freebsd-nq/sys/dev/cxgbe/t4_l2t.c

330 lines
8.6 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2012 Chelsio Communications, Inc.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* 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"
#include "opt_inet6.h"
#include <sys/param.h>
#include <sys/eventhandler.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/module.h>
#include <sys/bus.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/rwlock.h>
#include <sys/socket.h>
#include <sys/sbuf.h>
#include <netinet/in.h>
#include "common/common.h"
#include "common/t4_msg.h"
#include "t4_l2t.h"
/*
* Module locking notes: There is a RW lock protecting the L2 table as a
* whole plus a spinlock per L2T entry. Entry lookups and allocations happen
* under the protection of the table lock, individual entry changes happen
* while holding that entry's spinlock. The table lock nests outside the
* entry locks. Allocations of new entries take the table lock as writers so
* no other lookups can happen while allocating new entries. Entry updates
* take the table lock as readers so multiple entries can be updated in
* parallel. An L2T entry can be dropped by decrementing its reference count
* and therefore can happen in parallel with entry allocation but no entry
* can change state or increment its ref count during allocation as both of
* these perform lookups.
*
* Note: We do not take refereces to ifnets in this module because both
* the TOE and the sockets already hold references to the interfaces and the
* lifetime of an L2T entry is fully contained in the lifetime of the TOE.
*/
/*
* Allocate a free L2T entry. Must be called with l2t_data.lock held.
*/
struct l2t_entry *
t4_alloc_l2e(struct l2t_data *d)
{
struct l2t_entry *end, *e, **p;
rw_assert(&d->lock, RA_WLOCKED);
if (!atomic_load_acq_int(&d->nfree))
return (NULL);
/* there's definitely a free entry */
for (e = d->rover, end = &d->l2tab[d->l2t_size]; e != end; ++e)
if (atomic_load_acq_int(&e->refcnt) == 0)
goto found;
for (e = d->l2tab; atomic_load_acq_int(&e->refcnt); ++e)
continue;
found:
d->rover = e + 1;
atomic_subtract_int(&d->nfree, 1);
/*
* The entry we found may be an inactive entry that is
* presently in the hash table. We need to remove it.
*/
if (e->state < L2T_STATE_SWITCHING) {
for (p = &d->l2tab[e->hash].first; *p; p = &(*p)->next) {
if (*p == e) {
*p = e->next;
e->next = NULL;
break;
}
}
}
e->state = L2T_STATE_UNUSED;
return (e);
}
/*
* Write an L2T entry. Must be called with the entry locked.
* The write may be synchronous or asynchronous.
*/
int
t4_write_l2e(struct adapter *sc, struct l2t_entry *e, int sync)
{
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. MFC after: 2 months
2014-12-31 23:19:16 +00:00
struct wrq_cookie cookie;
struct cpl_l2t_write_req *req;
int idx = e->idx + sc->vres.l2t.start;
mtx_assert(&e->lock, MA_OWNED);
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. MFC after: 2 months
2014-12-31 23:19:16 +00:00
req = start_wrq_wr(&sc->sge.mgmtq, howmany(sizeof(*req), 16), &cookie);
if (req == NULL)
return (ENOMEM);
INIT_TP_WR(req, 0);
OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, idx |
V_SYNC_WR(sync) | V_TID_QID(sc->sge.fwq.abs_id)));
req->params = htons(V_L2T_W_PORT(e->lport) | V_L2T_W_NOREPLY(!sync));
req->l2t_idx = htons(idx);
req->vlan = htons(e->vlan);
memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
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. MFC after: 2 months
2014-12-31 23:19:16 +00:00
commit_wrq_wr(&sc->sge.mgmtq, req, &cookie);
if (sync && e->state != L2T_STATE_SWITCHING)
e->state = L2T_STATE_SYNC_WRITE;
return (0);
}
/*
* Allocate an L2T entry for use by a switching rule. Such need to be
* explicitly freed and while busy they are not on any hash chain, so normal
* address resolution updates do not see them.
*/
struct l2t_entry *
t4_l2t_alloc_switching(struct l2t_data *d)
{
struct l2t_entry *e;
rw_wlock(&d->lock);
e = t4_alloc_l2e(d);
if (e) {
mtx_lock(&e->lock); /* avoid race with t4_l2t_free */
e->state = L2T_STATE_SWITCHING;
atomic_store_rel_int(&e->refcnt, 1);
mtx_unlock(&e->lock);
}
rw_wunlock(&d->lock);
return e;
}
/*
* Sets/updates the contents of a switching L2T entry that has been allocated
* with an earlier call to @t4_l2t_alloc_switching.
*/
int
t4_l2t_set_switching(struct adapter *sc, struct l2t_entry *e, uint16_t vlan,
uint8_t port, uint8_t *eth_addr)
{
int rc;
e->vlan = vlan;
e->lport = port;
memcpy(e->dmac, eth_addr, ETHER_ADDR_LEN);
mtx_lock(&e->lock);
rc = t4_write_l2e(sc, e, 0);
mtx_unlock(&e->lock);
return (rc);
}
int
t4_init_l2t(struct adapter *sc, int flags)
{
int i, l2t_size;
struct l2t_data *d;
l2t_size = sc->vres.l2t.size;
if (l2t_size < 2) /* At least 1 bucket for IP and 1 for IPv6 */
return (EINVAL);
d = malloc(sizeof(*d) + l2t_size * sizeof (struct l2t_entry), M_CXGBE,
M_ZERO | flags);
if (!d)
return (ENOMEM);
d->l2t_size = l2t_size;
d->rover = d->l2tab;
atomic_store_rel_int(&d->nfree, l2t_size);
rw_init(&d->lock, "L2T");
for (i = 0; i < l2t_size; i++) {
struct l2t_entry *e = &d->l2tab[i];
e->idx = i;
e->state = L2T_STATE_UNUSED;
mtx_init(&e->lock, "L2T_E", NULL, MTX_DEF);
STAILQ_INIT(&e->wr_list);
atomic_store_rel_int(&e->refcnt, 0);
}
sc->l2t = d;
t4_register_cpl_handler(sc, CPL_L2T_WRITE_RPL, do_l2t_write_rpl);
return (0);
}
int
t4_free_l2t(struct l2t_data *d)
{
int i;
for (i = 0; i < d->l2t_size; i++)
mtx_destroy(&d->l2tab[i].lock);
rw_destroy(&d->lock);
free(d, M_CXGBE);
return (0);
}
int
do_l2t_write_rpl(struct sge_iq *iq, const struct rss_header *rss,
struct mbuf *m)
{
const struct cpl_l2t_write_rpl *rpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(rpl);
unsigned int idx = tid % L2T_SIZE;
if (__predict_false(rpl->status != CPL_ERR_NONE)) {
log(LOG_ERR,
"Unexpected L2T_WRITE_RPL (%u) for entry at hw_idx %u\n",
rpl->status, idx);
return (EINVAL);
}
return (0);
}
#ifdef SBUF_DRAIN
static inline unsigned int
vlan_prio(const struct l2t_entry *e)
{
return e->vlan >> 13;
}
static char
l2e_state(const struct l2t_entry *e)
{
switch (e->state) {
case L2T_STATE_VALID: return 'V'; /* valid, fast-path entry */
case L2T_STATE_STALE: return 'S'; /* needs revalidation, but usable */
case L2T_STATE_SYNC_WRITE: return 'W';
case L2T_STATE_RESOLVING: return STAILQ_EMPTY(&e->wr_list) ? 'R' : 'A';
case L2T_STATE_SWITCHING: return 'X';
default: return 'U';
}
}
int
sysctl_l2t(SYSCTL_HANDLER_ARGS)
{
struct adapter *sc = arg1;
struct l2t_data *l2t = sc->l2t;
struct l2t_entry *e;
struct sbuf *sb;
int rc, i, header = 0;
char ip[INET6_ADDRSTRLEN];
if (l2t == NULL)
return (ENXIO);
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);
e = &l2t->l2tab[0];
for (i = 0; i < l2t->l2t_size; i++, e++) {
mtx_lock(&e->lock);
if (e->state == L2T_STATE_UNUSED)
goto skip;
if (header == 0) {
sbuf_printf(sb, " Idx IP address "
"Ethernet address VLAN/P LP State Users Port");
header = 1;
}
if (e->state == L2T_STATE_SWITCHING)
ip[0] = 0;
else {
inet_ntop(e->ipv6 ? AF_INET6 : AF_INET, &e->addr[0],
&ip[0], sizeof(ip));
}
/*
* XXX: e->ifp may not be around.
* XXX: IPv6 addresses may not align properly in the output.
*/
sbuf_printf(sb, "\n%4u %-15s %02x:%02x:%02x:%02x:%02x:%02x %4d"
" %u %2u %c %5u %s",
e->idx, ip, e->dmac[0], e->dmac[1], e->dmac[2],
e->dmac[3], e->dmac[4], e->dmac[5],
e->vlan & 0xfff, vlan_prio(e), e->lport,
l2e_state(e), atomic_load_acq_int(&e->refcnt),
e->ifp->if_xname);
skip:
mtx_unlock(&e->lock);
}
rc = sbuf_finish(sb);
sbuf_delete(sb);
return (rc);
}
#endif