d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
freebsd-skq/sys/dev/cxgbe/t4_main.c

8476 lines
212 KiB
C
Raw Normal View History

cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*-
* Copyright (c) 2011 Chelsio Communications, Inc.
* All rights reserved.
* Written by: Navdeep Parhar <np@FreeBSD.org>
*
* 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_inet.h"
cxgbe(4): support for IPv6 TSO and LRO. Submitted by: bz (this is a modified version of that patch)
2012-06-29 19:51:06 +00:00
#include "opt_inet6.h"
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
#include <sys/param.h>
#include <sys/conf.h>
#include <sys/priv.h>
#include <sys/kernel.h>
#include <sys/bus.h>
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
#include <sys/systm.h>
#include <sys/counter.h>
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
#include <sys/module.h>
Resume tx immediately in response to an SGE egress update from the hardware. MFC after: 1 week
2011-03-05 03:06:38 +00:00
#include <sys/malloc.h>
#include <sys/queue.h>
#include <sys/taskqueue.h>
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
#include <sys/pciio.h>
#include <dev/pci/pcireg.h>
#include <dev/pci/pcivar.h>
#include <dev/pci/pci_private.h>
#include <sys/firmware.h>
Display holdoff timers and packet counts as a list of numbers. MFC after: 1 week
2011-03-09 21:07:09 +00:00
#include <sys/sbuf.h>
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
#include <sys/smp.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <net/if_types.h>
#include <net/if_dl.h>
Add missing header. The test for VLAN_CAPABILITIES later in the file doesn't make sense without it. MFC after: 3 days
2011-05-17 00:40:11 +00:00
#include <net/if_vlan_var.h>
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
#if defined(__i386__) || defined(__amd64__)
#include <vm/vm.h>
#include <vm/pmap.h>
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
#include "common/common.h"
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
#include "common/t4_msg.h"
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
#include "common/t4_regs.h"
#include "common/t4_regs_values.h"
#include "t4_ioctl.h"
L2 table code. This is enough to get the T4's switch + L2 rewrite filters working. (All other filters - switch without L2 info rewrite, steer, and drop - were already fully-functional). Some contrived examples of "switch" filters with L2 rewriting: # cxgbetool t4nex0 iport 0 dport 80 action switch vlan +9 eport 3 Intercept all packets received on physical port 0 with TCP port 80 as destination, insert a vlan tag with VID 9, and send them out of port 3. # cxgbetool t4nex0 sip 192.168.1.1/32 ivlan 5 action switch \ vlan =9 smac aa:bb:cc:dd:ee:ff eport 0 Intercept all packets (received on any port) with source IP address 192.168.1.1 and VLAN id 5, rewrite the VLAN id to 9, rewrite source mac to aa:bb:cc:dd:ee:ff, and send it out of port 0. MFC after: 1 week
2011-05-30 21:07:26 +00:00
#include "t4_l2t.h"
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
#include "t4_mp_ring.h"
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/* T4 bus driver interface */
static int t4_probe(device_t);
static int t4_attach(device_t);
static int t4_detach(device_t);
static device_method_t t4_methods[] = {
DEVMETHOD(device_probe, t4_probe),
DEVMETHOD(device_attach, t4_attach),
DEVMETHOD(device_detach, t4_detach),
- There's no need to overwrite the default device method with the default one. Interestingly, these are actually the default for quite some time (bus_generic_driver_added(9) since r52045 and bus_generic_print_child(9) since r52045) but even recently added device drivers do this unnecessarily. Discussed with: jhb, marcel - While at it, use DEVMETHOD_END. Discussed with: jhb - Also while at it, use __FBSDID.
2011-11-22 21:28:20 +00:00
DEVMETHOD_END
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
};
static driver_t t4_driver = {
"t4nex",
t4_methods,
sizeof(struct adapter)
};
/* T4 port (cxgbe) interface */
static int cxgbe_probe(device_t);
static int cxgbe_attach(device_t);
static int cxgbe_detach(device_t);
static device_method_t cxgbe_methods[] = {
DEVMETHOD(device_probe, cxgbe_probe),
DEVMETHOD(device_attach, cxgbe_attach),
DEVMETHOD(device_detach, cxgbe_detach),
{ 0, 0 }
};
static driver_t cxgbe_driver = {
"cxgbe",
cxgbe_methods,
sizeof(struct port_info)
};
static d_ioctl_t t4_ioctl;
static d_open_t t4_open;
static d_close_t t4_close;
static struct cdevsw t4_cdevsw = {
.d_version = D_VERSION,
.d_flags = 0,
.d_open = t4_open,
.d_close = t4_close,
.d_ioctl = t4_ioctl,
.d_name = "t4nex",
};
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/* T5 bus driver interface */
static int t5_probe(device_t);
static device_method_t t5_methods[] = {
DEVMETHOD(device_probe, t5_probe),
DEVMETHOD(device_attach, t4_attach),
DEVMETHOD(device_detach, t4_detach),
DEVMETHOD_END
};
static driver_t t5_driver = {
"t5nex",
t5_methods,
sizeof(struct adapter)
};
/* T5 port (cxl) interface */
static driver_t cxl_driver = {
"cxl",
cxgbe_methods,
sizeof(struct port_info)
};
static struct cdevsw t5_cdevsw = {
.d_version = D_VERSION,
.d_flags = 0,
.d_open = t4_open,
.d_close = t4_close,
.d_ioctl = t4_ioctl,
.d_name = "t5nex",
};
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/* ifnet + media interface */
static void cxgbe_init(void *);
static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
static int cxgbe_transmit(struct ifnet *, struct mbuf *);
static void cxgbe_qflush(struct ifnet *);
static int cxgbe_media_change(struct ifnet *);
static void cxgbe_media_status(struct ifnet *, struct ifmediareq *);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
/*
* Correct lock order when you need to acquire multiple locks is t4_list_lock,
* then ADAPTER_LOCK, then t4_uld_list_lock.
*/
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
static struct sx t4_list_lock;
MFC r259145: Unstaticize t4_list and t4_uld_list. This works around a clang annoyance[1] and allows kgdb to find these symbols. [1] http://lists.freebsd.org/pipermail/freebsd-hackers/2012-November/041166.html
2013-12-12 00:27:27 +00:00
SLIST_HEAD(, adapter) t4_list;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
static struct sx t4_uld_list_lock;
MFC r259145: Unstaticize t4_list and t4_uld_list. This works around a clang annoyance[1] and allows kgdb to find these symbols. [1] http://lists.freebsd.org/pipermail/freebsd-hackers/2012-November/041166.html
2013-12-12 00:27:27 +00:00
SLIST_HEAD(, uld_info) t4_uld_list;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
* Tunables. See tweak_tunables() too.
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
*
* Each tunable is set to a default value here if it's known at compile-time.
* Otherwise it is set to -1 as an indication to tweak_tunables() that it should
* provide a reasonable default when the driver is loaded.
*
* Tunables applicable to both T4 and T5 are under hw.cxgbe. Those specific to
* T5 are under hw.cxl.
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
*/
/*
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
* Number of queues for tx and rx, 10G and 1G, NIC and offload.
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
*/
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#define NTXQ_10G 16
static int t4_ntxq10g = -1;
TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq10g);
#define NRXQ_10G 8
static int t4_nrxq10g = -1;
TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq10g);
#define NTXQ_1G 4
static int t4_ntxq1g = -1;
TUNABLE_INT("hw.cxgbe.ntxq1g", &t4_ntxq1g);
#define NRXQ_1G 2
static int t4_nrxq1g = -1;
TUNABLE_INT("hw.cxgbe.nrxq1g", &t4_nrxq1g);
MFC r261558 Add a new sysctl, dev.cxgbe.N.rsrv_noflow, and a companion tunable, hw.cxgbe.rsrv_noflow. When set, queue 0 of the port is reserved for TX packets without a flowid. The hash value of packets with a flowid is bumped up by 1. The intent is to provide a private queue for link-level packets like LACP that is unlikely to overflow or suffer deep queue latency.
2014-04-15 08:08:44 +00:00
static int t4_rsrv_noflowq = 0;
TUNABLE_INT("hw.cxgbe.rsrv_noflowq", &t4_rsrv_noflowq);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#define NOFLDTXQ_10G 8
static int t4_nofldtxq10g = -1;
TUNABLE_INT("hw.cxgbe.nofldtxq10g", &t4_nofldtxq10g);
#define NOFLDRXQ_10G 2
static int t4_nofldrxq10g = -1;
TUNABLE_INT("hw.cxgbe.nofldrxq10g", &t4_nofldrxq10g);
#define NOFLDTXQ_1G 2
static int t4_nofldtxq1g = -1;
TUNABLE_INT("hw.cxgbe.nofldtxq1g", &t4_nofldtxq1g);
#define NOFLDRXQ_1G 1
static int t4_nofldrxq1g = -1;
TUNABLE_INT("hw.cxgbe.nofldrxq1g", &t4_nofldrxq1g);
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
#define NNMTXQ_10G 2
static int t4_nnmtxq10g = -1;
TUNABLE_INT("hw.cxgbe.nnmtxq10g", &t4_nnmtxq10g);
#define NNMRXQ_10G 2
static int t4_nnmrxq10g = -1;
TUNABLE_INT("hw.cxgbe.nnmrxq10g", &t4_nnmrxq10g);
#define NNMTXQ_1G 1
static int t4_nnmtxq1g = -1;
TUNABLE_INT("hw.cxgbe.nnmtxq1g", &t4_nnmtxq1g);
#define NNMRXQ_1G 1
static int t4_nnmrxq1g = -1;
TUNABLE_INT("hw.cxgbe.nnmrxq1g", &t4_nnmrxq1g);
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
* Holdoff parameters for 10G and 1G ports.
*/
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#define TMR_IDX_10G 1
static int t4_tmr_idx_10g = TMR_IDX_10G;
TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx_10g);
Change the default to not use packet counters to generate rx interrupts. Rely solely on the timer based mechanism. Update man page to reflect this change. MFC after: 1 week
2012-04-30 09:46:05 +00:00
#define PKTC_IDX_10G (-1)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int t4_pktc_idx_10g = PKTC_IDX_10G;
TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx_10g);
#define TMR_IDX_1G 1
static int t4_tmr_idx_1g = TMR_IDX_1G;
TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_1G", &t4_tmr_idx_1g);
Change the default to not use packet counters to generate rx interrupts. Rely solely on the timer based mechanism. Update man page to reflect this change. MFC after: 1 week
2012-04-30 09:46:05 +00:00
#define PKTC_IDX_1G (-1)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int t4_pktc_idx_1g = PKTC_IDX_1G;
TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_1G", &t4_pktc_idx_1g);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
* Size (# of entries) of each tx and rx queue.
*/
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static unsigned int t4_qsize_txq = TX_EQ_QSIZE;
TUNABLE_INT("hw.cxgbe.qsize_txq", &t4_qsize_txq);
static unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
TUNABLE_INT("hw.cxgbe.qsize_rxq", &t4_qsize_rxq);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/*
* Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
*/
static int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
TUNABLE_INT("hw.cxgbe.interrupt_types", &t4_intr_types);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
* Configuration file.
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
*/
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
#define DEFAULT_CF "default"
#define FLASH_CF "flash"
#define UWIRE_CF "uwire"
- Explain clearly why a different firmware is being installed (if/when it is being installed). Improve other error messages while here. - Select special FPGA specific configuration profile when appropriate. MFC after: 3 days
2013-04-11 19:39:40 +00:00
#define FPGA_CF "fpga"
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static char t4_cfg_file[32] = DEFAULT_CF;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
TUNABLE_STR("hw.cxgbe.config_file", t4_cfg_file, sizeof(t4_cfg_file));
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r271450: cxgbe(4): knobs to enable/disable PAUSE frame based flow control. Approved by: re (glebius)
2014-09-22 13:07:03 +00:00
/*
* PAUSE settings (bit 0, 1 = rx_pause, tx_pause respectively).
* rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
* tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
* mark or when signalled to do so, 0 to never emit PAUSE.
*/
static int t4_pause_settings = PAUSE_TX | PAUSE_RX;
TUNABLE_INT("hw.cxgbe.pause_settings", &t4_pause_settings);
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
/*
* Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
* encouraged respectively).
*/
static unsigned int t4_fw_install = 1;
TUNABLE_INT("hw.cxgbe.fw_install", &t4_fw_install);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
* ASIC features that will be used. Disable the ones you don't want so that the
* chip resources aren't wasted on features that will not be used.
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
*/
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int t4_linkcaps_allowed = 0; /* No DCBX, PPP, etc. by default */
TUNABLE_INT("hw.cxgbe.linkcaps_allowed", &t4_linkcaps_allowed);
static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC;
TUNABLE_INT("hw.cxgbe.niccaps_allowed", &t4_niccaps_allowed);
Instruct the firmware not to provision resources for TCP offload if the kernel is being built without TCP_OFFLOAD. But never override toecaps_allowed if it has been set manually.
2012-07-02 20:42:43 +00:00
static int t4_toecaps_allowed = -1;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
TUNABLE_INT("hw.cxgbe.toecaps_allowed", &t4_toecaps_allowed);
static int t4_rdmacaps_allowed = 0;
TUNABLE_INT("hw.cxgbe.rdmacaps_allowed", &t4_rdmacaps_allowed);
static int t4_iscsicaps_allowed = 0;
TUNABLE_INT("hw.cxgbe.iscsicaps_allowed", &t4_iscsicaps_allowed);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int t4_fcoecaps_allowed = 0;
TUNABLE_INT("hw.cxgbe.fcoecaps_allowed", &t4_fcoecaps_allowed);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static int t5_write_combine = 0;
TUNABLE_INT("hw.cxl.write_combine", &t5_write_combine);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
struct intrs_and_queues {
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
uint16_t intr_type; /* INTx, MSI, or MSI-X */
uint16_t nirq; /* Total # of vectors */
uint16_t intr_flags_10g;/* Interrupt flags for each 10G port */
uint16_t intr_flags_1g; /* Interrupt flags for each 1G port */
uint16_t ntxq10g; /* # of NIC txq's for each 10G port */
uint16_t nrxq10g; /* # of NIC rxq's for each 10G port */
uint16_t ntxq1g; /* # of NIC txq's for each 1G port */
uint16_t nrxq1g; /* # of NIC rxq's for each 1G port */
uint16_t rsrv_noflowq; /* Flag whether to reserve queue 0 */
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
uint16_t nofldtxq10g; /* # of TOE txq's for each 10G port */
uint16_t nofldrxq10g; /* # of TOE rxq's for each 10G port */
uint16_t nofldtxq1g; /* # of TOE txq's for each 1G port */
uint16_t nofldrxq1g; /* # of TOE rxq's for each 1G port */
#endif
#ifdef DEV_NETMAP
uint16_t nnmtxq10g; /* # of netmap txq's for each 10G port */
uint16_t nnmrxq10g; /* # of netmap rxq's for each 10G port */
uint16_t nnmtxq1g; /* # of netmap txq's for each 1G port */
uint16_t nnmrxq1g; /* # of netmap rxq's for each 1G port */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
};
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
struct filter_entry {
uint32_t valid:1; /* filter allocated and valid */
uint32_t locked:1; /* filter is administratively locked */
uint32_t pending:1; /* filter action is pending firmware reply */
uint32_t smtidx:8; /* Source MAC Table index for smac */
L2 table code. This is enough to get the T4's switch + L2 rewrite filters working. (All other filters - switch without L2 info rewrite, steer, and drop - were already fully-functional). Some contrived examples of "switch" filters with L2 rewriting: # cxgbetool t4nex0 iport 0 dport 80 action switch vlan +9 eport 3 Intercept all packets received on physical port 0 with TCP port 80 as destination, insert a vlan tag with VID 9, and send them out of port 3. # cxgbetool t4nex0 sip 192.168.1.1/32 ivlan 5 action switch \ vlan =9 smac aa:bb:cc:dd:ee:ff eport 0 Intercept all packets (received on any port) with source IP address 192.168.1.1 and VLAN id 5, rewrite the VLAN id to 9, rewrite source mac to aa:bb:cc:dd:ee:ff, and send it out of port 0. MFC after: 1 week
2011-05-30 21:07:26 +00:00
struct l2t_entry *l2t; /* Layer Two Table entry for dmac */
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
struct t4_filter_specification fs;
};
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static int map_bars_0_and_4(struct adapter *);
static int map_bar_2(struct adapter *);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static void setup_memwin(struct adapter *);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static int validate_mem_range(struct adapter *, uint32_t, int);
MFC r256459. cxgbe(4): Update T4 and T5 firmwares to 1.9.12.0 Approved by: re (glebius)
2013-10-20 15:24:44 +00:00
static int fwmtype_to_hwmtype(int);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static int validate_mt_off_len(struct adapter *, int, uint32_t, int,
uint32_t *);
static void memwin_info(struct adapter *, int, uint32_t *, uint32_t *);
static uint32_t position_memwin(struct adapter *, int, uint32_t);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int cfg_itype_and_nqueues(struct adapter *, int, int,
struct intrs_and_queues *);
static int prep_firmware(struct adapter *);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static int partition_resources(struct adapter *, const struct firmware *,
const char *);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int get_params__pre_init(struct adapter *);
static int get_params__post_init(struct adapter *);
cxgbe(4): Ask the card's firmware to pad up tiny CPLs by encapsulating them in a firmware message if it is able to do so. This works out better for one of the FIFOs in the chip. MFC after: 5 days
2013-02-26 00:27:27 +00:00
static int set_params__post_init(struct adapter *);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static void t4_set_desc(struct adapter *);
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
static void build_medialist(struct port_info *, struct ifmedia *);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int cxgbe_init_synchronized(struct port_info *);
static int cxgbe_uninit_synchronized(struct port_info *);
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
static int setup_intr_handlers(struct adapter *);
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
static void quiesce_txq(struct adapter *, struct sge_txq *);
static void quiesce_wrq(struct adapter *, struct sge_wrq *);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static void quiesce_iq(struct adapter *, struct sge_iq *);
static void quiesce_fl(struct adapter *, struct sge_fl *);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
driver_intr_t *, void *, char *);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int t4_free_irq(struct adapter *, struct irq *);
static void reg_block_dump(struct adapter *, uint8_t *, unsigned int,
unsigned int);
static void t4_get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
Backport some parts of r272200. - a lock to protect indirect register access - put code that deals with stats in a separate cxgbe_refresh_stats. This is a direct commit to stable/10.
2015-05-05 15:13:59 +00:00
static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static void cxgbe_tick(void *);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
static void cxgbe_vlan_config(void *, struct ifnet *, uint16_t);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int cpl_not_handled(struct sge_iq *, const struct rss_header *,
struct mbuf *);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
static int an_not_handled(struct sge_iq *, const struct rsp_ctrl *);
Allow for a different handler for each type of firmware message. MFC after: 2 weeks
2012-08-16 18:31:50 +00:00
static int fw_msg_not_handled(struct adapter *, const __be64 *);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int t4_sysctls(struct adapter *);
static int cxgbe_sysctls(struct port_info *);
Display holdoff timers and packet counts as a list of numbers. MFC after: 1 week
2011-03-09 21:07:09 +00:00
static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int sysctl_bitfield(SYSCTL_HANDLER_ARGS);
- Show the reason why link is down if this information is available. - Display the temperature and PHY firmware version of the BT PHY. MFC after: 1 day
2013-07-05 01:53:51 +00:00
static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
MFC r261558 Add a new sysctl, dev.cxgbe.N.rsrv_noflow, and a companion tunable, hw.cxgbe.rsrv_noflow. When set, queue 0 of the port is reserved for TX packets without a flowid. The hash value of packets with a flowid is bumped up by 1. The intent is to provide a private queue for link-level packets like LACP that is unlikely to overflow or suffer deep queue latency.
2014-04-15 08:08:44 +00:00
static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
MFC r271450: cxgbe(4): knobs to enable/disable PAUSE frame based flow control. Approved by: re (glebius)
2014-09-22 13:07:03 +00:00
static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
Display temperature sensor data. Shows -1 if sensor not available on the card. # sysctl dev.t4nex.0.temperature # sysctl dev.t5nex.0.temperature
2013-08-02 18:05:42 +00:00
static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
cxgbe: reduce diffs with other branches. Will help future MFCs from HEAD. MFC after: 3 days
2012-02-07 06:21:59 +00:00
#ifdef SBUF_DRAIN
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
Display a string instead of a numeric code in the linkdnrc sysctl. Submitted by: gnn@
2013-07-27 07:43:43 +00:00
static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
static int sysctl_tids(SYSCTL_HANDLER_ARGS);
static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
Cosmetic change (s/wrwc/wcwr/;s/WRWC/WCWR/). MFC after: 3 days.
2013-04-11 22:49:29 +00:00
static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
cxgbe: reduce diffs with other branches. Will help future MFCs from HEAD. MFC after: 3 days
2012-02-07 06:21:59 +00:00
#endif
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
static uint32_t fconf_to_mode(uint32_t);
static uint32_t mode_to_fconf(uint32_t);
static uint32_t fspec_to_fconf(struct t4_filter_specification *);
static int get_filter_mode(struct adapter *, uint32_t *);
static int set_filter_mode(struct adapter *, uint32_t);
Provide hit-count with rest of the information about a filter. MFC after: 1 week
2011-06-01 01:32:58 +00:00
static inline uint64_t get_filter_hits(struct adapter *, uint32_t);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
static int get_filter(struct adapter *, struct t4_filter *);
static int set_filter(struct adapter *, struct t4_filter *);
static int del_filter(struct adapter *, struct t4_filter *);
L2 table code. This is enough to get the T4's switch + L2 rewrite filters working. (All other filters - switch without L2 info rewrite, steer, and drop - were already fully-functional). Some contrived examples of "switch" filters with L2 rewriting: # cxgbetool t4nex0 iport 0 dport 80 action switch vlan +9 eport 3 Intercept all packets received on physical port 0 with TCP port 80 as destination, insert a vlan tag with VID 9, and send them out of port 3. # cxgbetool t4nex0 sip 192.168.1.1/32 ivlan 5 action switch \ vlan =9 smac aa:bb:cc:dd:ee:ff eport 0 Intercept all packets (received on any port) with source IP address 192.168.1.1 and VLAN id 5, rewrite the VLAN id to 9, rewrite source mac to aa:bb:cc:dd:ee:ff, and send it out of port 0. MFC after: 1 week
2011-05-30 21:07:26 +00:00
static void clear_filter(struct filter_entry *);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
static int set_filter_wr(struct adapter *, int);
static int del_filter_wr(struct adapter *, int);
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
static int get_sge_context(struct adapter *, struct t4_sge_context *);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
static int load_fw(struct adapter *, struct t4_data *);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
Add a driver ioctl to read a byte from any device on a port's i2c bus. This lets userspace read arbitrary information from the SFP+ modules etc. on this bus. Reading multiple bytes in the same transaction isn't possible right now. I'll update the driver once the chip's firmware supports this. MFC after: 3 days
2012-10-10 17:13:46 +00:00
static int read_i2c(struct adapter *, struct t4_i2c_data *);
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
static int set_sched_class(struct adapter *, struct t4_sched_params *);
static int set_sched_queue(struct adapter *, struct t4_sched_queue *);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int toe_capability(struct port_info *, int);
#endif
cxgbe(4): Ensure that the MOD_LOAD handler runs before either t4nex or t5nex attach to their devices. MFC after: 3 days
2013-04-11 17:50:50 +00:00
static int mod_event(module_t, int, void *);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
struct {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
uint16_t device;
char *desc;
} t4_pciids[] = {
Allow cxgbe(4) running within a VM to attach to its devices that have been exported via PCI passthrough. - Do not check for a specific physical function (PF) before claiming a device. Different PFs have different device-ids so this check is redundant anyway. - Obtain the PF# from the WHOAMI register instead of pci_get_function(). - Setup the memory windows using the real BAR0 address, not what the VM says it is. Obtained from: Chelsio Communications
2012-06-26 00:34:34 +00:00
{0xa000, "Chelsio Terminator 4 FPGA"},
{0x4400, "Chelsio T440-dbg"},
{0x4401, "Chelsio T420-CR"},
{0x4402, "Chelsio T422-CR"},
{0x4403, "Chelsio T440-CR"},
{0x4404, "Chelsio T420-BCH"},
{0x4405, "Chelsio T440-BCH"},
{0x4406, "Chelsio T440-CH"},
{0x4407, "Chelsio T420-SO"},
{0x4408, "Chelsio T420-CX"},
{0x4409, "Chelsio T420-BT"},
{0x440a, "Chelsio T404-BT"},
cxgbe(4): Add support for the T440-LP-CR card. This is the 4x10G low profile card with a QSFP+ transceiver. MFC after: 3 days
2012-12-22 07:47:07 +00:00
{0x440e, "Chelsio T440-LP-CR"},
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
}, t5_pciids[] = {
{0xb000, "Chelsio Terminator 5 FPGA"},
{0x5400, "Chelsio T580-dbg"},
Expand the list of devices claimed by cxgbe(4).
2013-07-27 00:53:07 +00:00
{0x5401, "Chelsio T520-CR"}, /* 2 x 10G */
{0x5402, "Chelsio T522-CR"}, /* 2 x 10G, 2 X 1G */
Attach to the 4x10G T540-CR card.
2013-07-11 19:09:31 +00:00
{0x5403, "Chelsio T540-CR"}, /* 4 x 10G */
Expand the list of devices claimed by cxgbe(4).
2013-07-27 00:53:07 +00:00
{0x5407, "Chelsio T520-SO"}, /* 2 x 10G, nomem */
{0x5409, "Chelsio T520-BT"}, /* 2 x 10GBaseT */
{0x540a, "Chelsio T504-BT"}, /* 4 x 1G */
{0x540d, "Chelsio T580-CR"}, /* 2 x 40G */
{0x540e, "Chelsio T540-LP-CR"}, /* 4 x 10G */
Attach to the T580 (2 x 40G) card. MFC after: 1 week.
2013-04-30 06:30:21 +00:00
{0x5410, "Chelsio T580-LP-CR"}, /* 2 x 40G */
Expand the list of devices claimed by cxgbe(4).
2013-07-27 00:53:07 +00:00
{0x5411, "Chelsio T520-LL-CR"}, /* 2 x 10G */
{0x5412, "Chelsio T560-CR"}, /* 1 x 40G, 2 x 10G */
{0x5414, "Chelsio T580-LP-SO-CR"}, /* 2 x 40G, nomem */
MFC 274402: Add device ID for the T502-BT (dual-port 1G) adapter.
2015-02-05 20:59:03 +00:00
{0x5415, "Chelsio T502-BT"}, /* 2 x 1G */
Add pciids of the T5 based cards. The ones that I haven't tested with cxgbe(4) are disabled for now. This will change. MFC after: 2 weeks
2013-04-11 23:40:05 +00:00
#ifdef notyet
{0x5404, "Chelsio T520-BCH"},
{0x5405, "Chelsio T540-BCH"},
{0x5406, "Chelsio T540-CH"},
Expand the list of devices claimed by cxgbe(4).
2013-07-27 00:53:07 +00:00
{0x5408, "Chelsio T520-CX"},
Add pciids of the T5 based cards. The ones that I haven't tested with cxgbe(4) are disabled for now. This will change. MFC after: 2 weeks
2013-04-11 23:40:05 +00:00
{0x540b, "Chelsio B520-SR"},
{0x540c, "Chelsio B504-BT"},
{0x540f, "Chelsio Amsterdam"},
Expand the list of devices claimed by cxgbe(4).
2013-07-27 00:53:07 +00:00
{0x5413, "Chelsio T580-CHR"},
Add pciids of the T5 based cards. The ones that I haven't tested with cxgbe(4) are disabled for now. This will change. MFC after: 2 weeks
2013-04-11 23:40:05 +00:00
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
};
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
/*
* service_iq() has an iq and needs the fl. Offset of fl from the iq should be
* exactly the same for both rxq and ofld_rxq.
*/
CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
#endif
Allow for a different handler for each type of firmware message. MFC after: 2 weeks
2012-08-16 18:31:50 +00:00
/* No easy way to include t4_msg.h before adapter.h so we check this way */
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
CTASSERT(nitems(((struct adapter *)0)->cpl_handler) == NUM_CPL_CMDS);
CTASSERT(nitems(((struct adapter *)0)->fw_msg_handler) == NUM_FW6_TYPES);
Allow for a different handler for each type of firmware message. MFC after: 2 weeks
2012-08-16 18:31:50 +00:00
MFC r263317, r263412, and r263451. r263317: cxgbe(4): significant rx rework. - More flexible cluster size selection, including the ability to fall back to a safe cluster size (PAGE_SIZE from zone_jumbop by default) in case an allocation of a larger size fails. - A single get_fl_payload() function that assembles the payload into an mbuf chain for any kind of freelist. This replaces two variants: one for freelists with buffer packing enabled and another for those without. - Buffer packing with any sized cluster. It was limited to 4K clusters only before this change. - Enable buffer packing for TOE rx queues as well. - Statistics and tunables to go with all these changes. The driver's man page will be updated separately. r263412: cxgbe(4): if_iqdrops statistic should include tunnel congestion drops. r263451: cxgbe(4): man page updates.
2014-05-06 06:49:39 +00:00
CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int
t4_probe(device_t dev)
{
int i;
uint16_t v = pci_get_vendor(dev);
uint16_t d = pci_get_device(dev);
Allow cxgbe(4) running within a VM to attach to its devices that have been exported via PCI passthrough. - Do not check for a specific physical function (PF) before claiming a device. Different PFs have different device-ids so this check is redundant anyway. - Obtain the PF# from the WHOAMI register instead of pci_get_function(). - Setup the memory windows using the real BAR0 address, not what the VM says it is. Obtained from: Chelsio Communications
2012-06-26 00:34:34 +00:00
uint8_t f = pci_get_function(dev);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (v != PCI_VENDOR_ID_CHELSIO)
return (ENXIO);
Allow cxgbe(4) running within a VM to attach to its devices that have been exported via PCI passthrough. - Do not check for a specific physical function (PF) before claiming a device. Different PFs have different device-ids so this check is redundant anyway. - Obtain the PF# from the WHOAMI register instead of pci_get_function(). - Setup the memory windows using the real BAR0 address, not what the VM says it is. Obtained from: Chelsio Communications
2012-06-26 00:34:34 +00:00
/* Attach only to PF0 of the FPGA */
if (d == 0xa000 && f != 0)
return (ENXIO);
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
for (i = 0; i < nitems(t4_pciids); i++) {
Allow cxgbe(4) running within a VM to attach to its devices that have been exported via PCI passthrough. - Do not check for a specific physical function (PF) before claiming a device. Different PFs have different device-ids so this check is redundant anyway. - Obtain the PF# from the WHOAMI register instead of pci_get_function(). - Setup the memory windows using the real BAR0 address, not what the VM says it is. Obtained from: Chelsio Communications
2012-06-26 00:34:34 +00:00
if (d == t4_pciids[i].device) {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
device_set_desc(dev, t4_pciids[i].desc);
return (BUS_PROBE_DEFAULT);
}
}
return (ENXIO);
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static int
t5_probe(device_t dev)
{
int i;
uint16_t v = pci_get_vendor(dev);
uint16_t d = pci_get_device(dev);
uint8_t f = pci_get_function(dev);
if (v != PCI_VENDOR_ID_CHELSIO)
return (ENXIO);
/* Attach only to PF0 of the FPGA */
if (d == 0xb000 && f != 0)
return (ENXIO);
for (i = 0; i < nitems(t5_pciids); i++) {
if (d == t5_pciids[i].device) {
device_set_desc(dev, t5_pciids[i].desc);
return (BUS_PROBE_DEFAULT);
}
}
return (ENXIO);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int
t4_attach(device_t dev)
{
struct adapter *sc;
int rc = 0, i, n10g, n1g, rqidx, tqidx;
struct intrs_and_queues iaq;
struct sge *s;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
int ofld_rqidx, ofld_tqidx;
#endif
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
int nm_rqidx, nm_tqidx;
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
sc = device_get_softc(dev);
sc->dev = dev;
pci_enable_busmaster(dev);
- Enable per-channel congestion notification. - Enable PCIe relaxed ordering for all egress queues and rx data buffers. MFC after: 3 days
2011-05-18 22:09:04 +00:00
if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
uint32_t v;
- Enable per-channel congestion notification. - Enable PCIe relaxed ordering for all egress queues and rx data buffers. MFC after: 3 days
2011-05-18 22:09:04 +00:00
pci_set_max_read_req(dev, 4096);
Align the PCI Express #defines with the style used for the PCI-X #defines. This also has the advantage that it makes the names more compact, iand also allows us to correct the non-uniform naming of the PCIM_LINK_* defines, making them all consistent amongst themselves. This is a mostly mechanical rename: s/PCIR_EXPRESS_/PCIER_/g s/PCIM_EXP_/PCIEM_/g s/PCIM_LINK_/PCIEM_LINK_/g When this is MFC'd, #defines will be added for the old names to assist out-of-tree drivers. Discussed with: jhb MFC after: 1 week
2012-09-18 22:04:59 +00:00
v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
MFC r274724: cxgbe(4): figure out the max payload size and save it for later.
2014-11-26 00:44:48 +00:00
sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
- Enable per-channel congestion notification. - Enable PCIe relaxed ordering for all egress queues and rx data buffers. MFC after: 3 days
2011-05-18 22:09:04 +00:00
}
Add support for packet-sniffing tracers to cxgbe(4). This works with all T4 and T5 based cards and is useful for analyzing TSO, LRO, TOE, and for general purpose monitoring without tapping any cxgbe or cxl ifnet directly. Tracers on the T4/T5 chips provide access to Ethernet frames exactly as they were received from or transmitted on the wire. On transmit, a tracer will capture a frame after TSO segmentation, hw VLAN tag insertion, hw L3 & L4 checksum insertion, etc. It will also capture frames generated by the TCP offload engine (TOE traffic is normally invisible to the kernel). On receive, a tracer will capture a frame before hw VLAN extraction, runt filtering, other badness filtering, before the steering/drop/L2-rewrite filters or the TOE have had a go at it, and of course before sw LRO in the driver. There are 4 tracers on a chip. A tracer can trace only in one direction (tx or rx). For now cxgbetool will set up tracers to capture the first 128B of every transmitted or received frame on a given port. This is a small subset of what the hardware can do. A pseudo ifnet with the same name as the nexus driver (t4nex0 or t5nex0) will be created for tracing. The data delivered to this ifnet is an additional copy made inside the chip. Normal delivery to cxgbe<n> or cxl<n> will be made as usual. /* watch cxl0, which is the first port hanging off t5nex0. */ # cxgbetool t5nex0 tracer 0 tx0 (watch what cxl0 is transmitting) # cxgbetool t5nex0 tracer 1 rx0 (watch what cxl0 is receiving) # cxgbetool t5nex0 tracer list # tcpdump -i t5nex0 <== all that cxl0 sees and puts on the wire If you were doing TSO, a tcpdump on cxl0 may have shown you ~64K "frames" with no L3/L4 checksum but this will show you the frames that were actually transmitted. /* all done */ # cxgbetool t5nex0 tracer 0 disable # cxgbetool t5nex0 tracer 1 disable # cxgbetool t5nex0 tracer list # ifconfig t5nex0 destroy
2013-07-26 22:04:11 +00:00
sc->traceq = -1;
mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
device_get_nameunit(dev));
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
snprintf(sc->lockname, sizeof(sc->lockname), "%s",
device_get_nameunit(dev));
mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_xlock(&t4_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SLIST_INSERT_HEAD(&t4_list, sc, link);
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_xunlock(&t4_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
TAILQ_INIT(&sc->sfl);
callout_init(&sc->sfl_callout, CALLOUT_MPSAFE);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Backport some parts of r272200. - a lock to protect indirect register access - put code that deals with stats in a separate cxgbe_refresh_stats. This is a direct commit to stable/10.
2015-05-05 15:13:59 +00:00
mtx_init(&sc->regwin_lock, "register and memory window", 0, MTX_DEF);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
rc = map_bars_0_and_4(sc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc != 0)
goto done; /* error message displayed already */
Allow cxgbe(4) running within a VM to attach to its devices that have been exported via PCI passthrough. - Do not check for a specific physical function (PF) before claiming a device. Different PFs have different device-ids so this check is redundant anyway. - Obtain the PF# from the WHOAMI register instead of pci_get_function(). - Setup the memory windows using the real BAR0 address, not what the VM says it is. Obtained from: Chelsio Communications
2012-06-26 00:34:34 +00:00
/*
* This is the real PF# to which we're attaching. Works from within PCI
* passthrough environments too, where pci_get_function() could return a
* different PF# depending on the passthrough configuration. We need to
* use the real PF# in all our communication with the firmware.
*/
sc->pf = G_SOURCEPF(t4_read_reg(sc, A_PL_WHOAMI));
sc->mbox = sc->pf;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
sc->an_handler = an_not_handled;
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
for (i = 0; i < nitems(sc->cpl_handler); i++)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
sc->cpl_handler[i] = cpl_not_handled;
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
for (i = 0; i < nitems(sc->fw_msg_handler); i++)
Allow for a different handler for each type of firmware message. MFC after: 2 weeks
2012-08-16 18:31:50 +00:00
sc->fw_msg_handler[i] = fw_msg_not_handled;
Add a routine (t4_set_tcb_field) to update arbitrary parts of a hardware TCB. Filters are programmed by modifying the TCB too (via a different routine) and the reply to any TCB update is delivered via a CPL_SET_TCB_RPL. Figure out whether the reply is for a filter-write or something else and route it appropriately. MFC after: 2 weeks
2012-08-16 20:15:29 +00:00
t4_register_cpl_handler(sc, CPL_SET_TCB_RPL, t4_filter_rpl);
Add support for packet-sniffing tracers to cxgbe(4). This works with all T4 and T5 based cards and is useful for analyzing TSO, LRO, TOE, and for general purpose monitoring without tapping any cxgbe or cxl ifnet directly. Tracers on the T4/T5 chips provide access to Ethernet frames exactly as they were received from or transmitted on the wire. On transmit, a tracer will capture a frame after TSO segmentation, hw VLAN tag insertion, hw L3 & L4 checksum insertion, etc. It will also capture frames generated by the TCP offload engine (TOE traffic is normally invisible to the kernel). On receive, a tracer will capture a frame before hw VLAN extraction, runt filtering, other badness filtering, before the steering/drop/L2-rewrite filters or the TOE have had a go at it, and of course before sw LRO in the driver. There are 4 tracers on a chip. A tracer can trace only in one direction (tx or rx). For now cxgbetool will set up tracers to capture the first 128B of every transmitted or received frame on a given port. This is a small subset of what the hardware can do. A pseudo ifnet with the same name as the nexus driver (t4nex0 or t5nex0) will be created for tracing. The data delivered to this ifnet is an additional copy made inside the chip. Normal delivery to cxgbe<n> or cxl<n> will be made as usual. /* watch cxl0, which is the first port hanging off t5nex0. */ # cxgbetool t5nex0 tracer 0 tx0 (watch what cxl0 is transmitting) # cxgbetool t5nex0 tracer 1 rx0 (watch what cxl0 is receiving) # cxgbetool t5nex0 tracer list # tcpdump -i t5nex0 <== all that cxl0 sees and puts on the wire If you were doing TSO, a tcpdump on cxl0 may have shown you ~64K "frames" with no L3/L4 checksum but this will show you the frames that were actually transmitted. /* all done */ # cxgbetool t5nex0 tracer 0 disable # cxgbetool t5nex0 tracer 1 disable # cxgbetool t5nex0 tracer list # ifconfig t5nex0 destroy
2013-07-26 22:04:11 +00:00
t4_register_cpl_handler(sc, CPL_TRACE_PKT, t4_trace_pkt);
t4_register_cpl_handler(sc, CPL_TRACE_PKT_T5, t5_trace_pkt);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
t4_init_sge_cpl_handlers(sc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/* Prepare the adapter for operation */
rc = -t4_prep_adapter(sc);
if (rc != 0) {
device_printf(dev, "failed to prepare adapter: %d.\n", rc);
goto done;
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/*
* Do this really early, with the memory windows set up even before the
* character device. The userland tool's register i/o and mem read
* will work even in "recovery mode".
*/
setup_memwin(sc);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
sc->cdev = make_dev(is_t4(sc) ? &t4_cdevsw : &t5_cdevsw,
device_get_unit(dev), UID_ROOT, GID_WHEEL, 0600, "%s",
device_get_nameunit(dev));
if (sc->cdev == NULL)
device_printf(dev, "failed to create nexus char device.\n");
else
sc->cdev->si_drv1 = sc;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/* Go no further if recovery mode has been requested. */
if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
device_printf(dev, "recovery mode.\n");
goto done;
}
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
#if defined(__i386__)
if ((cpu_feature & CPUID_CX8) == 0) {
device_printf(dev, "64 bit atomics not available.\n");
rc = ENOTSUP;
goto done;
}
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/* Prepare the firmware for operation */
rc = prep_firmware(sc);
if (rc != 0)
goto done; /* error message displayed already */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
rc = get_params__post_init(sc);
if (rc != 0)
goto done; /* error message displayed already */
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Ask the card's firmware to pad up tiny CPLs by encapsulating them in a firmware message if it is able to do so. This works out better for one of the FIFOs in the chip. MFC after: 5 days
2013-02-26 00:27:27 +00:00
rc = set_params__post_init(sc);
if (rc != 0)
goto done; /* error message displayed already */
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
rc = map_bar_2(sc);
if (rc != 0)
goto done; /* error message displayed already */
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
rc = t4_create_dma_tag(sc);
if (rc != 0)
goto done; /* error message displayed already */
/*
* First pass over all the ports - allocate VIs and initialize some
* basic parameters like mac address, port type, etc. We also figure
* out whether a port is 10G or 1G and use that information when
* calculating how many interrupts to attempt to allocate.
*/
n10g = n1g = 0;
for_each_port(sc, i) {
struct port_info *pi;
pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
sc->port[i] = pi;
/* These must be set before t4_port_init */
pi->adapter = sc;
pi->port_id = i;
/* Allocate the vi and initialize parameters like mac addr */
rc = -t4_port_init(pi, sc->mbox, sc->pf, 0);
if (rc != 0) {
device_printf(dev, "unable to initialize port %d: %d\n",
i, rc);
free(pi, M_CXGBE);
- Specialized ingress queues that take interrupts for other ingress queues. Try to have a set of these per port when possible, fall back to sharing a common pool between all ports otherwise. - One control queue per port (used to be one per hardware channel). - t4_eth_rx now handles Ethernet rx only. - sysctls to display pidx/cidx for some queues. MFC after: 1 week
2011-05-30 21:34:44 +00:00
sc->port[i] = NULL;
goto done;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
MFC r271450: cxgbe(4): knobs to enable/disable PAUSE frame based flow control. Approved by: re (glebius)
2014-09-22 13:07:03 +00:00
pi->link_cfg.requested_fc &= ~(PAUSE_TX | PAUSE_RX);
pi->link_cfg.requested_fc |= t4_pause_settings;
pi->link_cfg.fc &= ~(PAUSE_TX | PAUSE_RX);
pi->link_cfg.fc |= t4_pause_settings;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
rc = -t4_link_start(sc, sc->mbox, pi->tx_chan, &pi->link_cfg);
if (rc != 0) {
device_printf(dev, "port %d l1cfg failed: %d\n", i, rc);
free(pi, M_CXGBE);
sc->port[i] = NULL;
goto done;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
device_get_nameunit(dev), i);
mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
Add support for packet-sniffing tracers to cxgbe(4). This works with all T4 and T5 based cards and is useful for analyzing TSO, LRO, TOE, and for general purpose monitoring without tapping any cxgbe or cxl ifnet directly. Tracers on the T4/T5 chips provide access to Ethernet frames exactly as they were received from or transmitted on the wire. On transmit, a tracer will capture a frame after TSO segmentation, hw VLAN tag insertion, hw L3 & L4 checksum insertion, etc. It will also capture frames generated by the TCP offload engine (TOE traffic is normally invisible to the kernel). On receive, a tracer will capture a frame before hw VLAN extraction, runt filtering, other badness filtering, before the steering/drop/L2-rewrite filters or the TOE have had a go at it, and of course before sw LRO in the driver. There are 4 tracers on a chip. A tracer can trace only in one direction (tx or rx). For now cxgbetool will set up tracers to capture the first 128B of every transmitted or received frame on a given port. This is a small subset of what the hardware can do. A pseudo ifnet with the same name as the nexus driver (t4nex0 or t5nex0) will be created for tracing. The data delivered to this ifnet is an additional copy made inside the chip. Normal delivery to cxgbe<n> or cxl<n> will be made as usual. /* watch cxl0, which is the first port hanging off t5nex0. */ # cxgbetool t5nex0 tracer 0 tx0 (watch what cxl0 is transmitting) # cxgbetool t5nex0 tracer 1 rx0 (watch what cxl0 is receiving) # cxgbetool t5nex0 tracer list # tcpdump -i t5nex0 <== all that cxl0 sees and puts on the wire If you were doing TSO, a tcpdump on cxl0 may have shown you ~64K "frames" with no L3/L4 checksum but this will show you the frames that were actually transmitted. /* all done */ # cxgbetool t5nex0 tracer 0 disable # cxgbetool t5nex0 tracer 1 disable # cxgbetool t5nex0 tracer list # ifconfig t5nex0 destroy
2013-07-26 22:04:11 +00:00
sc->chan_map[pi->tx_chan] = i;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
- Provide accurate ifmedia information so that 40G ports/transceivers are displayed properly in ifconfig, etc. - Use the same number of tx and rx queues for a 40G port as for a 10G port. MFC after: 1 week
2013-04-30 05:51:52 +00:00
if (is_10G_port(pi) || is_40G_port(pi)) {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
n10g++;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
pi->tmr_idx = t4_tmr_idx_10g;
pi->pktc_idx = t4_pktc_idx_10g;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
} else {
n1g++;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
pi->tmr_idx = t4_tmr_idx_1g;
pi->pktc_idx = t4_pktc_idx_1g;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
pi->xact_addr_filt = -1;
- Show the reason why link is down if this information is available. - Display the temperature and PHY firmware version of the BT PHY. MFC after: 1 day
2013-07-05 01:53:51 +00:00
pi->linkdnrc = -1;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
pi->qsize_rxq = t4_qsize_rxq;
pi->qsize_txq = t4_qsize_txq;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
pi->dev = device_add_child(dev, is_t4(sc) ? "cxgbe" : "cxl", -1);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (pi->dev == NULL) {
device_printf(dev,
"failed to add device for port %d.\n", i);
rc = ENXIO;
goto done;
}
device_set_softc(pi->dev, pi);
}
/*
* Interrupt type, # of interrupts, # of rx/tx queues, etc.
*/
rc = cfg_itype_and_nqueues(sc, n10g, n1g, &iaq);
if (rc != 0)
goto done; /* error message displayed already */
sc->intr_type = iaq.intr_type;
sc->intr_count = iaq.nirq;
s = &sc->sge;
s->nrxq = n10g * iaq.nrxq10g + n1g * iaq.nrxq1g;
s->ntxq = n10g * iaq.ntxq10g + n1g * iaq.ntxq1g;
- Move all Ethernet specific items from sge_eq to sge_txq. sge_eq is now a suitable base for all kinds of egress queues. - Add control queues (sge_ctrlq) and allocate one of these per hardware channel. They can be used to program filters and steer traffic (and more). MFC after: 1 week
2011-04-19 22:08:28 +00:00
s->neq = s->ntxq + s->nrxq; /* the free list in an rxq is an eq */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
s->neq += sc->params.nports + 1;/* ctrl queues: 1 per port + 1 mgmt */
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
s->niq = s->nrxq + 1; /* 1 extra for firmware event queue */
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (is_offload(sc)) {
s->nofldrxq = n10g * iaq.nofldrxq10g + n1g * iaq.nofldrxq1g;
s->nofldtxq = n10g * iaq.nofldtxq10g + n1g * iaq.nofldtxq1g;
s->neq += s->nofldtxq + s->nofldrxq;
s->niq += s->nofldrxq;
s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
M_CXGBE, M_ZERO | M_WAITOK);
s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
M_CXGBE, M_ZERO | M_WAITOK);
}
#endif
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
s->nnmrxq = n10g * iaq.nnmrxq10g + n1g * iaq.nnmrxq1g;
s->nnmtxq = n10g * iaq.nnmtxq10g + n1g * iaq.nnmtxq1g;
s->neq += s->nnmtxq + s->nnmrxq;
s->niq += s->nnmrxq;
s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
M_CXGBE, M_ZERO | M_WAITOK);
s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
M_CXGBE, M_ZERO | M_WAITOK);
#endif
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
s->ctrlq = malloc(sc->params.nports * sizeof(struct sge_wrq), M_CXGBE,
- Move all Ethernet specific items from sge_eq to sge_txq. sge_eq is now a suitable base for all kinds of egress queues. - Add control queues (sge_ctrlq) and allocate one of these per hardware channel. They can be used to program filters and steer traffic (and more). MFC after: 1 week
2011-04-19 22:08:28 +00:00
M_ZERO | M_WAITOK);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
M_ZERO | M_WAITOK);
s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
M_ZERO | M_WAITOK);
s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
M_ZERO | M_WAITOK);
s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
M_ZERO | M_WAITOK);
sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
M_ZERO | M_WAITOK);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
t4_init_l2t(sc, M_WAITOK);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
* Second pass over the ports. This time we know the number of rx and
* tx queues that each port should get.
*/
rqidx = tqidx = 0;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
ofld_rqidx = ofld_tqidx = 0;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#endif
#ifdef DEV_NETMAP
nm_rqidx = nm_tqidx = 0;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
for_each_port(sc, i) {
struct port_info *pi = sc->port[i];
if (pi == NULL)
continue;
pi->first_rxq = rqidx;
pi->first_txq = tqidx;
- Provide accurate ifmedia information so that 40G ports/transceivers are displayed properly in ifconfig, etc. - Use the same number of tx and rx queues for a 40G port as for a 10G port. MFC after: 1 week
2013-04-30 05:51:52 +00:00
if (is_10G_port(pi) || is_40G_port(pi)) {
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
pi->flags |= iaq.intr_flags_10g;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
pi->nrxq = iaq.nrxq10g;
pi->ntxq = iaq.ntxq10g;
} else {
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
pi->flags |= iaq.intr_flags_1g;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
pi->nrxq = iaq.nrxq1g;
pi->ntxq = iaq.ntxq1g;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r261558 Add a new sysctl, dev.cxgbe.N.rsrv_noflow, and a companion tunable, hw.cxgbe.rsrv_noflow. When set, queue 0 of the port is reserved for TX packets without a flowid. The hash value of packets with a flowid is bumped up by 1. The intent is to provide a private queue for link-level packets like LACP that is unlikely to overflow or suffer deep queue latency.
2014-04-15 08:08:44 +00:00
if (pi->ntxq > 1)
pi->rsrv_noflowq = iaq.rsrv_noflowq ? 1 : 0;
else
pi->rsrv_noflowq = 0;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
rqidx += pi->nrxq;
tqidx += pi->ntxq;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (is_offload(sc)) {
pi->first_ofld_rxq = ofld_rqidx;
pi->first_ofld_txq = ofld_tqidx;
- Provide accurate ifmedia information so that 40G ports/transceivers are displayed properly in ifconfig, etc. - Use the same number of tx and rx queues for a 40G port as for a 10G port. MFC after: 1 week
2013-04-30 05:51:52 +00:00
if (is_10G_port(pi) || is_40G_port(pi)) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
pi->nofldrxq = iaq.nofldrxq10g;
pi->nofldtxq = iaq.nofldtxq10g;
} else {
pi->nofldrxq = iaq.nofldrxq1g;
pi->nofldtxq = iaq.nofldtxq1g;
}
ofld_rqidx += pi->nofldrxq;
ofld_tqidx += pi->nofldtxq;
}
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#endif
#ifdef DEV_NETMAP
pi->first_nm_rxq = nm_rqidx;
pi->first_nm_txq = nm_tqidx;
if (is_10G_port(pi) || is_40G_port(pi)) {
pi->nnmrxq = iaq.nnmrxq10g;
pi->nnmtxq = iaq.nnmtxq10g;
} else {
pi->nnmrxq = iaq.nnmrxq1g;
pi->nnmtxq = iaq.nnmtxq1g;
}
nm_rqidx += pi->nnmrxq;
nm_tqidx += pi->nnmtxq;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
rc = setup_intr_handlers(sc);
if (rc != 0) {
device_printf(dev,
"failed to setup interrupt handlers: %d\n", rc);
goto done;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
rc = bus_generic_attach(dev);
if (rc != 0) {
device_printf(dev,
"failed to attach all child ports: %d\n", rc);
goto done;
}
device_printf(dev,
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
"PCIe x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
sc->params.pci.width, sc->params.nports, sc->intr_count,
sc->intr_type == INTR_MSIX ? "MSI-X" :
(sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
t4_set_desc(sc);
done:
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (rc != 0 && sc->cdev) {
/* cdev was created and so cxgbetool works; recover that way. */
device_printf(dev,
"error during attach, adapter is now in recovery mode.\n");
rc = 0;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc != 0)
t4_detach(dev);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
else
t4_sysctls(sc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (rc);
}
/*
* Idempotent
*/
static int
t4_detach(device_t dev)
{
struct adapter *sc;
struct port_info *pi;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
int i, rc;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
sc = device_get_softc(dev);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (sc->flags & FULL_INIT_DONE)
t4_intr_disable(sc);
if (sc->cdev) {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
destroy_dev(sc->cdev);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
sc->cdev = NULL;
}
rc = bus_generic_detach(dev);
if (rc) {
device_printf(dev,
"failed to detach child devices: %d\n", rc);
return (rc);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
for (i = 0; i < sc->intr_count; i++)
t4_free_irq(sc, &sc->irq[i]);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
for (i = 0; i < MAX_NPORTS; i++) {
pi = sc->port[i];
if (pi) {
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->viid);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (pi->dev)
device_delete_child(dev, pi->dev);
mtx_destroy(&pi->pi_lock);
free(pi, M_CXGBE);
}
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (sc->flags & FULL_INIT_DONE)
adapter_full_uninit(sc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (sc->flags & FW_OK)
t4_fw_bye(sc, sc->mbox);
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
pci_release_msi(dev);
if (sc->regs_res)
bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
sc->regs_res);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (sc->udbs_res)
bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
sc->udbs_res);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (sc->msix_res)
bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
sc->msix_res);
L2 table code. This is enough to get the T4's switch + L2 rewrite filters working. (All other filters - switch without L2 info rewrite, steer, and drop - were already fully-functional). Some contrived examples of "switch" filters with L2 rewriting: # cxgbetool t4nex0 iport 0 dport 80 action switch vlan +9 eport 3 Intercept all packets received on physical port 0 with TCP port 80 as destination, insert a vlan tag with VID 9, and send them out of port 3. # cxgbetool t4nex0 sip 192.168.1.1/32 ivlan 5 action switch \ vlan =9 smac aa:bb:cc:dd:ee:ff eport 0 Intercept all packets (received on any port) with source IP address 192.168.1.1 and VLAN id 5, rewrite the VLAN id to 9, rewrite source mac to aa:bb:cc:dd:ee:ff, and send it out of port 0. MFC after: 1 week
2011-05-30 21:07:26 +00:00
if (sc->l2t)
t4_free_l2t(sc->l2t);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
free(sc->sge.ofld_rxq, M_CXGBE);
free(sc->sge.ofld_txq, M_CXGBE);
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#endif
#ifdef DEV_NETMAP
free(sc->sge.nm_rxq, M_CXGBE);
free(sc->sge.nm_txq, M_CXGBE);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
free(sc->irq, M_CXGBE);
free(sc->sge.rxq, M_CXGBE);
free(sc->sge.txq, M_CXGBE);
- Move all Ethernet specific items from sge_eq to sge_txq. sge_eq is now a suitable base for all kinds of egress queues. - Add control queues (sge_ctrlq) and allocate one of these per hardware channel. They can be used to program filters and steer traffic (and more). MFC after: 1 week
2011-04-19 22:08:28 +00:00
free(sc->sge.ctrlq, M_CXGBE);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
free(sc->sge.iqmap, M_CXGBE);
free(sc->sge.eqmap, M_CXGBE);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
free(sc->tids.ftid_tab, M_CXGBE);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
t4_destroy_dma_tag(sc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (mtx_initialized(&sc->sc_lock)) {
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_xlock(&t4_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SLIST_REMOVE(&t4_list, sc, adapter, link);
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_xunlock(&t4_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
mtx_destroy(&sc->sc_lock);
}
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (mtx_initialized(&sc->tids.ftid_lock))
mtx_destroy(&sc->tids.ftid_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (mtx_initialized(&sc->sfl_lock))
mtx_destroy(&sc->sfl_lock);
Add support for packet-sniffing tracers to cxgbe(4). This works with all T4 and T5 based cards and is useful for analyzing TSO, LRO, TOE, and for general purpose monitoring without tapping any cxgbe or cxl ifnet directly. Tracers on the T4/T5 chips provide access to Ethernet frames exactly as they were received from or transmitted on the wire. On transmit, a tracer will capture a frame after TSO segmentation, hw VLAN tag insertion, hw L3 & L4 checksum insertion, etc. It will also capture frames generated by the TCP offload engine (TOE traffic is normally invisible to the kernel). On receive, a tracer will capture a frame before hw VLAN extraction, runt filtering, other badness filtering, before the steering/drop/L2-rewrite filters or the TOE have had a go at it, and of course before sw LRO in the driver. There are 4 tracers on a chip. A tracer can trace only in one direction (tx or rx). For now cxgbetool will set up tracers to capture the first 128B of every transmitted or received frame on a given port. This is a small subset of what the hardware can do. A pseudo ifnet with the same name as the nexus driver (t4nex0 or t5nex0) will be created for tracing. The data delivered to this ifnet is an additional copy made inside the chip. Normal delivery to cxgbe<n> or cxl<n> will be made as usual. /* watch cxl0, which is the first port hanging off t5nex0. */ # cxgbetool t5nex0 tracer 0 tx0 (watch what cxl0 is transmitting) # cxgbetool t5nex0 tracer 1 rx0 (watch what cxl0 is receiving) # cxgbetool t5nex0 tracer list # tcpdump -i t5nex0 <== all that cxl0 sees and puts on the wire If you were doing TSO, a tcpdump on cxl0 may have shown you ~64K "frames" with no L3/L4 checksum but this will show you the frames that were actually transmitted. /* all done */ # cxgbetool t5nex0 tracer 0 disable # cxgbetool t5nex0 tracer 1 disable # cxgbetool t5nex0 tracer list # ifconfig t5nex0 destroy
2013-07-26 22:04:11 +00:00
if (mtx_initialized(&sc->ifp_lock))
mtx_destroy(&sc->ifp_lock);
Backport some parts of r272200. - a lock to protect indirect register access - put code that deals with stats in a separate cxgbe_refresh_stats. This is a direct commit to stable/10.
2015-05-05 15:13:59 +00:00
if (mtx_initialized(&sc->regwin_lock))
mtx_destroy(&sc->regwin_lock);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
bzero(sc, sizeof(*sc));
return (0);
}
static int
cxgbe_probe(device_t dev)
{
char buf[128];
struct port_info *pi = device_get_softc(dev);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
snprintf(buf, sizeof(buf), "port %d", pi->port_id);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
device_set_desc_copy(dev, buf);
return (BUS_PROBE_DEFAULT);
}
#define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
There are some high performance NICs that count statistics in hardware, and there are ifnets, that do that via counter(9). Provide a flag that would skip cache line trashing '+=' operation in ether_input(). Sponsored by: Netflix Sponsored by: Nginx, Inc. Reviewed by: melifaro, adrian Approved by: re (marius)
2013-10-09 19:04:40 +00:00
IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS)
cxgbe(4): support for IPv6 TSO and LRO. Submitted by: bz (this is a modified version of that patch)
2012-06-29 19:51:06 +00:00
#define T4_CAP_ENABLE (T4_CAP)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int
cxgbe_attach(device_t dev)
{
struct port_info *pi = device_get_softc(dev);
struct ifnet *ifp;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
char *s;
int n, o;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/* Allocate an ifnet and set it up */
ifp = if_alloc(IFT_ETHER);
if (ifp == NULL) {
device_printf(dev, "Cannot allocate ifnet\n");
return (ENOMEM);
}
pi->ifp = ifp;
ifp->if_softc = pi;
callout_init(&pi->tick, CALLOUT_MPSAFE);
if_initname(ifp, device_get_name(dev), device_get_unit(dev));
ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
ifp->if_init = cxgbe_init;
ifp->if_ioctl = cxgbe_ioctl;
ifp->if_transmit = cxgbe_transmit;
ifp->if_qflush = cxgbe_qflush;
ifp->if_capabilities = T4_CAP;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (is_offload(pi->adapter))
cxgbe/tom: List IFCAP_TOE6 as supported now that all the required pieces are in place. You still have to enable it explicitly, after loading the t4_tom KLD.
2013-01-26 01:06:27 +00:00
ifp->if_capabilities |= IFCAP_TOE;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ifp->if_capenable = T4_CAP_ENABLE;
cxgbe(4): support for IPv6 hardware checksumming (rx and tx).
2012-06-29 16:50:52 +00:00
ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r272190: cxgbe(4): explicitly set various if_hw_tso* values.
2014-11-12 19:53:35 +00:00
ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS;
ifp->if_hw_tsomaxsegsize = 65536;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/* Initialize ifmedia for this port */
ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
cxgbe_media_status);
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
build_medialist(pi, &pi->media);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
pi->vlan_c = EVENTHANDLER_REGISTER(vlan_config, cxgbe_vlan_config, ifp,
EVENTHANDLER_PRI_ANY);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ether_ifattach(ifp, pi->hw_addr);
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
n = 128;
s = malloc(n, M_CXGBE, M_WAITOK);
o = snprintf(s, n, "%d txq, %d rxq (NIC)", pi->ntxq, pi->nrxq);
MPASS(n > o);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (is_offload(pi->adapter)) {
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
o += snprintf(s + o, n - o, "; %d txq, %d rxq (TOE)",
pi->nofldtxq, pi->nofldrxq);
MPASS(n > o);
}
#endif
#ifdef DEV_NETMAP
o += snprintf(s + o, n - o, "; %d txq, %d rxq (netmap)", pi->nnmtxq,
pi->nnmrxq);
MPASS(n > o);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
#endif
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
device_printf(dev, "%s\n", s);
free(s, M_CXGBE);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
/* nm_media handled here to keep implementation private to this file */
ifmedia_init(&pi->nm_media, IFM_IMASK, cxgbe_media_change,
cxgbe_media_status);
build_medialist(pi, &pi->nm_media);
create_netmap_ifnet(pi); /* logs errors it something fails */
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe_sysctls(pi);
return (0);
}
static int
cxgbe_detach(device_t dev)
{
struct port_info *pi = device_get_softc(dev);
struct adapter *sc = pi->adapter;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
struct ifnet *ifp = pi->ifp;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/* Tell if_ioctl and if_init that the port is going away */
ADAPTER_LOCK(sc);
SET_DOOMED(pi);
wakeup(&sc->flags);
while (IS_BUSY(sc))
mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
SET_BUSY(sc);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
#ifdef INVARIANTS
sc->last_op = "t4detach";
sc->last_op_thr = curthread;
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ADAPTER_UNLOCK(sc);
Add support for packet-sniffing tracers to cxgbe(4). This works with all T4 and T5 based cards and is useful for analyzing TSO, LRO, TOE, and for general purpose monitoring without tapping any cxgbe or cxl ifnet directly. Tracers on the T4/T5 chips provide access to Ethernet frames exactly as they were received from or transmitted on the wire. On transmit, a tracer will capture a frame after TSO segmentation, hw VLAN tag insertion, hw L3 & L4 checksum insertion, etc. It will also capture frames generated by the TCP offload engine (TOE traffic is normally invisible to the kernel). On receive, a tracer will capture a frame before hw VLAN extraction, runt filtering, other badness filtering, before the steering/drop/L2-rewrite filters or the TOE have had a go at it, and of course before sw LRO in the driver. There are 4 tracers on a chip. A tracer can trace only in one direction (tx or rx). For now cxgbetool will set up tracers to capture the first 128B of every transmitted or received frame on a given port. This is a small subset of what the hardware can do. A pseudo ifnet with the same name as the nexus driver (t4nex0 or t5nex0) will be created for tracing. The data delivered to this ifnet is an additional copy made inside the chip. Normal delivery to cxgbe<n> or cxl<n> will be made as usual. /* watch cxl0, which is the first port hanging off t5nex0. */ # cxgbetool t5nex0 tracer 0 tx0 (watch what cxl0 is transmitting) # cxgbetool t5nex0 tracer 1 rx0 (watch what cxl0 is receiving) # cxgbetool t5nex0 tracer list # tcpdump -i t5nex0 <== all that cxl0 sees and puts on the wire If you were doing TSO, a tcpdump on cxl0 may have shown you ~64K "frames" with no L3/L4 checksum but this will show you the frames that were actually transmitted. /* all done */ # cxgbetool t5nex0 tracer 0 disable # cxgbetool t5nex0 tracer 1 disable # cxgbetool t5nex0 tracer list # ifconfig t5nex0 destroy
2013-07-26 22:04:11 +00:00
if (pi->flags & HAS_TRACEQ) {
sc->traceq = -1; /* cloner should not create ifnet */
t4_tracer_port_detach(sc);
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
if (pi->vlan_c)
EVENTHANDLER_DEREGISTER(vlan_config, pi->vlan_c);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
PORT_LOCK(pi);
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
callout_stop(&pi->tick);
PORT_UNLOCK(pi);
callout_drain(&pi->tick);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/* Let detach proceed even if these fail. */
cxgbe_uninit_synchronized(pi);
port_full_uninit(pi);
Resume tx immediately in response to an SGE egress update from the hardware. MFC after: 1 week
2011-03-05 03:06:38 +00:00
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ifmedia_removeall(&pi->media);
ether_ifdetach(pi->ifp);
if_free(pi->ifp);
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
/* XXXNM: equivalent of cxgbe_uninit_synchronized to ifdown nm_ifp */
destroy_netmap_ifnet(pi);
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ADAPTER_LOCK(sc);
CLR_BUSY(sc);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
wakeup(&sc->flags);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ADAPTER_UNLOCK(sc);
return (0);
}
static void
cxgbe_init(void *arg)
{
struct port_info *pi = arg;
struct adapter *sc = pi->adapter;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4init") != 0)
return;
cxgbe_init_synchronized(pi);
end_synchronized_op(sc, 0);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
static int
cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
{
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
int rc = 0, mtu, flags, can_sleep;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
struct port_info *pi = ifp->if_softc;
struct adapter *sc = pi->adapter;
struct ifreq *ifr = (struct ifreq *)data;
uint32_t mask;
switch (cmd) {
case SIOCSIFMTU:
mtu = ifr->ifr_mtu;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if ((mtu < ETHERMIN) || (mtu > ETHERMTU_JUMBO))
return (EINVAL);
rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4mtu");
if (rc)
return (rc);
ifp->if_mtu = mtu;
- Make note of interface MTU change if the rx queues exist, and not just when the interface is up. - Add a tunable to control the TOE's rx coalesce feature (enabled by default as it always has been). Consider the interface MTU or the coalesce size when deciding which cluster zone to use to fill the offload rx queue's free list. The tunable is: dev.{t4nex,t5nex}.<N>.toe.rx_coalesce MFC after: 1 day
2013-07-04 21:19:01 +00:00
if (pi->flags & PORT_INIT_DONE) {
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
t4_update_fl_bufsize(ifp);
- Make note of interface MTU change if the rx queues exist, and not just when the interface is up. - Add a tunable to control the TOE's rx coalesce feature (enabled by default as it always has been). Consider the interface MTU or the coalesce size when deciding which cluster zone to use to fill the offload rx queue's free list. The tunable is: dev.{t4nex,t5nex}.<N>.toe.rx_coalesce MFC after: 1 day
2013-07-04 21:19:01 +00:00
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
rc = update_mac_settings(ifp, XGMAC_MTU);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
end_synchronized_op(sc, 0);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
break;
case SIOCSIFFLAGS:
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
can_sleep = 0;
redo_sifflags:
rc = begin_synchronized_op(sc, pi,
can_sleep ? (SLEEP_OK | INTR_OK) : HOLD_LOCK, "t4flg");
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (rc)
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (ifp->if_flags & IFF_UP) {
if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
flags = pi->if_flags;
if ((ifp->if_flags ^ flags) &
(IFF_PROMISC | IFF_ALLMULTI)) {
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
if (can_sleep == 1) {
end_synchronized_op(sc, 0);
can_sleep = 0;
goto redo_sifflags;
}
rc = update_mac_settings(ifp,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
XGMAC_PROMISC | XGMAC_ALLMULTI);
}
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
} else {
if (can_sleep == 0) {
end_synchronized_op(sc, LOCK_HELD);
can_sleep = 1;
goto redo_sifflags;
}
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = cxgbe_init_synchronized(pi);
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
pi->if_flags = ifp->if_flags;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
if (can_sleep == 0) {
end_synchronized_op(sc, LOCK_HELD);
can_sleep = 1;
goto redo_sifflags;
}
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = cxgbe_uninit_synchronized(pi);
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
}
end_synchronized_op(sc, can_sleep ? 0 : LOCK_HELD);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
break;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
case SIOCADDMULTI:
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
case SIOCDELMULTI: /* these two are called with a mutex held :-( */
rc = begin_synchronized_op(sc, pi, HOLD_LOCK, "t4multi");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc)
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
return (rc);
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
rc = update_mac_settings(ifp, XGMAC_MCADDRS);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
end_synchronized_op(sc, LOCK_HELD);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
break;
case SIOCSIFCAP:
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4cap");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc)
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
mask = ifr->ifr_reqcap ^ ifp->if_capenable;
if (mask & IFCAP_TXCSUM) {
ifp->if_capenable ^= IFCAP_TXCSUM;
ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
- Assign (don't OR) the CSUM_XXX bits to csum_flags in the rx checksum code. - Fix TSO/TSO4 mixup. - Add IFCAP_LINKSTATE to the available/enabled capabilities.
2012-06-30 02:05:09 +00:00
if (IFCAP_TSO4 & ifp->if_capenable &&
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
!(IFCAP_TXCSUM & ifp->if_capenable)) {
cxgbe(4): support for IPv6 hardware checksumming (rx and tx).
2012-06-29 16:50:52 +00:00
ifp->if_capenable &= ~IFCAP_TSO4;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if_printf(ifp,
- Assign (don't OR) the CSUM_XXX bits to csum_flags in the rx checksum code. - Fix TSO/TSO4 mixup. - Add IFCAP_LINKSTATE to the available/enabled capabilities.
2012-06-30 02:05:09 +00:00
"tso4 disabled due to -txcsum.\n");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
}
cxgbe(4): support for IPv6 hardware checksumming (rx and tx).
2012-06-29 16:50:52 +00:00
if (mask & IFCAP_TXCSUM_IPV6) {
ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
if (IFCAP_TSO6 & ifp->if_capenable &&
!(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
ifp->if_capenable &= ~IFCAP_TSO6;
if_printf(ifp,
"tso6 disabled due to -txcsum6.\n");
}
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (mask & IFCAP_RXCSUM)
ifp->if_capenable ^= IFCAP_RXCSUM;
cxgbe(4): support for IPv6 hardware checksumming (rx and tx).
2012-06-29 16:50:52 +00:00
if (mask & IFCAP_RXCSUM_IPV6)
ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
/*
* Note that we leave CSUM_TSO alone (it is always set). The
* kernel takes both IFCAP_TSOx and CSUM_TSO into account before
* sending a TSO request our way, so it's sufficient to toggle
* IFCAP_TSOx only.
*/
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (mask & IFCAP_TSO4) {
cxgbe(4): support for IPv6 hardware checksumming (rx and tx).
2012-06-29 16:50:52 +00:00
if (!(IFCAP_TSO4 & ifp->if_capenable) &&
!(IFCAP_TXCSUM & ifp->if_capenable)) {
if_printf(ifp, "enable txcsum first.\n");
rc = EAGAIN;
goto fail;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ifp->if_capenable ^= IFCAP_TSO4;
cxgbe(4): support for IPv6 hardware checksumming (rx and tx).
2012-06-29 16:50:52 +00:00
}
if (mask & IFCAP_TSO6) {
if (!(IFCAP_TSO6 & ifp->if_capenable) &&
!(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
if_printf(ifp, "enable txcsum6 first.\n");
rc = EAGAIN;
goto fail;
}
ifp->if_capenable ^= IFCAP_TSO6;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
if (mask & IFCAP_LRO) {
cxgbe(4): support for IPv6 TSO and LRO. Submitted by: bz (this is a modified version of that patch)
2012-06-29 19:51:06 +00:00
#if defined(INET) || defined(INET6)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
int i;
struct sge_rxq *rxq;
ifp->if_capenable ^= IFCAP_LRO;
for_each_rxq(pi, i, rxq) {
if (ifp->if_capenable & IFCAP_LRO)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
rxq->iq.flags |= IQ_LRO_ENABLED;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
else
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
rxq->iq.flags &= ~IQ_LRO_ENABLED;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
#endif
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (mask & IFCAP_TOE) {
int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
rc = toe_capability(pi, enable);
if (rc != 0)
goto fail;
ifp->if_capenable ^= mask;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
#endif
if (mask & IFCAP_VLAN_HWTAGGING) {
ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (ifp->if_drv_flags & IFF_DRV_RUNNING)
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
rc = update_mac_settings(ifp, XGMAC_VLANEX);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
if (mask & IFCAP_VLAN_MTU) {
ifp->if_capenable ^= IFCAP_VLAN_MTU;
/* Need to find out how to disable auto-mtu-inflation */
}
if (mask & IFCAP_VLAN_HWTSO)
ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
if (mask & IFCAP_VLAN_HWCSUM)
ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
#ifdef VLAN_CAPABILITIES
VLAN_CAPABILITIES(ifp);
#endif
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
fail:
end_synchronized_op(sc, 0);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
break;
case SIOCSIFMEDIA:
case SIOCGIFMEDIA:
ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
break;
default:
rc = ether_ioctl(ifp, cmd, data);
}
return (rc);
}
static int
cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
{
struct port_info *pi = ifp->if_softc;
struct adapter *sc = pi->adapter;
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
struct sge_txq *txq;
void *items[1];
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
int rc;
M_ASSERTPKTHDR(m);
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
MPASS(m->m_nextpkt == NULL); /* not quite ready for this yet */
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (__predict_false(pi->link_cfg.link_ok == 0)) {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
m_freem(m);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (ENETDOWN);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
rc = parse_pkt(&m);
if (__predict_false(rc != 0)) {
MPASS(m == NULL); /* was freed already */
atomic_add_int(&pi->tx_parse_error, 1); /* rare, atomic is ok */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
/* Select a txq. */
txq = &sc->sge.txq[pi->first_txq];
if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
txq += ((m->m_pkthdr.flowid % (pi->ntxq - pi->rsrv_noflowq)) +
pi->rsrv_noflowq);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
items[0] = m;
rc = mp_ring_enqueue(txq->r, items, 1, 4096);
if (__predict_false(rc != 0))
m_freem(m);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (rc);
}
static void
cxgbe_qflush(struct ifnet *ifp)
{
struct port_info *pi = ifp->if_softc;
Fix a couple of bad races that can occur when a cxgbe interface is taken down. The ingress queue lock was unused and has been removed as part of these changes. - An in-flight egress update from the SGE must be handled before the queue that requested it is destroyed. Wait for the update to arrive. - Interrupt handlers must stop processing rx events for a queue before the queue is destroyed. Events that have not yet been processed should be ignored once the queue disappears. MFC after: 1 week
2011-04-15 03:09:27 +00:00
struct sge_txq *txq;
int i;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/* queues do not exist if !PORT_INIT_DONE. */
if (pi->flags & PORT_INIT_DONE) {
Fix a couple of bad races that can occur when a cxgbe interface is taken down. The ingress queue lock was unused and has been removed as part of these changes. - An in-flight egress update from the SGE must be handled before the queue that requested it is destroyed. Wait for the update to arrive. - Interrupt handlers must stop processing rx events for a queue before the queue is destroyed. Events that have not yet been processed should be ignored once the queue disappears. MFC after: 1 week
2011-04-15 03:09:27 +00:00
for_each_txq(pi, i, txq) {
TXQ_LOCK(txq);
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
txq->eq.flags &= ~EQ_ENABLED;
Fix a couple of bad races that can occur when a cxgbe interface is taken down. The ingress queue lock was unused and has been removed as part of these changes. - An in-flight egress update from the SGE must be handled before the queue that requested it is destroyed. Wait for the update to arrive. - Interrupt handlers must stop processing rx events for a queue before the queue is destroyed. Events that have not yet been processed should be ignored once the queue disappears. MFC after: 1 week
2011-04-15 03:09:27 +00:00
TXQ_UNLOCK(txq);
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
while (!mp_ring_is_idle(txq->r)) {
mp_ring_check_drainage(txq->r, 0);
pause("qflush", 1);
}
Fix a couple of bad races that can occur when a cxgbe interface is taken down. The ingress queue lock was unused and has been removed as part of these changes. - An in-flight egress update from the SGE must be handled before the queue that requested it is destroyed. Wait for the update to arrive. - Interrupt handlers must stop processing rx events for a queue before the queue is destroyed. Events that have not yet been processed should be ignored once the queue disappears. MFC after: 1 week
2011-04-15 03:09:27 +00:00
}
}
if_qflush(ifp);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
static int
cxgbe_media_change(struct ifnet *ifp)
{
struct port_info *pi = ifp->if_softc;
device_printf(pi->dev, "%s unimplemented.\n", __func__);
return (EOPNOTSUPP);
}
static void
cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
{
struct port_info *pi = ifp->if_softc;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
struct ifmedia *media = NULL;
struct ifmedia_entry *cur;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
int speed = pi->link_cfg.speed;
MFC r274456: Fix some bad interaction between cxgbe(4) and lacp lagg(4) that could leave a port permanently disabled when a copper cable is unplugged and then plugged right back in. lacp_linkstate goes looking for the current ifmedia on a link state change and it could get stale information from cxgbe(4) on a module unplug followed by replug. The fix is to process module events before link-state events within the driver, and to always rebuild the ifmedia list on a module change event (instead of rebuilding it lazily). Thanks to asomers@ for the problem report and detailed analysis to go with it.
2015-04-07 15:32:43 +00:00
#ifdef INVARIANTS
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
int data = (pi->port_type << 8) | pi->mod_type;
MFC r274456: Fix some bad interaction between cxgbe(4) and lacp lagg(4) that could leave a port permanently disabled when a copper cable is unplugged and then plugged right back in. lacp_linkstate goes looking for the current ifmedia on a link state change and it could get stale information from cxgbe(4) on a module unplug followed by replug. The fix is to process module events before link-state events within the driver, and to always rebuild the ifmedia list on a module change event (instead of rebuilding it lazily). Thanks to asomers@ for the problem report and detailed analysis to go with it.
2015-04-07 15:32:43 +00:00
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
if (ifp == pi->ifp)
media = &pi->media;
#ifdef DEV_NETMAP
else if (ifp == pi->nm_ifp)
media = &pi->nm_media;
#endif
MPASS(media != NULL);
cur = media->ifm_cur;
MFC r274456: Fix some bad interaction between cxgbe(4) and lacp lagg(4) that could leave a port permanently disabled when a copper cable is unplugged and then plugged right back in. lacp_linkstate goes looking for the current ifmedia on a link state change and it could get stale information from cxgbe(4) on a module unplug followed by replug. The fix is to process module events before link-state events within the driver, and to always rebuild the ifmedia list on a module change event (instead of rebuilding it lazily). Thanks to asomers@ for the problem report and detailed analysis to go with it.
2015-04-07 15:32:43 +00:00
MPASS(cur->ifm_data == data);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ifmr->ifm_status = IFM_AVALID;
if (!pi->link_cfg.link_ok)
return;
ifmr->ifm_status |= IFM_ACTIVE;
/* active and current will differ iff current media is autoselect. */
if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO)
return;
ifmr->ifm_active = IFM_ETHER | IFM_FDX;
if (speed == SPEED_10000)
ifmr->ifm_active |= IFM_10G_T;
else if (speed == SPEED_1000)
ifmr->ifm_active |= IFM_1000_T;
else if (speed == SPEED_100)
ifmr->ifm_active |= IFM_100_TX;
else if (speed == SPEED_10)
ifmr->ifm_active |= IFM_10_T;
else
KASSERT(0, ("%s: link up but speed unknown (%u)", __func__,
speed));
}
void
t4_fatal_err(struct adapter *sc)
{
t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
t4_intr_disable(sc);
log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
device_get_nameunit(sc->dev));
}
static int
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
map_bars_0_and_4(struct adapter *sc)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
{
sc->regs_rid = PCIR_BAR(0);
sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
&sc->regs_rid, RF_ACTIVE);
if (sc->regs_res == NULL) {
device_printf(sc->dev, "cannot map registers.\n");
return (ENXIO);
}
sc->bt = rman_get_bustag(sc->regs_res);
sc->bh = rman_get_bushandle(sc->regs_res);
sc->mmio_len = rman_get_size(sc->regs_res);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
setbit(&sc->doorbells, DOORBELL_KDB);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
sc->msix_rid = PCIR_BAR(4);
sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
&sc->msix_rid, RF_ACTIVE);
if (sc->msix_res == NULL) {
device_printf(sc->dev, "cannot map MSI-X BAR.\n");
return (ENXIO);
}
return (0);
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static int
map_bar_2(struct adapter *sc)
{
/*
* T4: only iWARP driver uses the userspace doorbells. There is no need
* to map it if RDMA is disabled.
*/
if (is_t4(sc) && sc->rdmacaps == 0)
return (0);
sc->udbs_rid = PCIR_BAR(2);
sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
&sc->udbs_rid, RF_ACTIVE);
if (sc->udbs_res == NULL) {
device_printf(sc->dev, "cannot map doorbell BAR.\n");
return (ENXIO);
}
sc->udbs_base = rman_get_virtual(sc->udbs_res);
if (is_t5(sc)) {
setbit(&sc->doorbells, DOORBELL_UDB);
#if defined(__i386__) || defined(__amd64__)
if (t5_write_combine) {
int rc;
/*
* Enable write combining on BAR2. This is the
* userspace doorbell BAR and is split into 128B
* (UDBS_SEG_SIZE) doorbell regions, each associated
* with an egress queue. The first 64B has the doorbell
* and the second 64B can be used to submit a tx work
* request with an implicit doorbell.
*/
rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
if (rc == 0) {
clrbit(&sc->doorbells, DOORBELL_UDB);
Cosmetic change (s/wrwc/wcwr/;s/WRWC/WCWR/). MFC after: 3 days.
2013-04-11 22:49:29 +00:00
setbit(&sc->doorbells, DOORBELL_WCWR);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
setbit(&sc->doorbells, DOORBELL_UDBWC);
} else {
device_printf(sc->dev,
"couldn't enable write combining: %d\n",
rc);
}
t4_write_reg(sc, A_SGE_STAT_CFG,
V_STATSOURCE_T5(7) | V_STATMODE(0));
}
#endif
}
return (0);
}
static const struct memwin t4_memwin[] = {
{ MEMWIN0_BASE, MEMWIN0_APERTURE },
{ MEMWIN1_BASE, MEMWIN1_APERTURE },
{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
};
static const struct memwin t5_memwin[] = {
{ MEMWIN0_BASE, MEMWIN0_APERTURE },
{ MEMWIN1_BASE, MEMWIN1_APERTURE },
{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
};
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static void
setup_memwin(struct adapter *sc)
{
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
const struct memwin *mw;
int i, n;
Allow cxgbe(4) running within a VM to attach to its devices that have been exported via PCI passthrough. - Do not check for a specific physical function (PF) before claiming a device. Different PFs have different device-ids so this check is redundant anyway. - Obtain the PF# from the WHOAMI register instead of pci_get_function(). - Setup the memory windows using the real BAR0 address, not what the VM says it is. Obtained from: Chelsio Communications
2012-06-26 00:34:34 +00:00
uint32_t bar0;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (is_t4(sc)) {
/*
* Read low 32b of bar0 indirectly via the hardware backdoor
* mechanism. Works from within PCI passthrough environments
* too, where rman_get_start() can return a different value. We
* need to program the T4 memory window decoders with the actual
* addresses that will be coming across the PCIe link.
*/
bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
mw = &t4_memwin[0];
n = nitems(t4_memwin);
} else {
/* T5 uses the relative offset inside the PCIe BAR */
bar0 = 0;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
mw = &t5_memwin[0];
n = nitems(t5_memwin);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
for (i = 0; i < n; i++, mw++) {
t4_write_reg(sc,
PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
(mw->base + bar0) | V_BIR(0) |
V_WINDOW(ilog2(mw->aperture) - 10));
}
Allow cxgbe(4) running within a VM to attach to its devices that have been exported via PCI passthrough. - Do not check for a specific physical function (PF) before claiming a device. Different PFs have different device-ids so this check is redundant anyway. - Obtain the PF# from the WHOAMI register instead of pci_get_function(). - Setup the memory windows using the real BAR0 address, not what the VM says it is. Obtained from: Chelsio Communications
2012-06-26 00:34:34 +00:00
/* flush */
t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/*
* Verify that the memory range specified by the addr/len pair is valid and lies
* entirely within a single region (EDCx or MCx).
*/
static int
validate_mem_range(struct adapter *sc, uint32_t addr, int len)
{
uint32_t em, addr_len, maddr, mlen;
/* Memory can only be accessed in naturally aligned 4 byte units */
if (addr & 3 || len & 3 || len == 0)
return (EINVAL);
/* Enabled memories */
em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
if (em & F_EDRAM0_ENABLE) {
addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
maddr = G_EDRAM0_BASE(addr_len) << 20;
mlen = G_EDRAM0_SIZE(addr_len) << 20;
if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
addr + len <= maddr + mlen)
return (0);
}
if (em & F_EDRAM1_ENABLE) {
addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
maddr = G_EDRAM1_BASE(addr_len) << 20;
mlen = G_EDRAM1_SIZE(addr_len) << 20;
if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
addr + len <= maddr + mlen)
return (0);
}
if (em & F_EXT_MEM_ENABLE) {
addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
maddr = G_EXT_MEM_BASE(addr_len) << 20;
mlen = G_EXT_MEM_SIZE(addr_len) << 20;
if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
addr + len <= maddr + mlen)
return (0);
}
if (!is_t4(sc) && em & F_EXT_MEM1_ENABLE) {
addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
maddr = G_EXT_MEM1_BASE(addr_len) << 20;
mlen = G_EXT_MEM1_SIZE(addr_len) << 20;
if (mlen > 0 && addr >= maddr && addr < maddr + mlen &&
addr + len <= maddr + mlen)
return (0);
}
return (EFAULT);
}
MFC r256459. cxgbe(4): Update T4 and T5 firmwares to 1.9.12.0 Approved by: re (glebius)
2013-10-20 15:24:44 +00:00
static int
fwmtype_to_hwmtype(int mtype)
{
switch (mtype) {
case FW_MEMTYPE_EDC0:
return (MEM_EDC0);
case FW_MEMTYPE_EDC1:
return (MEM_EDC1);
case FW_MEMTYPE_EXTMEM:
return (MEM_MC0);
case FW_MEMTYPE_EXTMEM1:
return (MEM_MC1);
default:
panic("%s: cannot translate fw mtype %d.", __func__, mtype);
}
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/*
* Verify that the memory range specified by the memtype/offset/len pair is
* valid and lies entirely within the memtype specified. The global address of
* the start of the range is returned in addr.
*/
static int
validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, int len,
uint32_t *addr)
{
uint32_t em, addr_len, maddr, mlen;
/* Memory can only be accessed in naturally aligned 4 byte units */
if (off & 3 || len & 3 || len == 0)
return (EINVAL);
em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
MFC r256459. cxgbe(4): Update T4 and T5 firmwares to 1.9.12.0 Approved by: re (glebius)
2013-10-20 15:24:44 +00:00
switch (fwmtype_to_hwmtype(mtype)) {
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
case MEM_EDC0:
if (!(em & F_EDRAM0_ENABLE))
return (EINVAL);
addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
maddr = G_EDRAM0_BASE(addr_len) << 20;
mlen = G_EDRAM0_SIZE(addr_len) << 20;
break;
case MEM_EDC1:
if (!(em & F_EDRAM1_ENABLE))
return (EINVAL);
addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
maddr = G_EDRAM1_BASE(addr_len) << 20;
mlen = G_EDRAM1_SIZE(addr_len) << 20;
break;
case MEM_MC:
if (!(em & F_EXT_MEM_ENABLE))
return (EINVAL);
addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
maddr = G_EXT_MEM_BASE(addr_len) << 20;
mlen = G_EXT_MEM_SIZE(addr_len) << 20;
break;
case MEM_MC1:
if (is_t4(sc) || !(em & F_EXT_MEM1_ENABLE))
return (EINVAL);
addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
maddr = G_EXT_MEM1_BASE(addr_len) << 20;
mlen = G_EXT_MEM1_SIZE(addr_len) << 20;
break;
default:
return (EINVAL);
}
if (mlen > 0 && off < mlen && off + len <= mlen) {
*addr = maddr + off; /* global address */
return (0);
}
return (EFAULT);
}
static void
memwin_info(struct adapter *sc, int win, uint32_t *base, uint32_t *aperture)
{
const struct memwin *mw;
if (is_t4(sc)) {
KASSERT(win >= 0 && win < nitems(t4_memwin),
("%s: incorrect memwin# (%d)", __func__, win));
mw = &t4_memwin[win];
} else {
KASSERT(win >= 0 && win < nitems(t5_memwin),
("%s: incorrect memwin# (%d)", __func__, win));
mw = &t5_memwin[win];
}
if (base != NULL)
*base = mw->base;
if (aperture != NULL)
*aperture = mw->aperture;
}
/*
* Positions the memory window such that it can be used to access the specified
* address in the chip's address space. The return value is the offset of addr
* from the start of the window.
*/
static uint32_t
position_memwin(struct adapter *sc, int n, uint32_t addr)
{
uint32_t start, pf;
uint32_t reg;
KASSERT(n >= 0 && n <= 3,
("%s: invalid window %d.", __func__, n));
KASSERT((addr & 3) == 0,
("%s: addr (0x%x) is not at a 4B boundary.", __func__, addr));
if (is_t4(sc)) {
pf = 0;
start = addr & ~0xf; /* start must be 16B aligned */
} else {
pf = V_PFNUM(sc->pf);
start = addr & ~0x7f; /* start must be 128B aligned */
}
reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, n);
t4_write_reg(sc, reg, start | pf);
t4_read_reg(sc, reg);
return (addr - start);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int
cfg_itype_and_nqueues(struct adapter *sc, int n10g, int n1g,
struct intrs_and_queues *iaq)
{
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
int rc, itype, navail, nrxq10g, nrxq1g, n;
int nofldrxq10g = 0, nofldrxq1g = 0;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
int nnmrxq10g = 0, nnmrxq1g = 0;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
bzero(iaq, sizeof(*iaq));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
iaq->ntxq10g = t4_ntxq10g;
iaq->ntxq1g = t4_ntxq1g;
iaq->nrxq10g = nrxq10g = t4_nrxq10g;
iaq->nrxq1g = nrxq1g = t4_nrxq1g;
MFC r261558 Add a new sysctl, dev.cxgbe.N.rsrv_noflow, and a companion tunable, hw.cxgbe.rsrv_noflow. When set, queue 0 of the port is reserved for TX packets without a flowid. The hash value of packets with a flowid is bumped up by 1. The intent is to provide a private queue for link-level packets like LACP that is unlikely to overflow or suffer deep queue latency.
2014-04-15 08:08:44 +00:00
iaq->rsrv_noflowq = t4_rsrv_noflowq;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Do not allocate extra vectors when adapter is not TOE capable (or toecaps have been disallowed by the user). + one very minor unrelated cleanup in t4_sge.c
2012-06-22 22:59:42 +00:00
if (is_offload(sc)) {
iaq->nofldtxq10g = t4_nofldtxq10g;
iaq->nofldtxq1g = t4_nofldtxq1g;
iaq->nofldrxq10g = nofldrxq10g = t4_nofldrxq10g;
iaq->nofldrxq1g = nofldrxq1g = t4_nofldrxq1g;
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
iaq->nnmtxq10g = t4_nnmtxq10g;
iaq->nnmtxq1g = t4_nnmtxq1g;
iaq->nnmrxq10g = nnmrxq10g = t4_nnmrxq10g;
iaq->nnmrxq1g = nnmrxq1g = t4_nnmrxq1g;
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
for (itype = INTR_MSIX; itype; itype >>= 1) {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if ((itype & t4_intr_types) == 0)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
continue; /* not allowed */
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
if (itype == INTR_MSIX)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
navail = pci_msix_count(sc->dev);
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
else if (itype == INTR_MSI)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
navail = pci_msi_count(sc->dev);
else
navail = 1;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
restart:
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (navail == 0)
continue;
iaq->intr_type = itype;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
iaq->intr_flags_10g = 0;
iaq->intr_flags_1g = 0;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/*
* Best option: an interrupt vector for errors, one for the
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
* firmware event queue, and one for every rxq (NIC, TOE, and
* netmap).
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
*/
iaq->nirq = T4_EXTRA_INTR;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
iaq->nirq += n10g * (nrxq10g + nofldrxq10g + nnmrxq10g);
iaq->nirq += n1g * (nrxq1g + nofldrxq1g + nnmrxq1g);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (iaq->nirq <= navail &&
(itype != INTR_MSI || powerof2(iaq->nirq))) {
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
iaq->intr_flags_10g = INTR_ALL;
iaq->intr_flags_1g = INTR_ALL;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
goto allocate;
}
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/*
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
* Second best option: a vector for errors, one for the firmware
* event queue, and vectors for either all the NIC rx queues or
* all the TOE rx queues. The queues that don't get vectors
* will forward their interrupts to those that do.
*
* Note: netmap rx queues cannot be created early and so they
* can't be setup to receive forwarded interrupts for others.
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
*/
iaq->nirq = T4_EXTRA_INTR;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
if (nrxq10g >= nofldrxq10g) {
iaq->intr_flags_10g = INTR_RXQ;
iaq->nirq += n10g * nrxq10g;
#ifdef DEV_NETMAP
iaq->nnmrxq10g = min(nnmrxq10g, nrxq10g);
#endif
} else {
iaq->intr_flags_10g = INTR_OFLD_RXQ;
iaq->nirq += n10g * nofldrxq10g;
#ifdef DEV_NETMAP
iaq->nnmrxq10g = min(nnmrxq10g, nofldrxq10g);
#endif
}
if (nrxq1g >= nofldrxq1g) {
iaq->intr_flags_1g = INTR_RXQ;
iaq->nirq += n1g * nrxq1g;
#ifdef DEV_NETMAP
iaq->nnmrxq1g = min(nnmrxq1g, nrxq1g);
#endif
} else {
iaq->intr_flags_1g = INTR_OFLD_RXQ;
iaq->nirq += n1g * nofldrxq1g;
#ifdef DEV_NETMAP
iaq->nnmrxq1g = min(nnmrxq1g, nofldrxq1g);
#endif
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (iaq->nirq <= navail &&
(itype != INTR_MSI || powerof2(iaq->nirq)))
goto allocate;
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/*
* Next best option: an interrupt vector for errors, one for the
* firmware event queue, and at least one per port. At this
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
* point we know we'll have to downsize nrxq and/or nofldrxq
* and/or nnmrxq to fit what's available to us.
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
*/
iaq->nirq = T4_EXTRA_INTR;
iaq->nirq += n10g + n1g;
if (iaq->nirq <= navail) {
int leftover = navail - iaq->nirq;
if (n10g > 0) {
int target = max(nrxq10g, nofldrxq10g);
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
iaq->intr_flags_10g = nrxq10g >= nofldrxq10g ?
INTR_RXQ : INTR_OFLD_RXQ;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
n = 1;
while (n < target && leftover >= n10g) {
leftover -= n10g;
iaq->nirq += n10g;
n++;
}
iaq->nrxq10g = min(n, nrxq10g);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
iaq->nofldrxq10g = min(n, nofldrxq10g);
#endif
#ifdef DEV_NETMAP
iaq->nnmrxq10g = min(n, nnmrxq10g);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
}
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (n1g > 0) {
int target = max(nrxq1g, nofldrxq1g);
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
iaq->intr_flags_1g = nrxq1g >= nofldrxq1g ?
INTR_RXQ : INTR_OFLD_RXQ;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
n = 1;
while (n < target && leftover >= n1g) {
leftover -= n1g;
iaq->nirq += n1g;
n++;
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
iaq->nrxq1g = min(n, nrxq1g);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
iaq->nofldrxq1g = min(n, nofldrxq1g);
#endif
#ifdef DEV_NETMAP
iaq->nnmrxq1g = min(n, nnmrxq1g);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (itype != INTR_MSI || powerof2(iaq->nirq))
goto allocate;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/*
* Least desirable option: one interrupt vector for everything.
*/
iaq->nirq = iaq->nrxq10g = iaq->nrxq1g = 1;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
iaq->intr_flags_10g = iaq->intr_flags_1g = 0;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Do not allocate extra vectors when adapter is not TOE capable (or toecaps have been disallowed by the user). + one very minor unrelated cleanup in t4_sge.c
2012-06-22 22:59:42 +00:00
if (is_offload(sc))
iaq->nofldrxq10g = iaq->nofldrxq1g = 1;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
iaq->nnmrxq10g = iaq->nnmrxq1g = 1;
#endif
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
allocate:
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
navail = iaq->nirq;
rc = 0;
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
if (itype == INTR_MSIX)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
rc = pci_alloc_msix(sc->dev, &navail);
Do not over-allocate MSI interrupts for the case where each ingress queue has its own interrupt. If the exact number that we need is not a power of 2 and we're using MSI, then switch to interrupt multiplexing. While here, replace the magic numbers with something more readable. MFC after: 3 days
2011-03-24 01:03:01 +00:00
else if (itype == INTR_MSI)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
rc = pci_alloc_msi(sc->dev, &navail);
if (rc == 0) {
if (navail == iaq->nirq)
return (0);
/*
* Didn't get the number requested. Use whatever number
* the kernel is willing to allocate (it's in navail).
*/
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
device_printf(sc->dev, "fewer vectors than requested, "
"type=%d, req=%d, rcvd=%d; will downshift req.\n",
itype, iaq->nirq, navail);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
pci_release_msi(sc->dev);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
goto restart;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
device_printf(sc->dev,
"failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
itype, rc, iaq->nirq, navail);
}
device_printf(sc->dev,
"failed to find a usable interrupt type. "
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
"allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
pci_msix_count(sc->dev), pci_msi_count(sc->dev));
return (ENXIO);
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
#define FW_VERSION(chip) ( \
- Include the T5 firmware with the driver. - Update the T4 firmware to the latest. - Minor reorganization and updates to the version macros, etc. Obtained from: Chelsio MFC after: 1 day
2013-07-03 23:52:15 +00:00
V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
#define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
struct fw_info {
uint8_t chip;
char *kld_name;
char *fw_mod_name;
struct fw_hdr fw_hdr; /* XXX: waste of space, need a sparse struct */
} fw_info[] = {
{
.chip = CHELSIO_T4,
.kld_name = "t4fw_cfg",
.fw_mod_name = "t4fw",
.fw_hdr = {
.chip = FW_HDR_CHIP_T4,
.fw_ver = htobe32_const(FW_VERSION(T4)),
.intfver_nic = FW_INTFVER(T4, NIC),
.intfver_vnic = FW_INTFVER(T4, VNIC),
.intfver_ofld = FW_INTFVER(T4, OFLD),
.intfver_ri = FW_INTFVER(T4, RI),
.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
.intfver_iscsi = FW_INTFVER(T4, ISCSI),
.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
.intfver_fcoe = FW_INTFVER(T4, FCOE),
},
}, {
.chip = CHELSIO_T5,
.kld_name = "t5fw_cfg",
.fw_mod_name = "t5fw",
.fw_hdr = {
.chip = FW_HDR_CHIP_T5,
.fw_ver = htobe32_const(FW_VERSION(T5)),
.intfver_nic = FW_INTFVER(T5, NIC),
.intfver_vnic = FW_INTFVER(T5, VNIC),
.intfver_ofld = FW_INTFVER(T5, OFLD),
.intfver_ri = FW_INTFVER(T5, RI),
.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
.intfver_iscsi = FW_INTFVER(T5, ISCSI),
.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
.intfver_fcoe = FW_INTFVER(T5, FCOE),
},
}
};
static struct fw_info *
find_fw_info(int chip)
{
int i;
for (i = 0; i < nitems(fw_info); i++) {
if (fw_info[i].chip == chip)
return (&fw_info[i]);
}
return (NULL);
}
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
/*
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
* Is the given firmware API compatible with the one the driver was compiled
* with?
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
*/
static int
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
{
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/* short circuit if it's the exact same firmware version */
if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
return (1);
/*
* XXX: Is this too conservative? Perhaps I should limit this to the
* features that are supported in the driver.
*/
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
#define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
return (1);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
#undef SAME_INTF
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
return (0);
}
- Explain clearly why a different firmware is being installed (if/when it is being installed). Improve other error messages while here. - Select special FPGA specific configuration profile when appropriate. MFC after: 3 days
2013-04-11 19:39:40 +00:00
/*
cxgbe(4): Never install a firmware if hw.cxgbe.fw_install is 0. MFC after: 1 week
2013-06-05 20:57:52 +00:00
* The firmware in the KLD is usable, but should it be installed? This routine
* explains itself in detail if it indicates the KLD firmware should be
* installed.
- Explain clearly why a different firmware is being installed (if/when it is being installed). Improve other error messages while here. - Select special FPGA specific configuration profile when appropriate. MFC after: 3 days
2013-04-11 19:39:40 +00:00
*/
static int
should_install_kld_fw(struct adapter *sc, int card_fw_usable, int k, int c)
{
const char *reason;
if (!card_fw_usable) {
reason = "incompatible or unusable";
goto install;
}
if (k > c) {
reason = "older than the version bundled with this driver";
goto install;
}
if (t4_fw_install == 2 && k != c) {
reason = "different than the version bundled with this driver";
goto install;
}
return (0);
install:
cxgbe(4): Never install a firmware if hw.cxgbe.fw_install is 0. MFC after: 1 week
2013-06-05 20:57:52 +00:00
if (t4_fw_install == 0) {
device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
"but the driver is prohibited from installing a different "
"firmware on the card.\n",
G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
return (0);
}
- Explain clearly why a different firmware is being installed (if/when it is being installed). Improve other error messages while here. - Select special FPGA specific configuration profile when appropriate. MFC after: 3 days
2013-04-11 19:39:40 +00:00
device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
"installing firmware %u.%u.%u.%u on card.\n",
G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
return (1);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
* Establish contact with the firmware and determine if we are the master driver
* or not, and whether we are responsible for chip initialization.
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
*/
static int
prep_firmware(struct adapter *sc)
{
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
const struct firmware *fw = NULL, *default_cfg;
int rc, pf, card_fw_usable, kld_fw_usable, need_fw_reset = 1;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
enum dev_state state;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
struct fw_info *fw_info;
struct fw_hdr *card_fw; /* fw on the card */
const struct fw_hdr *kld_fw; /* fw in the KLD */
const struct fw_hdr *drv_fw; /* fw header the driver was compiled
against */
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/* Contact firmware. */
rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
if (rc < 0 || state == DEV_STATE_ERR) {
rc = -rc;
device_printf(sc->dev,
"failed to connect to the firmware: %d, %d.\n", rc, state);
return (rc);
}
pf = rc;
if (pf == sc->mbox)
sc->flags |= MASTER_PF;
else if (state == DEV_STATE_UNINIT) {
/*
* We didn't get to be the master so we definitely won't be
* configuring the chip. It's a bug if someone else hasn't
* configured it already.
*/
device_printf(sc->dev, "couldn't be master(%d), "
"device not already initialized either(%d).\n", rc, state);
return (EDOOFUS);
}
/* This is the firmware whose headers the driver was compiled against */
fw_info = find_fw_info(chip_id(sc));
if (fw_info == NULL) {
device_printf(sc->dev,
"unable to look up firmware information for chip %d.\n",
chip_id(sc));
return (EINVAL);
}
drv_fw = &fw_info->fw_hdr;
/*
* The firmware KLD contains many modules. The KLD name is also the
* name of the module that contains the default config file.
*/
default_cfg = firmware_get(fw_info->kld_name);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
/* Read the header of the firmware on the card */
card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
rc = -t4_read_flash(sc, FLASH_FW_START,
sizeof (*card_fw) / sizeof (uint32_t), (uint32_t *)card_fw, 1);
if (rc == 0)
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
card_fw_usable = fw_compatible(drv_fw, (const void*)card_fw);
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
else {
device_printf(sc->dev,
"Unable to read card's firmware header: %d\n", rc);
card_fw_usable = 0;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
/* This is the firmware in the KLD */
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
fw = firmware_get(fw_info->fw_mod_name);
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
if (fw != NULL) {
kld_fw = (const void *)fw->data;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
kld_fw_usable = fw_compatible(drv_fw, kld_fw);
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
} else {
kld_fw = NULL;
kld_fw_usable = 0;
}
Upgrade the firmware on the card automatically if a better version is available. Downgrade only for a major version mismatch. MFC after: 1 week
2011-03-05 03:12:50 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (card_fw_usable && card_fw->fw_ver == drv_fw->fw_ver &&
cxgbe(4): Never install a firmware if hw.cxgbe.fw_install is 0. MFC after: 1 week
2013-06-05 20:57:52 +00:00
(!kld_fw_usable || kld_fw->fw_ver == drv_fw->fw_ver)) {
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/*
* Common case: the firmware on the card is an exact match and
* the KLD is an exact match too, or the KLD is
cxgbe(4): Never install a firmware if hw.cxgbe.fw_install is 0. MFC after: 1 week
2013-06-05 20:57:52 +00:00
* absent/incompatible. Note that t4_fw_install = 2 is ignored
* here -- use cxgbetool loadfw if you want to reinstall the
* same firmware as the one on the card.
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
*/
} else if (kld_fw_usable && state == DEV_STATE_UNINIT &&
- Explain clearly why a different firmware is being installed (if/when it is being installed). Improve other error messages while here. - Select special FPGA specific configuration profile when appropriate. MFC after: 3 days
2013-04-11 19:39:40 +00:00
should_install_kld_fw(sc, card_fw_usable, be32toh(kld_fw->fw_ver),
be32toh(card_fw->fw_ver))) {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe: Switch to a better way to install firmware. MFC after: 1 week
2013-05-03 20:09:17 +00:00
rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
if (rc != 0) {
device_printf(sc->dev,
"failed to install firmware: %d\n", rc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
goto done;
Upgrade the firmware on the card automatically if a better version is available. Downgrade only for a major version mismatch. MFC after: 1 week
2011-03-05 03:12:50 +00:00
}
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
/* Installed successfully, update the cached header too. */
memcpy(card_fw, kld_fw, sizeof(*card_fw));
card_fw_usable = 1;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
need_fw_reset = 0; /* already reset as part of load_fw */
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
}
Upgrade the firmware on the card automatically if a better version is available. Downgrade only for a major version mismatch. MFC after: 1 week
2011-03-05 03:12:50 +00:00
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
if (!card_fw_usable) {
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
uint32_t d, c, k;
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
d = ntohl(drv_fw->fw_ver);
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
c = ntohl(card_fw->fw_ver);
k = kld_fw ? ntohl(kld_fw->fw_ver) : 0;
device_printf(sc->dev, "Cannot find a usable firmware: "
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
"fw_install %d, chip state %d, "
"driver compiled with %d.%d.%d.%d, "
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
"card has %d.%d.%d.%d, KLD has %d.%d.%d.%d\n",
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
t4_fw_install, state,
G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d),
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c),
G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
rc = EINVAL;
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
goto done;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
/* We're using whatever's on the card and it's known to be good. */
sc->params.fw_vers = ntohl(card_fw->fw_ver);
snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
- Read all TP parameters in one place. - Read the filter mode, calculate various shifts, and use them properly during active open (in select_ntuple). MFC after: 1 day
2013-07-04 17:55:52 +00:00
t4_get_tp_version(sc, &sc->params.tp_vers);
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/* Reset device */
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (need_fw_reset &&
(rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST)) != 0) {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
if (rc != ETIMEDOUT && rc != EIO)
t4_fw_bye(sc, sc->mbox);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
goto done;
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
sc->flags |= FW_OK;
rc = get_params__pre_init(sc);
if (rc != 0)
goto done; /* error message displayed already */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/* Partition adapter resources as specified in the config file. */
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (state == DEV_STATE_UNINIT) {
KASSERT(sc->flags & MASTER_PF,
("%s: trying to change chip settings when not master.",
__func__));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
rc = partition_resources(sc, default_cfg, fw_info->kld_name);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (rc != 0)
goto done; /* error message displayed already */
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
t4_tweak_chip_settings(sc);
/* get basic stuff going */
rc = -t4_fw_initialize(sc, sc->mbox);
if (rc != 0) {
device_printf(sc->dev, "fw init failed: %d.\n", rc);
goto done;
}
Force the 404-BT card (4 x 1G) to use the "uwire" configuration file. MFC after: 3 days
2013-01-26 03:10:28 +00:00
} else {
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", pf);
sc->cfcsum = 0;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
done:
cxgbe(4): Consider all the API versions of the interfaces exported by the firmware (instead of just the main firmware version) when evaluating firmware compatibility. Document the new "hw.cxgbe.fw_install" knob being introduced here. This should fix kern/173584 too. Setting hw.cxgbe.fw_install=2 will mostly do what was requested in the PR but it's a bit more intelligent in that it won't reinstall the same firmware repeatedly if the knob is left set. PR: kern/173584 MFC after: 5 days
2013-02-26 20:35:54 +00:00
free(card_fw, M_CXGBE);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (fw != NULL)
firmware_put(fw, FIRMWARE_UNLOAD);
if (default_cfg != NULL)
firmware_put(default_cfg, FIRMWARE_UNLOAD);
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#define FW_PARAM_DEV(param) \
(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
#define FW_PARAM_PFVF(param) \
(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
/*
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
* Partition chip resources for use between various PFs, VFs, etc.
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
*/
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
static int
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
partition_resources(struct adapter *sc, const struct firmware *default_cfg,
const char *name_prefix)
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
{
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
const struct firmware *cfg = NULL;
int rc = 0;
struct fw_caps_config_cmd caps;
uint32_t mtype, moff, finicsum, cfcsum;
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/*
* Figure out what configuration file to use. Pick the default config
* file for the card if the user hasn't specified one explicitly.
*/
snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", t4_cfg_file);
if (strncmp(t4_cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
/* Card specific overrides go here. */
if (pci_get_device(sc->dev) == 0x440a)
snprintf(sc->cfg_file, sizeof(sc->cfg_file), UWIRE_CF);
- Explain clearly why a different firmware is being installed (if/when it is being installed). Improve other error messages while here. - Select special FPGA specific configuration profile when appropriate. MFC after: 3 days
2013-04-11 19:39:40 +00:00
if (is_fpga(sc))
snprintf(sc->cfg_file, sizeof(sc->cfg_file), FPGA_CF);
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/*
* We need to load another module if the profile is anything except
* "default" or "flash".
*/
if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) != 0 &&
strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
char s[32];
snprintf(s, sizeof(s), "%s_%s", name_prefix, sc->cfg_file);
cfg = firmware_get(s);
if (cfg == NULL) {
if (default_cfg != NULL) {
- Explain clearly why a different firmware is being installed (if/when it is being installed). Improve other error messages while here. - Select special FPGA specific configuration profile when appropriate. MFC after: 3 days
2013-04-11 19:39:40 +00:00
device_printf(sc->dev,
"unable to load module \"%s\" for "
"configuration profile \"%s\", will use "
"the default config file instead.\n",
s, sc->cfg_file);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
snprintf(sc->cfg_file, sizeof(sc->cfg_file),
"%s", DEFAULT_CF);
} else {
- Explain clearly why a different firmware is being installed (if/when it is being installed). Improve other error messages while here. - Select special FPGA specific configuration profile when appropriate. MFC after: 3 days
2013-04-11 19:39:40 +00:00
device_printf(sc->dev,
"unable to load module \"%s\" for "
"configuration profile \"%s\", will use "
"the config file on the card's flash "
"instead.\n", s, sc->cfg_file);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
snprintf(sc->cfg_file, sizeof(sc->cfg_file),
"%s", FLASH_CF);
}
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (strncmp(sc->cfg_file, DEFAULT_CF, sizeof(sc->cfg_file)) == 0 &&
default_cfg == NULL) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
device_printf(sc->dev,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
"default config file not available, will use the config "
"file on the card's flash instead.\n");
snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", FLASH_CF);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (strncmp(sc->cfg_file, FLASH_CF, sizeof(sc->cfg_file)) != 0) {
u_int cflen, i, n;
const uint32_t *cfdata;
uint32_t param, val, addr, off, mw_base, mw_aperture;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
KASSERT(cfg != NULL || default_cfg != NULL,
("%s: no config to upload", __func__));
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/*
* Ask the firmware where it wants us to upload the config file.
*/
param = FW_PARAM_DEV(CF);
rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
if (rc != 0) {
/* No support for config file? Shouldn't happen. */
device_printf(sc->dev,
"failed to query config file location: %d.\n", rc);
goto done;
}
mtype = G_FW_PARAMS_PARAM_Y(val);
moff = G_FW_PARAMS_PARAM_Z(val) << 16;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/*
* XXX: sheer laziness. We deliberately added 4 bytes of
* useless stuffing/comments at the end of the config file so
* it's ok to simply throw away the last remaining bytes when
* the config file is not an exact multiple of 4. This also
* helps with the validate_mt_off_len check.
*/
if (cfg != NULL) {
cflen = cfg->datasize & ~3;
cfdata = cfg->data;
} else {
cflen = default_cfg->datasize & ~3;
cfdata = default_cfg->data;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (cflen > FLASH_CFG_MAX_SIZE) {
device_printf(sc->dev,
"config file too long (%d, max allowed is %d). "
"Will try to use the config on the card, if any.\n",
cflen, FLASH_CFG_MAX_SIZE);
goto use_config_on_flash;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
if (rc != 0) {
device_printf(sc->dev,
"%s: addr (%d/0x%x) or len %d is not valid: %d. "
"Will try to use the config on the card, if any.\n",
__func__, mtype, moff, cflen, rc);
goto use_config_on_flash;
}
memwin_info(sc, 2, &mw_base, &mw_aperture);
while (cflen) {
off = position_memwin(sc, 2, addr);
n = min(cflen, mw_aperture - off);
for (i = 0; i < n; i += 4)
t4_write_reg(sc, mw_base + off + i, *cfdata++);
cflen -= n;
addr += n;
}
} else {
use_config_on_flash:
MFC r256459. cxgbe(4): Update T4 and T5 firmwares to 1.9.12.0 Approved by: re (glebius)
2013-10-20 15:24:44 +00:00
mtype = FW_MEMTYPE_FLASH;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
moff = t4_flash_cfg_addr(sc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
bzero(&caps, sizeof(caps));
caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
F_FW_CMD_REQUEST | F_FW_CMD_READ);
caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) | FW_LEN16(caps));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc != 0) {
device_printf(sc->dev,
- Explain clearly why a different firmware is being installed (if/when it is being installed). Improve other error messages while here. - Select special FPGA specific configuration profile when appropriate. MFC after: 3 days
2013-04-11 19:39:40 +00:00
"failed to pre-process config file: %d "
"(mtype %d, moff 0x%x).\n", rc, mtype, moff);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
goto done;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
finicsum = be32toh(caps.finicsum);
cfcsum = be32toh(caps.cfcsum);
if (finicsum != cfcsum) {
device_printf(sc->dev,
"WARNING: config file checksum mismatch: %08x %08x\n",
finicsum, cfcsum);
}
sc->cfcsum = cfcsum;
#define LIMIT_CAPS(x) do { \
caps.x &= htobe16(t4_##x##_allowed); \
} while (0)
/*
* Let the firmware know what features will (not) be used so it can tune
* things accordingly.
*/
LIMIT_CAPS(linkcaps);
LIMIT_CAPS(niccaps);
LIMIT_CAPS(toecaps);
LIMIT_CAPS(rdmacaps);
LIMIT_CAPS(iscsicaps);
LIMIT_CAPS(fcoecaps);
#undef LIMIT_CAPS
caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
if (rc != 0) {
device_printf(sc->dev,
"failed to process config file: %d.\n", rc);
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
done:
if (cfg != NULL)
firmware_put(cfg, FIRMWARE_UNLOAD);
return (rc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
/*
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
* Retrieve parameters that are needed (or nice to have) very early.
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
*/
static int
get_params__pre_init(struct adapter *sc)
{
int rc;
uint32_t param[2], val[2];
struct fw_devlog_cmd cmd;
struct devlog_params *dlog = &sc->params.devlog;
param[0] = FW_PARAM_DEV(PORTVEC);
param[1] = FW_PARAM_DEV(CCLK);
rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
if (rc != 0) {
device_printf(sc->dev,
"failed to query parameters (pre_init): %d.\n", rc);
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
sc->params.portvec = val[0];
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
sc->params.nports = bitcount32(val[0]);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
sc->params.vpd.cclk = val[1];
/* Read device log parameters. */
bzero(&cmd, sizeof(cmd));
cmd.op_to_write = htobe32(V_FW_CMD_OP(FW_DEVLOG_CMD) |
F_FW_CMD_REQUEST | F_FW_CMD_READ);
cmd.retval_len16 = htobe32(FW_LEN16(cmd));
rc = -t4_wr_mbox(sc, sc->mbox, &cmd, sizeof(cmd), &cmd);
if (rc != 0) {
device_printf(sc->dev,
"failed to get devlog parameters: %d.\n", rc);
bzero(dlog, sizeof (*dlog));
rc = 0; /* devlog isn't critical for device operation */
} else {
val[0] = be32toh(cmd.memtype_devlog_memaddr16_devlog);
dlog->memtype = G_FW_DEVLOG_CMD_MEMTYPE_DEVLOG(val[0]);
dlog->start = G_FW_DEVLOG_CMD_MEMADDR16_DEVLOG(val[0]) << 4;
dlog->size = be32toh(cmd.memsize_devlog);
}
return (rc);
}
/*
* Retrieve various parameters that are of interest to the driver. The device
* has been initialized by the firmware at this point.
*/
static int
get_params__post_init(struct adapter *sc)
{
int rc;
uint32_t param[7], val[7];
struct fw_caps_config_cmd caps;
param[0] = FW_PARAM_PFVF(IQFLINT_START);
param[1] = FW_PARAM_PFVF(EQ_START);
param[2] = FW_PARAM_PFVF(FILTER_START);
param[3] = FW_PARAM_PFVF(FILTER_END);
cxgbe(4): Updates to the hardware L2 table management code. - Add full support for IPv6 addresses. - Read the size of the L2 table during attach. Do not assume that PCIe physical function 4 of the card has all of the table to itself. - Use FNV instead of Jenkins to hash L3 addresses and drop the private copy of jhash.h from the driver. MFC after: 1 week
2013-01-14 20:36:22 +00:00
param[4] = FW_PARAM_PFVF(L2T_START);
param[5] = FW_PARAM_PFVF(L2T_END);
rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (rc != 0) {
device_printf(sc->dev,
"failed to query parameters (post_init): %d.\n", rc);
return (rc);
}
sc->sge.iq_start = val[0];
sc->sge.eq_start = val[1];
sc->tids.ftid_base = val[2];
sc->tids.nftids = val[3] - val[2] + 1;
MFC r259382: Read card capabilities after firmware initialization, instead of setting them up as part of firmware initialization (which the driver gets to do only if it's the master driver). Read the range of tids available for the ETHOFLD functionality if it's enabled. New is_ftid() and is_etid() functions to test whether a tid falls within the range of filter tids or ETHOFLD tids respectively.
2014-05-06 07:21:50 +00:00
sc->params.ftid_min = val[2];
sc->params.ftid_max = val[3];
cxgbe(4): Updates to the hardware L2 table management code. - Add full support for IPv6 addresses. - Read the size of the L2 table during attach. Do not assume that PCIe physical function 4 of the card has all of the table to itself. - Use FNV instead of Jenkins to hash L3 addresses and drop the private copy of jhash.h from the driver. MFC after: 1 week
2013-01-14 20:36:22 +00:00
sc->vres.l2t.start = val[4];
sc->vres.l2t.size = val[5] - val[4] + 1;
KASSERT(sc->vres.l2t.size <= L2T_SIZE,
("%s: L2 table size (%u) larger than expected (%u)",
__func__, sc->vres.l2t.size, L2T_SIZE));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/* get capabilites */
bzero(&caps, sizeof(caps));
caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
F_FW_CMD_REQUEST | F_FW_CMD_READ);
caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
if (rc != 0) {
device_printf(sc->dev,
"failed to get card capabilities: %d.\n", rc);
return (rc);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r259382: Read card capabilities after firmware initialization, instead of setting them up as part of firmware initialization (which the driver gets to do only if it's the master driver). Read the range of tids available for the ETHOFLD functionality if it's enabled. New is_ftid() and is_etid() functions to test whether a tid falls within the range of filter tids or ETHOFLD tids respectively.
2014-05-06 07:21:50 +00:00
#define READ_CAPS(x) do { \
sc->x = htobe16(caps.x); \
} while (0)
READ_CAPS(linkcaps);
READ_CAPS(niccaps);
READ_CAPS(toecaps);
READ_CAPS(rdmacaps);
READ_CAPS(iscsicaps);
READ_CAPS(fcoecaps);
if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
param[0] = FW_PARAM_PFVF(ETHOFLD_START);
param[1] = FW_PARAM_PFVF(ETHOFLD_END);
param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
if (rc != 0) {
device_printf(sc->dev,
"failed to query NIC parameters: %d.\n", rc);
return (rc);
}
sc->tids.etid_base = val[0];
sc->params.etid_min = val[0];
sc->tids.netids = val[1] - val[0] + 1;
sc->params.netids = sc->tids.netids;
sc->params.eo_wr_cred = val[2];
sc->params.ethoffload = 1;
}
if (sc->toecaps) {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/* query offload-related parameters */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
param[0] = FW_PARAM_DEV(NTID);
param[1] = FW_PARAM_PFVF(SERVER_START);
param[2] = FW_PARAM_PFVF(SERVER_END);
param[3] = FW_PARAM_PFVF(TDDP_START);
param[4] = FW_PARAM_PFVF(TDDP_END);
param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc != 0) {
device_printf(sc->dev,
"failed to query TOE parameters: %d.\n", rc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
sc->tids.ntids = val[0];
sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
sc->tids.stid_base = val[1];
sc->tids.nstids = val[2] - val[1] + 1;
sc->vres.ddp.start = val[3];
sc->vres.ddp.size = val[4] - val[3] + 1;
sc->params.ofldq_wr_cred = val[5];
sc->params.offload = 1;
}
MFC r259382: Read card capabilities after firmware initialization, instead of setting them up as part of firmware initialization (which the driver gets to do only if it's the master driver). Read the range of tids available for the ETHOFLD functionality if it's enabled. New is_ftid() and is_etid() functions to test whether a tid falls within the range of filter tids or ETHOFLD tids respectively.
2014-05-06 07:21:50 +00:00
if (sc->rdmacaps) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
param[0] = FW_PARAM_PFVF(STAG_START);
param[1] = FW_PARAM_PFVF(STAG_END);
param[2] = FW_PARAM_PFVF(RQ_START);
param[3] = FW_PARAM_PFVF(RQ_END);
param[4] = FW_PARAM_PFVF(PBL_START);
param[5] = FW_PARAM_PFVF(PBL_END);
rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc != 0) {
device_printf(sc->dev,
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
"failed to query RDMA parameters(1): %d.\n", rc);
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
sc->vres.stag.start = val[0];
sc->vres.stag.size = val[1] - val[0] + 1;
sc->vres.rq.start = val[2];
sc->vres.rq.size = val[3] - val[2] + 1;
sc->vres.pbl.start = val[4];
sc->vres.pbl.size = val[5] - val[4] + 1;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
param[0] = FW_PARAM_PFVF(SQRQ_START);
param[1] = FW_PARAM_PFVF(SQRQ_END);
param[2] = FW_PARAM_PFVF(CQ_START);
param[3] = FW_PARAM_PFVF(CQ_END);
param[4] = FW_PARAM_PFVF(OCQ_START);
param[5] = FW_PARAM_PFVF(OCQ_END);
Use correct mailbox and PCIe PF number when querying RDMA parameters.
2013-08-26 19:02:52 +00:00
rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (rc != 0) {
device_printf(sc->dev,
"failed to query RDMA parameters(2): %d.\n", rc);
return (rc);
}
sc->vres.qp.start = val[0];
sc->vres.qp.size = val[1] - val[0] + 1;
sc->vres.cq.start = val[2];
sc->vres.cq.size = val[3] - val[2] + 1;
sc->vres.ocq.start = val[4];
sc->vres.ocq.size = val[5] - val[4] + 1;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
MFC r259382: Read card capabilities after firmware initialization, instead of setting them up as part of firmware initialization (which the driver gets to do only if it's the master driver). Read the range of tids available for the ETHOFLD functionality if it's enabled. New is_ftid() and is_etid() functions to test whether a tid falls within the range of filter tids or ETHOFLD tids respectively.
2014-05-06 07:21:50 +00:00
if (sc->iscsicaps) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
param[0] = FW_PARAM_PFVF(ISCSI_START);
param[1] = FW_PARAM_PFVF(ISCSI_END);
rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc != 0) {
device_printf(sc->dev,
"failed to query iSCSI parameters: %d.\n", rc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
sc->vres.iscsi.start = val[0];
sc->vres.iscsi.size = val[1] - val[0] + 1;
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
/*
* We've got the params we wanted to query via the firmware. Now grab
* some others directly from the chip.
*/
rc = t4_read_chip_settings(sc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (rc);
}
cxgbe(4): Ask the card's firmware to pad up tiny CPLs by encapsulating them in a firmware message if it is able to do so. This works out better for one of the FIFOs in the chip. MFC after: 5 days
2013-02-26 00:27:27 +00:00
static int
set_params__post_init(struct adapter *sc)
{
uint32_t param, val;
There is no need for elaborate queries and error checking when trying to set FW4MSG_ENCAP. MFC after: 3 days
2013-04-11 21:15:35 +00:00
/* ask for encapsulated CPLs */
cxgbe(4): Ask the card's firmware to pad up tiny CPLs by encapsulating them in a firmware message if it is able to do so. This works out better for one of the FIFOs in the chip. MFC after: 5 days
2013-02-26 00:27:27 +00:00
param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
There is no need for elaborate queries and error checking when trying to set FW4MSG_ENCAP. MFC after: 3 days
2013-04-11 21:15:35 +00:00
val = 1;
(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
cxgbe(4): Ask the card's firmware to pad up tiny CPLs by encapsulating them in a firmware message if it is able to do so. This works out better for one of the FIFOs in the chip. MFC after: 5 days
2013-02-26 00:27:27 +00:00
There is no need for elaborate queries and error checking when trying to set FW4MSG_ENCAP. MFC after: 3 days
2013-04-11 21:15:35 +00:00
return (0);
cxgbe(4): Ask the card's firmware to pad up tiny CPLs by encapsulating them in a firmware message if it is able to do so. This works out better for one of the FIFOs in the chip. MFC after: 5 days
2013-02-26 00:27:27 +00:00
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#undef FW_PARAM_PFVF
#undef FW_PARAM_DEV
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static void
t4_set_desc(struct adapter *sc)
{
char buf[128];
struct adapter_params *p = &sc->params;
Display P/N information in the description. Submitted by: gnn MFC after: 3 days
2013-08-20 18:22:04 +00:00
snprintf(buf, sizeof(buf), "Chelsio %s %sNIC (rev %d), S/N:%s, "
"P/N:%s, E/C:%s", p->vpd.id, is_offload(sc) ? "R" : "",
chip_rev(sc), p->vpd.sn, p->vpd.pn, p->vpd.ec);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
device_set_desc_copy(sc->dev, buf);
}
static void
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
build_medialist(struct port_info *pi, struct ifmedia *media)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
{
int data, m;
PORT_LOCK(pi);
ifmedia_removeall(media);
m = IFM_ETHER | IFM_FDX;
data = (pi->port_type << 8) | pi->mod_type;
switch(pi->port_type) {
case FW_PORT_TYPE_BT_XFI:
case FW_PORT_TYPE_BT_XAUI:
ifmedia_add(media, m | IFM_10G_T, data, NULL);
/* fall through */
case FW_PORT_TYPE_BT_SGMII:
ifmedia_add(media, m | IFM_1000_T, data, NULL);
ifmedia_add(media, m | IFM_100_TX, data, NULL);
ifmedia_add(media, IFM_ETHER | IFM_AUTO, data, NULL);
ifmedia_set(media, IFM_ETHER | IFM_AUTO);
break;
case FW_PORT_TYPE_CX4:
ifmedia_add(media, m | IFM_10G_CX4, data, NULL);
ifmedia_set(media, m | IFM_10G_CX4);
break;
MFC r261533, r261536, r261537, and r263457. r261533: cxgbe(4): Use the port's tx channel to identify it to t4_clr_port_stats. r261536: cxgbe(4): The T5 allows for a different freelist starvation threshold for queues with buffer packing. Use the correct value to calculate a freelist's low water mark. r261537: cxgbe(4): Use the rx channel map (instead of the tx channel map) as the congestion channel map. r263457: cxgbe(4): Recognize the "spider" configuration where a T5 card's 40G QSFP port is presented as 4 distinct 10G SFP+ ports to the driver.
2014-05-06 02:22:52 +00:00
case FW_PORT_TYPE_QSFP_10G:
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
case FW_PORT_TYPE_SFP:
case FW_PORT_TYPE_FIBER_XFI:
case FW_PORT_TYPE_FIBER_XAUI:
switch (pi->mod_type) {
case FW_PORT_MOD_TYPE_LR:
ifmedia_add(media, m | IFM_10G_LR, data, NULL);
ifmedia_set(media, m | IFM_10G_LR);
break;
case FW_PORT_MOD_TYPE_SR:
ifmedia_add(media, m | IFM_10G_SR, data, NULL);
ifmedia_set(media, m | IFM_10G_SR);
break;
case FW_PORT_MOD_TYPE_LRM:
ifmedia_add(media, m | IFM_10G_LRM, data, NULL);
ifmedia_set(media, m | IFM_10G_LRM);
break;
case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
ifmedia_add(media, m | IFM_10G_TWINAX, data, NULL);
ifmedia_set(media, m | IFM_10G_TWINAX);
break;
case FW_PORT_MOD_TYPE_NONE:
m &= ~IFM_FDX;
ifmedia_add(media, m | IFM_NONE, data, NULL);
ifmedia_set(media, m | IFM_NONE);
break;
case FW_PORT_MOD_TYPE_NA:
case FW_PORT_MOD_TYPE_ER:
default:
- Provide accurate ifmedia information so that 40G ports/transceivers are displayed properly in ifconfig, etc. - Use the same number of tx and rx queues for a 40G port as for a 10G port. MFC after: 1 week
2013-04-30 05:51:52 +00:00
device_printf(pi->dev,
"unknown port_type (%d), mod_type (%d)\n",
pi->port_type, pi->mod_type);
ifmedia_add(media, m | IFM_UNKNOWN, data, NULL);
ifmedia_set(media, m | IFM_UNKNOWN);
break;
}
break;
case FW_PORT_TYPE_QSFP:
switch (pi->mod_type) {
case FW_PORT_MOD_TYPE_LR:
ifmedia_add(media, m | IFM_40G_LR4, data, NULL);
ifmedia_set(media, m | IFM_40G_LR4);
break;
case FW_PORT_MOD_TYPE_SR:
ifmedia_add(media, m | IFM_40G_SR4, data, NULL);
ifmedia_set(media, m | IFM_40G_SR4);
break;
Deal correctly with 40G ports that don't have any transceiver plugged in. Do not claim that they have unknown tranceivers. MFC after: 3 days
2013-05-13 20:00:03 +00:00
- Provide accurate ifmedia information so that 40G ports/transceivers are displayed properly in ifconfig, etc. - Use the same number of tx and rx queues for a 40G port as for a 10G port. MFC after: 1 week
2013-04-30 05:51:52 +00:00
case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
ifmedia_add(media, m | IFM_40G_CR4, data, NULL);
ifmedia_set(media, m | IFM_40G_CR4);
break;
Deal correctly with 40G ports that don't have any transceiver plugged in. Do not claim that they have unknown tranceivers. MFC after: 3 days
2013-05-13 20:00:03 +00:00
case FW_PORT_MOD_TYPE_NONE:
m &= ~IFM_FDX;
ifmedia_add(media, m | IFM_NONE, data, NULL);
ifmedia_set(media, m | IFM_NONE);
break;
- Provide accurate ifmedia information so that 40G ports/transceivers are displayed properly in ifconfig, etc. - Use the same number of tx and rx queues for a 40G port as for a 10G port. MFC after: 1 week
2013-04-30 05:51:52 +00:00
default:
device_printf(pi->dev,
"unknown port_type (%d), mod_type (%d)\n",
pi->port_type, pi->mod_type);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ifmedia_add(media, m | IFM_UNKNOWN, data, NULL);
ifmedia_set(media, m | IFM_UNKNOWN);
break;
}
break;
default:
- Provide accurate ifmedia information so that 40G ports/transceivers are displayed properly in ifconfig, etc. - Use the same number of tx and rx queues for a 40G port as for a 10G port. MFC after: 1 week
2013-04-30 05:51:52 +00:00
device_printf(pi->dev,
"unknown port_type (%d), mod_type (%d)\n", pi->port_type,
pi->mod_type);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ifmedia_add(media, m | IFM_UNKNOWN, data, NULL);
ifmedia_set(media, m | IFM_UNKNOWN);
break;
}
PORT_UNLOCK(pi);
}
Program the MAC exact match table in batches of 7 addresses at a time when possible. This is more efficient than one at a time. Submitted by: gnn MFC after: 3 days
2012-02-08 00:36:36 +00:00
#define FW_MAC_EXACT_CHUNK 7
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
* Program the port's XGMAC based on parameters in ifnet. The caller also
* indicates which parameters should be programmed (the rest are left alone).
*/
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
int
update_mac_settings(struct ifnet *ifp, int flags)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
{
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
int rc = 0;
struct port_info *pi = ifp->if_softc;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
struct adapter *sc = pi->adapter;
int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
uint16_t viid = 0xffff;
int16_t *xact_addr_filt = NULL;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
ASSERT_SYNCHRONIZED_OP(sc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
KASSERT(flags, ("%s: not told what to update.", __func__));
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
if (ifp == pi->ifp) {
viid = pi->viid;
xact_addr_filt = &pi->xact_addr_filt;
}
#ifdef DEV_NETMAP
else if (ifp == pi->nm_ifp) {
viid = pi->nm_viid;
xact_addr_filt = &pi->nm_xact_addr_filt;
}
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (flags & XGMAC_MTU)
mtu = ifp->if_mtu;
if (flags & XGMAC_PROMISC)
promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
if (flags & XGMAC_ALLMULTI)
allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
if (flags & XGMAC_VLANEX)
vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
rc = -t4_set_rxmode(sc, sc->mbox, viid, mtu, promisc, allmulti,
1, vlanex, false);
if (rc) {
if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
rc);
return (rc);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
if (flags & XGMAC_UCADDR) {
uint8_t ucaddr[ETHER_ADDR_LEN];
bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
rc = t4_change_mac(sc, sc->mbox, viid, *xact_addr_filt, ucaddr,
true, true);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc < 0) {
rc = -rc;
if_printf(ifp, "change_mac failed: %d\n", rc);
return (rc);
} else {
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
*xact_addr_filt = rc;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
rc = 0;
}
}
if (flags & XGMAC_MCADDRS) {
Program the MAC exact match table in batches of 7 addresses at a time when possible. This is more efficient than one at a time. Submitted by: gnn MFC after: 3 days
2012-02-08 00:36:36 +00:00
const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
int del = 1;
uint64_t hash = 0;
struct ifmultiaddr *ifma;
Program the MAC exact match table in batches of 7 addresses at a time when possible. This is more efficient than one at a time. Submitted by: gnn MFC after: 3 days
2012-02-08 00:36:36 +00:00
int i = 0, j;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if_maddr_rlock(ifp);
TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
Fix inverted test that resulted in incorrect multicast hw programming.
2012-07-03 06:56:11 +00:00
if (ifma->ifma_addr->sa_family != AF_LINK)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
continue;
MFC r272051: cxgbe(4): Verify that the addresses in if_multiaddrs really are multicast addresses. (The chip doesn't really care, it's just that it needs to be told explicitly if unicast DMACs are checked for "hits" in the hash that is used after the TCAM entries are all used up).
2015-05-03 08:17:37 +00:00
mcaddr[i] =
Program the MAC exact match table in batches of 7 addresses at a time when possible. This is more efficient than one at a time. Submitted by: gnn MFC after: 3 days
2012-02-08 00:36:36 +00:00
LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
MFC r272051: cxgbe(4): Verify that the addresses in if_multiaddrs really are multicast addresses. (The chip doesn't really care, it's just that it needs to be told explicitly if unicast DMACs are checked for "hits" in the hash that is used after the TCAM entries are all used up).
2015-05-03 08:17:37 +00:00
MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
i++;
Program the MAC exact match table in batches of 7 addresses at a time when possible. This is more efficient than one at a time. Submitted by: gnn MFC after: 3 days
2012-02-08 00:36:36 +00:00
if (i == FW_MAC_EXACT_CHUNK) {
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
rc = t4_alloc_mac_filt(sc, sc->mbox, viid, del,
i, mcaddr, NULL, &hash, 0);
Program the MAC exact match table in batches of 7 addresses at a time when possible. This is more efficient than one at a time. Submitted by: gnn MFC after: 3 days
2012-02-08 00:36:36 +00:00
if (rc < 0) {
rc = -rc;
for (j = 0; j < i; j++) {
if_printf(ifp,
"failed to add mc address"
" %02x:%02x:%02x:"
"%02x:%02x:%02x rc=%d\n",
mcaddr[j][0], mcaddr[j][1],
mcaddr[j][2], mcaddr[j][3],
mcaddr[j][4], mcaddr[j][5],
rc);
}
goto mcfail;
}
del = 0;
i = 0;
}
}
if (i > 0) {
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
rc = t4_alloc_mac_filt(sc, sc->mbox, viid, del, i,
mcaddr, NULL, &hash, 0);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc < 0) {
rc = -rc;
Program the MAC exact match table in batches of 7 addresses at a time when possible. This is more efficient than one at a time. Submitted by: gnn MFC after: 3 days
2012-02-08 00:36:36 +00:00
for (j = 0; j < i; j++) {
if_printf(ifp,
"failed to add mc address"
" %02x:%02x:%02x:"
"%02x:%02x:%02x rc=%d\n",
mcaddr[j][0], mcaddr[j][1],
mcaddr[j][2], mcaddr[j][3],
mcaddr[j][4], mcaddr[j][5],
rc);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
goto mcfail;
}
}
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
rc = -t4_set_addr_hash(sc, sc->mbox, viid, 0, hash, 0);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc != 0)
if_printf(ifp, "failed to set mc address hash: %d", rc);
mcfail:
if_maddr_runlock(ifp);
}
return (rc);
}
MFC r278371: cxgbe(4): a change to the synchronization rules within the the driver. This is purely cosmetic because the new rules are already followed.
2015-04-08 01:07:51 +00:00
/*
* {begin|end}_synchronized_op must be called from the same thread.
*/
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
int
begin_synchronized_op(struct adapter *sc, struct port_info *pi, int flags,
char *wmesg)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
{
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
int rc, pri;
#ifdef WITNESS
/* the caller thinks it's ok to sleep, but is it really? */
if (flags & SLEEP_OK)
pause("t4slptst", 1);
#endif
if (INTR_OK)
pri = PCATCH;
else
pri = 0;
ADAPTER_LOCK(sc);
for (;;) {
if (pi && IS_DOOMED(pi)) {
rc = ENXIO;
goto done;
}
if (!IS_BUSY(sc)) {
rc = 0;
break;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (!(flags & SLEEP_OK)) {
rc = EBUSY;
goto done;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
rc = EINTR;
goto done;
}
}
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
SET_BUSY(sc);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
#ifdef INVARIANTS
sc->last_op = wmesg;
sc->last_op_thr = curthread;
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
done:
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (!(flags & HOLD_LOCK) || rc)
ADAPTER_UNLOCK(sc);
return (rc);
}
MFC r278371: cxgbe(4): a change to the synchronization rules within the the driver. This is purely cosmetic because the new rules are already followed.
2015-04-08 01:07:51 +00:00
/*
* {begin|end}_synchronized_op must be called from the same thread.
*/
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
void
end_synchronized_op(struct adapter *sc, int flags)
{
if (flags & LOCK_HELD)
ADAPTER_LOCK_ASSERT_OWNED(sc);
else
ADAPTER_LOCK(sc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
CLR_BUSY(sc);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
wakeup(&sc->flags);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ADAPTER_UNLOCK(sc);
}
static int
cxgbe_init_synchronized(struct port_info *pi)
{
struct adapter *sc = pi->adapter;
struct ifnet *ifp = pi->ifp;
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
int rc = 0, i;
struct sge_txq *txq;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
ASSERT_SYNCHRONIZED_OP(sc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (isset(&sc->open_device_map, pi->port_id)) {
KASSERT(ifp->if_drv_flags & IFF_DRV_RUNNING,
("mismatch between open_device_map and if_drv_flags"));
return (0); /* already running */
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (!(sc->flags & FULL_INIT_DONE) &&
((rc = adapter_full_init(sc)) != 0))
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (rc); /* error message displayed already */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (!(pi->flags & PORT_INIT_DONE) &&
((rc = port_full_init(pi)) != 0))
return (rc); /* error message displayed already */
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
rc = update_mac_settings(ifp, XGMAC_ALL);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (rc)
goto done; /* error message displayed already */
rc = -t4_enable_vi(sc, sc->mbox, pi->viid, true, true);
if (rc != 0) {
if_printf(ifp, "enable_vi failed: %d\n", rc);
goto done;
}
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
/*
* Can't fail from this point onwards. Review cxgbe_uninit_synchronized
* if this changes.
*/
for_each_txq(pi, i, txq) {
TXQ_LOCK(txq);
txq->eq.flags |= EQ_ENABLED;
TXQ_UNLOCK(txq);
}
Add support for packet-sniffing tracers to cxgbe(4). This works with all T4 and T5 based cards and is useful for analyzing TSO, LRO, TOE, and for general purpose monitoring without tapping any cxgbe or cxl ifnet directly. Tracers on the T4/T5 chips provide access to Ethernet frames exactly as they were received from or transmitted on the wire. On transmit, a tracer will capture a frame after TSO segmentation, hw VLAN tag insertion, hw L3 & L4 checksum insertion, etc. It will also capture frames generated by the TCP offload engine (TOE traffic is normally invisible to the kernel). On receive, a tracer will capture a frame before hw VLAN extraction, runt filtering, other badness filtering, before the steering/drop/L2-rewrite filters or the TOE have had a go at it, and of course before sw LRO in the driver. There are 4 tracers on a chip. A tracer can trace only in one direction (tx or rx). For now cxgbetool will set up tracers to capture the first 128B of every transmitted or received frame on a given port. This is a small subset of what the hardware can do. A pseudo ifnet with the same name as the nexus driver (t4nex0 or t5nex0) will be created for tracing. The data delivered to this ifnet is an additional copy made inside the chip. Normal delivery to cxgbe<n> or cxl<n> will be made as usual. /* watch cxl0, which is the first port hanging off t5nex0. */ # cxgbetool t5nex0 tracer 0 tx0 (watch what cxl0 is transmitting) # cxgbetool t5nex0 tracer 1 rx0 (watch what cxl0 is receiving) # cxgbetool t5nex0 tracer list # tcpdump -i t5nex0 <== all that cxl0 sees and puts on the wire If you were doing TSO, a tcpdump on cxl0 may have shown you ~64K "frames" with no L3/L4 checksum but this will show you the frames that were actually transmitted. /* all done */ # cxgbetool t5nex0 tracer 0 disable # cxgbetool t5nex0 tracer 1 disable # cxgbetool t5nex0 tracer list # ifconfig t5nex0 destroy
2013-07-26 22:04:11 +00:00
/*
* The first iq of the first port to come up is used for tracing.
*/
if (sc->traceq < 0) {
sc->traceq = sc->sge.rxq[pi->first_rxq].iq.abs_id;
t4_write_reg(sc, is_t4(sc) ? A_MPS_TRC_RSS_CONTROL :
A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
V_QUEUENUMBER(sc->traceq));
pi->flags |= HAS_TRACEQ;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/* all ok */
setbit(&sc->open_device_map, pi->port_id);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
PORT_LOCK(pi);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ifp->if_drv_flags |= IFF_DRV_RUNNING;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
PORT_UNLOCK(pi);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
callout_reset(&pi->tick, hz, cxgbe_tick, pi);
done:
if (rc != 0)
cxgbe_uninit_synchronized(pi);
return (rc);
}
/*
* Idempotent.
*/
static int
cxgbe_uninit_synchronized(struct port_info *pi)
{
struct adapter *sc = pi->adapter;
struct ifnet *ifp = pi->ifp;
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
int rc, i;
struct sge_txq *txq;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
ASSERT_SYNCHRONIZED_OP(sc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
* Disable the VI so that all its data in either direction is discarded
* by the MPS. Leave everything else (the queues, interrupts, and 1Hz
* tick) intact as the TP can deliver negative advice or data that it's
* holding in its RAM (for an offloaded connection) even after the VI is
* disabled.
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
*/
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
rc = -t4_enable_vi(sc, sc->mbox, pi->viid, false, false);
if (rc) {
if_printf(ifp, "disable_vi failed: %d\n", rc);
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
for_each_txq(pi, i, txq) {
TXQ_LOCK(txq);
txq->eq.flags &= ~EQ_ENABLED;
TXQ_UNLOCK(txq);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
clrbit(&sc->open_device_map, pi->port_id);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
PORT_LOCK(pi);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
PORT_UNLOCK(pi);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
pi->link_cfg.link_ok = 0;
pi->link_cfg.speed = 0;
- Show the reason why link is down if this information is available. - Display the temperature and PHY firmware version of the BT PHY. MFC after: 1 day
2013-07-05 01:53:51 +00:00
pi->linkdnrc = -1;
t4_os_link_changed(sc, pi->port_id, 0, -1);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (0);
}
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
/*
* It is ok for this function to fail midway and return right away. t4_detach
* will walk the entire sc->irq list and clean up whatever is valid.
*/
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
setup_intr_handlers(struct adapter *sc)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
{
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
int rc, rid, p, q;
- Specialized ingress queues that take interrupts for other ingress queues. Try to have a set of these per port when possible, fall back to sharing a common pool between all ports otherwise. - One control queue per port (used to be one per hardware channel). - t4_eth_rx now handles Ethernet rx only. - sysctls to display pidx/cidx for some queues. MFC after: 1 week
2011-05-30 21:34:44 +00:00
char s[8];
struct irq *irq;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
struct port_info *pi;
struct sge_rxq *rxq;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
struct sge_ofld_rxq *ofld_rxq;
#endif
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
struct sge_nm_rxq *nm_rxq;
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
* Setup interrupts.
*/
- Specialized ingress queues that take interrupts for other ingress queues. Try to have a set of these per port when possible, fall back to sharing a common pool between all ports otherwise. - One control queue per port (used to be one per hardware channel). - t4_eth_rx now handles Ethernet rx only. - sysctls to display pidx/cidx for some queues. MFC after: 1 week
2011-05-30 21:34:44 +00:00
irq = &sc->irq[0];
rid = sc->intr_type == INTR_INTX ? 0 : 1;
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
if (sc->intr_count == 1)
return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
- Specialized ingress queues that take interrupts for other ingress queues. Try to have a set of these per port when possible, fall back to sharing a common pool between all ports otherwise. - One control queue per port (used to be one per hardware channel). - t4_eth_rx now handles Ethernet rx only. - sysctls to display pidx/cidx for some queues. MFC after: 1 week
2011-05-30 21:34:44 +00:00
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
/* Multiple interrupts. */
KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
("%s: too few intr.", __func__));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
/* The first one is always error intr */
rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
if (rc != 0)
return (rc);
irq++;
rid++;
- Specialized ingress queues that take interrupts for other ingress queues. Try to have a set of these per port when possible, fall back to sharing a common pool between all ports otherwise. - One control queue per port (used to be one per hardware channel). - t4_eth_rx now handles Ethernet rx only. - sysctls to display pidx/cidx for some queues. MFC after: 1 week
2011-05-30 21:34:44 +00:00
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
/* The second one is always the firmware event queue */
rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sc->sge.fwq, "evt");
if (rc != 0)
return (rc);
irq++;
rid++;
- Specialized ingress queues that take interrupts for other ingress queues. Try to have a set of these per port when possible, fall back to sharing a common pool between all ports otherwise. - One control queue per port (used to be one per hardware channel). - t4_eth_rx now handles Ethernet rx only. - sysctls to display pidx/cidx for some queues. MFC after: 1 week
2011-05-30 21:34:44 +00:00
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
for_each_port(sc, p) {
pi = sc->port[p];
- Specialized ingress queues that take interrupts for other ingress queues. Try to have a set of these per port when possible, fall back to sharing a common pool between all ports otherwise. - One control queue per port (used to be one per hardware channel). - t4_eth_rx now handles Ethernet rx only. - sysctls to display pidx/cidx for some queues. MFC after: 1 week
2011-05-30 21:34:44 +00:00
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
if (pi->flags & INTR_RXQ) {
for_each_rxq(pi, q, rxq) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
snprintf(s, sizeof(s), "%d.%d", p, q);
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
rc = t4_alloc_irq(sc, irq, rid, t4_intr, rxq,
s);
if (rc != 0)
return (rc);
- Specialized ingress queues that take interrupts for other ingress queues. Try to have a set of these per port when possible, fall back to sharing a common pool between all ports otherwise. - One control queue per port (used to be one per hardware channel). - t4_eth_rx now handles Ethernet rx only. - sysctls to display pidx/cidx for some queues. MFC after: 1 week
2011-05-30 21:34:44 +00:00
irq++;
rid++;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
if (pi->flags & INTR_OFLD_RXQ) {
for_each_ofld_rxq(pi, q, ofld_rxq) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
snprintf(s, sizeof(s), "%d,%d", p, q);
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
rc = t4_alloc_irq(sc, irq, rid, t4_intr,
ofld_rxq, s);
if (rc != 0)
return (rc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
irq++;
rid++;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
if (pi->flags & INTR_NM_RXQ) {
for_each_nm_rxq(pi, q, nm_rxq) {
snprintf(s, sizeof(s), "%d-%d", p, q);
rc = t4_alloc_irq(sc, irq, rid, t4_nm_intr,
nm_rxq, s);
if (rc != 0)
return (rc);
irq++;
rid++;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
MPASS(irq == &sc->irq[sc->intr_count]);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
return (0);
}
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
int
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
adapter_full_init(struct adapter *sc)
{
int rc, i;
MFC r278372: cxgbe(4): adapter_full_init is always a synchronized operation.
2015-04-08 01:09:36 +00:00
ASSERT_SYNCHRONIZED_OP(sc);
Install interrupt handlers early, during attach, for the reason explained in r239913 by jhb. MFC after: 1 week
2012-09-13 09:18:13 +00:00
ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
KASSERT((sc->flags & FULL_INIT_DONE) == 0,
("%s: FULL_INIT_DONE already", __func__));
/*
* queues that belong to the adapter (not any particular port).
*/
rc = t4_setup_adapter_queues(sc);
if (rc != 0)
goto done;
for (i = 0; i < nitems(sc->tq); i++) {
sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
taskqueue_thread_enqueue, &sc->tq[i]);
if (sc->tq[i] == NULL) {
device_printf(sc->dev,
"failed to allocate task queue %d\n", i);
rc = ENOMEM;
goto done;
}
taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
device_get_nameunit(sc->dev), i);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
t4_intr_enable(sc);
sc->flags |= FULL_INIT_DONE;
done:
if (rc != 0)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
adapter_full_uninit(sc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (rc);
}
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
int
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
adapter_full_uninit(struct adapter *sc)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
{
int i;
ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
- Move all Ethernet specific items from sge_eq to sge_txq. sge_eq is now a suitable base for all kinds of egress queues. - Add control queues (sge_ctrlq) and allocate one of these per hardware channel. They can be used to program filters and steer traffic (and more). MFC after: 1 week
2011-04-19 22:08:28 +00:00
t4_teardown_adapter_queues(sc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
taskqueue_free(sc->tq[i]);
sc->tq[i] = NULL;
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
sc->flags &= ~FULL_INIT_DONE;
return (0);
}
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
int
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
port_full_init(struct port_info *pi)
{
struct adapter *sc = pi->adapter;
struct ifnet *ifp = pi->ifp;
uint16_t *rss;
struct sge_rxq *rxq;
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
int rc, i, j;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
ASSERT_SYNCHRONIZED_OP(sc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
KASSERT((pi->flags & PORT_INIT_DONE) == 0,
("%s: PORT_INIT_DONE already", __func__));
sysctl_ctx_init(&pi->ctx);
pi->flags |= PORT_SYSCTL_CTX;
/*
* Allocate tx/rx/fl queues for this port.
*/
rc = t4_setup_port_queues(pi);
if (rc != 0)
goto done; /* error message displayed already */
/*
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
* Setup RSS for this port. Save a copy of the RSS table for later use.
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
*/
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
rss = malloc(pi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
for (i = 0; i < pi->rss_size;) {
for_each_rxq(pi, j, rxq) {
rss[i++] = rxq->iq.abs_id;
if (i == pi->rss_size)
break;
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
rc = -t4_config_rss_range(sc, sc->mbox, pi->viid, 0, pi->rss_size, rss,
pi->rss_size);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (rc != 0) {
if_printf(ifp, "rss_config failed: %d\n", rc);
goto done;
}
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
pi->rss = rss;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
pi->flags |= PORT_INIT_DONE;
done:
if (rc != 0)
port_full_uninit(pi);
return (rc);
}
/*
* Idempotent.
*/
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
int
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
port_full_uninit(struct port_info *pi)
{
struct adapter *sc = pi->adapter;
int i;
struct sge_rxq *rxq;
struct sge_txq *txq;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
struct sge_ofld_rxq *ofld_rxq;
struct sge_wrq *ofld_txq;
#endif
if (pi->flags & PORT_INIT_DONE) {
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
/* Need to quiesce queues. */
quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
for_each_txq(pi, i, txq) {
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
quiesce_txq(sc, txq);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
for_each_ofld_txq(pi, i, ofld_txq) {
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
quiesce_wrq(sc, ofld_txq);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
#endif
for_each_rxq(pi, i, rxq) {
quiesce_iq(sc, &rxq->iq);
quiesce_fl(sc, &rxq->fl);
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
for_each_ofld_rxq(pi, i, ofld_rxq) {
quiesce_iq(sc, &ofld_rxq->iq);
quiesce_fl(sc, &ofld_rxq->fl);
}
#endif
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
free(pi->rss, M_CXGBE);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
t4_teardown_port_queues(pi);
pi->flags &= ~PORT_INIT_DONE;
return (0);
}
static void
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
quiesce_txq(struct adapter *sc, struct sge_txq *txq)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
{
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
struct sge_eq *eq = &txq->eq;
struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
(void) sc; /* unused */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
#ifdef INVARIANTS
TXQ_LOCK(txq);
MPASS((eq->flags & EQ_ENABLED) == 0);
TXQ_UNLOCK(txq);
#endif
/* Wait for the mp_ring to empty. */
while (!mp_ring_is_idle(txq->r)) {
mp_ring_check_drainage(txq->r, 0);
pause("rquiesce", 1);
}
/* Then wait for the hardware to finish. */
while (spg->cidx != htobe16(eq->pidx))
pause("equiesce", 1);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
/* Finally, wait for the driver to reclaim all descriptors. */
while (eq->cidx != eq->pidx)
pause("dquiesce", 1);
}
static void
quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
{
/* XXXTX */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
static void
quiesce_iq(struct adapter *sc, struct sge_iq *iq)
{
(void) sc; /* unused */
/* Synchronize with the interrupt handler */
while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
pause("iqfree", 1);
}
static void
quiesce_fl(struct adapter *sc, struct sge_fl *fl)
{
mtx_lock(&sc->sfl_lock);
FL_LOCK(fl);
fl->flags |= FL_DOOMED;
FL_UNLOCK(fl);
mtx_unlock(&sc->sfl_lock);
callout_drain(&sc->sfl_callout);
KASSERT((fl->flags & FL_STARVING) == 0,
("%s: still starving", __func__));
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int
t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
driver_intr_t *handler, void *arg, char *name)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
{
int rc;
irq->rid = rid;
irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
RF_SHAREABLE | RF_ACTIVE);
if (irq->res == NULL) {
device_printf(sc->dev,
"failed to allocate IRQ for rid %d, name %s.\n", rid, name);
return (ENOMEM);
}
rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
NULL, handler, arg, &irq->tag);
if (rc != 0) {
device_printf(sc->dev,
"failed to setup interrupt for rid %d, name %s: %d\n",
rid, name, rc);
} else if (name)
bus_describe_intr(sc->dev, irq->res, irq->tag, name);
return (rc);
}
static int
t4_free_irq(struct adapter *sc, struct irq *irq)
{
if (irq->tag)
bus_teardown_intr(sc->dev, irq->res, irq->tag);
if (irq->res)
bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
bzero(irq, sizeof(*irq));
return (0);
}
static void
reg_block_dump(struct adapter *sc, uint8_t *buf, unsigned int start,
unsigned int end)
{
uint32_t *p = (uint32_t *)(buf + start);
for ( ; start <= end; start += sizeof(uint32_t))
*p++ = t4_read_reg(sc, start);
}
static void
t4_get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
{
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
int i, n;
const unsigned int *reg_ranges;
static const unsigned int t4_reg_ranges[] = {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
0x1008, 0x1108,
0x1180, 0x11b4,
0x11fc, 0x123c,
0x1300, 0x173c,
0x1800, 0x18fc,
0x3000, 0x30d8,
0x30e0, 0x5924,
0x5960, 0x59d4,
0x5a00, 0x5af8,
0x6000, 0x6098,
0x6100, 0x6150,
0x6200, 0x6208,
0x6240, 0x6248,
0x6280, 0x6338,
0x6370, 0x638c,
0x6400, 0x643c,
0x6500, 0x6524,
0x6a00, 0x6a38,
0x6a60, 0x6a78,
0x6b00, 0x6b84,
0x6bf0, 0x6c84,
0x6cf0, 0x6d84,
0x6df0, 0x6e84,
0x6ef0, 0x6f84,
0x6ff0, 0x7084,
0x70f0, 0x7184,
0x71f0, 0x7284,
0x72f0, 0x7384,
0x73f0, 0x7450,
0x7500, 0x7530,
0x7600, 0x761c,
0x7680, 0x76cc,
0x7700, 0x7798,
0x77c0, 0x77fc,
0x7900, 0x79fc,
0x7b00, 0x7c38,
0x7d00, 0x7efc,
0x8dc0, 0x8e1c,
0x8e30, 0x8e78,
0x8ea0, 0x8f6c,
0x8fc0, 0x9074,
0x90fc, 0x90fc,
0x9400, 0x9458,
0x9600, 0x96bc,
0x9800, 0x9808,
0x9820, 0x983c,
0x9850, 0x9864,
0x9c00, 0x9c6c,
0x9c80, 0x9cec,
0x9d00, 0x9d6c,
0x9d80, 0x9dec,
0x9e00, 0x9e6c,
0x9e80, 0x9eec,
0x9f00, 0x9f6c,
0x9f80, 0x9fec,
0xd004, 0xd03c,
0xdfc0, 0xdfe0,
0xe000, 0xea7c,
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
0xf000, 0x11110,
0x11118, 0x11190,
Do not read registers with read side effects while performing a register dump for cxgbetool.
2012-06-22 08:37:33 +00:00
0x19040, 0x1906c,
0x19078, 0x19080,
0x1908c, 0x19124,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
0x19150, 0x191b0,
0x191d0, 0x191e8,
0x19238, 0x1924c,
0x193f8, 0x19474,
0x19490, 0x194f8,
0x19800, 0x19f30,
0x1a000, 0x1a06c,
0x1a0b0, 0x1a120,
0x1a128, 0x1a138,
0x1a190, 0x1a1c4,
0x1a1fc, 0x1a1fc,
0x1e040, 0x1e04c,
Do not read registers with read side effects while performing a register dump for cxgbetool.
2012-06-22 08:37:33 +00:00
0x1e284, 0x1e28c,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
0x1e2c0, 0x1e2c0,
0x1e2e0, 0x1e2e0,
0x1e300, 0x1e384,
0x1e3c0, 0x1e3c8,
0x1e440, 0x1e44c,
Do not read registers with read side effects while performing a register dump for cxgbetool.
2012-06-22 08:37:33 +00:00
0x1e684, 0x1e68c,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
0x1e6c0, 0x1e6c0,
0x1e6e0, 0x1e6e0,
0x1e700, 0x1e784,
0x1e7c0, 0x1e7c8,
0x1e840, 0x1e84c,
Do not read registers with read side effects while performing a register dump for cxgbetool.
2012-06-22 08:37:33 +00:00
0x1ea84, 0x1ea8c,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
0x1eac0, 0x1eac0,
0x1eae0, 0x1eae0,
0x1eb00, 0x1eb84,
0x1ebc0, 0x1ebc8,
0x1ec40, 0x1ec4c,
Do not read registers with read side effects while performing a register dump for cxgbetool.
2012-06-22 08:37:33 +00:00
0x1ee84, 0x1ee8c,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
0x1eec0, 0x1eec0,
0x1eee0, 0x1eee0,
0x1ef00, 0x1ef84,
0x1efc0, 0x1efc8,
0x1f040, 0x1f04c,
Do not read registers with read side effects while performing a register dump for cxgbetool.
2012-06-22 08:37:33 +00:00
0x1f284, 0x1f28c,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
0x1f2c0, 0x1f2c0,
0x1f2e0, 0x1f2e0,
0x1f300, 0x1f384,
0x1f3c0, 0x1f3c8,
0x1f440, 0x1f44c,
Do not read registers with read side effects while performing a register dump for cxgbetool.
2012-06-22 08:37:33 +00:00
0x1f684, 0x1f68c,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
0x1f6c0, 0x1f6c0,
0x1f6e0, 0x1f6e0,
0x1f700, 0x1f784,
0x1f7c0, 0x1f7c8,
0x1f840, 0x1f84c,
Do not read registers with read side effects while performing a register dump for cxgbetool.
2012-06-22 08:37:33 +00:00
0x1fa84, 0x1fa8c,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
0x1fac0, 0x1fac0,
0x1fae0, 0x1fae0,
0x1fb00, 0x1fb84,
0x1fbc0, 0x1fbc8,
0x1fc40, 0x1fc4c,
Do not read registers with read side effects while performing a register dump for cxgbetool.
2012-06-22 08:37:33 +00:00
0x1fe84, 0x1fe8c,
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
0x1fec0, 0x1fec0,
0x1fee0, 0x1fee0,
0x1ff00, 0x1ff84,
0x1ffc0, 0x1ffc8,
0x20000, 0x2002c,
0x20100, 0x2013c,
0x20190, 0x201c8,
0x20200, 0x20318,
0x20400, 0x20528,
0x20540, 0x20614,
0x21000, 0x21040,
0x2104c, 0x21060,
0x210c0, 0x210ec,
0x21200, 0x21268,
0x21270, 0x21284,
0x212fc, 0x21388,
0x21400, 0x21404,
0x21500, 0x21518,
0x2152c, 0x2153c,
0x21550, 0x21554,
0x21600, 0x21600,
0x21608, 0x21628,
0x21630, 0x2163c,
0x21700, 0x2171c,
0x21780, 0x2178c,
0x21800, 0x21c38,
0x21c80, 0x21d7c,
0x21e00, 0x21e04,
0x22000, 0x2202c,
0x22100, 0x2213c,
0x22190, 0x221c8,
0x22200, 0x22318,
0x22400, 0x22528,
0x22540, 0x22614,
0x23000, 0x23040,
0x2304c, 0x23060,
0x230c0, 0x230ec,
0x23200, 0x23268,
0x23270, 0x23284,
0x232fc, 0x23388,
0x23400, 0x23404,
0x23500, 0x23518,
0x2352c, 0x2353c,
0x23550, 0x23554,
0x23600, 0x23600,
0x23608, 0x23628,
0x23630, 0x2363c,
0x23700, 0x2371c,
0x23780, 0x2378c,
0x23800, 0x23c38,
0x23c80, 0x23d7c,
0x23e00, 0x23e04,
0x24000, 0x2402c,
0x24100, 0x2413c,
0x24190, 0x241c8,
0x24200, 0x24318,
0x24400, 0x24528,
0x24540, 0x24614,
0x25000, 0x25040,
0x2504c, 0x25060,
0x250c0, 0x250ec,
0x25200, 0x25268,
0x25270, 0x25284,
0x252fc, 0x25388,
0x25400, 0x25404,
0x25500, 0x25518,
0x2552c, 0x2553c,
0x25550, 0x25554,
0x25600, 0x25600,
0x25608, 0x25628,
0x25630, 0x2563c,
0x25700, 0x2571c,
0x25780, 0x2578c,
0x25800, 0x25c38,
0x25c80, 0x25d7c,
0x25e00, 0x25e04,
0x26000, 0x2602c,
0x26100, 0x2613c,
0x26190, 0x261c8,
0x26200, 0x26318,
0x26400, 0x26528,
0x26540, 0x26614,
0x27000, 0x27040,
0x2704c, 0x27060,
0x270c0, 0x270ec,
0x27200, 0x27268,
0x27270, 0x27284,
0x272fc, 0x27388,
0x27400, 0x27404,
0x27500, 0x27518,
0x2752c, 0x2753c,
0x27550, 0x27554,
0x27600, 0x27600,
0x27608, 0x27628,
0x27630, 0x2763c,
0x27700, 0x2771c,
0x27780, 0x2778c,
0x27800, 0x27c38,
0x27c80, 0x27d7c,
0x27e00, 0x27e04
};
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static const unsigned int t5_reg_ranges[] = {
0x1008, 0x1148,
0x1180, 0x11b4,
0x11fc, 0x123c,
0x1280, 0x173c,
0x1800, 0x18fc,
0x3000, 0x3028,
0x3060, 0x30d8,
0x30e0, 0x30fc,
0x3140, 0x357c,
0x35a8, 0x35cc,
0x35ec, 0x35ec,
0x3600, 0x5624,
0x56cc, 0x575c,
0x580c, 0x5814,
0x5890, 0x58bc,
0x5940, 0x59dc,
0x59fc, 0x5a18,
0x5a60, 0x5a9c,
0x5b94, 0x5bfc,
0x6000, 0x6040,
0x6058, 0x614c,
0x7700, 0x7798,
0x77c0, 0x78fc,
0x7b00, 0x7c54,
0x7d00, 0x7efc,
0x8dc0, 0x8de0,
0x8df8, 0x8e84,
0x8ea0, 0x8f84,
0x8fc0, 0x90f8,
0x9400, 0x9470,
0x9600, 0x96f4,
0x9800, 0x9808,
0x9820, 0x983c,
0x9850, 0x9864,
0x9c00, 0x9c6c,
0x9c80, 0x9cec,
0x9d00, 0x9d6c,
0x9d80, 0x9dec,
0x9e00, 0x9e6c,
0x9e80, 0x9eec,
0x9f00, 0x9f6c,
0x9f80, 0xa020,
0xd004, 0xd03c,
0xdfc0, 0xdfe0,
0xe000, 0x11088,
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
0x1109c, 0x11110,
0x11118, 0x1117c,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x11190, 0x11204,
0x19040, 0x1906c,
0x19078, 0x19080,
0x1908c, 0x19124,
0x19150, 0x191b0,
0x191d0, 0x191e8,
0x19238, 0x19290,
0x193f8, 0x19474,
0x19490, 0x194cc,
0x194f0, 0x194f8,
0x19c00, 0x19c60,
0x19c94, 0x19e10,
0x19e50, 0x19f34,
0x19f40, 0x19f50,
0x19f90, 0x19fe4,
0x1a000, 0x1a06c,
0x1a0b0, 0x1a120,
0x1a128, 0x1a138,
0x1a190, 0x1a1c4,
0x1a1fc, 0x1a1fc,
0x1e008, 0x1e00c,
0x1e040, 0x1e04c,
0x1e284, 0x1e290,
0x1e2c0, 0x1e2c0,
0x1e2e0, 0x1e2e0,
0x1e300, 0x1e384,
0x1e3c0, 0x1e3c8,
0x1e408, 0x1e40c,
0x1e440, 0x1e44c,
0x1e684, 0x1e690,
0x1e6c0, 0x1e6c0,
0x1e6e0, 0x1e6e0,
0x1e700, 0x1e784,
0x1e7c0, 0x1e7c8,
0x1e808, 0x1e80c,
0x1e840, 0x1e84c,
0x1ea84, 0x1ea90,
0x1eac0, 0x1eac0,
0x1eae0, 0x1eae0,
0x1eb00, 0x1eb84,
0x1ebc0, 0x1ebc8,
0x1ec08, 0x1ec0c,
0x1ec40, 0x1ec4c,
0x1ee84, 0x1ee90,
0x1eec0, 0x1eec0,
0x1eee0, 0x1eee0,
0x1ef00, 0x1ef84,
0x1efc0, 0x1efc8,
0x1f008, 0x1f00c,
0x1f040, 0x1f04c,
0x1f284, 0x1f290,
0x1f2c0, 0x1f2c0,
0x1f2e0, 0x1f2e0,
0x1f300, 0x1f384,
0x1f3c0, 0x1f3c8,
0x1f408, 0x1f40c,
0x1f440, 0x1f44c,
0x1f684, 0x1f690,
0x1f6c0, 0x1f6c0,
0x1f6e0, 0x1f6e0,
0x1f700, 0x1f784,
0x1f7c0, 0x1f7c8,
0x1f808, 0x1f80c,
0x1f840, 0x1f84c,
0x1fa84, 0x1fa90,
0x1fac0, 0x1fac0,
0x1fae0, 0x1fae0,
0x1fb00, 0x1fb84,
0x1fbc0, 0x1fbc8,
0x1fc08, 0x1fc0c,
0x1fc40, 0x1fc4c,
0x1fe84, 0x1fe90,
0x1fec0, 0x1fec0,
0x1fee0, 0x1fee0,
0x1ff00, 0x1ff84,
0x1ffc0, 0x1ffc8,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x30000, 0x30030,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x30100, 0x30144,
0x30190, 0x301d0,
0x30200, 0x30318,
0x30400, 0x3052c,
0x30540, 0x3061c,
0x30800, 0x30834,
0x308c0, 0x30908,
0x30910, 0x309ac,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x30a00, 0x30a2c,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x30a44, 0x30a50,
0x30a74, 0x30c24,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x30d00, 0x30d00,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x30d08, 0x30d14,
0x30d1c, 0x30d20,
0x30d3c, 0x30d50,
0x31200, 0x3120c,
0x31220, 0x31220,
0x31240, 0x31240,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x31600, 0x3160c,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x31a00, 0x31a1c,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x31e00, 0x31e20,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x31e38, 0x31e3c,
0x31e80, 0x31e80,
0x31e88, 0x31ea8,
0x31eb0, 0x31eb4,
0x31ec8, 0x31ed4,
0x31fb8, 0x32004,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x32200, 0x32200,
0x32208, 0x32240,
0x32248, 0x32280,
0x32288, 0x322c0,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x322c8, 0x322fc,
0x32600, 0x32630,
0x32a00, 0x32abc,
0x32b00, 0x32b70,
0x33000, 0x33048,
0x33060, 0x3309c,
0x330f0, 0x33148,
0x33160, 0x3319c,
0x331f0, 0x332e4,
0x332f8, 0x333e4,
0x333f8, 0x33448,
0x33460, 0x3349c,
0x334f0, 0x33548,
0x33560, 0x3359c,
0x335f0, 0x336e4,
0x336f8, 0x337e4,
0x337f8, 0x337fc,
0x33814, 0x33814,
0x3382c, 0x3382c,
0x33880, 0x3388c,
0x338e8, 0x338ec,
0x33900, 0x33948,
0x33960, 0x3399c,
0x339f0, 0x33ae4,
0x33af8, 0x33b10,
0x33b28, 0x33b28,
0x33b3c, 0x33b50,
0x33bf0, 0x33c10,
0x33c28, 0x33c28,
0x33c3c, 0x33c50,
0x33cf0, 0x33cfc,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x34000, 0x34030,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x34100, 0x34144,
0x34190, 0x341d0,
0x34200, 0x34318,
0x34400, 0x3452c,
0x34540, 0x3461c,
0x34800, 0x34834,
0x348c0, 0x34908,
0x34910, 0x349ac,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x34a00, 0x34a2c,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x34a44, 0x34a50,
0x34a74, 0x34c24,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x34d00, 0x34d00,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x34d08, 0x34d14,
0x34d1c, 0x34d20,
0x34d3c, 0x34d50,
0x35200, 0x3520c,
0x35220, 0x35220,
0x35240, 0x35240,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x35600, 0x3560c,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x35a00, 0x35a1c,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x35e00, 0x35e20,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x35e38, 0x35e3c,
0x35e80, 0x35e80,
0x35e88, 0x35ea8,
0x35eb0, 0x35eb4,
0x35ec8, 0x35ed4,
0x35fb8, 0x36004,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x36200, 0x36200,
0x36208, 0x36240,
0x36248, 0x36280,
0x36288, 0x362c0,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x362c8, 0x362fc,
0x36600, 0x36630,
0x36a00, 0x36abc,
0x36b00, 0x36b70,
0x37000, 0x37048,
0x37060, 0x3709c,
0x370f0, 0x37148,
0x37160, 0x3719c,
0x371f0, 0x372e4,
0x372f8, 0x373e4,
0x373f8, 0x37448,
0x37460, 0x3749c,
0x374f0, 0x37548,
0x37560, 0x3759c,
0x375f0, 0x376e4,
0x376f8, 0x377e4,
0x377f8, 0x377fc,
0x37814, 0x37814,
0x3782c, 0x3782c,
0x37880, 0x3788c,
0x378e8, 0x378ec,
0x37900, 0x37948,
0x37960, 0x3799c,
0x379f0, 0x37ae4,
0x37af8, 0x37b10,
0x37b28, 0x37b28,
0x37b3c, 0x37b50,
0x37bf0, 0x37c10,
0x37c28, 0x37c28,
0x37c3c, 0x37c50,
0x37cf0, 0x37cfc,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x38000, 0x38030,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x38100, 0x38144,
0x38190, 0x381d0,
0x38200, 0x38318,
0x38400, 0x3852c,
0x38540, 0x3861c,
0x38800, 0x38834,
0x388c0, 0x38908,
0x38910, 0x389ac,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x38a00, 0x38a2c,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x38a44, 0x38a50,
0x38a74, 0x38c24,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x38d00, 0x38d00,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x38d08, 0x38d14,
0x38d1c, 0x38d20,
0x38d3c, 0x38d50,
0x39200, 0x3920c,
0x39220, 0x39220,
0x39240, 0x39240,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x39600, 0x3960c,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x39a00, 0x39a1c,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x39e00, 0x39e20,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x39e38, 0x39e3c,
0x39e80, 0x39e80,
0x39e88, 0x39ea8,
0x39eb0, 0x39eb4,
0x39ec8, 0x39ed4,
0x39fb8, 0x3a004,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x3a200, 0x3a200,
0x3a208, 0x3a240,
0x3a248, 0x3a280,
0x3a288, 0x3a2c0,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x3a2c8, 0x3a2fc,
0x3a600, 0x3a630,
0x3aa00, 0x3aabc,
0x3ab00, 0x3ab70,
0x3b000, 0x3b048,
0x3b060, 0x3b09c,
0x3b0f0, 0x3b148,
0x3b160, 0x3b19c,
0x3b1f0, 0x3b2e4,
0x3b2f8, 0x3b3e4,
0x3b3f8, 0x3b448,
0x3b460, 0x3b49c,
0x3b4f0, 0x3b548,
0x3b560, 0x3b59c,
0x3b5f0, 0x3b6e4,
0x3b6f8, 0x3b7e4,
0x3b7f8, 0x3b7fc,
0x3b814, 0x3b814,
0x3b82c, 0x3b82c,
0x3b880, 0x3b88c,
0x3b8e8, 0x3b8ec,
0x3b900, 0x3b948,
0x3b960, 0x3b99c,
0x3b9f0, 0x3bae4,
0x3baf8, 0x3bb10,
0x3bb28, 0x3bb28,
0x3bb3c, 0x3bb50,
0x3bbf0, 0x3bc10,
0x3bc28, 0x3bc28,
0x3bc3c, 0x3bc50,
0x3bcf0, 0x3bcfc,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x3c000, 0x3c030,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x3c100, 0x3c144,
0x3c190, 0x3c1d0,
0x3c200, 0x3c318,
0x3c400, 0x3c52c,
0x3c540, 0x3c61c,
0x3c800, 0x3c834,
0x3c8c0, 0x3c908,
0x3c910, 0x3c9ac,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x3ca00, 0x3ca2c,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x3ca44, 0x3ca50,
0x3ca74, 0x3cc24,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x3cd00, 0x3cd00,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x3cd08, 0x3cd14,
0x3cd1c, 0x3cd20,
0x3cd3c, 0x3cd50,
0x3d200, 0x3d20c,
0x3d220, 0x3d220,
0x3d240, 0x3d240,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x3d600, 0x3d60c,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x3da00, 0x3da1c,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x3de00, 0x3de20,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x3de38, 0x3de3c,
0x3de80, 0x3de80,
0x3de88, 0x3dea8,
0x3deb0, 0x3deb4,
0x3dec8, 0x3ded4,
0x3dfb8, 0x3e004,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x3e200, 0x3e200,
0x3e208, 0x3e240,
0x3e248, 0x3e280,
0x3e288, 0x3e2c0,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x3e2c8, 0x3e2fc,
0x3e600, 0x3e630,
0x3ea00, 0x3eabc,
0x3eb00, 0x3eb70,
0x3f000, 0x3f048,
0x3f060, 0x3f09c,
0x3f0f0, 0x3f148,
0x3f160, 0x3f19c,
0x3f1f0, 0x3f2e4,
0x3f2f8, 0x3f3e4,
0x3f3f8, 0x3f448,
0x3f460, 0x3f49c,
0x3f4f0, 0x3f548,
0x3f560, 0x3f59c,
0x3f5f0, 0x3f6e4,
0x3f6f8, 0x3f7e4,
0x3f7f8, 0x3f7fc,
0x3f814, 0x3f814,
0x3f82c, 0x3f82c,
0x3f880, 0x3f88c,
0x3f8e8, 0x3f8ec,
0x3f900, 0x3f948,
0x3f960, 0x3f99c,
0x3f9f0, 0x3fae4,
0x3faf8, 0x3fb10,
0x3fb28, 0x3fb28,
0x3fb3c, 0x3fb50,
0x3fbf0, 0x3fc10,
0x3fc28, 0x3fc28,
0x3fc3c, 0x3fc50,
0x3fcf0, 0x3fcfc,
0x40000, 0x4000c,
0x40040, 0x40068,
0x4007c, 0x40144,
0x40180, 0x4018c,
0x40200, 0x40298,
0x402ac, 0x4033c,
0x403f8, 0x403fc,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x41304, 0x413c4,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x41400, 0x4141c,
0x41480, 0x414d0,
0x44000, 0x44078,
0x440c0, 0x44278,
0x442c0, 0x44478,
0x444c0, 0x44678,
0x446c0, 0x44878,
0x448c0, 0x449fc,
0x45000, 0x45068,
0x45080, 0x45084,
0x450a0, 0x450b0,
0x45200, 0x45268,
0x45280, 0x45284,
0x452a0, 0x452b0,
0x460c0, 0x460e4,
0x47000, 0x4708c,
0x47200, 0x47250,
0x47400, 0x47420,
0x47600, 0x47618,
0x47800, 0x47814,
0x48000, 0x4800c,
0x48040, 0x48068,
0x4807c, 0x48144,
0x48180, 0x4818c,
0x48200, 0x48298,
0x482ac, 0x4833c,
0x483f8, 0x483fc,
Update T5 register ranges. This is so that regdump skips over registers with read side-effects. MFC after: 3 days
2013-06-27 18:59:07 +00:00
0x49304, 0x493c4,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
0x49400, 0x4941c,
0x49480, 0x494d0,
0x4c000, 0x4c078,
0x4c0c0, 0x4c278,
0x4c2c0, 0x4c478,
0x4c4c0, 0x4c678,
0x4c6c0, 0x4c878,
0x4c8c0, 0x4c9fc,
0x4d000, 0x4d068,
0x4d080, 0x4d084,
0x4d0a0, 0x4d0b0,
0x4d200, 0x4d268,
0x4d280, 0x4d284,
0x4d2a0, 0x4d2b0,
0x4e0c0, 0x4e0e4,
0x4f000, 0x4f08c,
0x4f200, 0x4f250,
0x4f400, 0x4f420,
0x4f600, 0x4f618,
0x4f800, 0x4f814,
0x50000, 0x500cc,
0x50400, 0x50400,
0x50800, 0x508cc,
0x50c00, 0x50c00,
0x51000, 0x5101c,
0x51300, 0x51308,
};
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (is_t4(sc)) {
reg_ranges = &t4_reg_ranges[0];
n = nitems(t4_reg_ranges);
} else {
reg_ranges = &t5_reg_ranges[0];
n = nitems(t5_reg_ranges);
}
regs->version = chip_id(sc) | chip_rev(sc) << 10;
for (i = 0; i < n; i += 2)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
reg_block_dump(sc, buf, reg_ranges[i], reg_ranges[i + 1]);
}
static void
Backport some parts of r272200. - a lock to protect indirect register access - put code that deals with stats in a separate cxgbe_refresh_stats. This is a direct commit to stable/10.
2015-05-05 15:13:59 +00:00
cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
{
struct ifnet *ifp = pi->ifp;
struct sge_txq *txq;
int i, drops;
struct port_stats *s = &pi->stats;
Backport some parts of r272200. - a lock to protect indirect register access - put code that deals with stats in a separate cxgbe_refresh_stats. This is a direct commit to stable/10.
2015-05-05 15:13:59 +00:00
struct timeval tv;
const struct timeval interval = {0, 250000}; /* 250ms */
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Backport some parts of r272200. - a lock to protect indirect register access - put code that deals with stats in a separate cxgbe_refresh_stats. This is a direct commit to stable/10.
2015-05-05 15:13:59 +00:00
getmicrotime(&tv);
timevalsub(&tv, &interval);
if (timevalcmp(&tv, &pi->last_refreshed, <))
return;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r263317, r263412, and r263451. r263317: cxgbe(4): significant rx rework. - More flexible cluster size selection, including the ability to fall back to a safe cluster size (PAGE_SIZE from zone_jumbop by default) in case an allocation of a larger size fails. - A single get_fl_payload() function that assembles the payload into an mbuf chain for any kind of freelist. This replaces two variants: one for freelists with buffer packing enabled and another for those without. - Buffer packing with any sized cluster. It was limited to 4K clusters only before this change. - Enable buffer packing for TOE rx queues as well. - Statistics and tunables to go with all these changes. The driver's man page will be updated separately. r263412: cxgbe(4): if_iqdrops statistic should include tunnel congestion drops. r263451: cxgbe(4): man page updates.
2014-05-06 06:49:39 +00:00
t4_get_port_stats(sc, pi->tx_chan, s);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
ifp->if_opackets = s->tx_frames - s->tx_pause;
ifp->if_ipackets = s->rx_frames - s->rx_pause;
ifp->if_obytes = s->tx_octets - s->tx_pause * 64;
ifp->if_ibytes = s->rx_octets - s->rx_pause * 64;
ifp->if_omcasts = s->tx_mcast_frames - s->tx_pause;
ifp->if_imcasts = s->rx_mcast_frames - s->rx_pause;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ifp->if_iqdrops = s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
if_iqdrops should include frames truncated within the chip. MFC after: 2 weeks
2012-08-14 22:15:12 +00:00
s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
s->rx_trunc3;
Backport some parts of r272200. - a lock to protect indirect register access - put code that deals with stats in a separate cxgbe_refresh_stats. This is a direct commit to stable/10.
2015-05-05 15:13:59 +00:00
for (i = 0; i < NCHAN; i++) {
MFC r263317, r263412, and r263451. r263317: cxgbe(4): significant rx rework. - More flexible cluster size selection, including the ability to fall back to a safe cluster size (PAGE_SIZE from zone_jumbop by default) in case an allocation of a larger size fails. - A single get_fl_payload() function that assembles the payload into an mbuf chain for any kind of freelist. This replaces two variants: one for freelists with buffer packing enabled and another for those without. - Buffer packing with any sized cluster. It was limited to 4K clusters only before this change. - Enable buffer packing for TOE rx queues as well. - Statistics and tunables to go with all these changes. The driver's man page will be updated separately. r263412: cxgbe(4): if_iqdrops statistic should include tunnel congestion drops. r263451: cxgbe(4): man page updates.
2014-05-06 06:49:39 +00:00
if (pi->rx_chan_map & (1 << i)) {
uint32_t v;
Backport some parts of r272200. - a lock to protect indirect register access - put code that deals with stats in a separate cxgbe_refresh_stats. This is a direct commit to stable/10.
2015-05-05 15:13:59 +00:00
mtx_lock(&sc->regwin_lock);
MFC r263317, r263412, and r263451. r263317: cxgbe(4): significant rx rework. - More flexible cluster size selection, including the ability to fall back to a safe cluster size (PAGE_SIZE from zone_jumbop by default) in case an allocation of a larger size fails. - A single get_fl_payload() function that assembles the payload into an mbuf chain for any kind of freelist. This replaces two variants: one for freelists with buffer packing enabled and another for those without. - Buffer packing with any sized cluster. It was limited to 4K clusters only before this change. - Enable buffer packing for TOE rx queues as well. - Statistics and tunables to go with all these changes. The driver's man page will be updated separately. r263412: cxgbe(4): if_iqdrops statistic should include tunnel congestion drops. r263451: cxgbe(4): man page updates.
2014-05-06 06:49:39 +00:00
t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
1, A_TP_MIB_TNL_CNG_DROP_0 + i);
Backport some parts of r272200. - a lock to protect indirect register access - put code that deals with stats in a separate cxgbe_refresh_stats. This is a direct commit to stable/10.
2015-05-05 15:13:59 +00:00
mtx_unlock(&sc->regwin_lock);
MFC r263317, r263412, and r263451. r263317: cxgbe(4): significant rx rework. - More flexible cluster size selection, including the ability to fall back to a safe cluster size (PAGE_SIZE from zone_jumbop by default) in case an allocation of a larger size fails. - A single get_fl_payload() function that assembles the payload into an mbuf chain for any kind of freelist. This replaces two variants: one for freelists with buffer packing enabled and another for those without. - Buffer packing with any sized cluster. It was limited to 4K clusters only before this change. - Enable buffer packing for TOE rx queues as well. - Statistics and tunables to go with all these changes. The driver's man page will be updated separately. r263412: cxgbe(4): if_iqdrops statistic should include tunnel congestion drops. r263451: cxgbe(4): man page updates.
2014-05-06 06:49:39 +00:00
ifp->if_iqdrops += v;
}
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
drops = s->tx_drop;
for_each_txq(pi, i, txq)
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
drops += counter_u64_fetch(txq->r->drops);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ifp->if_snd.ifq_drops = drops;
ifp->if_oerrors = s->tx_error_frames;
ifp->if_ierrors = s->rx_jabber + s->rx_runt + s->rx_too_long +
s->rx_fcs_err + s->rx_len_err;
Backport some parts of r272200. - a lock to protect indirect register access - put code that deals with stats in a separate cxgbe_refresh_stats. This is a direct commit to stable/10.
2015-05-05 15:13:59 +00:00
getmicrotime(&pi->last_refreshed);
}
static void
cxgbe_tick(void *arg)
{
struct port_info *pi = arg;
struct adapter *sc = pi->adapter;
struct ifnet *ifp = pi->ifp;
PORT_LOCK(pi);
if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
PORT_UNLOCK(pi);
return; /* without scheduling another callout */
}
cxgbe_refresh_stats(sc, pi);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
callout_schedule(&pi->tick, hz);
PORT_UNLOCK(pi);
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
static void
cxgbe_vlan_config(void *arg, struct ifnet *ifp, uint16_t vid)
{
struct ifnet *vlan;
Temporary fix for kern/172364. PR: kern/172364 MFC after: 3 days
2012-10-12 21:58:21 +00:00
if (arg != ifp || ifp->if_type != IFT_ETHER)
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
return;
vlan = VLAN_DEVAT(ifp, vid);
VLAN_SETCOOKIE(vlan, ifp);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cpl_not_handled(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
{
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#ifdef INVARIANTS
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
panic("%s: opcode 0x%02x on iq %p with payload %p",
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
__func__, rss->opcode, iq, m);
#else
Allow for a different handler for each type of firmware message. MFC after: 2 weeks
2012-08-16 18:31:50 +00:00
log(LOG_ERR, "%s: opcode 0x%02x on iq %p with payload %p\n",
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
__func__, rss->opcode, iq, m);
m_freem(m);
#endif
return (EDOOFUS);
}
int
t4_register_cpl_handler(struct adapter *sc, int opcode, cpl_handler_t h)
{
uintptr_t *loc, new;
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
if (opcode >= nitems(sc->cpl_handler))
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (EINVAL);
new = h ? (uintptr_t)h : (uintptr_t)cpl_not_handled;
loc = (uintptr_t *) &sc->cpl_handler[opcode];
atomic_store_rel_ptr(loc, new);
return (0);
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
static int
an_not_handled(struct sge_iq *iq, const struct rsp_ctrl *ctrl)
{
#ifdef INVARIANTS
panic("%s: async notification on iq %p (ctrl %p)", __func__, iq, ctrl);
#else
Allow for a different handler for each type of firmware message. MFC after: 2 weeks
2012-08-16 18:31:50 +00:00
log(LOG_ERR, "%s: async notification on iq %p (ctrl %p)\n",
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
__func__, iq, ctrl);
#endif
return (EDOOFUS);
}
int
t4_register_an_handler(struct adapter *sc, an_handler_t h)
{
uintptr_t *loc, new;
new = h ? (uintptr_t)h : (uintptr_t)an_not_handled;
loc = (uintptr_t *) &sc->an_handler;
atomic_store_rel_ptr(loc, new);
return (0);
}
Allow for a different handler for each type of firmware message. MFC after: 2 weeks
2012-08-16 18:31:50 +00:00
static int
fw_msg_not_handled(struct adapter *sc, const __be64 *rpl)
{
Prefer __containerof() over __member2struct(). The former works better with qualifiers, but also properly type checks the input pointer.
2012-10-19 13:26:40 +00:00
const struct cpl_fw6_msg *cpl =
__containerof(rpl, struct cpl_fw6_msg, data[0]);
Allow for a different handler for each type of firmware message. MFC after: 2 weeks
2012-08-16 18:31:50 +00:00
#ifdef INVARIANTS
panic("%s: fw_msg type %d", __func__, cpl->type);
#else
log(LOG_ERR, "%s: fw_msg type %d\n", __func__, cpl->type);
#endif
return (EDOOFUS);
}
int
t4_register_fw_msg_handler(struct adapter *sc, int type, fw_msg_handler_t h)
{
uintptr_t *loc, new;
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
if (type >= nitems(sc->fw_msg_handler))
Allow for a different handler for each type of firmware message. MFC after: 2 weeks
2012-08-16 18:31:50 +00:00
return (EINVAL);
cxgbe(4): Ask the card's firmware to pad up tiny CPLs by encapsulating them in a firmware message if it is able to do so. This works out better for one of the FIFOs in the chip. MFC after: 5 days
2013-02-26 00:27:27 +00:00
/*
* These are dispatched by the handler for FW{4|6}_CPL_MSG using the CPL
* handler dispatch table. Reject any attempt to install a handler for
* this subtype.
*/
if (type == FW_TYPE_RSSCPL || type == FW6_TYPE_RSSCPL)
return (EINVAL);
Allow for a different handler for each type of firmware message. MFC after: 2 weeks
2012-08-16 18:31:50 +00:00
new = h ? (uintptr_t)h : (uintptr_t)fw_msg_not_handled;
loc = (uintptr_t *) &sc->fw_msg_handler[type];
atomic_store_rel_ptr(loc, new);
return (0);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int
t4_sysctls(struct adapter *sc)
{
struct sysctl_ctx_list *ctx;
struct sysctl_oid *oid;
struct sysctl_oid_list *children, *c0;
static char *caps[] = {
"\20\1PPP\2QFC\3DCBX", /* caps[0] linkcaps */
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
"\20\1NIC\2VM\3IDS\4UM\5UM_ISGL" /* caps[1] niccaps */
"\6HASHFILTER\7ETHOFLD",
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
"\20\1TOE", /* caps[2] toecaps */
"\20\1RDDP\2RDMAC", /* caps[3] rdmacaps */
"\20\1INITIATOR_PDU\2TARGET_PDU" /* caps[4] iscsicaps */
"\3INITIATOR_CNXOFLD\4TARGET_CNXOFLD"
"\5INITIATOR_SSNOFLD\6TARGET_SSNOFLD",
"\20\1INITIATOR\2TARGET\3CTRL_OFLD" /* caps[5] fcoecaps */
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
"\4PO_INITIAOR\5PO_TARGET"
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
};
Cosmetic change (s/wrwc/wcwr/;s/WRWC/WCWR/). MFC after: 3 days.
2013-04-11 22:49:29 +00:00
static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ctx = device_get_sysctl_ctx(sc->dev);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/*
* dev.t4nex.X.
*/
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
oid = device_get_sysctl_tree(sc->dev);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
c0 = children = SYSCTL_CHILDREN(oid);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MFC r260210 (by adrian@): Add an option to enable or disable the small RX packet copying that is done to improve performance of small frames. When doing RX packing, the RX copying isn't necessarily required.
2014-05-06 04:22:06 +00:00
sc->sc_do_rxcopy = 1;
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
&sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
sc->params.nports, "# of ports");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
NULL, chip_rev(sc), "chip hardware revision");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
MFC r263710, r273377, r273378, r273423 and r273455: - De-vnet hash sizes and hash masks. - Fix multiple issues related to arguments passed to SYSCTL macros. Sponsored by: Mellanox Technologies
2014-10-27 14:38:00 +00:00
CTLFLAG_RD, sc->fw_version, 0, "firmware version");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
MFC r263710, r273377, r273378, r273423 and r273455: - De-vnet hash sizes and hash masks. - Fix multiple issues related to arguments passed to SYSCTL macros. Sponsored by: Mellanox Technologies
2014-10-27 14:38:00 +00:00
CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
sc->cfcsum, "config file checksum");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
CTLTYPE_STRING | CTLFLAG_RD, doorbells, sc->doorbells,
sysctl_bitfield, "A", "available doorbells");
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkcaps",
CTLTYPE_STRING | CTLFLAG_RD, caps[0], sc->linkcaps,
sysctl_bitfield, "A", "available link capabilities");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "niccaps",
CTLTYPE_STRING | CTLFLAG_RD, caps[1], sc->niccaps,
sysctl_bitfield, "A", "available NIC capabilities");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "toecaps",
CTLTYPE_STRING | CTLFLAG_RD, caps[2], sc->toecaps,
sysctl_bitfield, "A", "available TCP offload capabilities");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdmacaps",
CTLTYPE_STRING | CTLFLAG_RD, caps[3], sc->rdmacaps,
sysctl_bitfield, "A", "available RDMA capabilities");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "iscsicaps",
CTLTYPE_STRING | CTLFLAG_RD, caps[4], sc->iscsicaps,
sysctl_bitfield, "A", "available iSCSI capabilities");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoecaps",
CTLTYPE_STRING | CTLFLAG_RD, caps[5], sc->fcoecaps,
sysctl_bitfield, "A", "available FCoE capabilities");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
sc->params.vpd.cclk, "core clock frequency (in KHz)");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Display holdoff timers and packet counts as a list of numbers. MFC after: 1 week
2011-03-09 21:07:09 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
CTLTYPE_STRING | CTLFLAG_RD, sc->sge.timer_val,
sizeof(sc->sge.timer_val), sysctl_int_array, "A",
"interrupt holdoff timer values (us)");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Display holdoff timers and packet counts as a list of numbers. MFC after: 1 week
2011-03-09 21:07:09 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
CTLTYPE_STRING | CTLFLAG_RD, sc->sge.counter_val,
sizeof(sc->sge.counter_val), sysctl_int_array, "A",
"interrupt holdoff packet counter values");
Add a sysctl to get the number of filters available. sysctl dev.t4nex.<N>.nfilters sysctl dev.t5nex.<N>.nfilters MFC after: 3 days
2013-07-01 17:31:04 +00:00
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
NULL, sc->tids.nftids, "number of filters");
Display temperature sensor data. Shows -1 if sensor not available on the card. # sysctl dev.t4nex.0.temperature # sysctl dev.t5nex.0.temperature
2013-08-02 18:05:42 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
MFC r264621 use correct (integer) type for the temperature sysctl Reviewed by: np, scottl Obtained from: Netflix
2014-04-21 17:17:23 +00:00
CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
Display temperature sensor data. Shows -1 if sensor not available on the card. # sysctl dev.t4nex.0.temperature # sysctl dev.t5nex.0.temperature
2013-08-02 18:05:42 +00:00
"chip temperature (in Celsius)");
Display SGE tunables in the sysctl tree. dev.t5nex.0.fl_pktshift: payload DMA offset in rx buffer (bytes) dev.t5nex.0.fl_pad: payload pad boundary (bytes) dev.t5nex.0.spg_len: status page size (bytes) dev.t5nex.0.cong_drop: congestion drop setting Discussed with: scottl
2013-07-31 05:12:51 +00:00
t4_sge_sysctls(sc, ctx, children);
Merge r254386 from user/np/cxl_tuning. Add an INET|INET6 check missing in said revision. r254386: Flush inactive LRO entries periodically.
2013-08-29 06:26:22 +00:00
sc->lro_timeout = 100;
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
&sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
cxgbe: reduce diffs with other branches. Will help future MFCs from HEAD. MFC after: 3 days
2012-02-07 06:21:59 +00:00
#ifdef SBUF_DRAIN
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/*
* dev.t4nex.X.misc. Marked CTLFLAG_SKIP to avoid information overload.
*/
oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
CTLFLAG_RD | CTLFLAG_SKIP, NULL,
"logs and miscellaneous information");
children = SYSCTL_CHILDREN(oid);
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_cctrl, "A", "congestion control");
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_cim_la, "A", "CIM logic analyzer");
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (is_t5(sc)) {
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
}
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_cim_qcfg, "A", "CIM queue configuration");
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_cpl_stats, "A", "CPL statistics");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
MFC r276728: cxgbe(4): fix the description of a strange bunch of counters.
2015-04-08 00:13:17 +00:00
sysctl_ddp_stats, "A", "non-TCP DDP statistics");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
sysctl_devlog, "A", "firmware's device log");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_fcoe_stats, "A", "FCoE statistics");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_hw_sched, "A", "hardware scheduler ");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_l2t, "A", "hardware L2 table");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_lb_stats, "A", "loopback statistics");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_meminfo, "A", "memory regions");
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_mps_tcam, "A", "MPS TCAM entries");
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_path_mtus, "A", "path MTUs");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_pm_stats, "A", "PM statistics");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_rdma_stats, "A", "RDMA statistics");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_tcp_stats, "A", "TCP statistics");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_tids, "A", "TID information");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_tp_err_stats, "A", "TP error statistics");
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_tp_la, "A", "TP logic analyzer");
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_tx_rate, "A", "Tx rate");
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
sysctl_ulprx_la, "A", "ULPRX logic analyzer");
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (is_t5(sc)) {
Cosmetic change (s/wrwc/wcwr/;s/WRWC/WCWR/). MFC after: 3 days.
2013-04-11 22:49:29 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
Cosmetic change (s/wrwc/wcwr/;s/WRWC/WCWR/). MFC after: 3 days.
2013-04-11 22:49:29 +00:00
sysctl_wcwr_stats, "A", "write combined work requests");
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
}
cxgbe: reduce diffs with other branches. Will help future MFCs from HEAD. MFC after: 3 days
2012-02-07 06:21:59 +00:00
#endif
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (is_offload(sc)) {
/*
* dev.t4nex.X.toe.
*/
oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
NULL, "TOE parameters");
children = SYSCTL_CHILDREN(oid);
sc->tt.sndbuf = 256 * 1024;
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
&sc->tt.sndbuf, 0, "max hardware send buffer size");
sc->tt.ddp = 0;
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
&sc->tt.ddp, 0, "DDP allowed");
Initialize various DDP parameters in the main cxgbe(4) driver: - Setup multiple DDP page sizes. When the driver attempts DDP it will try to combine physically contiguous pages into regions of these sizes. - Set the indicate size such that the payload carried in the indicate can be copied in the header mbuf (and the 16K rx buffer can be recycled). - Set DDP threshold to the max payload that the chip will coalesce and deliver to the driver (this is ~16K by default, which is also why the offload rx queue is backed by 16K buffers). If the chip is able to coalesce up to the max it's allowed to, it's a good sign that the peer is transmitting in bulk without any TCP PSH. MFC after: 2 weeks
2012-08-16 22:33:56 +00:00
sc->tt.indsz = G_INDICATESIZE(t4_read_reg(sc, A_TP_PARA_REG5));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "indsz", CTLFLAG_RW,
&sc->tt.indsz, 0, "DDP max indicate size allowed");
Initialize various DDP parameters in the main cxgbe(4) driver: - Setup multiple DDP page sizes. When the driver attempts DDP it will try to combine physically contiguous pages into regions of these sizes. - Set the indicate size such that the payload carried in the indicate can be copied in the header mbuf (and the 16K rx buffer can be recycled). - Set DDP threshold to the max payload that the chip will coalesce and deliver to the driver (this is ~16K by default, which is also why the offload rx queue is backed by 16K buffers). If the chip is able to coalesce up to the max it's allowed to, it's a good sign that the peer is transmitting in bulk without any TCP PSH. MFC after: 2 weeks
2012-08-16 22:33:56 +00:00
sc->tt.ddp_thres =
G_RXCOALESCESIZE(t4_read_reg(sc, A_TP_PARA_REG2));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp_thres", CTLFLAG_RW,
&sc->tt.ddp_thres, 0, "DDP threshold");
- Make note of interface MTU change if the rx queues exist, and not just when the interface is up. - Add a tunable to control the TOE's rx coalesce feature (enabled by default as it always has been). Consider the interface MTU or the coalesce size when deciding which cluster zone to use to fill the offload rx queue's free list. The tunable is: dev.{t4nex,t5nex}.<N>.toe.rx_coalesce MFC after: 1 day
2013-07-04 21:19:01 +00:00
sc->tt.rx_coalesce = 1;
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
MFC r276597: cxgbe/tom: do not engage the TOE's payload chopper for payload < 2 MSS or for 10Gbps ports.
2015-04-08 00:35:12 +00:00
sc->tt.tx_align = 1;
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
#endif
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (0);
}
static int
cxgbe_sysctls(struct port_info *pi)
{
struct sysctl_ctx_list *ctx;
struct sysctl_oid *oid;
struct sysctl_oid_list *children;
MFC r259382: Read card capabilities after firmware initialization, instead of setting them up as part of firmware initialization (which the driver gets to do only if it's the master driver). Read the range of tids available for the ETHOFLD functionality if it's enabled. New is_ftid() and is_etid() functions to test whether a tid falls within the range of filter tids or ETHOFLD tids respectively.
2014-05-06 07:21:50 +00:00
struct adapter *sc = pi->adapter;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ctx = device_get_sysctl_ctx(pi->dev);
/*
* dev.cxgbe.X.
*/
oid = device_get_sysctl_tree(pi->dev);
children = SYSCTL_CHILDREN(oid);
Display a string instead of a numeric code in the linkdnrc sysctl. Submitted by: gnn@
2013-07-27 07:43:43 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
- Show the reason why link is down if this information is available. - Display the temperature and PHY firmware version of the BT PHY. MFC after: 1 day
2013-07-05 01:53:51 +00:00
if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
"PHY temperature (in Celsius)");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
"PHY firmware version");
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
&pi->nrxq, 0, "# of rx queues");
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
&pi->ntxq, 0, "# of tx queues");
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
&pi->first_rxq, 0, "index of first rx queue");
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
&pi->first_txq, 0, "index of first tx queue");
MFC r261558 Add a new sysctl, dev.cxgbe.N.rsrv_noflow, and a companion tunable, hw.cxgbe.rsrv_noflow. When set, queue 0 of the port is reserved for TX packets without a flowid. The hash value of packets with a flowid is bumped up by 1. The intent is to provide a private queue for link-level packets like LACP that is unlikely to overflow or suffer deep queue latency.
2014-04-15 08:08:44 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq", CTLTYPE_INT |
CTLFLAG_RW, pi, 0, sysctl_noflowq, "IU",
"Reserve queue 0 for non-flowid packets");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
MFC r259382: Read card capabilities after firmware initialization, instead of setting them up as part of firmware initialization (which the driver gets to do only if it's the master driver). Read the range of tids available for the ETHOFLD functionality if it's enabled. New is_ftid() and is_etid() functions to test whether a tid falls within the range of filter tids or ETHOFLD tids respectively.
2014-05-06 07:21:50 +00:00
if (is_offload(sc)) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
&pi->nofldrxq, 0,
"# of rx queues for offloaded TCP connections");
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
&pi->nofldtxq, 0,
"# of tx queues for offloaded TCP connections");
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
CTLFLAG_RD, &pi->first_ofld_rxq, 0,
"index of first TOE rx queue");
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
CTLFLAG_RD, &pi->first_ofld_txq, 0,
"index of first TOE tx queue");
}
#endif
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
&pi->nnmrxq, 0, "# of rx queues for netmap");
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
&pi->nnmtxq, 0, "# of tx queues for netmap");
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
CTLFLAG_RD, &pi->first_nm_rxq, 0,
"index of first netmap rx queue");
SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
CTLFLAG_RD, &pi->first_nm_txq, 0,
"index of first netmap tx queue");
#endif
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_holdoff_tmr_idx, "I",
"holdoff timer index");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_holdoff_pktc_idx, "I",
"holdoff packet counter index");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_qsize_rxq, "I",
"rx queue size");
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_qsize_txq, "I",
"tx queue size");
MFC r271450: cxgbe(4): knobs to enable/disable PAUSE frame based flow control. Approved by: re (glebius)
2014-09-22 13:07:03 +00:00
SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
CTLTYPE_STRING | CTLFLAG_RW, pi, PAUSE_TX, sysctl_pause_settings,
"A", "PAUSE settings (bit 0 = rx_pause, bit 1 = tx_pause)");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
/*
* dev.cxgbe.X.stats.
*/
oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
NULL, "port statistics");
children = SYSCTL_CHILDREN(oid);
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
&pi->tx_parse_error, 0,
"# of tx packets with invalid length or # of segments");
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
#define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
MFC r259382: Read card capabilities after firmware initialization, instead of setting them up as part of firmware initialization (which the driver gets to do only if it's the master driver). Read the range of tids available for the ETHOFLD functionality if it's enabled. New is_ftid() and is_etid() functions to test whether a tid falls within the range of filter tids or ETHOFLD tids respectively.
2014-05-06 07:21:50 +00:00
CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
sysctl_handle_t4_reg64, "QU", desc)
SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
"# of tx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
"# of tx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
"# of tx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
"# of tx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
"# of tx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
"# of tx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
"# of tx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
"# of frames received with bad FCS",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
"# of frames received with length error",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
"# of rx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
"# of rx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
"# of rx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
"# of rx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
"# of rx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
"# of rx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
"# of rx frames in this range",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
#undef SYSCTL_ADD_T4_REG64
#define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
&pi->stats.name, desc)
/* We get these from port_stats and they may be stale by upto 1s */
SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
"# drops due to buffer-group 0 overflows");
SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
"# drops due to buffer-group 1 overflows");
SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
"# drops due to buffer-group 2 overflows");
SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
"# drops due to buffer-group 3 overflows");
SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
"# of buffer-group 0 truncated packets");
SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
"# of buffer-group 1 truncated packets");
SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
"# of buffer-group 2 truncated packets");
SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
"# of buffer-group 3 truncated packets");
#undef SYSCTL_ADD_T4_PORTSTAT
return (0);
}
Display holdoff timers and packet counts as a list of numbers. MFC after: 1 week
2011-03-09 21:07:09 +00:00
static int
sysctl_int_array(SYSCTL_HANDLER_ARGS)
{
MFC r280403: cxgbe(4): Do not call sbuf_trim on an sbuf with a drain function.
2015-04-08 01:25:26 +00:00
int rc, *i, space = 0;
Display holdoff timers and packet counts as a list of numbers. MFC after: 1 week
2011-03-09 21:07:09 +00:00
struct sbuf sb;
sbuf_new(&sb, NULL, 32, SBUF_AUTOEXTEND);
MFC r280403: cxgbe(4): Do not call sbuf_trim on an sbuf with a drain function.
2015-04-08 01:25:26 +00:00
for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
if (space)
sbuf_printf(&sb, " ");
sbuf_printf(&sb, "%d", *i);
space = 1;
}
Display holdoff timers and packet counts as a list of numbers. MFC after: 1 week
2011-03-09 21:07:09 +00:00
sbuf_finish(&sb);
rc = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
sbuf_delete(&sb);
return (rc);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int
sysctl_bitfield(SYSCTL_HANDLER_ARGS)
{
int rc;
struct sbuf *sb;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return(rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
if (sb == NULL)
return (ENOMEM);
sbuf_printf(sb, "%b", (int)arg2, (char *)arg1);
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
- Show the reason why link is down if this information is available. - Display the temperature and PHY firmware version of the BT PHY. MFC after: 1 day
2013-07-05 01:53:51 +00:00
static int
sysctl_btphy(SYSCTL_HANDLER_ARGS)
{
struct port_info *pi = arg1;
int op = arg2;
struct adapter *sc = pi->adapter;
u_int v;
int rc;
rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4btt");
if (rc)
return (rc);
/* XXX: magic numbers */
rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
&v);
end_synchronized_op(sc, 0);
if (rc)
return (rc);
if (op == 0)
v /= 256;
rc = sysctl_handle_int(oidp, &v, 0, req);
return (rc);
}
MFC r261558 Add a new sysctl, dev.cxgbe.N.rsrv_noflow, and a companion tunable, hw.cxgbe.rsrv_noflow. When set, queue 0 of the port is reserved for TX packets without a flowid. The hash value of packets with a flowid is bumped up by 1. The intent is to provide a private queue for link-level packets like LACP that is unlikely to overflow or suffer deep queue latency.
2014-04-15 08:08:44 +00:00
static int
sysctl_noflowq(SYSCTL_HANDLER_ARGS)
{
struct port_info *pi = arg1;
int rc, val;
val = pi->rsrv_noflowq;
rc = sysctl_handle_int(oidp, &val, 0, req);
if (rc != 0 || req->newptr == NULL)
return (rc);
if ((val >= 1) && (pi->ntxq > 1))
pi->rsrv_noflowq = 1;
else
pi->rsrv_noflowq = 0;
return (rc);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int
sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
{
struct port_info *pi = arg1;
struct adapter *sc = pi->adapter;
int idx, rc, i;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
struct sge_rxq *rxq;
On-the-fly changes to the interrupt coalescing timer should apply to the TOE rx queues too. MFC after: 1 day
2013-07-04 20:17:39 +00:00
#ifdef TCP_OFFLOAD
struct sge_ofld_rxq *ofld_rxq;
#endif
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
uint8_t v;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
idx = pi->tmr_idx;
rc = sysctl_handle_int(oidp, &idx, 0, req);
if (rc != 0 || req->newptr == NULL)
return (rc);
if (idx < 0 || idx >= SGE_NTIMERS)
return (EINVAL);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
"t4tmr");
if (rc)
return (rc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(pi->pktc_idx != -1);
for_each_rxq(pi, i, rxq) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#ifdef atomic_store_rel_8
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
atomic_store_rel_8(&rxq->iq.intr_params, v);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#else
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rxq->iq.intr_params = v;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
On-the-fly changes to the interrupt coalescing timer should apply to the TOE rx queues too. MFC after: 1 day
2013-07-04 20:17:39 +00:00
#ifdef TCP_OFFLOAD
for_each_ofld_rxq(pi, i, ofld_rxq) {
#ifdef atomic_store_rel_8
atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
#else
ofld_rxq->iq.intr_params = v;
#endif
}
#endif
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
pi->tmr_idx = idx;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
end_synchronized_op(sc, LOCK_HELD);
return (0);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
static int
sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
{
struct port_info *pi = arg1;
struct adapter *sc = pi->adapter;
int idx, rc;
idx = pi->pktc_idx;
rc = sysctl_handle_int(oidp, &idx, 0, req);
if (rc != 0 || req->newptr == NULL)
return (rc);
if (idx < -1 || idx >= SGE_NCOUNTERS)
return (EINVAL);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
"t4pktc");
if (rc)
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (pi->flags & PORT_INIT_DONE)
rc = EBUSY; /* cannot be changed once the queues are created */
else
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
pi->pktc_idx = idx;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
end_synchronized_op(sc, LOCK_HELD);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (rc);
}
static int
sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
{
struct port_info *pi = arg1;
struct adapter *sc = pi->adapter;
int qsize, rc;
qsize = pi->qsize_rxq;
rc = sysctl_handle_int(oidp, &qsize, 0, req);
if (rc != 0 || req->newptr == NULL)
return (rc);
if (qsize < 128 || (qsize & 7))
return (EINVAL);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
"t4rxqs");
if (rc)
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (pi->flags & PORT_INIT_DONE)
rc = EBUSY; /* cannot be changed once the queues are created */
else
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
pi->qsize_rxq = qsize;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
end_synchronized_op(sc, LOCK_HELD);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (rc);
}
static int
sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
{
struct port_info *pi = arg1;
struct adapter *sc = pi->adapter;
int qsize, rc;
qsize = pi->qsize_txq;
rc = sysctl_handle_int(oidp, &qsize, 0, req);
if (rc != 0 || req->newptr == NULL)
return (rc);
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
if (qsize < 128 || qsize > 65536)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (EINVAL);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, pi, HOLD_LOCK | SLEEP_OK | INTR_OK,
"t4txqs");
if (rc)
return (rc);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (pi->flags & PORT_INIT_DONE)
rc = EBUSY; /* cannot be changed once the queues are created */
else
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
pi->qsize_txq = qsize;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
end_synchronized_op(sc, LOCK_HELD);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
return (rc);
}
MFC r271450: cxgbe(4): knobs to enable/disable PAUSE frame based flow control. Approved by: re (glebius)
2014-09-22 13:07:03 +00:00
static int
sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
{
struct port_info *pi = arg1;
struct adapter *sc = pi->adapter;
struct link_config *lc = &pi->link_cfg;
int rc;
if (req->newptr == NULL) {
struct sbuf *sb;
static char *bits = "\20\1PAUSE_RX\2PAUSE_TX";
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return(rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
if (sb == NULL)
return (ENOMEM);
sbuf_printf(sb, "%b", lc->fc & (PAUSE_TX | PAUSE_RX), bits);
rc = sbuf_finish(sb);
sbuf_delete(sb);
} else {
char s[2];
int n;
s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX));
s[1] = 0;
rc = sysctl_handle_string(oidp, s, sizeof(s), req);
if (rc != 0)
return(rc);
if (s[1] != 0)
return (EINVAL);
if (s[0] < '0' || s[0] > '9')
return (EINVAL); /* not a number */
n = s[0] - '0';
if (n & ~(PAUSE_TX | PAUSE_RX))
return (EINVAL); /* some other bit is set too */
rc = begin_synchronized_op(sc, pi, SLEEP_OK | INTR_OK, "t4PAUSE");
if (rc)
return (rc);
if ((lc->requested_fc & (PAUSE_TX | PAUSE_RX)) != n) {
int link_ok = lc->link_ok;
lc->requested_fc &= ~(PAUSE_TX | PAUSE_RX);
lc->requested_fc |= n;
rc = -t4_link_start(sc, sc->mbox, pi->tx_chan, lc);
lc->link_ok = link_ok; /* restore */
}
end_synchronized_op(sc, 0);
}
return (rc);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int
sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
int reg = arg2;
uint64_t val;
val = t4_read_reg64(sc, reg);
return (sysctl_handle_64(oidp, &val, 0, req));
}
Display temperature sensor data. Shows -1 if sensor not available on the card. # sysctl dev.t4nex.0.temperature # sysctl dev.t5nex.0.temperature
2013-08-02 18:05:42 +00:00
static int
sysctl_temperature(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
int rc, t;
uint32_t param, val;
rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
if (rc)
return (rc);
param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
end_synchronized_op(sc, 0);
if (rc)
return (rc);
/* unknown is returned as 0 but we display -1 in that case */
t = val == 0 ? -1 : val;
rc = sysctl_handle_int(oidp, &t, 0, req);
return (rc);
}
cxgbe: reduce diffs with other branches. Will help future MFCs from HEAD. MFC after: 3 days
2012-02-07 06:21:59 +00:00
#ifdef SBUF_DRAIN
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int
sysctl_cctrl(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc, i;
uint16_t incr[NMTUS][NCCTRL_WIN];
static const char *dec_fac[] = {
"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
"0.9375"
};
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
if (sb == NULL)
return (ENOMEM);
t4_read_cong_tbl(sc, incr);
for (i = 0; i < NCCTRL_WIN; ++i) {
sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
incr[5][i], incr[6][i], incr[7][i]);
sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
incr[8][i], incr[9][i], incr[10][i], incr[11][i],
incr[12][i], incr[13][i], incr[14][i], incr[15][i],
sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
}
rc = sbuf_finish(sb);
sbuf_delete(sb);
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
return (rc);
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI", /* ibq's */
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI", /* obq's */
"SGE0-RX", "SGE1-RX" /* additional obq's (T5 onwards) */
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
};
static int
sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc, i, n, qid = arg2;
uint32_t *buf, *p;
char *qtype;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
u_int cim_num_obq = is_t4(sc) ? CIM_NUM_OBQ : CIM_NUM_OBQ_T5;
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
("%s: bad qid %d\n", __func__, qid));
if (qid < CIM_NUM_IBQ) {
/* inbound queue */
qtype = "IBQ";
n = 4 * CIM_IBQ_SIZE;
buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
rc = t4_read_cim_ibq(sc, qid, buf, n);
} else {
/* outbound queue */
qtype = "OBQ";
qid -= CIM_NUM_IBQ;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
n = 4 * cim_num_obq * CIM_OBQ_SIZE;
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
rc = t4_read_cim_obq(sc, qid, buf, n);
}
if (rc < 0) {
rc = -rc;
goto done;
}
n = rc * sizeof(uint32_t); /* rc has # of words actually read */
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
goto done;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
if (sb == NULL) {
rc = ENOMEM;
goto done;
}
sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
for (i = 0, p = buf; i < n; i += 16, p += 4)
sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
p[2], p[3]);
rc = sbuf_finish(sb);
sbuf_delete(sb);
done:
free(buf, M_CXGBE);
return (rc);
}
static int
sysctl_cim_la(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
u_int cfg;
struct sbuf *sb;
uint32_t *buf, *p;
int rc;
rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
if (rc != 0)
return (rc);
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
if (sb == NULL)
return (ENOMEM);
buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
M_ZERO | M_WAITOK);
rc = -t4_cim_read_la(sc, buf, NULL);
if (rc != 0)
goto done;
sbuf_printf(sb, "Status Data PC%s",
cfg & F_UPDBGLACAPTPCONLY ? "" :
" LS0Stat LS0Addr LS0Data");
KASSERT((sc->params.cim_la_size & 7) == 0,
("%s: p will walk off the end of buf", __func__));
for (p = buf; p < &buf[sc->params.cim_la_size]; p += 8) {
if (cfg & F_UPDBGLACAPTPCONLY) {
sbuf_printf(sb, "\n %02x %08x %08x", p[5] & 0xff,
p[6], p[7]);
sbuf_printf(sb, "\n %02x %02x%06x %02x%06x",
(p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
p[4] & 0xff, p[5] >> 8);
sbuf_printf(sb, "\n %02x %x%07x %x%07x",
(p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
p[1] & 0xf, p[2] >> 4);
} else {
sbuf_printf(sb,
"\n %02x %x%07x %x%07x %08x %08x "
"%08x%08x%08x%08x",
(p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
p[6], p[7]);
}
}
rc = sbuf_finish(sb);
sbuf_delete(sb);
done:
free(buf, M_CXGBE);
return (rc);
}
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
static int
sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
u_int i;
struct sbuf *sb;
uint32_t *buf, *p;
int rc;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
if (sb == NULL)
return (ENOMEM);
buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
M_ZERO | M_WAITOK);
t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
p = buf;
for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
p[1], p[0]);
}
sbuf_printf(sb, "\n\nCnt ID Tag UE Data RDY VLD");
for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
sbuf_printf(sb, "\n%3u %2u %x %u %08x%08x %u %u",
(p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
(p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
(p[1] >> 2) | ((p[2] & 3) << 30),
(p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
p[0] & 1);
}
rc = sbuf_finish(sb);
sbuf_delete(sb);
free(buf, M_CXGBE);
return (rc);
}
static int
sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
u_int i;
struct sbuf *sb;
uint32_t *buf, *p;
int rc;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
if (sb == NULL)
return (ENOMEM);
buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
M_ZERO | M_WAITOK);
t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
p = buf;
sbuf_printf(sb, "Cntl ID DataBE Addr Data");
for (i = 0; i < CIM_MALA_SIZE; i++, p += 6) {
sbuf_printf(sb, "\n %02x %02x %04x %08x %08x%08x%08x%08x",
(p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
p[4], p[3], p[2], p[1], p[0]);
}
sbuf_printf(sb, "\n\nCntl ID Data");
for (i = 0; i < CIM_MALA_SIZE; i++, p += 6) {
sbuf_printf(sb, "\n %02x %02x %08x%08x%08x%08x",
(p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
}
rc = sbuf_finish(sb);
sbuf_delete(sb);
free(buf, M_CXGBE);
return (rc);
}
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
static int
sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc, i;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
uint16_t thres[CIM_NUM_IBQ];
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
if (is_t4(sc)) {
cim_num_obq = CIM_NUM_OBQ;
ibq_rdaddr = A_UP_IBQ_0_RDADDR;
obq_rdaddr = A_UP_OBQ_0_REALADDR;
} else {
cim_num_obq = CIM_NUM_OBQ_T5;
ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
}
nq = CIM_NUM_IBQ + cim_num_obq;
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
if (rc == 0)
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
if (rc != 0)
return (rc);
t4_read_cimq_cfg(sc, base, size, thres);
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
if (sb == NULL)
return (ENOMEM);
sbuf_printf(sb, "Queue Base Size Thres RdPtr WrPtr SOP EOP Avail");
for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
sbuf_printf(sb, "\n%7s %5x %5u %5u %6x %4x %4u %4u %5u",
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
G_QUEREMFLITS(p[2]) * 16);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
for ( ; i < nq; i++, p += 4, wr += 2)
sbuf_printf(sb, "\n%7s %5x %5u %12x %4x %4u %4u %5u", qname[i],
cxgbe(4): Add sysctls to extract debug information from the chip: dev.t4nex.X.misc.cim_la logic analyzer dump dev.t4nex.X.misc.cim_qcfg queue configuration dev.t4nex.X.misc.cim_ibq_xxx inbound queues dev.t4nex.X.misc.cim_obq_xxx outbound queues Obtained from: Chelsio MFC after: 1 week
2013-02-21 20:13:15 +00:00
base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
G_QUEREMFLITS(p[2]) * 16);
rc = sbuf_finish(sb);
sbuf_delete(sb);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (rc);
}
static int
sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc;
struct tp_cpl_stats stats;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
t4_tp_get_cpl_stats(sc, &stats);
sbuf_printf(sb, " channel 0 channel 1 channel 2 "
"channel 3\n");
sbuf_printf(sb, "CPL requests: %10u %10u %10u %10u\n",
stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
sbuf_printf(sb, "CPL responses: %10u %10u %10u %10u",
stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
static int
sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc;
struct tp_usm_stats stats;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return(rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
t4_get_usm_stats(sc, &stats);
sbuf_printf(sb, "Frames: %u\n", stats.frames);
sbuf_printf(sb, "Octets: %ju\n", stats.octets);
sbuf_printf(sb, "Drops: %u", stats.drops);
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
const char *devlog_level_strings[] = {
[FW_DEVLOG_LEVEL_EMERG] = "EMERG",
[FW_DEVLOG_LEVEL_CRIT] = "CRIT",
[FW_DEVLOG_LEVEL_ERR] = "ERR",
[FW_DEVLOG_LEVEL_NOTICE] = "NOTICE",
[FW_DEVLOG_LEVEL_INFO] = "INFO",
[FW_DEVLOG_LEVEL_DEBUG] = "DEBUG"
};
const char *devlog_facility_strings[] = {
[FW_DEVLOG_FACILITY_CORE] = "CORE",
MFC r268706: cxgbe(4): Display CF facility correctly in the device log.
2014-07-18 00:31:06 +00:00
[FW_DEVLOG_FACILITY_CF] = "CF",
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
[FW_DEVLOG_FACILITY_SCHED] = "SCHED",
[FW_DEVLOG_FACILITY_TIMER] = "TIMER",
[FW_DEVLOG_FACILITY_RES] = "RES",
[FW_DEVLOG_FACILITY_HW] = "HW",
[FW_DEVLOG_FACILITY_FLR] = "FLR",
[FW_DEVLOG_FACILITY_DMAQ] = "DMAQ",
[FW_DEVLOG_FACILITY_PHY] = "PHY",
[FW_DEVLOG_FACILITY_MAC] = "MAC",
[FW_DEVLOG_FACILITY_PORT] = "PORT",
[FW_DEVLOG_FACILITY_VI] = "VI",
[FW_DEVLOG_FACILITY_FILTER] = "FILTER",
[FW_DEVLOG_FACILITY_ACL] = "ACL",
[FW_DEVLOG_FACILITY_TM] = "TM",
[FW_DEVLOG_FACILITY_QFC] = "QFC",
[FW_DEVLOG_FACILITY_DCB] = "DCB",
[FW_DEVLOG_FACILITY_ETH] = "ETH",
[FW_DEVLOG_FACILITY_OFLD] = "OFLD",
[FW_DEVLOG_FACILITY_RI] = "RI",
[FW_DEVLOG_FACILITY_ISCSI] = "ISCSI",
[FW_DEVLOG_FACILITY_FCOE] = "FCOE",
[FW_DEVLOG_FACILITY_FOISCSI] = "FOISCSI",
[FW_DEVLOG_FACILITY_FOFCOE] = "FOFCOE"
};
static int
sysctl_devlog(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct devlog_params *dparams = &sc->params.devlog;
struct fw_devlog_e *buf, *e;
MFC r256459. cxgbe(4): Update T4 and T5 firmwares to 1.9.12.0 Approved by: re (glebius)
2013-10-20 15:24:44 +00:00
int i, j, rc, nentries, first = 0, m;
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
struct sbuf *sb;
uint64_t ftstamp = UINT64_MAX;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (dparams->start == 0) {
MFC r256459. cxgbe(4): Update T4 and T5 firmwares to 1.9.12.0 Approved by: re (glebius)
2013-10-20 15:24:44 +00:00
dparams->memtype = FW_MEMTYPE_EDC0;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
dparams->start = 0x84000;
dparams->size = 32768;
}
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
nentries = dparams->size / sizeof(struct fw_devlog_e);
buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
if (buf == NULL)
return (ENOMEM);
MFC r256459. cxgbe(4): Update T4 and T5 firmwares to 1.9.12.0 Approved by: re (glebius)
2013-10-20 15:24:44 +00:00
m = fwmtype_to_hwmtype(dparams->memtype);
rc = -t4_mem_read(sc, m, dparams->start, dparams->size, (void *)buf);
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
if (rc != 0)
goto done;
for (i = 0; i < nentries; i++) {
e = &buf[i];
if (e->timestamp == 0)
break; /* end */
e->timestamp = be64toh(e->timestamp);
e->seqno = be32toh(e->seqno);
for (j = 0; j < 8; j++)
e->params[j] = be32toh(e->params[j]);
if (e->timestamp < ftstamp) {
ftstamp = e->timestamp;
first = i;
}
}
if (buf[first].timestamp == 0)
goto done; /* nothing in the log */
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
goto done;
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (sb == NULL) {
rc = ENOMEM;
goto done;
}
sbuf_printf(sb, "%10s %15s %8s %8s %s\n",
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
"Seq#", "Tstamp", "Level", "Facility", "Message");
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
i = first;
do {
e = &buf[i];
if (e->timestamp == 0)
break; /* end */
sbuf_printf(sb, "%10d %15ju %8s %8s ",
e->seqno, e->timestamp,
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
(e->level < nitems(devlog_level_strings) ?
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
devlog_level_strings[e->level] : "UNKNOWN"),
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
(e->facility < nitems(devlog_facility_strings) ?
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
devlog_facility_strings[e->facility] : "UNKNOWN"));
sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
e->params[2], e->params[3], e->params[4],
e->params[5], e->params[6], e->params[7]);
if (++i == nentries)
i = 0;
} while (i != first);
rc = sbuf_finish(sb);
sbuf_delete(sb);
done:
free(buf, M_CXGBE);
return (rc);
}
static int
sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc;
struct tp_fcoe_stats stats[4];
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
t4_get_fcoe_stats(sc, 0, &stats[0]);
t4_get_fcoe_stats(sc, 1, &stats[1]);
t4_get_fcoe_stats(sc, 2, &stats[2]);
t4_get_fcoe_stats(sc, 3, &stats[3]);
sbuf_printf(sb, " channel 0 channel 1 "
"channel 2 channel 3\n");
sbuf_printf(sb, "octetsDDP: %16ju %16ju %16ju %16ju\n",
stats[0].octetsDDP, stats[1].octetsDDP, stats[2].octetsDDP,
stats[3].octetsDDP);
sbuf_printf(sb, "framesDDP: %16u %16u %16u %16u\n", stats[0].framesDDP,
stats[1].framesDDP, stats[2].framesDDP, stats[3].framesDDP);
sbuf_printf(sb, "framesDrop: %16u %16u %16u %16u",
stats[0].framesDrop, stats[1].framesDrop, stats[2].framesDrop,
stats[3].framesDrop);
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
static int
sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc, i;
unsigned int map, kbps, ipg, mode;
unsigned int pace_tab[NTX_SCHED];
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
t4_read_pace_tbl(sc, pace_tab);
sbuf_printf(sb, "Scheduler Mode Channel Rate (Kbps) "
"Class IPG (0.1 ns) Flow IPG (us)");
for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
t4_get_tx_sched(sc, i, &kbps, &ipg);
sbuf_printf(sb, "\n %u %-5s %u ", i,
(mode & (1 << i)) ? "flow" : "class", map & 3);
if (kbps)
sbuf_printf(sb, "%9u ", kbps);
else
sbuf_printf(sb, " disabled ");
if (ipg)
sbuf_printf(sb, "%13u ", ipg);
else
sbuf_printf(sb, " disabled ");
if (pace_tab[i])
sbuf_printf(sb, "%10u", pace_tab[i]);
else
sbuf_printf(sb, " disabled");
}
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
static int
sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc, i, j;
uint64_t *p0, *p1;
struct lb_port_stats s[2];
static const char *stat_name[] = {
"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
"Frames128To255:", "Frames256To511:", "Frames512To1023:",
"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
"BG2FramesTrunc:", "BG3FramesTrunc:"
};
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
if (sb == NULL)
return (ENOMEM);
memset(s, 0, sizeof(s));
for (i = 0; i < 4; i += 2) {
t4_get_lb_stats(sc, i, &s[0]);
t4_get_lb_stats(sc, i + 1, &s[1]);
p0 = &s[0].octets;
p1 = &s[1].octets;
sbuf_printf(sb, "%s Loopback %u"
" Loopback %u", i == 0 ? "" : "\n", i, i + 1);
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
for (j = 0; j < nitems(stat_name); j++)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
*p0++, *p1++);
}
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
Display a string instead of a numeric code in the linkdnrc sysctl. Submitted by: gnn@
2013-07-27 07:43:43 +00:00
static int
sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
{
int rc = 0;
struct port_info *pi = arg1;
struct sbuf *sb;
static const char *linkdnreasons[] = {
"non-specific", "remote fault", "autoneg failed", "reserved3",
"PHY overheated", "unknown", "rx los", "reserved7"
};
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return(rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
if (sb == NULL)
return (ENOMEM);
if (pi->linkdnrc < 0)
sbuf_printf(sb, "n/a");
else if (pi->linkdnrc < nitems(linkdnreasons))
sbuf_printf(sb, "%s", linkdnreasons[pi->linkdnrc]);
else
sbuf_printf(sb, "%d", pi->linkdnrc);
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
struct mem_desc {
unsigned int base;
unsigned int limit;
unsigned int idx;
};
static int
mem_desc_cmp(const void *a, const void *b)
{
return ((const struct mem_desc *)a)->base -
((const struct mem_desc *)b)->base;
}
static void
mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
unsigned int to)
{
unsigned int size;
size = to - from + 1;
if (size == 0)
return;
/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
}
static int
sysctl_meminfo(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc, i, n;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
uint32_t lo, hi, used, alloc;
static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static const char *region[] = {
"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
"RQUDP region:", "PBL region:", "TXPBL region:",
"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
"On-chip queues:"
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
};
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
struct mem_desc avail[4];
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
struct mem_desc mem[nitems(region) + 3]; /* up to 3 holes */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
struct mem_desc *md = mem;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
if (sb == NULL)
return (ENOMEM);
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
for (i = 0; i < nitems(mem); i++) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
mem[i].limit = 0;
mem[i].idx = i;
}
/* Find and sort the populated memory ranges */
i = 0;
lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
if (lo & F_EDRAM0_ENABLE) {
hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
avail[i].base = G_EDRAM0_BASE(hi) << 20;
avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
avail[i].idx = 0;
i++;
}
if (lo & F_EDRAM1_ENABLE) {
hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
avail[i].base = G_EDRAM1_BASE(hi) << 20;
avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
avail[i].idx = 1;
i++;
}
if (lo & F_EXT_MEM_ENABLE) {
hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
avail[i].base = G_EXT_MEM_BASE(hi) << 20;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
avail[i].limit = avail[i].base +
(G_EXT_MEM_SIZE(hi) << 20);
avail[i].idx = is_t4(sc) ? 2 : 3; /* Call it MC for T4 */
i++;
}
if (!is_t4(sc) && lo & F_EXT_MEM1_ENABLE) {
hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
avail[i].limit = avail[i].base +
(G_EXT_MEM1_SIZE(hi) << 20);
avail[i].idx = 4;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
i++;
}
if (!i) /* no memory available */
return 0;
qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
/* the next few have explicit upper bounds */
md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
md->limit = md->base - 1 +
t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
md++;
md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
md->limit = md->base - 1 +
t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
md++;
if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
hi = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
md->limit = (sc->tids.ntids - hi) * 16 + md->base - 1;
} else {
md->base = 0;
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
md->idx = nitems(region); /* hide it */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
md++;
#define ulp_region(reg) \
md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
ulp_region(RX_ISCSI);
ulp_region(RX_TDDP);
ulp_region(TX_TPT);
ulp_region(RX_STAG);
ulp_region(RX_RQ);
ulp_region(RX_RQUDP);
ulp_region(RX_PBL);
ulp_region(TX_PBL);
#undef ulp_region
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
md->base = 0;
md->idx = nitems(region);
if (!is_t4(sc) && t4_read_reg(sc, A_SGE_CONTROL2) & F_VFIFO_ENABLE) {
md->base = G_BASEADDR(t4_read_reg(sc, A_SGE_DBVFIFO_BADDR));
md->limit = md->base + (G_DBVFIFO_SIZE((t4_read_reg(sc,
A_SGE_DBVFIFO_SIZE))) << 2) - 1;
}
md++;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
md->limit = md->base + sc->tids.ntids - 1;
md++;
md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
md->limit = md->base + sc->tids.ntids - 1;
md++;
md->base = sc->vres.ocq.start;
if (sc->vres.ocq.size)
md->limit = md->base + sc->vres.ocq.size - 1;
else
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
md->idx = nitems(region); /* hide it */
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
md++;
/* add any address-space holes, there can be up to 3 */
for (n = 0; n < i - 1; n++)
if (avail[n].limit < avail[n + 1].base)
(md++)->base = avail[n].limit;
if (avail[n].limit)
(md++)->base = avail[n].limit;
n = md - mem;
qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
for (lo = 0; lo < i; lo++)
mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
avail[lo].limit - 1);
sbuf_printf(sb, "\n");
for (i = 0; i < n; i++) {
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
if (mem[i].idx >= nitems(region))
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
continue; /* skip holes */
if (!mem[i].limit)
mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
mem_region_show(sb, region[mem[i].idx], mem[i].base,
mem[i].limit);
}
sbuf_printf(sb, "\n");
lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
mem_region_show(sb, "uP RAM:", lo, hi);
lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
mem_region_show(sb, "uP Extmem2:", lo, hi);
lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
G_PMRXMAXPAGE(lo),
t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
(lo & F_PMRXNUMCHN) ? 2 : 1);
lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
G_PMTXMAXPAGE(lo),
hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
sbuf_printf(sb, "%u p-structs\n",
t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
for (i = 0; i < 4; i++) {
lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (is_t4(sc)) {
used = G_USED(lo);
alloc = G_ALLOC(lo);
} else {
used = G_T5_USED(lo);
alloc = G_T5_ALLOC(lo);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
i, used, alloc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
for (i = 0; i < 4; i++) {
lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
if (is_t4(sc)) {
used = G_USED(lo);
alloc = G_ALLOC(lo);
} else {
used = G_T5_USED(lo);
alloc = G_T5_ALLOC(lo);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
sbuf_printf(sb,
"\nLoopback %d using %u pages out of %u allocated",
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
i, used, alloc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
static inline void
tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
{
*mask = x | y;
y = htobe64(y);
memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
}
static int
sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc, i, n;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
if (sb == NULL)
return (ENOMEM);
sbuf_printf(sb,
"Idx Ethernet address Mask Vld Ports PF"
" VF Replication P0 P1 P2 P3 ML");
n = is_t4(sc) ? NUM_MPS_CLS_SRAM_L_INSTANCES :
NUM_MPS_T5_CLS_SRAM_L_INSTANCES;
for (i = 0; i < n; i++) {
uint64_t tcamx, tcamy, mask;
uint32_t cls_lo, cls_hi;
uint8_t addr[ETHER_ADDR_LEN];
tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
if (tcamx & tcamy)
continue;
tcamxy2valmask(tcamx, tcamy, addr, &mask);
sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
" %c %#x%4u%4d", i, addr[0], addr[1], addr[2],
addr[3], addr[4], addr[5], (uintmax_t)mask,
(cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
G_PORTMAP(cls_hi), G_PF(cls_lo),
(cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
if (cls_lo & F_REPLICATE) {
struct fw_ldst_cmd ldst_cmd;
memset(&ldst_cmd, 0, sizeof(ldst_cmd));
ldst_cmd.op_to_addrspace =
htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
F_FW_CMD_REQUEST | F_FW_CMD_READ |
V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
ldst_cmd.u.mps.fid_ctl =
htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
V_FW_LDST_CMD_CTL(i));
rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
"t4mps");
if (rc)
break;
rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
sizeof(ldst_cmd), &ldst_cmd);
end_synchronized_op(sc, 0);
if (rc != 0) {
sbuf_printf(sb,
" ------------ error %3u ------------", rc);
rc = 0;
} else {
sbuf_printf(sb, " %08x %08x %08x %08x",
be32toh(ldst_cmd.u.mps.rplc127_96),
be32toh(ldst_cmd.u.mps.rplc95_64),
be32toh(ldst_cmd.u.mps.rplc63_32),
be32toh(ldst_cmd.u.mps.rplc31_0));
}
} else
sbuf_printf(sb, "%36s", "");
sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
}
if (rc)
(void) sbuf_finish(sb);
else
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int
sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc;
uint16_t mtus[NMTUS];
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
t4_read_mtu_tbl(sc, mtus, NULL);
sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
mtus[14], mtus[15]);
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
static int
sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc, i;
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
uint32_t cnt[PM_NSTATS];
uint64_t cyc[PM_NSTATS];
static const char *rx_stats[] = {
"Read:", "Write bypass:", "Write mem:", "Flush:"
};
static const char *tx_stats[] = {
"Read:", "Write bypass:", "Write mem:", "Bypass + mem:"
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
};
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
t4_pmtx_get_stats(sc, cnt, cyc);
sbuf_printf(sb, " Tx pcmds Tx bytes");
for (i = 0; i < ARRAY_SIZE(tx_stats); i++)
sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], cnt[i],
cyc[i]);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
t4_pmrx_get_stats(sc, cnt, cyc);
sbuf_printf(sb, "\n Rx pcmds Rx bytes");
for (i = 0; i < ARRAY_SIZE(rx_stats); i++)
sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], cnt[i],
cyc[i]);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
static int
sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc;
struct tp_rdma_stats stats;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
t4_tp_get_rdma_stats(sc, &stats);
sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
static int
sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc;
struct tp_tcp_stats v4, v6;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
t4_tp_get_tcp_stats(sc, &v4, &v6);
sbuf_printf(sb,
" IP IPv6\n");
sbuf_printf(sb, "OutRsts: %20u %20u\n",
v4.tcpOutRsts, v6.tcpOutRsts);
sbuf_printf(sb, "InSegs: %20ju %20ju\n",
v4.tcpInSegs, v6.tcpInSegs);
sbuf_printf(sb, "OutSegs: %20ju %20ju\n",
v4.tcpOutSegs, v6.tcpOutSegs);
sbuf_printf(sb, "RetransSegs: %20ju %20ju",
v4.tcpRetransSegs, v6.tcpRetransSegs);
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
static int
sysctl_tids(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc;
struct tid_info *t = &sc->tids;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
if (t->natids) {
sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
t->atids_in_use);
}
if (t->ntids) {
if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
uint32_t b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
if (b) {
sbuf_printf(sb, "TID range: 0-%u, %u-%u", b - 1,
t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4,
t->ntids - 1);
} else {
sbuf_printf(sb, "TID range: %u-%u",
t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4,
t->ntids - 1);
}
} else
sbuf_printf(sb, "TID range: 0-%u", t->ntids - 1);
sbuf_printf(sb, ", in use: %u\n",
atomic_load_acq_int(&t->tids_in_use));
}
if (t->nstids) {
sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
t->stid_base + t->nstids - 1, t->stids_in_use);
}
if (t->nftids) {
sbuf_printf(sb, "FTID range: %u-%u\n", t->ftid_base,
t->ftid_base + t->nftids - 1);
}
MFC r259382: Read card capabilities after firmware initialization, instead of setting them up as part of firmware initialization (which the driver gets to do only if it's the master driver). Read the range of tids available for the ETHOFLD functionality if it's enabled. New is_ftid() and is_etid() functions to test whether a tid falls within the range of filter tids or ETHOFLD tids respectively.
2014-05-06 07:21:50 +00:00
if (t->netids) {
sbuf_printf(sb, "ETID range: %u-%u\n", t->etid_base,
t->etid_base + t->netids - 1);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
static int
sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc;
struct tp_err_stats stats;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
t4_tp_get_err_stats(sc, &stats);
sbuf_printf(sb, " channel 0 channel 1 channel 2 "
"channel 3\n");
sbuf_printf(sb, "macInErrs: %10u %10u %10u %10u\n",
stats.macInErrs[0], stats.macInErrs[1], stats.macInErrs[2],
stats.macInErrs[3]);
sbuf_printf(sb, "hdrInErrs: %10u %10u %10u %10u\n",
stats.hdrInErrs[0], stats.hdrInErrs[1], stats.hdrInErrs[2],
stats.hdrInErrs[3]);
sbuf_printf(sb, "tcpInErrs: %10u %10u %10u %10u\n",
stats.tcpInErrs[0], stats.tcpInErrs[1], stats.tcpInErrs[2],
stats.tcpInErrs[3]);
sbuf_printf(sb, "tcp6InErrs: %10u %10u %10u %10u\n",
stats.tcp6InErrs[0], stats.tcp6InErrs[1], stats.tcp6InErrs[2],
stats.tcp6InErrs[3]);
sbuf_printf(sb, "tnlCongDrops: %10u %10u %10u %10u\n",
stats.tnlCongDrops[0], stats.tnlCongDrops[1], stats.tnlCongDrops[2],
stats.tnlCongDrops[3]);
sbuf_printf(sb, "tnlTxDrops: %10u %10u %10u %10u\n",
stats.tnlTxDrops[0], stats.tnlTxDrops[1], stats.tnlTxDrops[2],
stats.tnlTxDrops[3]);
sbuf_printf(sb, "ofldVlanDrops: %10u %10u %10u %10u\n",
stats.ofldVlanDrops[0], stats.ofldVlanDrops[1],
stats.ofldVlanDrops[2], stats.ofldVlanDrops[3]);
sbuf_printf(sb, "ofldChanDrops: %10u %10u %10u %10u\n\n",
stats.ofldChanDrops[0], stats.ofldChanDrops[1],
stats.ofldChanDrops[2], stats.ofldChanDrops[3]);
sbuf_printf(sb, "ofldNoNeigh: %u\nofldCongDefer: %u",
stats.ofldNoNeigh, stats.ofldCongDefer);
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
struct field_desc {
const char *name;
u_int start;
u_int width;
};
static void
field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
{
char buf[32];
int line_size = 0;
while (f->name) {
uint64_t mask = (1ULL << f->width) - 1;
int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
((uintmax_t)v >> f->start) & mask);
if (line_size + len >= 79) {
line_size = 8;
sbuf_printf(sb, "\n ");
}
sbuf_printf(sb, "%s ", buf);
line_size += len + 1;
f++;
}
sbuf_printf(sb, "\n");
}
static struct field_desc tp_la0[] = {
{ "RcfOpCodeOut", 60, 4 },
{ "State", 56, 4 },
{ "WcfState", 52, 4 },
{ "RcfOpcSrcOut", 50, 2 },
{ "CRxError", 49, 1 },
{ "ERxError", 48, 1 },
{ "SanityFailed", 47, 1 },
{ "SpuriousMsg", 46, 1 },
{ "FlushInputMsg", 45, 1 },
{ "FlushInputCpl", 44, 1 },
{ "RssUpBit", 43, 1 },
{ "RssFilterHit", 42, 1 },
{ "Tid", 32, 10 },
{ "InitTcb", 31, 1 },
{ "LineNumber", 24, 7 },
{ "Emsg", 23, 1 },
{ "EdataOut", 22, 1 },
{ "Cmsg", 21, 1 },
{ "CdataOut", 20, 1 },
{ "EreadPdu", 19, 1 },
{ "CreadPdu", 18, 1 },
{ "TunnelPkt", 17, 1 },
{ "RcfPeerFin", 16, 1 },
{ "RcfReasonOut", 12, 4 },
{ "TxCchannel", 10, 2 },
{ "RcfTxChannel", 8, 2 },
{ "RxEchannel", 6, 2 },
{ "RcfRxChannel", 5, 1 },
{ "RcfDataOutSrdy", 4, 1 },
{ "RxDvld", 3, 1 },
{ "RxOoDvld", 2, 1 },
{ "RxCongestion", 1, 1 },
{ "TxCongestion", 0, 1 },
{ NULL }
};
static struct field_desc tp_la1[] = {
{ "CplCmdIn", 56, 8 },
{ "CplCmdOut", 48, 8 },
{ "ESynOut", 47, 1 },
{ "EAckOut", 46, 1 },
{ "EFinOut", 45, 1 },
{ "ERstOut", 44, 1 },
{ "SynIn", 43, 1 },
{ "AckIn", 42, 1 },
{ "FinIn", 41, 1 },
{ "RstIn", 40, 1 },
{ "DataIn", 39, 1 },
{ "DataInVld", 38, 1 },
{ "PadIn", 37, 1 },
{ "RxBufEmpty", 36, 1 },
{ "RxDdp", 35, 1 },
{ "RxFbCongestion", 34, 1 },
{ "TxFbCongestion", 33, 1 },
{ "TxPktSumSrdy", 32, 1 },
{ "RcfUlpType", 28, 4 },
{ "Eread", 27, 1 },
{ "Ebypass", 26, 1 },
{ "Esave", 25, 1 },
{ "Static0", 24, 1 },
{ "Cread", 23, 1 },
{ "Cbypass", 22, 1 },
{ "Csave", 21, 1 },
{ "CPktOut", 20, 1 },
{ "RxPagePoolFull", 18, 2 },
{ "RxLpbkPkt", 17, 1 },
{ "TxLpbkPkt", 16, 1 },
{ "RxVfValid", 15, 1 },
{ "SynLearned", 14, 1 },
{ "SetDelEntry", 13, 1 },
{ "SetInvEntry", 12, 1 },
{ "CpcmdDvld", 11, 1 },
{ "CpcmdSave", 10, 1 },
{ "RxPstructsFull", 8, 2 },
{ "EpcmdDvld", 7, 1 },
{ "EpcmdFlush", 6, 1 },
{ "EpcmdTrimPrefix", 5, 1 },
{ "EpcmdTrimPostfix", 4, 1 },
{ "ERssIp4Pkt", 3, 1 },
{ "ERssIp6Pkt", 2, 1 },
{ "ERssTcpUdpPkt", 1, 1 },
{ "ERssFceFipPkt", 0, 1 },
{ NULL }
};
static struct field_desc tp_la2[] = {
{ "CplCmdIn", 56, 8 },
{ "MpsVfVld", 55, 1 },
{ "MpsPf", 52, 3 },
{ "MpsVf", 44, 8 },
{ "SynIn", 43, 1 },
{ "AckIn", 42, 1 },
{ "FinIn", 41, 1 },
{ "RstIn", 40, 1 },
{ "DataIn", 39, 1 },
{ "DataInVld", 38, 1 },
{ "PadIn", 37, 1 },
{ "RxBufEmpty", 36, 1 },
{ "RxDdp", 35, 1 },
{ "RxFbCongestion", 34, 1 },
{ "TxFbCongestion", 33, 1 },
{ "TxPktSumSrdy", 32, 1 },
{ "RcfUlpType", 28, 4 },
{ "Eread", 27, 1 },
{ "Ebypass", 26, 1 },
{ "Esave", 25, 1 },
{ "Static0", 24, 1 },
{ "Cread", 23, 1 },
{ "Cbypass", 22, 1 },
{ "Csave", 21, 1 },
{ "CPktOut", 20, 1 },
{ "RxPagePoolFull", 18, 2 },
{ "RxLpbkPkt", 17, 1 },
{ "TxLpbkPkt", 16, 1 },
{ "RxVfValid", 15, 1 },
{ "SynLearned", 14, 1 },
{ "SetDelEntry", 13, 1 },
{ "SetInvEntry", 12, 1 },
{ "CpcmdDvld", 11, 1 },
{ "CpcmdSave", 10, 1 },
{ "RxPstructsFull", 8, 2 },
{ "EpcmdDvld", 7, 1 },
{ "EpcmdFlush", 6, 1 },
{ "EpcmdTrimPrefix", 5, 1 },
{ "EpcmdTrimPostfix", 4, 1 },
{ "ERssIp4Pkt", 3, 1 },
{ "ERssIp6Pkt", 2, 1 },
{ "ERssTcpUdpPkt", 1, 1 },
{ "ERssFceFipPkt", 0, 1 },
{ NULL }
};
static void
tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
{
field_desc_show(sb, *p, tp_la0);
}
static void
tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
{
if (idx)
sbuf_printf(sb, "\n");
field_desc_show(sb, p[0], tp_la0);
if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
field_desc_show(sb, p[1], tp_la0);
}
static void
tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
{
if (idx)
sbuf_printf(sb, "\n");
field_desc_show(sb, p[0], tp_la0);
if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
}
static int
sysctl_tp_la(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
uint64_t *buf, *p;
int rc;
u_int i, inc;
void (*show_func)(struct sbuf *, uint64_t *, int);
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
if (sb == NULL)
return (ENOMEM);
buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
t4_tp_read_la(sc, buf, NULL);
p = buf;
switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
case 2:
inc = 2;
show_func = tp_la_show2;
break;
case 3:
inc = 2;
show_func = tp_la_show3;
break;
default:
inc = 1;
show_func = tp_la_show;
}
for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
(*show_func)(sb, p, i);
rc = sbuf_finish(sb);
sbuf_delete(sb);
free(buf, M_CXGBE);
return (rc);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int
sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc;
u64 nrate[NCHAN], orate[NCHAN];
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
if (sb == NULL)
return (ENOMEM);
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
t4_get_chan_txrate(sc, nrate, orate);
sbuf_printf(sb, " channel 0 channel 1 channel 2 "
"channel 3\n");
sbuf_printf(sb, "NIC B/s: %10ju %10ju %10ju %10ju\n",
nrate[0], nrate[1], nrate[2], nrate[3]);
sbuf_printf(sb, "Offload B/s: %10ju %10ju %10ju %10ju",
orate[0], orate[1], orate[2], orate[3]);
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
rc = sbuf_finish(sb);
sbuf_delete(sb);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
return (rc);
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
cxgbe(4): Some more debug sysctls. These work on both T4 and T5 based cards. dev.t5nex.0.misc.cim_ma_la: CIM MA logic analyzer dev.t5nex.0.misc.cim_pif_la: CIM PIF logic analyzer dev.t5nex.0.misc.mps_tcam: MPS TCAM entries dev.t5nex.0.misc.tp_la: TP logic analyzer dev.t5nex.0.misc.ulprx_la: ULPRX logic analyzer Obtained from: Chelsio MFC after: 1 week
2013-06-01 02:07:37 +00:00
static int
sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct sbuf *sb;
uint32_t *buf, *p;
int rc, i;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
if (sb == NULL)
return (ENOMEM);
buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
M_ZERO | M_WAITOK);
t4_ulprx_read_la(sc, buf);
p = buf;
sbuf_printf(sb, " Pcmd Type Message"
" Data");
for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
sbuf_printf(sb, "\n%08x%08x %4x %08x %08x%08x%08x%08x",
p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
}
rc = sbuf_finish(sb);
sbuf_delete(sb);
free(buf, M_CXGBE);
return (rc);
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static int
Cosmetic change (s/wrwc/wcwr/;s/WRWC/WCWR/). MFC after: 3 days.
2013-04-11 22:49:29 +00:00
sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
{
struct adapter *sc = arg1;
struct sbuf *sb;
int rc, v;
rc = sysctl_wire_old_buffer(req, 0);
if (rc != 0)
return (rc);
sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
if (sb == NULL)
return (ENOMEM);
v = t4_read_reg(sc, A_SGE_STAT_CFG);
if (G_STATSOURCE_T5(v) == 7) {
if (G_STATMODE(v) == 0) {
Get rid of a couple of stray \n's. MFC after: 3 days.
2013-04-11 21:17:49 +00:00
sbuf_printf(sb, "total %d, incomplete %d",
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
t4_read_reg(sc, A_SGE_STAT_TOTAL),
t4_read_reg(sc, A_SGE_STAT_MATCH));
} else if (G_STATMODE(v) == 1) {
Get rid of a couple of stray \n's. MFC after: 3 days.
2013-04-11 21:17:49 +00:00
sbuf_printf(sb, "total %d, data overflow %d",
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
t4_read_reg(sc, A_SGE_STAT_TOTAL),
t4_read_reg(sc, A_SGE_STAT_MATCH));
}
}
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
cxgbe: reduce diffs with other branches. Will help future MFCs from HEAD. MFC after: 3 days
2012-02-07 06:21:59 +00:00
#endif
Firmware device log. # sysctl dev.t4nex.0.devlog MFC after: mdf's sysctl+sbuf changes are MFC'd
2011-05-31 23:49:13 +00:00
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
static uint32_t
fconf_to_mode(uint32_t fconf)
{
uint32_t mode;
mode = T4_FILTER_IPv4 | T4_FILTER_IPv6 | T4_FILTER_IP_SADDR |
T4_FILTER_IP_DADDR | T4_FILTER_IP_SPORT | T4_FILTER_IP_DPORT;
if (fconf & F_FRAGMENTATION)
mode |= T4_FILTER_IP_FRAGMENT;
if (fconf & F_MPSHITTYPE)
mode |= T4_FILTER_MPS_HIT_TYPE;
if (fconf & F_MACMATCH)
mode |= T4_FILTER_MAC_IDX;
if (fconf & F_ETHERTYPE)
mode |= T4_FILTER_ETH_TYPE;
if (fconf & F_PROTOCOL)
mode |= T4_FILTER_IP_PROTO;
if (fconf & F_TOS)
mode |= T4_FILTER_IP_TOS;
if (fconf & F_VLAN)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
mode |= T4_FILTER_VLAN;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
if (fconf & F_VNIC_ID)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
mode |= T4_FILTER_VNIC;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
if (fconf & F_PORT)
mode |= T4_FILTER_PORT;
if (fconf & F_FCOE)
mode |= T4_FILTER_FCoE;
return (mode);
}
static uint32_t
mode_to_fconf(uint32_t mode)
{
uint32_t fconf = 0;
if (mode & T4_FILTER_IP_FRAGMENT)
fconf |= F_FRAGMENTATION;
if (mode & T4_FILTER_MPS_HIT_TYPE)
fconf |= F_MPSHITTYPE;
if (mode & T4_FILTER_MAC_IDX)
fconf |= F_MACMATCH;
if (mode & T4_FILTER_ETH_TYPE)
fconf |= F_ETHERTYPE;
if (mode & T4_FILTER_IP_PROTO)
fconf |= F_PROTOCOL;
if (mode & T4_FILTER_IP_TOS)
fconf |= F_TOS;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (mode & T4_FILTER_VLAN)
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
fconf |= F_VLAN;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (mode & T4_FILTER_VNIC)
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
fconf |= F_VNIC_ID;
if (mode & T4_FILTER_PORT)
fconf |= F_PORT;
if (mode & T4_FILTER_FCoE)
fconf |= F_FCOE;
return (fconf);
}
static uint32_t
fspec_to_fconf(struct t4_filter_specification *fs)
{
uint32_t fconf = 0;
if (fs->val.frag || fs->mask.frag)
fconf |= F_FRAGMENTATION;
if (fs->val.matchtype || fs->mask.matchtype)
fconf |= F_MPSHITTYPE;
if (fs->val.macidx || fs->mask.macidx)
fconf |= F_MACMATCH;
if (fs->val.ethtype || fs->mask.ethtype)
fconf |= F_ETHERTYPE;
if (fs->val.proto || fs->mask.proto)
fconf |= F_PROTOCOL;
if (fs->val.tos || fs->mask.tos)
fconf |= F_TOS;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (fs->val.vlan_vld || fs->mask.vlan_vld)
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
fconf |= F_VLAN;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (fs->val.vnic_vld || fs->mask.vnic_vld)
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
fconf |= F_VNIC_ID;
if (fs->val.iport || fs->mask.iport)
fconf |= F_PORT;
if (fs->val.fcoe || fs->mask.fcoe)
fconf |= F_FCOE;
return (fconf);
}
static int
get_filter_mode(struct adapter *sc, uint32_t *mode)
{
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
int rc;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
uint32_t fconf;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
"t4getfm");
if (rc)
return (rc);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
t4_read_indirect(sc, A_TP_PIO_ADDR, A_TP_PIO_DATA, &fconf, 1,
A_TP_VLAN_PRI_MAP);
- Read all TP parameters in one place. - Read the filter mode, calculate various shifts, and use them properly during active open (in select_ntuple). MFC after: 1 day
2013-07-04 17:55:52 +00:00
if (sc->params.tp.vlan_pri_map != fconf) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
log(LOG_WARNING, "%s: cached filter mode out of sync %x %x.\n",
- Read all TP parameters in one place. - Read the filter mode, calculate various shifts, and use them properly during active open (in select_ntuple). MFC after: 1 day
2013-07-04 17:55:52 +00:00
device_get_nameunit(sc->dev), sc->params.tp.vlan_pri_map,
fconf);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
MFC r278485: cxgbe(4): allow the SET_FILTER_MODE ioctl to change the mode when it's safe to do so.
2015-04-08 03:55:02 +00:00
*mode = fconf_to_mode(fconf);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
end_synchronized_op(sc, LOCK_HELD);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
return (0);
}
static int
set_filter_mode(struct adapter *sc, uint32_t mode)
{
uint32_t fconf;
int rc;
fconf = mode_to_fconf(mode);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
"t4setfm");
if (rc)
return (rc);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
if (sc->tids.ftids_in_use > 0) {
rc = EBUSY;
goto done;
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
if (uld_active(sc, ULD_TOM)) {
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
rc = EBUSY;
goto done;
}
#endif
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
rc = -t4_set_filter_mode(sc, fconf);
done:
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
end_synchronized_op(sc, LOCK_HELD);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
return (rc);
}
Provide hit-count with rest of the information about a filter. MFC after: 1 week
2011-06-01 01:32:58 +00:00
static inline uint64_t
get_filter_hits(struct adapter *sc, uint32_t fid)
{
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
uint32_t mw_base, off, tcb_base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
Provide hit-count with rest of the information about a filter. MFC after: 1 week
2011-06-01 01:32:58 +00:00
uint64_t hits;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
memwin_info(sc, 0, &mw_base, NULL);
off = position_memwin(sc, 0,
Provide hit-count with rest of the information about a filter. MFC after: 1 week
2011-06-01 01:32:58 +00:00
tcb_base + (fid + sc->tids.ftid_base) * TCB_SIZE);
cxgbe(4): Provide accurate hit count for filters on T5 cards. The location within the TCB and the size have both changed. MFC after: 1 week
2013-06-04 02:25:25 +00:00
if (is_t4(sc)) {
hits = t4_read_reg64(sc, mw_base + off + 16);
hits = be64toh(hits);
} else {
hits = t4_read_reg(sc, mw_base + off + 24);
hits = be32toh(hits);
}
Provide hit-count with rest of the information about a filter. MFC after: 1 week
2011-06-01 01:32:58 +00:00
cxgbe(4): Provide accurate hit count for filters on T5 cards. The location within the TCB and the size have both changed. MFC after: 1 week
2013-06-04 02:25:25 +00:00
return (hits);
Provide hit-count with rest of the information about a filter. MFC after: 1 week
2011-06-01 01:32:58 +00:00
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
static int
get_filter(struct adapter *sc, struct t4_filter *t)
{
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
int i, rc, nfilters = sc->tids.nftids;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
struct filter_entry *f;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, NULL, HOLD_LOCK | SLEEP_OK | INTR_OK,
"t4getf");
if (rc)
return (rc);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
if (sc->tids.ftids_in_use == 0 || sc->tids.ftid_tab == NULL ||
t->idx >= nfilters) {
t->idx = 0xffffffff;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
goto done;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
f = &sc->tids.ftid_tab[t->idx];
for (i = t->idx; i < nfilters; i++, f++) {
if (f->valid) {
t->idx = i;
L2 table code. This is enough to get the T4's switch + L2 rewrite filters working. (All other filters - switch without L2 info rewrite, steer, and drop - were already fully-functional). Some contrived examples of "switch" filters with L2 rewriting: # cxgbetool t4nex0 iport 0 dport 80 action switch vlan +9 eport 3 Intercept all packets received on physical port 0 with TCP port 80 as destination, insert a vlan tag with VID 9, and send them out of port 3. # cxgbetool t4nex0 sip 192.168.1.1/32 ivlan 5 action switch \ vlan =9 smac aa:bb:cc:dd:ee:ff eport 0 Intercept all packets (received on any port) with source IP address 192.168.1.1 and VLAN id 5, rewrite the VLAN id to 9, rewrite source mac to aa:bb:cc:dd:ee:ff, and send it out of port 0. MFC after: 1 week
2011-05-30 21:07:26 +00:00
t->l2tidx = f->l2t ? f->l2t->idx : 0;
t->smtidx = f->smtidx;
Provide hit-count with rest of the information about a filter. MFC after: 1 week
2011-06-01 01:32:58 +00:00
if (f->fs.hitcnts)
t->hits = get_filter_hits(sc, t->idx);
else
t->hits = UINT64_MAX;
L2 table code. This is enough to get the T4's switch + L2 rewrite filters working. (All other filters - switch without L2 info rewrite, steer, and drop - were already fully-functional). Some contrived examples of "switch" filters with L2 rewriting: # cxgbetool t4nex0 iport 0 dport 80 action switch vlan +9 eport 3 Intercept all packets received on physical port 0 with TCP port 80 as destination, insert a vlan tag with VID 9, and send them out of port 3. # cxgbetool t4nex0 sip 192.168.1.1/32 ivlan 5 action switch \ vlan =9 smac aa:bb:cc:dd:ee:ff eport 0 Intercept all packets (received on any port) with source IP address 192.168.1.1 and VLAN id 5, rewrite the VLAN id to 9, rewrite source mac to aa:bb:cc:dd:ee:ff, and send it out of port 0. MFC after: 1 week
2011-05-30 21:07:26 +00:00
t->fs = f->fs;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
goto done;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
}
t->idx = 0xffffffff;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
done:
end_synchronized_op(sc, LOCK_HELD);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
return (0);
}
static int
set_filter(struct adapter *sc, struct t4_filter *t)
{
unsigned int nfilters, nports;
struct filter_entry *f;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
int i, rc;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setf");
if (rc)
return (rc);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
nfilters = sc->tids.nftids;
nports = sc->params.nports;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (nfilters == 0) {
rc = ENOTSUP;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (!(sc->flags & FULL_INIT_DONE)) {
rc = EAGAIN;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (t->idx >= nfilters) {
rc = EINVAL;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
/* Validate against the global filter mode */
- Read all TP parameters in one place. - Read the filter mode, calculate various shifts, and use them properly during active open (in select_ntuple). MFC after: 1 day
2013-07-04 17:55:52 +00:00
if ((sc->params.tp.vlan_pri_map | fspec_to_fconf(&t->fs)) !=
sc->params.tp.vlan_pri_map) {
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = E2BIG;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (t->fs.action == FILTER_SWITCH && t->fs.eport >= nports) {
rc = EINVAL;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (t->fs.val.iport >= nports) {
rc = EINVAL;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
/* Can't specify an iq if not steering to it */
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (!t->fs.dirsteer && t->fs.iq) {
rc = EINVAL;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
/* IPv6 filter idx must be 4 aligned */
if (t->fs.type == 1 &&
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
((t->idx & 0x3) || t->idx + 4 >= nfilters)) {
rc = EINVAL;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
if (sc->tids.ftid_tab == NULL) {
KASSERT(sc->tids.ftids_in_use == 0,
("%s: no memory allocated but filters_in_use > 0",
__func__));
sc->tids.ftid_tab = malloc(sizeof (struct filter_entry) *
nfilters, M_CXGBE, M_NOWAIT | M_ZERO);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (sc->tids.ftid_tab == NULL) {
rc = ENOMEM;
goto done;
}
mtx_init(&sc->tids.ftid_lock, "T4 filters", 0, MTX_DEF);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
for (i = 0; i < 4; i++) {
f = &sc->tids.ftid_tab[t->idx + i];
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (f->pending || f->valid) {
rc = EBUSY;
goto done;
}
if (f->locked) {
rc = EPERM;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
if (t->fs.type == 0)
break;
}
f = &sc->tids.ftid_tab[t->idx];
f->fs = t->fs;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = set_filter_wr(sc, t->idx);
done:
end_synchronized_op(sc, 0);
if (rc == 0) {
mtx_lock(&sc->tids.ftid_lock);
for (;;) {
if (f->pending == 0) {
rc = f->valid ? 0 : EIO;
break;
}
if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
PCATCH, "t4setfw", 0)) {
rc = EINPROGRESS;
break;
}
}
mtx_unlock(&sc->tids.ftid_lock);
}
return (rc);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
static int
del_filter(struct adapter *sc, struct t4_filter *t)
{
unsigned int nfilters;
struct filter_entry *f;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
int rc;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4delf");
if (rc)
return (rc);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
nfilters = sc->tids.nftids;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (nfilters == 0) {
rc = ENOTSUP;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
if (sc->tids.ftid_tab == NULL || sc->tids.ftids_in_use == 0 ||
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
t->idx >= nfilters) {
rc = EINVAL;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (!(sc->flags & FULL_INIT_DONE)) {
rc = EAGAIN;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
f = &sc->tids.ftid_tab[t->idx];
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
if (f->pending) {
rc = EBUSY;
goto done;
}
if (f->locked) {
rc = EPERM;
goto done;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
if (f->valid) {
t->fs = f->fs; /* extra info for the caller */
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = del_filter_wr(sc, t->idx);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
done:
end_synchronized_op(sc, 0);
if (rc == 0) {
mtx_lock(&sc->tids.ftid_lock);
for (;;) {
if (f->pending == 0) {
rc = f->valid ? EIO : 0;
break;
}
if (mtx_sleep(&sc->tids.ftid_tab, &sc->tids.ftid_lock,
PCATCH, "t4delfw", 0)) {
rc = EINPROGRESS;
break;
}
}
mtx_unlock(&sc->tids.ftid_lock);
}
return (rc);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
static void
L2 table code. This is enough to get the T4's switch + L2 rewrite filters working. (All other filters - switch without L2 info rewrite, steer, and drop - were already fully-functional). Some contrived examples of "switch" filters with L2 rewriting: # cxgbetool t4nex0 iport 0 dport 80 action switch vlan +9 eport 3 Intercept all packets received on physical port 0 with TCP port 80 as destination, insert a vlan tag with VID 9, and send them out of port 3. # cxgbetool t4nex0 sip 192.168.1.1/32 ivlan 5 action switch \ vlan =9 smac aa:bb:cc:dd:ee:ff eport 0 Intercept all packets (received on any port) with source IP address 192.168.1.1 and VLAN id 5, rewrite the VLAN id to 9, rewrite source mac to aa:bb:cc:dd:ee:ff, and send it out of port 0. MFC after: 1 week
2011-05-30 21:07:26 +00:00
clear_filter(struct filter_entry *f)
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
{
L2 table code. This is enough to get the T4's switch + L2 rewrite filters working. (All other filters - switch without L2 info rewrite, steer, and drop - were already fully-functional). Some contrived examples of "switch" filters with L2 rewriting: # cxgbetool t4nex0 iport 0 dport 80 action switch vlan +9 eport 3 Intercept all packets received on physical port 0 with TCP port 80 as destination, insert a vlan tag with VID 9, and send them out of port 3. # cxgbetool t4nex0 sip 192.168.1.1/32 ivlan 5 action switch \ vlan =9 smac aa:bb:cc:dd:ee:ff eport 0 Intercept all packets (received on any port) with source IP address 192.168.1.1 and VLAN id 5, rewrite the VLAN id to 9, rewrite source mac to aa:bb:cc:dd:ee:ff, and send it out of port 0. MFC after: 1 week
2011-05-30 21:07:26 +00:00
if (f->l2t)
t4_l2t_release(f->l2t);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
bzero(f, sizeof (*f));
}
static int
set_filter_wr(struct adapter *sc, int fidx)
{
struct filter_entry *f = &sc->tids.ftid_tab[fidx];
struct fw_filter_wr *fwr;
unsigned int ftid;
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
struct wrq_cookie cookie;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
ASSERT_SYNCHRONIZED_OP(sc);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
L2 table code. This is enough to get the T4's switch + L2 rewrite filters working. (All other filters - switch without L2 info rewrite, steer, and drop - were already fully-functional). Some contrived examples of "switch" filters with L2 rewriting: # cxgbetool t4nex0 iport 0 dport 80 action switch vlan +9 eport 3 Intercept all packets received on physical port 0 with TCP port 80 as destination, insert a vlan tag with VID 9, and send them out of port 3. # cxgbetool t4nex0 sip 192.168.1.1/32 ivlan 5 action switch \ vlan =9 smac aa:bb:cc:dd:ee:ff eport 0 Intercept all packets (received on any port) with source IP address 192.168.1.1 and VLAN id 5, rewrite the VLAN id to 9, rewrite source mac to aa:bb:cc:dd:ee:ff, and send it out of port 0. MFC after: 1 week
2011-05-30 21:07:26 +00:00
if (f->fs.newdmac || f->fs.newvlan) {
/* This filter needs an L2T entry; allocate one. */
f->l2t = t4_l2t_alloc_switching(sc->l2t);
if (f->l2t == NULL)
return (EAGAIN);
if (t4_l2t_set_switching(sc, f->l2t, f->fs.vlan, f->fs.eport,
f->fs.dmac)) {
t4_l2t_release(f->l2t);
f->l2t = NULL;
return (ENOMEM);
}
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
ftid = sc->tids.ftid_base + fidx;
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
if (fwr == NULL)
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
return (ENOMEM);
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
bzero(fwr, sizeof(*fwr));
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
fwr->op_pkd = htobe32(V_FW_WR_OP(FW_FILTER_WR));
fwr->len16_pkd = htobe32(FW_LEN16(*fwr));
fwr->tid_to_iq =
htobe32(V_FW_FILTER_WR_TID(ftid) |
V_FW_FILTER_WR_RQTYPE(f->fs.type) |
V_FW_FILTER_WR_NOREPLY(0) |
V_FW_FILTER_WR_IQ(f->fs.iq));
fwr->del_filter_to_l2tix =
htobe32(V_FW_FILTER_WR_RPTTID(f->fs.rpttid) |
V_FW_FILTER_WR_DROP(f->fs.action == FILTER_DROP) |
V_FW_FILTER_WR_DIRSTEER(f->fs.dirsteer) |
V_FW_FILTER_WR_MASKHASH(f->fs.maskhash) |
V_FW_FILTER_WR_DIRSTEERHASH(f->fs.dirsteerhash) |
V_FW_FILTER_WR_LPBK(f->fs.action == FILTER_SWITCH) |
V_FW_FILTER_WR_DMAC(f->fs.newdmac) |
V_FW_FILTER_WR_SMAC(f->fs.newsmac) |
V_FW_FILTER_WR_INSVLAN(f->fs.newvlan == VLAN_INSERT ||
f->fs.newvlan == VLAN_REWRITE) |
V_FW_FILTER_WR_RMVLAN(f->fs.newvlan == VLAN_REMOVE ||
f->fs.newvlan == VLAN_REWRITE) |
V_FW_FILTER_WR_HITCNTS(f->fs.hitcnts) |
V_FW_FILTER_WR_TXCHAN(f->fs.eport) |
V_FW_FILTER_WR_PRIO(f->fs.prio) |
L2 table code. This is enough to get the T4's switch + L2 rewrite filters working. (All other filters - switch without L2 info rewrite, steer, and drop - were already fully-functional). Some contrived examples of "switch" filters with L2 rewriting: # cxgbetool t4nex0 iport 0 dport 80 action switch vlan +9 eport 3 Intercept all packets received on physical port 0 with TCP port 80 as destination, insert a vlan tag with VID 9, and send them out of port 3. # cxgbetool t4nex0 sip 192.168.1.1/32 ivlan 5 action switch \ vlan =9 smac aa:bb:cc:dd:ee:ff eport 0 Intercept all packets (received on any port) with source IP address 192.168.1.1 and VLAN id 5, rewrite the VLAN id to 9, rewrite source mac to aa:bb:cc:dd:ee:ff, and send it out of port 0. MFC after: 1 week
2011-05-30 21:07:26 +00:00
V_FW_FILTER_WR_L2TIX(f->l2t ? f->l2t->idx : 0));
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
fwr->ethtype = htobe16(f->fs.val.ethtype);
fwr->ethtypem = htobe16(f->fs.mask.ethtype);
fwr->frag_to_ovlan_vldm =
(V_FW_FILTER_WR_FRAG(f->fs.val.frag) |
V_FW_FILTER_WR_FRAGM(f->fs.mask.frag) |
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
V_FW_FILTER_WR_IVLAN_VLD(f->fs.val.vlan_vld) |
V_FW_FILTER_WR_OVLAN_VLD(f->fs.val.vnic_vld) |
V_FW_FILTER_WR_IVLAN_VLDM(f->fs.mask.vlan_vld) |
V_FW_FILTER_WR_OVLAN_VLDM(f->fs.mask.vnic_vld));
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
fwr->smac_sel = 0;
fwr->rx_chan_rx_rpl_iq = htobe16(V_FW_FILTER_WR_RX_CHAN(0) |
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
V_FW_FILTER_WR_RX_RPL_IQ(sc->sge.fwq.abs_id));
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
fwr->maci_to_matchtypem =
htobe32(V_FW_FILTER_WR_MACI(f->fs.val.macidx) |
V_FW_FILTER_WR_MACIM(f->fs.mask.macidx) |
V_FW_FILTER_WR_FCOE(f->fs.val.fcoe) |
V_FW_FILTER_WR_FCOEM(f->fs.mask.fcoe) |
V_FW_FILTER_WR_PORT(f->fs.val.iport) |
V_FW_FILTER_WR_PORTM(f->fs.mask.iport) |
V_FW_FILTER_WR_MATCHTYPE(f->fs.val.matchtype) |
V_FW_FILTER_WR_MATCHTYPEM(f->fs.mask.matchtype));
fwr->ptcl = f->fs.val.proto;
fwr->ptclm = f->fs.mask.proto;
fwr->ttyp = f->fs.val.tos;
fwr->ttypm = f->fs.mask.tos;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
fwr->ivlan = htobe16(f->fs.val.vlan);
fwr->ivlanm = htobe16(f->fs.mask.vlan);
fwr->ovlan = htobe16(f->fs.val.vnic);
fwr->ovlanm = htobe16(f->fs.mask.vnic);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
bcopy(f->fs.val.dip, fwr->lip, sizeof (fwr->lip));
bcopy(f->fs.mask.dip, fwr->lipm, sizeof (fwr->lipm));
bcopy(f->fs.val.sip, fwr->fip, sizeof (fwr->fip));
bcopy(f->fs.mask.sip, fwr->fipm, sizeof (fwr->fipm));
fwr->lp = htobe16(f->fs.val.dport);
fwr->lpm = htobe16(f->fs.mask.dport);
fwr->fp = htobe16(f->fs.val.sport);
fwr->fpm = htobe16(f->fs.mask.sport);
if (f->fs.newsmac)
bcopy(f->fs.smac, fwr->sma, sizeof (fwr->sma));
f->pending = 1;
sc->tids.ftids_in_use++;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (0);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
static int
del_filter_wr(struct adapter *sc, int fidx)
{
struct filter_entry *f = &sc->tids.ftid_tab[fidx];
struct fw_filter_wr *fwr;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
unsigned int ftid;
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
struct wrq_cookie cookie;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
ftid = sc->tids.ftid_base + fidx;
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
fwr = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*fwr), 16), &cookie);
if (fwr == NULL)
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
return (ENOMEM);
bzero(fwr, sizeof (*fwr));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
t4_mk_filtdelwr(ftid, fwr, sc->sge.fwq.abs_id);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
f->pending = 1;
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
commit_wrq_wr(&sc->sge.mgmtq, fwr, &cookie);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (0);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
Add a routine (t4_set_tcb_field) to update arbitrary parts of a hardware TCB. Filters are programmed by modifying the TCB too (via a different routine) and the reply to any TCB update is delivered via a CPL_SET_TCB_RPL. Figure out whether the reply is for a filter-write or something else and route it appropriately. MFC after: 2 weeks
2012-08-16 20:15:29 +00:00
int
t4_filter_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
{
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
struct adapter *sc = iq->adapter;
const struct cpl_set_tcb_rpl *rpl = (const void *)(rss + 1);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
unsigned int idx = GET_TID(rpl);
MFC r259382: Read card capabilities after firmware initialization, instead of setting them up as part of firmware initialization (which the driver gets to do only if it's the master driver). Read the range of tids available for the ETHOFLD functionality if it's enabled. New is_ftid() and is_etid() functions to test whether a tid falls within the range of filter tids or ETHOFLD tids respectively.
2014-05-06 07:21:50 +00:00
unsigned int rc;
struct filter_entry *f;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
KASSERT(m == NULL, ("%s: payload with opcode %02x", __func__,
rss->opcode));
MFC r259382: Read card capabilities after firmware initialization, instead of setting them up as part of firmware initialization (which the driver gets to do only if it's the master driver). Read the range of tids available for the ETHOFLD functionality if it's enabled. New is_ftid() and is_etid() functions to test whether a tid falls within the range of filter tids or ETHOFLD tids respectively.
2014-05-06 07:21:50 +00:00
if (is_ftid(sc, idx)) {
idx -= sc->tids.ftid_base;
f = &sc->tids.ftid_tab[idx];
rc = G_COOKIE(rpl->cookie);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
mtx_lock(&sc->tids.ftid_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (rc == FW_FILTER_WR_FLT_ADDED) {
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
KASSERT(f->pending, ("%s: filter[%u] isn't pending.",
__func__, idx));
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
f->smtidx = (be64toh(rpl->oldval) >> 24) & 0xff;
f->pending = 0; /* asynchronous setup completed */
f->valid = 1;
Acquire the adapter lock before updating fields of the filter structure. Submitted by: gnn (different version) MFC after: 3 days
2012-02-07 09:39:46 +00:00
} else {
if (rc != FW_FILTER_WR_FLT_DELETED) {
/* Add or delete failed, display an error */
log(LOG_ERR,
"filter %u setup failed with error %u\n",
idx, rc);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
Acquire the adapter lock before updating fields of the filter structure. Submitted by: gnn (different version) MFC after: 3 days
2012-02-07 09:39:46 +00:00
clear_filter(f);
sc->tids.ftids_in_use--;
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
wakeup(&sc->tids.ftid_tab);
mtx_unlock(&sc->tids.ftid_lock);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (0);
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
}
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
static int
get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
{
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
int rc;
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
if (cntxt->cid > M_CTXTQID)
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
return (EINVAL);
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
return (EINVAL);
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
Do not hold locks around hardware context reads. MFC after: 3 days
2013-02-09 00:35:28 +00:00
rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
if (rc)
return (rc);
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
if (sc->flags & FW_OK) {
Do not hold locks around hardware context reads. MFC after: 3 days
2013-02-09 00:35:28 +00:00
rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
&cntxt->data[0]);
if (rc == 0)
goto done;
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
}
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
/*
* Read via firmware failed or wasn't even attempted. Read directly via
* the backdoor.
*/
Do not hold locks around hardware context reads. MFC after: 3 days
2013-02-09 00:35:28 +00:00
rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
done:
end_synchronized_op(sc, 0);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
return (rc);
}
static int
load_fw(struct adapter *sc, struct t4_data *fw)
{
int rc;
uint8_t *fw_data;
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
if (rc)
return (rc);
if (sc->flags & FULL_INIT_DONE) {
rc = EBUSY;
goto done;
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
}
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
if (fw_data == NULL) {
rc = ENOMEM;
goto done;
}
rc = copyin(fw->data, fw_data, fw->len);
if (rc == 0)
rc = -t4_load_fw(sc, fw_data, fw->len);
free(fw_data, M_CXGBE);
done:
end_synchronized_op(sc, 0);
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
return (rc);
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
{
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
uint32_t addr, off, remaining, i, n;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
uint32_t *buf, *b;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
uint32_t mw_base, mw_aperture;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
int rc;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
uint8_t *dst;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
rc = validate_mem_range(sc, mr->addr, mr->len);
if (rc != 0)
return (rc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
memwin_info(sc, win, &mw_base, &mw_aperture);
buf = b = malloc(min(mr->len, mw_aperture), M_CXGBE, M_WAITOK);
addr = mr->addr;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
remaining = mr->len;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
dst = (void *)mr->data;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
while (remaining) {
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
off = position_memwin(sc, win, addr);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
/* number of bytes that we'll copy in the inner loop */
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
n = min(remaining, mw_aperture - off);
for (i = 0; i < n; i += 4)
*b++ = t4_read_reg(sc, mw_base + off + i);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
rc = copyout(buf, dst, n);
if (rc != 0)
break;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
b = buf;
dst += n;
remaining -= n;
addr += n;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
free(buf, M_CXGBE);
return (rc);
}
Add a driver ioctl to read a byte from any device on a port's i2c bus. This lets userspace read arbitrary information from the SFP+ modules etc. on this bus. Reading multiple bytes in the same transaction isn't possible right now. I'll update the driver once the chip's firmware supports this. MFC after: 3 days
2012-10-10 17:13:46 +00:00
static int
read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
{
int rc;
if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
return (EINVAL);
MFC r268640 and r268989. r268640: Allow multi-byte reads in the private CHELSIO_T4_GET_I2C ioctl. The firmware allows up to 48B to be read this way but the driver limits itself to 8B at a time to remain compatible with old cxgbetool binaries. r268989: Add missing newline to an error message.
2014-07-25 00:30:55 +00:00
if (i2cd->len > sizeof(i2cd->data))
return (EFBIG);
Add a driver ioctl to read a byte from any device on a port's i2c bus. This lets userspace read arbitrary information from the SFP+ modules etc. on this bus. Reading multiple bytes in the same transaction isn't possible right now. I'll update the driver once the chip's firmware supports this. MFC after: 3 days
2012-10-10 17:13:46 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
if (rc)
return (rc);
Add a driver ioctl to read a byte from any device on a port's i2c bus. This lets userspace read arbitrary information from the SFP+ modules etc. on this bus. Reading multiple bytes in the same transaction isn't possible right now. I'll update the driver once the chip's firmware supports this. MFC after: 3 days
2012-10-10 17:13:46 +00:00
rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
MFC r268640 and r268989. r268640: Allow multi-byte reads in the private CHELSIO_T4_GET_I2C ioctl. The firmware allows up to 48B to be read this way but the driver limits itself to 8B at a time to remain compatible with old cxgbetool binaries. r268989: Add missing newline to an error message.
2014-07-25 00:30:55 +00:00
i2cd->offset, i2cd->len, &i2cd->data[0]);
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
end_synchronized_op(sc, 0);
Add a driver ioctl to read a byte from any device on a port's i2c bus. This lets userspace read arbitrary information from the SFP+ modules etc. on this bus. Reading multiple bytes in the same transaction isn't possible right now. I'll update the driver once the chip's firmware supports this. MFC after: 3 days
2012-10-10 17:13:46 +00:00
return (rc);
}
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
static int
in_range(int val, int lo, int hi)
{
return (val < 0 || (val <= hi && val >= lo));
}
static int
set_sched_class(struct adapter *sc, struct t4_sched_params *p)
{
int fw_subcmd, fw_type, rc;
rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsc");
if (rc)
return (rc);
if (!(sc->flags & FULL_INIT_DONE)) {
rc = EAGAIN;
goto done;
}
/*
* Translate the cxgbetool parameters into T4 firmware parameters. (The
* sub-command and type are in common locations.)
*/
if (p->subcmd == SCHED_CLASS_SUBCMD_CONFIG)
fw_subcmd = FW_SCHED_SC_CONFIG;
else if (p->subcmd == SCHED_CLASS_SUBCMD_PARAMS)
fw_subcmd = FW_SCHED_SC_PARAMS;
else {
rc = EINVAL;
goto done;
}
if (p->type == SCHED_CLASS_TYPE_PACKET)
fw_type = FW_SCHED_TYPE_PKTSCHED;
else {
rc = EINVAL;
goto done;
}
if (fw_subcmd == FW_SCHED_SC_CONFIG) {
/* Vet our parameters ..*/
if (p->u.config.minmax < 0) {
rc = EINVAL;
goto done;
}
/* And pass the request to the firmware ...*/
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
rc = -t4_sched_config(sc, fw_type, p->u.config.minmax, 1);
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
goto done;
}
if (fw_subcmd == FW_SCHED_SC_PARAMS) {
int fw_level;
int fw_mode;
int fw_rateunit;
int fw_ratemode;
if (p->u.params.level == SCHED_CLASS_LEVEL_CL_RL)
fw_level = FW_SCHED_PARAMS_LEVEL_CL_RL;
else if (p->u.params.level == SCHED_CLASS_LEVEL_CL_WRR)
fw_level = FW_SCHED_PARAMS_LEVEL_CL_WRR;
else if (p->u.params.level == SCHED_CLASS_LEVEL_CH_RL)
fw_level = FW_SCHED_PARAMS_LEVEL_CH_RL;
else {
rc = EINVAL;
goto done;
}
if (p->u.params.mode == SCHED_CLASS_MODE_CLASS)
fw_mode = FW_SCHED_PARAMS_MODE_CLASS;
else if (p->u.params.mode == SCHED_CLASS_MODE_FLOW)
fw_mode = FW_SCHED_PARAMS_MODE_FLOW;
else {
rc = EINVAL;
goto done;
}
if (p->u.params.rateunit == SCHED_CLASS_RATEUNIT_BITS)
fw_rateunit = FW_SCHED_PARAMS_UNIT_BITRATE;
else if (p->u.params.rateunit == SCHED_CLASS_RATEUNIT_PKTS)
fw_rateunit = FW_SCHED_PARAMS_UNIT_PKTRATE;
else {
rc = EINVAL;
goto done;
}
if (p->u.params.ratemode == SCHED_CLASS_RATEMODE_REL)
fw_ratemode = FW_SCHED_PARAMS_RATE_REL;
else if (p->u.params.ratemode == SCHED_CLASS_RATEMODE_ABS)
fw_ratemode = FW_SCHED_PARAMS_RATE_ABS;
else {
rc = EINVAL;
goto done;
}
/* Vet our parameters ... */
if (!in_range(p->u.params.channel, 0, 3) ||
!in_range(p->u.params.cl, 0, is_t4(sc) ? 15 : 16) ||
!in_range(p->u.params.minrate, 0, 10000000) ||
!in_range(p->u.params.maxrate, 0, 10000000) ||
!in_range(p->u.params.weight, 0, 100)) {
rc = ERANGE;
goto done;
}
/*
* Translate any unset parameters into the firmware's
* nomenclature and/or fail the call if the parameters
* are required ...
*/
if (p->u.params.rateunit < 0 || p->u.params.ratemode < 0 ||
p->u.params.channel < 0 || p->u.params.cl < 0) {
rc = EINVAL;
goto done;
}
if (p->u.params.minrate < 0)
p->u.params.minrate = 0;
if (p->u.params.maxrate < 0) {
if (p->u.params.level == SCHED_CLASS_LEVEL_CL_RL ||
p->u.params.level == SCHED_CLASS_LEVEL_CH_RL) {
rc = EINVAL;
goto done;
} else
p->u.params.maxrate = 0;
}
if (p->u.params.weight < 0) {
if (p->u.params.level == SCHED_CLASS_LEVEL_CL_WRR) {
rc = EINVAL;
goto done;
} else
p->u.params.weight = 0;
}
if (p->u.params.pktsize < 0) {
if (p->u.params.level == SCHED_CLASS_LEVEL_CL_RL ||
p->u.params.level == SCHED_CLASS_LEVEL_CH_RL) {
rc = EINVAL;
goto done;
} else
p->u.params.pktsize = 0;
}
/* See what the firmware thinks of the request ... */
rc = -t4_sched_params(sc, fw_type, fw_level, fw_mode,
fw_rateunit, fw_ratemode, p->u.params.channel,
p->u.params.cl, p->u.params.minrate, p->u.params.maxrate,
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
p->u.params.weight, p->u.params.pktsize, 1);
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
goto done;
}
rc = EINVAL;
done:
end_synchronized_op(sc, 0);
return (rc);
}
static int
set_sched_queue(struct adapter *sc, struct t4_sched_queue *p)
{
struct port_info *pi = NULL;
struct sge_txq *txq;
uint32_t fw_mnem, fw_queue, fw_class;
int i, rc;
rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4setsq");
if (rc)
return (rc);
if (!(sc->flags & FULL_INIT_DONE)) {
rc = EAGAIN;
goto done;
}
if (p->port >= sc->params.nports) {
rc = EINVAL;
goto done;
}
pi = sc->port[p->port];
if (!in_range(p->queue, 0, pi->ntxq - 1) || !in_range(p->cl, 0, 7)) {
rc = EINVAL;
goto done;
}
/*
* Create a template for the FW_PARAMS_CMD mnemonic and value (TX
* Scheduling Class in this case).
*/
fw_mnem = (V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DMAQ) |
V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DMAQ_EQ_SCHEDCLASS_ETH));
fw_class = p->cl < 0 ? 0xffffffff : p->cl;
/*
* If op.queue is non-negative, then we're only changing the scheduling
* on a single specified TX queue.
*/
if (p->queue >= 0) {
txq = &sc->sge.txq[pi->first_txq + p->queue];
fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
&fw_class);
goto done;
}
/*
* Change the scheduling on all the TX queues for the
* interface.
*/
for_each_txq(pi, i, txq) {
fw_queue = (fw_mnem | V_FW_PARAMS_PARAM_YZ(txq->eq.cntxt_id));
rc = -t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &fw_queue,
&fw_class);
if (rc)
goto done;
}
rc = 0;
done:
end_synchronized_op(sc, 0);
return (rc);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
int
t4_os_find_pci_capability(struct adapter *sc, int cap)
{
Simplify t4_os_find_pci_capability. MFC after: 3 days
2011-05-19 19:37:41 +00:00
int i;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
Simplify t4_os_find_pci_capability. MFC after: 3 days
2011-05-19 19:37:41 +00:00
return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
int
t4_os_pci_save_state(struct adapter *sc)
{
device_t dev;
struct pci_devinfo *dinfo;
dev = sc->dev;
dinfo = device_get_ivars(dev);
pci_cfg_save(dev, dinfo, 0);
return (0);
}
int
t4_os_pci_restore_state(struct adapter *sc)
{
device_t dev;
struct pci_devinfo *dinfo;
dev = sc->dev;
dinfo = device_get_ivars(dev);
pci_cfg_restore(dev, dinfo);
return (0);
}
Be sure to stay within the bounds of the mod_str array when displaying the transceiver type.
2011-03-05 04:19:38 +00:00
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
void
t4_os_portmod_changed(const struct adapter *sc, int idx)
{
struct port_info *pi = sc->port[idx];
static const char *mod_str[] = {
Update header and related code for firmware 1.3.8 MFC after: 3 days
2011-04-01 00:40:24 +00:00
NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
};
MFC r274456: Fix some bad interaction between cxgbe(4) and lacp lagg(4) that could leave a port permanently disabled when a copper cable is unplugged and then plugged right back in. lacp_linkstate goes looking for the current ifmedia on a link state change and it could get stale information from cxgbe(4) on a module unplug followed by replug. The fix is to process module events before link-state events within the driver, and to always rebuild the ifmedia list on a module change event (instead of rebuilding it lazily). Thanks to asomers@ for the problem report and detailed analysis to go with it.
2015-04-07 15:32:43 +00:00
build_medialist(pi, &pi->media);
#ifdef DEV_NETMAP
build_medialist(pi, &pi->nm_media);
#endif
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
if_printf(pi->ifp, "transceiver unplugged.\n");
Update header and related code for firmware 1.3.8 MFC after: 3 days
2011-04-01 00:40:24 +00:00
else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
if_printf(pi->ifp, "unknown transceiver inserted.\n");
else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
if_printf(pi->ifp, "unsupported transceiver inserted.\n");
Use native FreeBSD facilities everywhere except the shared code in common/ MFC after: 1 week
2012-09-13 09:10:10 +00:00
else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if_printf(pi->ifp, "%s transceiver inserted.\n",
mod_str[pi->mod_type]);
Be sure to stay within the bounds of the mod_str array when displaying the transceiver type.
2011-03-05 04:19:38 +00:00
} else {
if_printf(pi->ifp, "transceiver (type %d) inserted.\n",
pi->mod_type);
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
void
- Show the reason why link is down if this information is available. - Display the temperature and PHY firmware version of the BT PHY. MFC after: 1 day
2013-07-05 01:53:51 +00:00
t4_os_link_changed(struct adapter *sc, int idx, int link_stat, int reason)
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
{
struct port_info *pi = sc->port[idx];
struct ifnet *ifp = pi->ifp;
if (link_stat) {
- Show the reason why link is down if this information is available. - Display the temperature and PHY firmware version of the BT PHY. MFC after: 1 day
2013-07-05 01:53:51 +00:00
pi->linkdnrc = -1;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
ifp->if_baudrate = IF_Mbps(pi->link_cfg.speed);
if_link_state_change(ifp, LINK_STATE_UP);
- Show the reason why link is down if this information is available. - Display the temperature and PHY firmware version of the BT PHY. MFC after: 1 day
2013-07-05 01:53:51 +00:00
} else {
if (reason >= 0)
pi->linkdnrc = reason;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if_link_state_change(ifp, LINK_STATE_DOWN);
- Show the reason why link is down if this information is available. - Display the temperature and PHY firmware version of the BT PHY. MFC after: 1 day
2013-07-05 01:53:51 +00:00
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
void
t4_iterate(void (*func)(struct adapter *, void *), void *arg)
{
struct adapter *sc;
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_slock(&t4_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SLIST_FOREACH(sc, &t4_list, link) {
/*
* func should not make any assumptions about what state sc is
* in - the only guarantee is that sc->sc_lock is a valid lock.
*/
func(sc, arg);
}
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_sunlock(&t4_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
static int
t4_open(struct cdev *dev, int flags, int type, struct thread *td)
{
return (0);
}
static int
t4_close(struct cdev *dev, int flags, int type, struct thread *td)
{
return (0);
}
static int
t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
struct thread *td)
{
int rc;
struct adapter *sc = dev->si_drv1;
rc = priv_check(td, PRIV_DRIVER);
if (rc != 0)
return (rc);
switch (cmd) {
Modify read/write ioctls to work with 64 bit registers too. MFC after: 3 days
2011-04-07 07:10:42 +00:00
case CHELSIO_T4_GETREG: {
struct t4_reg *edata = (struct t4_reg *)data;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
return (EFAULT);
Modify read/write ioctls to work with 64 bit registers too. MFC after: 3 days
2011-04-07 07:10:42 +00:00
if (edata->size == 4)
edata->val = t4_read_reg(sc, edata->addr);
else if (edata->size == 8)
edata->val = t4_read_reg64(sc, edata->addr);
else
return (EINVAL);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
break;
}
Modify read/write ioctls to work with 64 bit registers too. MFC after: 3 days
2011-04-07 07:10:42 +00:00
case CHELSIO_T4_SETREG: {
struct t4_reg *edata = (struct t4_reg *)data;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
return (EFAULT);
Modify read/write ioctls to work with 64 bit registers too. MFC after: 3 days
2011-04-07 07:10:42 +00:00
if (edata->size == 4) {
if (edata->val & 0xffffffff00000000)
return (EINVAL);
t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
} else if (edata->size == 8)
t4_write_reg64(sc, edata->addr, edata->val);
else
return (EINVAL);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
break;
}
case CHELSIO_T4_REGDUMP: {
struct t4_regdump *regs = (struct t4_regdump *)data;
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
int reglen = is_t4(sc) ? T4_REGDUMP_SIZE : T5_REGDUMP_SIZE;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
uint8_t *buf;
if (regs->len < reglen) {
regs->len = reglen; /* hint to the caller */
return (ENOBUFS);
}
regs->len = reglen;
buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
t4_get_regs(sc, regs, buf);
rc = copyout(buf, regs->data, reglen);
free(buf, M_CXGBE);
break;
}
T4 packet filtering/steering. - Enable 5-tuple and every-packet lookup. - Setup the default filter mode to allow filtering/steering based on IP protocol, ingress port, inner VLAN ID, IP frag, FCoE, and MPS match type; all combined together. You can also filter based on MAC index, Ethernet type, IP TOS/IPv6 Traffic Class, and outer VLAN ID but you'll have to modify the default filter mode and exclude some of the match-fields in it. IPv4 and IPv6 SIP/DIP/SPORT/DPORT are always available in all filter rules. - Add driver ioctls to get/set the global filter mode. - Add driver ioctls to program and delete hardware filters. A couple of the "switch" actions that rewrite Ethernet and VLAN information and switch the packet out of another port may not work as the L2 code is not yet in place. Everything else, including all "drop" and "pass" rules with RSS or absolute qid, should work. Obtained from: Chelsio Communications
2011-05-05 02:04:56 +00:00
case CHELSIO_T4_GET_FILTER_MODE:
rc = get_filter_mode(sc, (uint32_t *)data);
break;
case CHELSIO_T4_SET_FILTER_MODE:
rc = set_filter_mode(sc, *(uint32_t *)data);
break;
case CHELSIO_T4_GET_FILTER:
rc = get_filter(sc, (struct t4_filter *)data);
break;
case CHELSIO_T4_SET_FILTER:
rc = set_filter(sc, (struct t4_filter *)data);
break;
case CHELSIO_T4_DEL_FILTER:
rc = del_filter(sc, (struct t4_filter *)data);
break;
- driver ioctl to get SGE context for any given queue. - sysctls to display the context id, cidx, and pidx of all kinds of queues. MFC after: 3 days
2011-06-11 04:50:54 +00:00
case CHELSIO_T4_GET_SGE_CONTEXT:
rc = get_sge_context(sc, (struct t4_sge_context *)data);
break;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
case CHELSIO_T4_LOAD_FW:
rc = load_fw(sc, (struct t4_data *)data);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
break;
case CHELSIO_T4_GET_MEM:
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
break;
Add a driver ioctl to read a byte from any device on a port's i2c bus. This lets userspace read arbitrary information from the SFP+ modules etc. on this bus. Reading multiple bytes in the same transaction isn't possible right now. I'll update the driver once the chip's firmware supports this. MFC after: 3 days
2012-10-10 17:13:46 +00:00
case CHELSIO_T4_GET_I2C:
rc = read_i2c(sc, (struct t4_i2c_data *)data);
break;
Add a driver ioctl to clear a port's MAC statistics. Submitted by: gnn@ MFC after: 3 days
2012-10-10 19:27:40 +00:00
case CHELSIO_T4_CLEAR_STATS: {
cxgbe: Do a more thorough job in the CLEAR_STATS ioctl. MFC after: 3 days
2013-01-16 23:49:55 +00:00
int i;
Add a driver ioctl to clear a port's MAC statistics. Submitted by: gnn@ MFC after: 3 days
2012-10-10 19:27:40 +00:00
u_int port_id = *(uint32_t *)data;
cxgbe: Do a more thorough job in the CLEAR_STATS ioctl. MFC after: 3 days
2013-01-16 23:49:55 +00:00
struct port_info *pi;
Add a driver ioctl to clear a port's MAC statistics. Submitted by: gnn@ MFC after: 3 days
2012-10-10 19:27:40 +00:00
if (port_id >= sc->params.nports)
return (EINVAL);
MFC r261533, r261536, r261537, and r263457. r261533: cxgbe(4): Use the port's tx channel to identify it to t4_clr_port_stats. r261536: cxgbe(4): The T5 allows for a different freelist starvation threshold for queues with buffer packing. Use the correct value to calculate a freelist's low water mark. r261537: cxgbe(4): Use the rx channel map (instead of the tx channel map) as the congestion channel map. r263457: cxgbe(4): Recognize the "spider" configuration where a T5 card's 40G QSFP port is presented as 4 distinct 10G SFP+ ports to the driver.
2014-05-06 02:22:52 +00:00
pi = sc->port[port_id];
Add a driver ioctl to clear a port's MAC statistics. Submitted by: gnn@ MFC after: 3 days
2012-10-10 19:27:40 +00:00
cxgbe: Do a more thorough job in the CLEAR_STATS ioctl. MFC after: 3 days
2013-01-16 23:49:55 +00:00
/* MAC stats */
MFC r261533, r261536, r261537, and r263457. r261533: cxgbe(4): Use the port's tx channel to identify it to t4_clr_port_stats. r261536: cxgbe(4): The T5 allows for a different freelist starvation threshold for queues with buffer packing. Use the correct value to calculate a freelist's low water mark. r261537: cxgbe(4): Use the rx channel map (instead of the tx channel map) as the congestion channel map. r263457: cxgbe(4): Recognize the "spider" configuration where a T5 card's 40G QSFP port is presented as 4 distinct 10G SFP+ ports to the driver.
2014-05-06 02:22:52 +00:00
t4_clr_port_stats(sc, pi->tx_chan);
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
pi->tx_parse_error = 0;
cxgbe: Do a more thorough job in the CLEAR_STATS ioctl. MFC after: 3 days
2013-01-16 23:49:55 +00:00
if (pi->flags & PORT_INIT_DONE) {
struct sge_rxq *rxq;
struct sge_txq *txq;
struct sge_wrq *wrq;
for_each_rxq(pi, i, rxq) {
#if defined(INET) || defined(INET6)
rxq->lro.lro_queued = 0;
rxq->lro.lro_flushed = 0;
#endif
rxq->rxcsum = 0;
rxq->vlan_extraction = 0;
}
for_each_txq(pi, i, txq) {
txq->txcsum = 0;
txq->tso_wrs = 0;
txq->vlan_insertion = 0;
txq->imm_wrs = 0;
txq->sgl_wrs = 0;
txq->txpkt_wrs = 0;
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
txq->txpkts0_wrs = 0;
txq->txpkts1_wrs = 0;
txq->txpkts0_pkts = 0;
txq->txpkts1_pkts = 0;
mp_ring_reset_stats(txq->r);
cxgbe: Do a more thorough job in the CLEAR_STATS ioctl. MFC after: 3 days
2013-01-16 23:49:55 +00:00
}
#ifdef TCP_OFFLOAD
/* nothing to clear for each ofld_rxq */
for_each_ofld_txq(pi, i, wrq) {
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
wrq->tx_wrs_direct = 0;
wrq->tx_wrs_copied = 0;
cxgbe: Do a more thorough job in the CLEAR_STATS ioctl. MFC after: 3 days
2013-01-16 23:49:55 +00:00
}
#endif
wrq = &sc->sge.ctrlq[pi->port_id];
MFC r276480, r276485, r276498, r277225, r277226, r277227, r277230, r277637, and r283149 (by emaste@). r276485 is the real change here, the rest deal with the fallout of mp_ring's reliance on 64b atomics. Use the incorrectly spelled 'eigth' from struct pkthdr in this branch instead of MFC'ing r261733, which would have renamed the field of a public structure in a -STABLE branch. --- r276480: Temporarily unplug cxgbe(4) from !amd64 builds. r276485: cxgbe(4): major tx rework. a) Front load as much work as possible in if_transmit, before any driver lock or software queue has to get involved. b) Replace buf_ring with a brand new mp_ring (multiproducer ring). This is specifically for the tx multiqueue model where one of the if_transmit producer threads becomes the consumer and other producers carry on as usual. mp_ring is implemented as standalone code and it should be possible to use it in any driver with tx multiqueue. It also has: - the ability to enqueue/dequeue multiple items. This might become significant if packet batching is ever implemented. - an abdication mechanism to allow a thread to give up writing tx descriptors and have another if_transmit thread take over. A thread that's writing tx descriptors can end up doing so for an unbounded time period if a) there are other if_transmit threads continuously feeding the sofware queue, and b) the chip keeps up with whatever the thread is throwing at it. - accurate statistics about interesting events even when the stats come at the expense of additional branches/conditional code. The NIC txq lock is uncontested on the fast path at this point. I've left it there for synchronization with the control events (interface up/down, modload/unload). c) Add support for "type 1" coalescing work request in the normal NIC tx path. This work request is optimized for frames with a single item in the DMA gather list. These are very common when forwarding packets. Note that netmap tx in cxgbe already uses these "type 1" work requests. d) Do not request automatic cidx updates every 32 descriptors. Instead, request updates via bits in individual work requests (still every 32 descriptors approximately). Also, request an automatic final update when the queue idles after activity. This means NIC tx reclaim is still performed lazily but it will catch up quickly as soon as the queue idles. This seems to be the best middle ground and I'll probably do something similar for netmap tx as well. e) Implement a faster tx path for WRQs (used by TOE tx and control queues, _not_ by the normal NIC tx). Allow work requests to be written directly to the hardware descriptor ring if room is available. I will convert t4_tom and iw_cxgbe modules to this faster style gradually. r276498: cxgbe(4): remove buf_ring specific restriction on the txq size. r277225: Make cxgbe(4) buildable with the gcc in base. r277226: Allow cxgbe(4) to be built on i386. Driver attach will succeed only on a subset of i386 systems. r277227: Plug cxgbe(4) back into !powerpc && !arm builds, instead of building it on amd64 only. r277230: Build cxgbe(4) on powerpc64 too. r277637: Make sure the compiler flag to get cxgbe(4) to compile with gcc is used only when gcc is being used. This is what r277225 should have been.
2015-06-06 09:28:40 +00:00
wrq->tx_wrs_direct = 0;
wrq->tx_wrs_copied = 0;
cxgbe: Do a more thorough job in the CLEAR_STATS ioctl. MFC after: 3 days
2013-01-16 23:49:55 +00:00
}
Add a driver ioctl to clear a port's MAC statistics. Submitted by: gnn@ MFC after: 3 days
2012-10-10 19:27:40 +00:00
break;
}
MFC r257654, r257772, r258441, r258689, r258698, r258879, r259048, and r259103. r257654: cxgbe(4): Exclude MPS_RPLC_MAP_CTL (0x11114) from the register dump. Turns out it's a write-only register with strange side effects on read. r257772: cxgbe(4): Tidy up the display for payload memory statistics (pm_stats). r258441: cxgbe(4): update the internal list of device features. r258689: Disable an assertion that relies on some code[1] that isn't in HEAD yet. r258698: cxgbetool: "modinfo" command to display SFP+ module information. r258879: cxgbe(4): T4_SET_SCHED_CLASS and T4_SET_SCHED_QUEUE ioctls to program scheduling classes in the chip and to bind tx queue(s) to a scheduling class respectively. These can be used for various kinds of tx traffic throttling (to force selected tx queues to drain at a fixed Kbps rate, or a % of the port's total bandwidth, or at a fixed pps rate, etc.). r259048: Two new cxgbetool subcommands to set up scheduler classes and to bind them to NIC queues. r259103: cxgbe(4): save a copy of the RSS map for each port for the driver's use.
2013-12-09 22:40:22 +00:00
case CHELSIO_T4_SCHED_CLASS:
rc = set_sched_class(sc, (struct t4_sched_params *)data);
break;
case CHELSIO_T4_SCHED_QUEUE:
rc = set_sched_queue(sc, (struct t4_sched_queue *)data);
break;
Add support for packet-sniffing tracers to cxgbe(4). This works with all T4 and T5 based cards and is useful for analyzing TSO, LRO, TOE, and for general purpose monitoring without tapping any cxgbe or cxl ifnet directly. Tracers on the T4/T5 chips provide access to Ethernet frames exactly as they were received from or transmitted on the wire. On transmit, a tracer will capture a frame after TSO segmentation, hw VLAN tag insertion, hw L3 & L4 checksum insertion, etc. It will also capture frames generated by the TCP offload engine (TOE traffic is normally invisible to the kernel). On receive, a tracer will capture a frame before hw VLAN extraction, runt filtering, other badness filtering, before the steering/drop/L2-rewrite filters or the TOE have had a go at it, and of course before sw LRO in the driver. There are 4 tracers on a chip. A tracer can trace only in one direction (tx or rx). For now cxgbetool will set up tracers to capture the first 128B of every transmitted or received frame on a given port. This is a small subset of what the hardware can do. A pseudo ifnet with the same name as the nexus driver (t4nex0 or t5nex0) will be created for tracing. The data delivered to this ifnet is an additional copy made inside the chip. Normal delivery to cxgbe<n> or cxl<n> will be made as usual. /* watch cxl0, which is the first port hanging off t5nex0. */ # cxgbetool t5nex0 tracer 0 tx0 (watch what cxl0 is transmitting) # cxgbetool t5nex0 tracer 1 rx0 (watch what cxl0 is receiving) # cxgbetool t5nex0 tracer list # tcpdump -i t5nex0 <== all that cxl0 sees and puts on the wire If you were doing TSO, a tcpdump on cxl0 may have shown you ~64K "frames" with no L3/L4 checksum but this will show you the frames that were actually transmitted. /* all done */ # cxgbetool t5nex0 tracer 0 disable # cxgbetool t5nex0 tracer 1 disable # cxgbetool t5nex0 tracer list # ifconfig t5nex0 destroy
2013-07-26 22:04:11 +00:00
case CHELSIO_T4_GET_TRACER:
rc = t4_get_tracer(sc, (struct t4_tracer *)data);
break;
case CHELSIO_T4_SET_TRACER:
rc = t4_set_tracer(sc, (struct t4_tracer *)data);
break;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
default:
rc = EINVAL;
}
return (rc);
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
void
t4_iscsi_init(struct ifnet *ifp, unsigned int tag_mask,
const unsigned int *pgsz_order)
{
struct port_info *pi = ifp->if_softc;
struct adapter *sc = pi->adapter;
t4_write_reg(sc, A_ULP_RX_ISCSI_TAGMASK, tag_mask);
t4_write_reg(sc, A_ULP_RX_ISCSI_PSZ, V_HPZ0(pgsz_order[0]) |
V_HPZ1(pgsz_order[1]) | V_HPZ2(pgsz_order[2]) |
V_HPZ3(pgsz_order[3]));
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
static int
toe_capability(struct port_info *pi, int enable)
{
int rc;
struct adapter *sc = pi->adapter;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
ASSERT_SYNCHRONIZED_OP(sc);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (!is_offload(sc))
return (ENODEV);
if (enable) {
MFC r278342: cxgbe(4): fix a test made while enabling TOE.
2015-04-08 01:05:50 +00:00
/*
* We need the port's queues around so that we're able to send
* and receive CPLs to/from the TOE even if the ifnet for this
* port has never been UP'd administratively.
*/
if (!(pi->flags & PORT_INIT_DONE)) {
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
rc = cxgbe_init_synchronized(pi);
if (rc)
return (rc);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (isset(&sc->offload_map, pi->port_id))
return (0);
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
if (!uld_active(sc, ULD_TOM)) {
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
rc = t4_activate_uld(sc, ULD_TOM);
if (rc == EAGAIN) {
log(LOG_WARNING,
"You must kldload t4_tom.ko before trying "
"to enable TOE on a cxgbe interface.\n");
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (rc != 0)
return (rc);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
KASSERT(sc->tom_softc != NULL,
("%s: TOM activated but softc NULL", __func__));
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
KASSERT(uld_active(sc, ULD_TOM),
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
("%s: TOM activated but flag not set", __func__));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
/* Activate iWARP and iSCSI too, if the modules are loaded. */
if (!uld_active(sc, ULD_IWARP))
(void) t4_activate_uld(sc, ULD_IWARP);
if (!uld_active(sc, ULD_ISCSI))
(void) t4_activate_uld(sc, ULD_ISCSI);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
setbit(&sc->offload_map, pi->port_id);
} else {
if (!isset(&sc->offload_map, pi->port_id))
return (0);
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
KASSERT(uld_active(sc, ULD_TOM),
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
("%s: TOM never initialized?", __func__));
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
clrbit(&sc->offload_map, pi->port_id);
}
return (0);
}
/*
* Add an upper layer driver to the global list.
*/
int
t4_register_uld(struct uld_info *ui)
{
int rc = 0;
struct uld_info *u;
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_xlock(&t4_uld_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SLIST_FOREACH(u, &t4_uld_list, link) {
if (u->uld_id == ui->uld_id) {
rc = EEXIST;
goto done;
}
}
SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
ui->refcount = 0;
done:
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_xunlock(&t4_uld_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (rc);
}
int
t4_unregister_uld(struct uld_info *ui)
{
int rc = EINVAL;
struct uld_info *u;
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_xlock(&t4_uld_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SLIST_FOREACH(u, &t4_uld_list, link) {
if (u == ui) {
if (ui->refcount > 0) {
rc = EBUSY;
goto done;
}
SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
rc = 0;
goto done;
}
}
done:
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_xunlock(&t4_uld_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (rc);
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
int
t4_activate_uld(struct adapter *sc, int id)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
{
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
int rc;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
struct uld_info *ui;
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
ASSERT_SYNCHRONIZED_OP(sc);
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
if (id < 0 || id > ULD_MAX)
return (EINVAL);
rc = EAGAIN; /* kldoad the module with this ULD and try again. */
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_slock(&t4_uld_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
SLIST_FOREACH(ui, &t4_uld_list, link) {
if (ui->uld_id == id) {
MFC r272183: Make sure the adapter's management queue and the event queue are available before any uppper layer driver (TOE, iWARP, or iSCSI) registers with the base cxgbe(4) driver.
2015-05-03 18:54:17 +00:00
if (!(sc->flags & FULL_INIT_DONE)) {
rc = adapter_full_init(sc);
if (rc != 0)
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
break;
MFC r272183: Make sure the adapter's management queue and the event queue are available before any uppper layer driver (TOE, iWARP, or iSCSI) registers with the base cxgbe(4) driver.
2015-05-03 18:54:17 +00:00
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
rc = ui->activate(sc);
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
if (rc == 0) {
setbit(&sc->active_ulds, id);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
ui->refcount++;
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
}
break;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
}
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_sunlock(&t4_uld_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (rc);
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
int
t4_deactivate_uld(struct adapter *sc, int id)
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
{
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
int rc;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
struct uld_info *ui;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
cxgbe(4): Add functions to help synchronize "slow" operations (those not on the fast data path) and use them instead of frobbing the adapter lock and busy flag directly. Other changes made while reworking all slow operations: - Wait for the reply to a filter request (add/delete). This guarantees that the operation is complete by the time the ioctl returns. - Tidy up the tid_info structure. - Do not allow the tx queue size to be set to something that's not a power of 2. MFC after: 1 week
2013-01-10 23:56:50 +00:00
ASSERT_SYNCHRONIZED_OP(sc);
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
if (id < 0 || id > ULD_MAX)
return (EINVAL);
rc = ENXIO;
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_slock(&t4_uld_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
SLIST_FOREACH(ui, &t4_uld_list, link) {
if (ui->uld_id == id) {
rc = ui->deactivate(sc);
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
if (rc == 0) {
clrbit(&sc->active_ulds, id);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
ui->refcount--;
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
}
break;
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_sunlock(&t4_uld_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
return (rc);
}
MFC r278239 and r278374. r278239: cxgbe(4): reserve id for iSCSI upper layer driver. r278374: cxgbe(4): tidy up some of the interaction between the Upper Layer Drivers (ULDs) and the base if_cxgbe driver. Track the per-adapter activation of ULDs in a new "active_ulds" field. This was done pretty arbitrarily before this change -- via TOM_INIT_DONE in adapter->flags for TOM, and the (1 << MAX_NPORTS) bit in adapter->offload_map for iWARP. iWARP and hw-accelerated iSCSI rely on the TOE (supported by the TOM ULD). The rules are: a) If the iWARP and/or iSCSI ULDs are available when TOE is enabled then iWARP and/or iSCSI are enabled too. b) When the iWARP and iSCSI modules are loaded they go looking for adapters with TOE enabled and enable themselves on that adapter. c) You cannot deactivate or unload the TOM module from underneath iWARP or iSCSI. Any such attempt will fail with EBUSY.
2015-06-06 18:00:36 +00:00
int
uld_active(struct adapter *sc, int uld_id)
{
MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
return (isset(&sc->active_ulds, uld_id));
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
/*
* Come up with reasonable defaults for some of the tunables, provided they're
* not set by the user (in which case we'll use the values as is).
*/
static void
tweak_tunables(void)
{
int nc = mp_ncpus; /* our snapshot of the number of CPUs */
if (t4_ntxq10g < 1)
t4_ntxq10g = min(nc, NTXQ_10G);
if (t4_ntxq1g < 1)
t4_ntxq1g = min(nc, NTXQ_1G);
if (t4_nrxq10g < 1)
t4_nrxq10g = min(nc, NRXQ_10G);
if (t4_nrxq1g < 1)
t4_nrxq1g = min(nc, NRXQ_1G);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (t4_nofldtxq10g < 1)
t4_nofldtxq10g = min(nc, NOFLDTXQ_10G);
if (t4_nofldtxq1g < 1)
t4_nofldtxq1g = min(nc, NOFLDTXQ_1G);
if (t4_nofldrxq10g < 1)
t4_nofldrxq10g = min(nc, NOFLDRXQ_10G);
if (t4_nofldrxq1g < 1)
t4_nofldrxq1g = min(nc, NOFLDRXQ_1G);
Instruct the firmware not to provision resources for TCP offload if the kernel is being built without TCP_OFFLOAD. But never override toecaps_allowed if it has been set manually.
2012-07-02 20:42:43 +00:00
if (t4_toecaps_allowed == -1)
t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
#else
if (t4_toecaps_allowed == -1)
t4_toecaps_allowed = 0;
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
MFC r266571, r266757, r268536, r269076, r269364, r269366, r269411, r269413, r269428, r269440, r269537, r269644, r269731, and the cxgbe portion of r270063. r266571: cxgbe(4): Remove stray if_up from the code that creates the tracing ifnet. r266757: cxgbe(4): netmap support for Terminator 5 (T5) based 10G/40G cards. Netmap gets its own hardware-assisted virtual interface and won't take over or disrupt the "normal" interface in any way. You can use both simultaneously. For kernels with DEV_NETMAP, cxgbe(4) carves out an ncxl<N> interface (note the 'n' prefix) in the hardware to accompany each cxl<N> interface. These two ifnet's per port share the same wire but really are separate interfaces in the hardware and software. Each gets its own L2 MAC addresses (unicast and multicast), MTU, checksum caps, etc. You should run netmap on the 'n' interfaces only, that's what they are for. With this, pkt-gen is able to transmit > 45Mpps out of a single 40G port of a T580 card. 2 port tx is at ~56Mpps total (28M + 28M) as of now. Single port receive is at 33Mpps but this is very much a work in progress. I expect it to be closer to 40Mpps once done. In any case the current effort can already saturate multiple 10G ports of a T5 card at the smallest legal packet size. T4 gear is totally untested. trantor:~# ./pkt-gen -i ncxl0 -f tx -D 00:07:43:ab:cd:ef 881.952141 main [1621] interface is ncxl0 881.952250 extract_ip_range [275] range is 10.0.0.1:0 to 10.0.0.1:0 881.952253 extract_ip_range [275] range is 10.1.0.1:0 to 10.1.0.1:0 881.962540 main [1804] mapped 334980KB at 0x801dff000 Sending on netmap:ncxl0: 4 queues, 1 threads and 1 cpus. 10.0.0.1 -> 10.1.0.1 (00:00:00:00:00:00 -> 00:07:43:ab:cd:ef) 881.962562 main [1882] Sending 512 packets every 0.000000000 s 881.962563 main [1884] Wait 2 secs for phy reset 884.088516 main [1886] Ready... 884.088535 nm_open [457] overriding ifname ncxl0 ringid 0x0 flags 0x1 884.088607 sender_body [996] start 884.093246 sender_body [1064] drop copy 885.090435 main_thread [1418] 45206353 pps (45289533 pkts in 1001840 usec) 886.091600 main_thread [1418] 45322792 pps (45375593 pkts in 1001165 usec) 887.092435 main_thread [1418] 45313992 pps (45351784 pkts in 1000834 usec) 888.094434 main_thread [1418] 45315765 pps (45406397 pkts in 1002000 usec) 889.095434 main_thread [1418] 45333218 pps (45378551 pkts in 1001000 usec) 890.097434 main_thread [1418] 45315247 pps (45405877 pkts in 1002000 usec) 891.099434 main_thread [1418] 45326515 pps (45417168 pkts in 1002000 usec) 892.101434 main_thread [1418] 45333039 pps (45423705 pkts in 1002000 usec) 893.103434 main_thread [1418] 45324105 pps (45414708 pkts in 1001999 usec) 894.105434 main_thread [1418] 45318042 pps (45408723 pkts in 1002001 usec) 895.106434 main_thread [1418] 45332430 pps (45377762 pkts in 1001000 usec) 896.107434 main_thread [1418] 45338072 pps (45383410 pkts in 1001000 usec) ... r268536: cxgbe(4): Add an iSCSI softc to the adapter structure. r269076: Some hooks in cxgbe(4) for the offloaded iSCSI driver. r269364: Improve compliance with style.Makefile(5). r269366: List one file per line in the Makefiles. This makes it easier to read diffs when a file is added or removed. r269411: cxgbe(4): minor optimizations in ingress queue processing. Reorganize struct sge_iq. Make the iq entry size a compile time constant. While here, eliminate RX_FL_ESIZE and use EQ_ESIZE directly. r269413: cxgbe(4): Fix an off by one error when looking for the BAR2 doorbell address of an egress queue. r269428: cxgbe(4): some optimizations in freelist handling. r269440: cxgbe(4): Remove an unused version of t4_enable_vi. r269537: cxgbe(4): Do not run any sleepable code in the SIOCSIFFLAGS handler when IFF_PROMISC or IFF_ALLMULTI is being flipped. bpf(4) holds its global mutex around ifpromisc in at least the bpf_dtor path. r269644: cxgbe(4): Let caller specify whether it's ok to sleep in t4_sched_config and t4_sched_params. r269731: cxgbe(4): Do not poke T4-only registers on a T5 (and vice versa). Relnotes: Yes (native netmap support for Chelsio T4/T5 cards)
2014-08-21 19:54:02 +00:00
#ifdef DEV_NETMAP
if (t4_nnmtxq10g < 1)
t4_nnmtxq10g = min(nc, NNMTXQ_10G);
if (t4_nnmtxq1g < 1)
t4_nnmtxq1g = min(nc, NNMTXQ_1G);
if (t4_nnmrxq10g < 1)
t4_nnmrxq10g = min(nc, NNMRXQ_10G);
if (t4_nnmrxq1g < 1)
t4_nnmrxq1g = min(nc, NNMRXQ_1G);
#endif
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
if (t4_tmr_idx_10g < 0 || t4_tmr_idx_10g >= SGE_NTIMERS)
t4_tmr_idx_10g = TMR_IDX_10G;
if (t4_pktc_idx_10g < -1 || t4_pktc_idx_10g >= SGE_NCOUNTERS)
t4_pktc_idx_10g = PKTC_IDX_10G;
if (t4_tmr_idx_1g < 0 || t4_tmr_idx_1g >= SGE_NTIMERS)
t4_tmr_idx_1g = TMR_IDX_1G;
if (t4_pktc_idx_1g < -1 || t4_pktc_idx_1g >= SGE_NCOUNTERS)
t4_pktc_idx_1g = PKTC_IDX_1G;
if (t4_qsize_txq < 128)
t4_qsize_txq = 128;
if (t4_qsize_rxq < 128)
t4_qsize_rxq = 128;
while (t4_qsize_rxq & 7)
t4_qsize_rxq++;
t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
}
MFC r268971 and r269032. r268971: Simplify r267600, there's no need to distinguish between allocated and inlined mbufs. r269032: cxgbe(4): Keep track of the clusters that have to be freed by the custom free routine (rxb_free) in the driver. Fail MOD_UNLOAD with EBUSY if any such cluster has been handed up to the kernel but hasn't been freed yet. This prevents a panic later when the cluster finally needs to be freed but rxb_free is gone from the kernel.
2014-07-31 23:04:41 +00:00
static struct sx mlu; /* mod load unload */
SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
cxgbe shouldn't directly know of the UMA zones where network buffers come from. MFC after: 1 week
2011-03-08 03:04:07 +00:00
static int
cxgbe(4): Ensure that the MOD_LOAD handler runs before either t4nex or t5nex attach to their devices. MFC after: 3 days
2013-04-11 17:50:50 +00:00
mod_event(module_t mod, int cmd, void *arg)
cxgbe shouldn't directly know of the UMA zones where network buffers come from. MFC after: 1 week
2011-03-08 03:04:07 +00:00
{
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
int rc = 0;
cxgbe(4): Ensure that the MOD_LOAD handler runs before either t4nex or t5nex attach to their devices. MFC after: 3 days
2013-04-11 17:50:50 +00:00
static int loaded = 0;
cxgbe shouldn't directly know of the UMA zones where network buffers come from. MFC after: 1 week
2011-03-08 03:04:07 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
switch (cmd) {
case MOD_LOAD:
MFC r268971 and r269032. r268971: Simplify r267600, there's no need to distinguish between allocated and inlined mbufs. r269032: cxgbe(4): Keep track of the clusters that have to be freed by the custom free routine (rxb_free) in the driver. Fail MOD_UNLOAD with EBUSY if any such cluster has been handed up to the kernel but hasn't been freed yet. This prevents a panic later when the cluster finally needs to be freed but rxb_free is gone from the kernel.
2014-07-31 23:04:41 +00:00
sx_xlock(&mlu);
if (loaded++ == 0) {
t4_sge_modload();
sx_init(&t4_list_lock, "T4/T5 adapters");
SLIST_INIT(&t4_list);
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
MFC r268971 and r269032. r268971: Simplify r267600, there's no need to distinguish between allocated and inlined mbufs. r269032: cxgbe(4): Keep track of the clusters that have to be freed by the custom free routine (rxb_free) in the driver. Fail MOD_UNLOAD with EBUSY if any such cluster has been handed up to the kernel but hasn't been freed yet. This prevents a panic later when the cluster finally needs to be freed but rxb_free is gone from the kernel.
2014-07-31 23:04:41 +00:00
sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
SLIST_INIT(&t4_uld_list);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
MFC r268971 and r269032. r268971: Simplify r267600, there's no need to distinguish between allocated and inlined mbufs. r269032: cxgbe(4): Keep track of the clusters that have to be freed by the custom free routine (rxb_free) in the driver. Fail MOD_UNLOAD with EBUSY if any such cluster has been handed up to the kernel but hasn't been freed yet. This prevents a panic later when the cluster finally needs to be freed but rxb_free is gone from the kernel.
2014-07-31 23:04:41 +00:00
t4_tracer_modload();
tweak_tunables();
}
sx_xunlock(&mlu);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
break;
cxgbe shouldn't directly know of the UMA zones where network buffers come from. MFC after: 1 week
2011-03-08 03:04:07 +00:00
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
case MOD_UNLOAD:
MFC r268971 and r269032. r268971: Simplify r267600, there's no need to distinguish between allocated and inlined mbufs. r269032: cxgbe(4): Keep track of the clusters that have to be freed by the custom free routine (rxb_free) in the driver. Fail MOD_UNLOAD with EBUSY if any such cluster has been handed up to the kernel but hasn't been freed yet. This prevents a panic later when the cluster finally needs to be freed but rxb_free is gone from the kernel.
2014-07-31 23:04:41 +00:00
sx_xlock(&mlu);
if (--loaded == 0) {
int tries;
sx_slock(&t4_list_lock);
if (!SLIST_EMPTY(&t4_list)) {
rc = EBUSY;
sx_sunlock(&t4_list_lock);
goto done_unload;
}
#ifdef TCP_OFFLOAD
sx_slock(&t4_uld_list_lock);
if (!SLIST_EMPTY(&t4_uld_list)) {
rc = EBUSY;
sx_sunlock(&t4_uld_list_lock);
sx_sunlock(&t4_list_lock);
goto done_unload;
}
#endif
tries = 0;
while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
uprintf("%ju clusters with custom free routine "
"still is use.\n", t4_sge_extfree_refs());
pause("t4unload", 2 * hz);
}
- Updated TOE support in the kernel. - Stateful TCP offload drivers for Terminator 3 and 4 (T3 and T4) ASICs. These are available as t3_tom and t4_tom modules that augment cxgb(4) and cxgbe(4) respectively. The cxgb/cxgbe drivers continue to work as usual with or without these extra features. - iWARP driver for Terminator 3 ASIC (kernel verbs). T4 iWARP in the works and will follow soon. Build-tested with make universe. 30s overview ============ What interfaces support TCP offload? Look for TOE4 and/or TOE6 in the capabilities of an interface: # ifconfig -m | grep TOE Enable/disable TCP offload on an interface (just like any other ifnet capability): # ifconfig cxgbe0 toe # ifconfig cxgbe0 -toe Which connections are offloaded? Look for toe4 and/or toe6 in the output of netstat and sockstat: # netstat -np tcp | grep toe # sockstat -46c | grep toe Reviewed by: bz, gnn Sponsored by: Chelsio communications. MFC after: ~3 months (after 9.1, and after ensuring MFC is feasible)
2012-06-19 07:34:13 +00:00
#ifdef TCP_OFFLOAD
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_sunlock(&t4_uld_list_lock);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
#endif
Change t4_list_lock and t4_uld_list_lock from mutexes to sx'es. - tom_uninit had to be reworked not to hold the adapter lock (a mutex) around t4_deactivate_uld, which acquires the uld_list_lock. - the ifc_match for the interface cloner that creates the tracer ifnet had to be reworked as the kernel calls ifc_match with the global if_cloners_mtx held.
2013-08-28 20:59:22 +00:00
sx_sunlock(&t4_list_lock);
MFC r268971 and r269032. r268971: Simplify r267600, there's no need to distinguish between allocated and inlined mbufs. r269032: cxgbe(4): Keep track of the clusters that have to be freed by the custom free routine (rxb_free) in the driver. Fail MOD_UNLOAD with EBUSY if any such cluster has been handed up to the kernel but hasn't been freed yet. This prevents a panic later when the cluster finally needs to be freed but rxb_free is gone from the kernel.
2014-07-31 23:04:41 +00:00
if (t4_sge_extfree_refs() == 0) {
t4_tracer_modunload();
#ifdef TCP_OFFLOAD
sx_destroy(&t4_uld_list_lock);
#endif
sx_destroy(&t4_list_lock);
t4_sge_modunload();
loaded = 0;
} else {
rc = EBUSY;
loaded++; /* undo earlier decrement */
}
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
}
MFC r268971 and r269032. r268971: Simplify r267600, there's no need to distinguish between allocated and inlined mbufs. r269032: cxgbe(4): Keep track of the clusters that have to be freed by the custom free routine (rxb_free) in the driver. Fail MOD_UNLOAD with EBUSY if any such cluster has been handed up to the kernel but hasn't been freed yet. This prevents a panic later when the cluster finally needs to be freed but rxb_free is gone from the kernel.
2014-07-31 23:04:41 +00:00
done_unload:
sx_xunlock(&mlu);
Many updates to cxgbe(4) - Device configuration via plain text config file. Also able to operate when not attached to the chip as the master driver. - Generic "work request" queue that serves as the base for both ctrl and ofld tx queues. - Generic interrupt handler routine that can process any event on any kind of ingress queue (via a dispatch table). - A couple of new driver ioctls. cxgbetool can now install a firmware to the card ("loadfw" command) and can read the card's memory ("memdump" and "tcb" commands). - Lots of assorted information within dev.t4nex.X.misc.* This is primarily for debugging and won't show up in sysctl -a. - Code to manage the L2 tables on the chip. - Updates to cxgbe(4) man page to go with the tunables that have changed. - Updates to the shared code in common/ - Updates to the driver-firmware interface (now at fw 1.4.16.0) MFC after: 1 month
2011-12-16 02:09:51 +00:00
break;
}
return (rc);
cxgbe shouldn't directly know of the UMA zones where network buffers come from. MFC after: 1 week
2011-03-08 03:04:07 +00:00
}
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
static devclass_t t4_devclass, t5_devclass;
static devclass_t cxgbe_devclass, cxl_devclass;
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Ensure that the MOD_LOAD handler runs before either t4nex or t5nex attach to their devices. MFC after: 3 days
2013-04-11 17:50:50 +00:00
DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
MODULE_VERSION(t4nex, 1);
Add dependencies on the firmware, which allows the loading of the cxgb and cxgbe modules. Reviewed and approved by: np MFC after: 1 week
2013-05-16 13:07:02 +00:00
MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
cxgbe(4): Ensure that the MOD_LOAD handler runs before either t4nex or t5nex attach to their devices. MFC after: 3 days
2013-04-11 17:50:50 +00:00
DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
MODULE_VERSION(t5nex, 1);
Add dependencies on the firmware, which allows the loading of the cxgb and cxgbe modules. Reviewed and approved by: np MFC after: 1 week
2013-05-16 13:07:02 +00:00
MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
cxgbe(4) - NIC driver for Chelsio T4 (Terminator 4) based 10Gb/1Gb adapters. MFC after: 3 weeks
2011-02-18 08:00:26 +00:00
DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
MODULE_VERSION(cxgbe, 1);
cxgbe(4): Add support for Chelsio's Terminator 5 (aka T5) ASIC. This includes support for the NIC and TOE features of the 40G, 10G, and 1G/100M cards based on the T5. The ASIC is mostly backward compatible with the Terminator 4 so cxgbe(4) has been updated instead of writing a brand new driver. T5 cards will show up as cxl (short for cxlgb) ports attached to the t5nex bus driver. Sponsored by: Chelsio
2013-03-30 02:26:20 +00:00
DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
MODULE_VERSION(cxl, 1);
Reference in New Issue Copy Permalink