Author: np
Date: Mon Jan 14 20:36:22 2013
New Revision: 245434
URL: http://svnweb.freebsd.org/changeset/base/245434
Log:
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
Deleted:
head/sys/dev/cxgbe/common/jhash.h
Modified:
head/sys/dev/cxgbe/offload.h
head/sys/dev/cxgbe/t4_l2t.c
head/sys/dev/cxgbe/t4_l2t.h
head/sys/dev/cxgbe/t4_main.c
head/sys/dev/cxgbe/tom/t4_tom_l2t.c
head/sys/modules/cxgbe/tom/Makefile
Modified: head/sys/dev/cxgbe/offload.h
==============================================================================
--- head/sys/dev/cxgbe/offload.h Mon Jan 14 20:31:45 2013
(r245433)
+++ head/sys/dev/cxgbe/offload.h Mon Jan 14 20:36:22 2013
(r245434)
@@ -104,8 +104,8 @@ struct tid_info {
};
struct t4_range {
- unsigned int start;
- unsigned int size;
+ u_int start;
+ u_int size;
};
struct t4_virt_res { /* virtualized HW resources */
@@ -117,6 +117,7 @@ struct t4_virt_res {
struct t4_range qp;
struct t4_range cq;
struct t4_range ocq;
+ struct t4_range l2t;
};
#ifdef TCP_OFFLOAD
Modified: head/sys/dev/cxgbe/t4_l2t.c
==============================================================================
--- head/sys/dev/cxgbe/t4_l2t.c Mon Jan 14 20:31:45 2013 (r245433)
+++ head/sys/dev/cxgbe/t4_l2t.c Mon Jan 14 20:36:22 2013 (r245434)
@@ -42,7 +42,6 @@ __FBSDID("$FreeBSD$");
#include <netinet/in.h>
#include "common/common.h"
-#include "common/jhash.h"
#include "common/t4_msg.h"
#include "t4_l2t.h"
@@ -78,7 +77,7 @@ t4_alloc_l2e(struct l2t_data *d)
return (NULL);
/* there's definitely a free entry */
- for (e = d->rover, end = &d->l2tab[L2T_SIZE]; e != end; ++e)
+ for (e = d->rover, end = &d->l2tab[d->l2t_size]; e != end; ++e)
if (atomic_load_acq_int(&e->refcnt) == 0)
goto found;
@@ -115,6 +114,7 @@ t4_write_l2e(struct adapter *sc, struct
{
struct wrqe *wr;
struct cpl_l2t_write_req *req;
+ int idx = e->idx + sc->vres.l2t.start;
mtx_assert(&e->lock, MA_OWNED);
@@ -124,10 +124,10 @@ t4_write_l2e(struct adapter *sc, struct
req = wrtod(wr);
INIT_TP_WR(req, 0);
- OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, e->idx |
+ 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(e->idx);
+ req->l2t_idx = htons(idx);
req->vlan = htons(e->vlan);
memcpy(req->dst_mac, e->dmac, sizeof(req->dst_mac));
@@ -183,18 +183,24 @@ t4_l2t_set_switching(struct adapter *sc,
int
t4_init_l2t(struct adapter *sc, int flags)
{
- int i;
+ int i, l2t_size;
struct l2t_data *d;
- d = malloc(sizeof(*d), M_CXGBE, M_ZERO | flags);
+ 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);
+ atomic_store_rel_int(&d->nfree, l2t_size);
rw_init(&d->lock, "L2T");
- for (i = 0; i < L2T_SIZE; i++) {
+ for (i = 0; i < l2t_size; i++) {
struct l2t_entry *e = &d->l2tab[i];
e->idx = i;
@@ -215,7 +221,7 @@ t4_free_l2t(struct l2t_data *d)
{
int i;
- for (i = 0; i < L2T_SIZE; i++)
+ for (i = 0; i < d->l2t_size; i++)
mtx_destroy(&d->l2tab[i].lock);
rw_destroy(&d->lock);
free(d, M_CXGBE);
@@ -229,11 +235,11 @@ do_l2t_write_rpl(struct sge_iq *iq, cons
{
const struct cpl_l2t_write_rpl *rpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(rpl);
- unsigned int idx = tid & (L2T_SIZE - 1);
+ unsigned int idx = tid % L2T_SIZE;
if (__predict_false(rpl->status != CPL_ERR_NONE)) {
log(LOG_ERR,
- "Unexpected L2T_WRITE_RPL status %u for entry %u\n",
+ "Unexpected L2T_WRITE_RPL (%u) for entry at hw_idx %u\n",
rpl->status, idx);
return (EINVAL);
}
@@ -269,7 +275,7 @@ sysctl_l2t(SYSCTL_HANDLER_ARGS)
struct l2t_entry *e;
struct sbuf *sb;
int rc, i, header = 0;
- char ip[60];
+ char ip[INET6_ADDRSTRLEN];
if (l2t == NULL)
return (ENXIO);
@@ -283,7 +289,7 @@ sysctl_l2t(SYSCTL_HANDLER_ARGS)
return (ENOMEM);
e = &l2t->l2tab[0];
- for (i = 0; i < L2T_SIZE; i++, e++) {
+ for (i = 0; i < l2t->l2t_size; i++, e++) {
mtx_lock(&e->lock);
if (e->state == L2T_STATE_UNUSED)
goto skip;
@@ -295,11 +301,15 @@ sysctl_l2t(SYSCTL_HANDLER_ARGS)
}
if (e->state == L2T_STATE_SWITCHING)
ip[0] = 0;
- else
- snprintf(ip, sizeof(ip), "%s",
- inet_ntoa(*(struct in_addr *)&e->addr));
+ else {
+ inet_ntop(e->ipv6 ? AF_INET6 : AF_INET, &e->addr[0],
+ &ip[0], sizeof(ip));
+ }
- /* XXX: e->ifp may not be around */
+ /*
+ * 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],
Modified: head/sys/dev/cxgbe/t4_l2t.h
==============================================================================
--- head/sys/dev/cxgbe/t4_l2t.h Mon Jan 14 20:31:45 2013 (r245433)
+++ head/sys/dev/cxgbe/t4_l2t.h Mon Jan 14 20:36:22 2013 (r245434)
@@ -60,7 +60,7 @@ enum {
struct l2t_entry {
uint16_t state; /* entry state */
uint16_t idx; /* entry index */
- uint32_t addr; /* next hop IP address */
+ uint32_t addr[4]; /* next hop IP or IPv6 address */
struct ifnet *ifp; /* outgoing interface */
uint16_t smt_idx; /* SMT index */
uint16_t vlan; /* VLAN TCI (id: 0-11, prio: 13-15) */
@@ -70,15 +70,17 @@ struct l2t_entry {
struct mtx lock;
volatile int refcnt; /* entry reference count */
uint16_t hash; /* hash bucket the entry is on */
+ uint8_t ipv6; /* entry is for an IPv6 address */
uint8_t lport; /* associated offload logical port */
uint8_t dmac[ETHER_ADDR_LEN]; /* next hop's MAC address */
};
struct l2t_data {
struct rwlock lock;
+ u_int l2t_size;
volatile int nfree; /* number of free entries */
struct l2t_entry *rover;/* starting point for next allocation */
- struct l2t_entry l2tab[L2T_SIZE];
+ struct l2t_entry l2tab[];
};
Modified: head/sys/dev/cxgbe/t4_main.c
==============================================================================
--- head/sys/dev/cxgbe/t4_main.c Mon Jan 14 20:31:45 2013
(r245433)
+++ head/sys/dev/cxgbe/t4_main.c Mon Jan 14 20:36:22 2013
(r245434)
@@ -1866,7 +1866,9 @@ get_params__post_init(struct adapter *sc
param[1] = FW_PARAM_PFVF(EQ_START);
param[2] = FW_PARAM_PFVF(FILTER_START);
param[3] = FW_PARAM_PFVF(FILTER_END);
- rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
+ 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);
if (rc != 0) {
device_printf(sc->dev,
"failed to query parameters (post_init): %d.\n", rc);
@@ -1877,6 +1879,11 @@ get_params__post_init(struct adapter *sc
sc->sge.eq_start = val[1];
sc->tids.ftid_base = val[2];
sc->tids.nftids = val[3] - val[2] + 1;
+ 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));
/* get capabilites */
bzero(&caps, sizeof(caps));
Modified: head/sys/dev/cxgbe/tom/t4_tom_l2t.c
==============================================================================
--- head/sys/dev/cxgbe/tom/t4_tom_l2t.c Mon Jan 14 20:31:45 2013
(r245433)
+++ head/sys/dev/cxgbe/tom/t4_tom_l2t.c Mon Jan 14 20:36:22 2013
(r245434)
@@ -27,6 +27,7 @@
__FBSDID("$FreeBSD$");
#include "opt_inet.h"
+#include "opt_inet6.h"
#ifdef TCP_OFFLOAD
#include <sys/param.h>
@@ -34,6 +35,7 @@ __FBSDID("$FreeBSD$");
#include <sys/kernel.h>
#include <sys/module.h>
#include <sys/bus.h>
+#include <sys/fnv_hash.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/rwlock.h>
@@ -48,28 +50,89 @@ __FBSDID("$FreeBSD$");
#include <netinet/toecore.h>
#include "common/common.h"
-#include "common/jhash.h"
#include "common/t4_msg.h"
#include "tom/t4_tom_l2t.h"
#include "tom/t4_tom.h"
#define VLAN_NONE 0xfff
-#define SA(x) ((struct sockaddr *)(x))
-#define SIN(x) ((struct sockaddr_in *)(x))
-#define SINADDR(x) (SIN(x)->sin_addr.s_addr)
-
static inline void
l2t_hold(struct l2t_data *d, struct l2t_entry *e)
{
+
if (atomic_fetchadd_int(&e->refcnt, 1) == 0) /* 0 -> 1 transition */
atomic_subtract_int(&d->nfree, 1);
}
-static inline unsigned int
-arp_hash(const uint32_t key, int ifindex)
+static inline u_int
+l2_hash(struct l2t_data *d, const struct sockaddr *sa, int ifindex)
+{
+ u_int hash, half = d->l2t_size / 2, start = 0;
+ const void *key;
+ size_t len;
+
+ KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6,
+ ("%s: sa %p has unexpected sa_family %d", __func__, sa,
+ sa->sa_family));
+
+ if (sa->sa_family == AF_INET) {
+ const struct sockaddr_in *sin = (const void *)sa;
+
+ key = &sin->sin_addr;
+ len = sizeof(sin->sin_addr);
+ } else {
+ const struct sockaddr_in6 *sin6 = (const void *)sa;
+
+ key = &sin6->sin6_addr;
+ len = sizeof(sin6->sin6_addr);
+ start = half;
+ }
+
+ hash = fnv_32_buf(key, len, FNV1_32_INIT);
+ hash = fnv_32_buf(&ifindex, sizeof(ifindex), hash);
+ hash %= half;
+
+ return (hash + start);
+}
+
+static inline int
+l2_cmp(const struct sockaddr *sa, struct l2t_entry *e)
+{
+
+ KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6,
+ ("%s: sa %p has unexpected sa_family %d", __func__, sa,
+ sa->sa_family));
+
+ if (sa->sa_family == AF_INET) {
+ const struct sockaddr_in *sin = (const void *)sa;
+
+ return (e->addr[0] != sin->sin_addr.s_addr);
+ } else {
+ const struct sockaddr_in6 *sin6 = (const void *)sa;
+
+ return (memcmp(&e->addr[0], &sin6->sin6_addr, sizeof(e->addr)));
+ }
+}
+
+static inline void
+l2_store(const struct sockaddr *sa, struct l2t_entry *e)
{
- return jhash_2words(key, ifindex, 0) & (L2T_SIZE - 1);
+
+ KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6,
+ ("%s: sa %p has unexpected sa_family %d", __func__, sa,
+ sa->sa_family));
+
+ if (sa->sa_family == AF_INET) {
+ const struct sockaddr_in *sin = (const void *)sa;
+
+ e->addr[0] = sin->sin_addr.s_addr;
+ e->ipv6 = 0;
+ } else {
+ const struct sockaddr_in6 *sin6 = (const void *)sa;
+
+ memcpy(&e->addr[0], &sin6->sin6_addr, sizeof(e->addr));
+ e->ipv6 = 1;
+ }
}
/*
@@ -100,7 +163,7 @@ send_pending(struct adapter *sc, struct
static void
resolution_failed_for_wr(struct wrqe *wr)
{
- log(LOG_ERR, "%s: leaked work request %p, wr_len %d", __func__, wr,
+ log(LOG_ERR, "%s: leaked work request %p, wr_len %d\n", __func__, wr,
wr->wr_len);
/* free(wr, M_CXGBE); */
@@ -175,15 +238,25 @@ resolve_entry(struct adapter *sc, struct
struct tom_data *td = sc->tom_softc;
struct toedev *tod = &td->tod;
struct sockaddr_in sin = {0};
+ struct sockaddr_in6 sin6 = {0};
+ struct sockaddr *sa;
uint8_t dmac[ETHER_ADDR_LEN];
uint16_t vtag = VLAN_NONE;
int rc;
- sin.sin_family = AF_INET;
- sin.sin_len = sizeof(struct sockaddr_in);
- SINADDR(&sin) = e->addr;
+ if (e->ipv6 == 0) {
+ sin.sin_family = AF_INET;
+ sin.sin_len = sizeof(struct sockaddr_in);
+ sin.sin_addr.s_addr = e->addr[0];
+ sa = (void *)&sin;
+ } else {
+ sin6.sin6_family = AF_INET6;
+ sin6.sin6_len = sizeof(struct sockaddr_in6);
+ memcpy(&sin6.sin6_addr, &e->addr[0], sizeof(e->addr));
+ sa = (void *)&sin6;
+ }
- rc = toe_l2_resolve(tod, e->ifp, SA(&sin), dmac, &vtag);
+ rc = toe_l2_resolve(tod, e->ifp, sa, dmac, &vtag);
if (rc == EWOULDBLOCK)
return (rc);
@@ -263,7 +336,7 @@ do_l2t_write_rpl2(struct sge_iq *iq, con
struct adapter *sc = iq->adapter;
const struct cpl_l2t_write_rpl *rpl = (const void *)(rss + 1);
unsigned int tid = GET_TID(rpl);
- unsigned int idx = tid & (L2T_SIZE - 1);
+ unsigned int idx = tid % L2T_SIZE;
int rc;
rc = do_l2t_write_rpl(iq, rss, m);
@@ -271,7 +344,7 @@ do_l2t_write_rpl2(struct sge_iq *iq, con
return (rc);
if (tid & F_SYNC_WR) {
- struct l2t_entry *e = &sc->l2t->l2tab[idx];
+ struct l2t_entry *e = &sc->l2t->l2tab[idx - sc->vres.l2t.start];
mtx_lock(&e->lock);
if (e->state != L2T_STATE_SWITCHING) {
@@ -310,21 +383,22 @@ t4_l2t_get(struct port_info *pi, struct
{
struct l2t_entry *e;
struct l2t_data *d = pi->adapter->l2t;
- uint32_t addr = SINADDR(sa);
- int hash = arp_hash(addr, ifp->if_index);
- unsigned int smt_idx = pi->port_id;
+ u_int hash, smt_idx = pi->port_id;
- if (sa->sa_family != AF_INET)
- return (NULL); /* XXX: no IPv6 support right now */
+ KASSERT(sa->sa_family == AF_INET || sa->sa_family == AF_INET6,
+ ("%s: sa %p has unexpected sa_family %d", __func__, sa,
+ sa->sa_family));
#ifndef VLAN_TAG
if (ifp->if_type == IFT_L2VLAN)
return (NULL);
#endif
+ hash = l2_hash(d, sa, ifp->if_index);
rw_wlock(&d->lock);
for (e = d->l2tab[hash].first; e; e = e->next) {
- if (e->addr == addr && e->ifp == ifp && e->smt_idx == smt_idx) {
+ if (l2_cmp(sa, e) == 0 && e->ifp == ifp &&
+ e->smt_idx == smt_idx) {
l2t_hold(d, e);
goto done;
}
@@ -338,7 +412,7 @@ t4_l2t_get(struct port_info *pi, struct
d->l2tab[hash].first = e;
e->state = L2T_STATE_RESOLVING;
- e->addr = addr;
+ l2_store(sa, e);
e->ifp = ifp;
e->smt_idx = smt_idx;
e->hash = hash;
@@ -368,14 +442,14 @@ t4_l2_update(struct toedev *tod, struct
struct adapter *sc = tod->tod_softc;
struct l2t_entry *e;
struct l2t_data *d = sc->l2t;
- uint32_t addr = SINADDR(sa);
- int hash = arp_hash(addr, ifp->if_index);
+ u_int hash;
KASSERT(d != NULL, ("%s: no L2 table", __func__));
+ hash = l2_hash(d, sa, ifp->if_index);
rw_rlock(&d->lock);
for (e = d->l2tab[hash].first; e; e = e->next) {
- if (e->addr == addr && e->ifp == ifp) {
+ if (l2_cmp(sa, e) == 0 && e->ifp == ifp) {
mtx_lock(&e->lock);
if (atomic_load_acq_int(&e->refcnt))
goto found;
Modified: head/sys/modules/cxgbe/tom/Makefile
==============================================================================
--- head/sys/modules/cxgbe/tom/Makefile Mon Jan 14 20:31:45 2013
(r245433)
+++ head/sys/modules/cxgbe/tom/Makefile Mon Jan 14 20:36:22 2013
(r245434)
@@ -10,15 +10,20 @@ CXGBE = ${.CURDIR}/../../../dev/cxgbe
KMOD = t4_tom
SRCS = t4_tom.c t4_connect.c t4_listen.c t4_cpl_io.c t4_tom_l2t.c t4_ddp.c
SRCS+= device_if.h bus_if.h pci_if.h
-SRCS+= opt_inet.h
+SRCS+= opt_inet.h opt_inet6.h
CFLAGS+= -I${CXGBE}
.if !defined(KERNBUILDDIR)
.if ${MK_INET_SUPPORT} != "no"
opt_inet.h:
- echo "#define INET 1" > ${.TARGET}
- echo "#define TCP_OFFLOAD 1" >> ${.TARGET}
+ @echo "#define INET 1" > ${.TARGET}
+ @echo "#define TCP_OFFLOAD 1" >> ${.TARGET}
+.endif
+
+.if ${MK_INET6_SUPPORT} != "no"
+opt_inet6.h:
+ @echo "#define INET6 1" > ${.TARGET}
.endif
.endif
_______________________________________________
svn-src-head@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/svn-src-head
To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
