_handle_int(oidp, &result, 0, req);
if (error || !req->newptr)
return (error);
/*
* This value will cause a hex dump of the
* first 32 16-bit words of the EEPROM to
* the screen.
*/
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
if (result == 1) {
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
adapter = (struct adapter *)arg1;
lem_print_nvm_info(adapter);
}
return (error);
}
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
static void
lem_print_nvm_info(struct adapter *adapter)
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
{
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
u16 eeprom_data;
int i, j, row = 0;
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
/* Its a bit crude, but it gets the job done */
printf("\nInterface EEPROM Dump:\n");
printf("Offset\n0x0000 ");
for (i = 0, j = 0; i < 32; i++, j++) {
if (j == 8) { /* Make the offset block */
j = 0; ++row;
printf("\n0x00%x0 ",row);
}
e1000_read_nvm(&adapter->hw, i, 1, &eeprom_data);
printf("%04x ", eeprom_data);
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
Tweak the stats exported by the e1000 drivers: - Add a single sysctl procedure to all three drivers to read an arbitrary register (the register is passed as arg2). Use it to replace existing routines in igb(4) that used a separate routine for each register, and to add support for missing stats in em(4) and lem(4). - Move the 'rx_overruns' and 'watchdog_timeouts' stats out of the MAC stats section as they are driver stats, not MAC counters. - Simplify the code that creates per-queue stats in igb(4) to use a single loop and remove duplicated code. - Properly read all 64 bits of the 'good octets received/transmitted' in em(4) and lem(4). - Actually read the interrupt count registers in em(4), and drop the 'host to card' sysctl stats from em(4) as they are not implemented in any of the hardware this driver supports. - Restore several stats to em(4) that were lost in the earlier stats conversion including per-queue stats. - Export several MAC stats in em(4) that were exported in igb(4) but not in em(4). - Export stats in lem(4) using individual sysctls as in em(4) and igb(4). Reviewed by: jfv MFC after: 1 week
2010-09-20 16:04:44 +00:00
printf("\n");
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
}
static int
lem_sysctl_int_delay(SYSCTL_HANDLER_ARGS)
{
struct em_int_delay_info *info;
struct adapter *adapter;
u32 regval;
int error;
int usecs;
int ticks;
info = (struct em_int_delay_info *)arg1;
usecs = info->value;
error = sysctl_handle_int(oidp, &usecs, 0, req);
if (error != 0 || req->newptr == NULL)
return (error);
if (usecs < 0 || usecs > EM_TICKS_TO_USECS(65535))
return (EINVAL);
info->value = usecs;
ticks = EM_USECS_TO_TICKS(usecs);
adapter = info->adapter;
EM_CORE_LOCK(adapter);
regval = E1000_READ_OFFSET(&adapter->hw, info->offset);
regval = (regval & ~0xffff) | (ticks & 0xffff);
/* Handle a few special cases. */
switch (info->offset) {
case E1000_RDTR:
break;
case E1000_TIDV:
if (ticks == 0) {
adapter->txd_cmd &= ~E1000_TXD_CMD_IDE;
/* Don't write 0 into the TIDV register. */
regval++;
} else
adapter->txd_cmd |= E1000_TXD_CMD_IDE;
break;
}
E1000_WRITE_OFFSET(&adapter->hw, info->offset, regval);
EM_CORE_UNLOCK(adapter);
return (0);
}
static void
lem_add_int_delay_sysctl(struct adapter *adapter, const char *name,
const char *description, struct em_int_delay_info *info,
int offset, int value)
{
info->adapter = adapter;
info->offset = offset;
info->value = value;
SYSCTL_ADD_PROC(device_get_sysctl_ctx(adapter->dev),
SYSCTL_CHILDREN(device_get_sysctl_tree(adapter->dev)),
OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW,
info, 0, lem_sysctl_int_delay, "I", description);
}
Sync the lem code up with the vlan and other fixes in em. Delete a unneeded test from the beginning of em_xmit. CRITICAL: shared code fix for 82574, a mutex might not be released, this can cause hangs.
2010-11-01 20:19:25 +00:00
static void
lem_set_flow_cntrl(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);
}
Update to igb and em: em revision 7.0.0: - Using driver devclass, seperate legacy (pre-pcie) code into a seperate source file. This will at least help protect against regression issues. It compiles along with em, and is transparent to end use, devices in each appear to be 'emX'. When using em in a modular form this also allows the legacy stuff to be defined out. - Add tx and rx rings as in igb, in the 82574 this becomes actual multiqueue for the first time (2 queues) while in other PCIE adapters its just make code cleaner. - Add RX mbuf handling logic that matches igb, this will eliminate packet drops due to temporary mbuf shortage. igb revision 1.9.3: - Following the ixgbe code, use a new approach in what was called 'get_buf', the routine now has been made independent of rxeof, it now does the update to the engine TDT register, this design allows temporary mbuf resources to become non-critical, not requiring a packet to be discarded, instead it just returns and does not increment the tail pointer. - With the above change it was also unnecessary to keep 'spare' maps around, since we do not have the discard issue. - Performance tweaks and improvements to the code also. MFC in a week
2010-03-29 23:36:34 +00:00
static void
lem_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