/*
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
*
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2011, 2012, Intel Corporation.
*
* This file is part of Portals
* http://sourceforge.net/projects/sandiaportals/
#if defined(__KERNEL__) && defined(LNET_ROUTER)
-#define LNET_NRB_TINY 1024
-#define LNET_NRB_SMALL 8192
-#define LNET_NRB_LARGE 512
+#define LNET_NRB_TINY_MIN 512 /* min value for each CPT */
+#define LNET_NRB_TINY (LNET_NRB_TINY_MIN * 4)
+#define LNET_NRB_SMALL_MIN 4096 /* min value for each CPT */
+#define LNET_NRB_SMALL (LNET_NRB_SMALL_MIN * 4)
+#define LNET_NRB_LARGE_MIN 256 /* min value for each CPT */
+#define LNET_NRB_LARGE (LNET_NRB_LARGE_MIN * 4)
static char *forwarding = "";
CFS_MODULE_PARM(forwarding, "s", charp, 0444,
static int check_routers_before_use = 0;
CFS_MODULE_PARM(check_routers_before_use, "i", int, 0444,
- "Assume routers are down and ping them before use");
+ "Assume routers are down and ping them before use");
-static int avoid_asym_router_failure = 0;
-CFS_MODULE_PARM(avoid_asym_router_failure, "i", int, 0444,
- "Avoid asymmetrical failures: reserved, use at your own risk");
+static int avoid_asym_router_failure = 1;
+CFS_MODULE_PARM(avoid_asym_router_failure, "i", int, 0644,
+ "Avoid asymmetrical router failures (0 to disable)");
-static int dead_router_check_interval = 0;
-CFS_MODULE_PARM(dead_router_check_interval, "i", int, 0444,
- "Seconds between dead router health checks (<= 0 to disable)");
+static int dead_router_check_interval = 60;
+CFS_MODULE_PARM(dead_router_check_interval, "i", int, 0644,
+ "Seconds between dead router health checks (<= 0 to disable)");
-static int live_router_check_interval = 0;
-CFS_MODULE_PARM(live_router_check_interval, "i", int, 0444,
- "Seconds between live router health checks (<= 0 to disable)");
+static int live_router_check_interval = 60;
+CFS_MODULE_PARM(live_router_check_interval, "i", int, 0644,
+ "Seconds between live router health checks (<= 0 to disable)");
static int router_ping_timeout = 50;
-CFS_MODULE_PARM(router_ping_timeout, "i", int, 0444,
- "Seconds to wait for the reply to a router health query");
+CFS_MODULE_PARM(router_ping_timeout, "i", int, 0644,
+ "Seconds to wait for the reply to a router health query");
int
lnet_peers_start_down(void)
lp->lp_notify = 1;
lp->lp_notifylnd |= notifylnd;
if (lp->lp_alive)
- lp->lp_ping_version = LNET_PROTO_PING_UNKNOWN; /* reset */
+ lp->lp_ping_feats = LNET_PING_FEAT_INVAL; /* reset */
- CDEBUG(D_NET, "set %s %d\n", libcfs_nid2str(lp->lp_nid), alive);
+ CDEBUG(D_NET, "set %s %d\n", libcfs_nid2str(lp->lp_nid), alive);
}
void
lp->lp_notify = 0;
if (notifylnd && ni->ni_lnd->lnd_notify != NULL) {
- LNET_UNLOCK();
+ lnet_net_unlock(lp->lp_cpt);
- /* A new notification could happen now; I'll handle it
- * when control returns to me */
+ /* A new notification could happen now; I'll handle it
+ * when control returns to me */
- (ni->ni_lnd->lnd_notify)(ni, lp->lp_nid, alive);
+ (ni->ni_lnd->lnd_notify)(ni, lp->lp_nid, alive);
- LNET_LOCK();
- }
- }
+ lnet_net_lock(lp->lp_cpt);
+ }
+ }
- lp->lp_notifying = 0;
+ lp->lp_notifying = 0;
}
static void
lnet_rtr_addref_locked(lnet_peer_t *lp)
{
- LASSERT (lp->lp_refcount > 0);
- LASSERT (lp->lp_rtr_refcount >= 0);
+ LASSERT(lp->lp_refcount > 0);
+ LASSERT(lp->lp_rtr_refcount >= 0);
+ /* lnet_net_lock must be exclusively locked */
lp->lp_rtr_refcount++;
if (lp->lp_rtr_refcount == 1) {
cfs_list_t *pos;
static void
lnet_rtr_decref_locked(lnet_peer_t *lp)
{
- LASSERT (lp->lp_refcount > 0);
- LASSERT (lp->lp_rtr_refcount > 0);
+ LASSERT(lp->lp_refcount > 0);
+ LASSERT(lp->lp_rtr_refcount > 0);
- lp->lp_rtr_refcount--;
- if (lp->lp_rtr_refcount == 0) {
+ /* lnet_net_lock must be exclusively locked */
+ lp->lp_rtr_refcount--;
+ if (lp->lp_rtr_refcount == 0) {
LASSERT(cfs_list_empty(&lp->lp_routes));
- if (lp->lp_rcd != NULL) {
- cfs_list_add(&lp->lp_rcd->rcd_list,
+ if (lp->lp_rcd != NULL) {
+ cfs_list_add(&lp->lp_rcd->rcd_list,
&the_lnet.ln_rcd_deathrow);
lp->lp_rcd = NULL;
}
lnet_remotenet_t *
lnet_find_net_locked (__u32 net)
{
- lnet_remotenet_t *rnet;
- cfs_list_t *tmp;
+ lnet_remotenet_t *rnet;
+ cfs_list_t *tmp;
+ cfs_list_t *rn_list;
- LASSERT (!the_lnet.ln_shutdown);
+ LASSERT(!the_lnet.ln_shutdown);
- cfs_list_for_each (tmp, &the_lnet.ln_remote_nets) {
- rnet = cfs_list_entry(tmp, lnet_remotenet_t, lrn_list);
+ rn_list = lnet_net2rnethash(net);
+ cfs_list_for_each(tmp, rn_list) {
+ rnet = cfs_list_entry(tmp, lnet_remotenet_t, lrn_list);
- if (rnet->lrn_net == net)
- return rnet;
- }
- return NULL;
+ if (rnet->lrn_net == net)
+ return rnet;
+ }
+ return NULL;
}
static void lnet_shuffle_seed(void)
cfs_list_add(&route->lr_list, e);
cfs_list_add(&route->lr_gwlist, &route->lr_gateway->lp_routes);
- the_lnet.ln_remote_nets_version++;
- lnet_rtr_addref_locked(route->lr_gateway);
+ the_lnet.ln_remote_nets_version++;
+ lnet_rtr_addref_locked(route->lr_gateway);
}
int
route->lr_hops = hops;
route->lr_net = net;
- LNET_LOCK();
+ lnet_net_lock(LNET_LOCK_EX);
- rc = lnet_nid2peer_locked(&route->lr_gateway, gateway);
- if (rc != 0) {
- LNET_UNLOCK();
+ rc = lnet_nid2peer_locked(&route->lr_gateway, gateway, LNET_LOCK_EX);
+ if (rc != 0) {
+ lnet_net_unlock(LNET_LOCK_EX);
- LIBCFS_FREE(route, sizeof(*route));
- LIBCFS_FREE(rnet, sizeof(*rnet));
+ LIBCFS_FREE(route, sizeof(*route));
+ LIBCFS_FREE(rnet, sizeof(*rnet));
if (rc == -EHOSTUNREACH) { /* gateway is not on a local net */
- return 0; /* ignore the route entry */
+ return 0; /* ignore the route entry */
} else {
CERROR("Error %d creating route %s %d %s\n", rc,
libcfs_net2str(net), hops,
rnet2 = lnet_find_net_locked(net);
if (rnet2 == NULL) {
/* new network */
- cfs_list_add_tail(&rnet->lrn_list, &the_lnet.ln_remote_nets);
+ cfs_list_add_tail(&rnet->lrn_list, lnet_net2rnethash(net));
rnet2 = rnet;
}
lnet_add_route_to_rnet(rnet2, route);
ni = route->lr_gateway->lp_ni;
- LNET_UNLOCK();
+ lnet_net_unlock(LNET_LOCK_EX);
/* XXX Assume alive */
if (ni->ni_lnd->lnd_notify != NULL)
(ni->ni_lnd->lnd_notify)(ni, gateway, 1);
- LNET_LOCK();
+ lnet_net_lock(LNET_LOCK_EX);
}
/* -1 for notify or !add_route */
lnet_peer_decref_locked(route->lr_gateway);
- LNET_UNLOCK();
+ lnet_net_unlock(LNET_LOCK_EX);
if (!add_route)
LIBCFS_FREE(route, sizeof(*route));
- if (rnet != rnet2)
- LIBCFS_FREE(rnet, sizeof(*rnet));
+ if (rnet != rnet2)
+ LIBCFS_FREE(rnet, sizeof(*rnet));
- return 0;
+ return 0;
}
int
-lnet_check_routes (void)
+lnet_check_routes(void)
{
- lnet_remotenet_t *rnet;
- lnet_route_t *route;
- lnet_route_t *route2;
- cfs_list_t *e1;
- cfs_list_t *e2;
+ lnet_remotenet_t *rnet;
+ lnet_route_t *route;
+ lnet_route_t *route2;
+ cfs_list_t *e1;
+ cfs_list_t *e2;
+ int cpt;
+ cfs_list_t *rn_list;
+ int i;
- LNET_LOCK();
+ cpt = lnet_net_lock_current();
- cfs_list_for_each (e1, &the_lnet.ln_remote_nets) {
- rnet = cfs_list_entry(e1, lnet_remotenet_t, lrn_list);
+ for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
+ rn_list = &the_lnet.ln_remote_nets_hash[i];
+ cfs_list_for_each(e1, rn_list) {
+ rnet = cfs_list_entry(e1, lnet_remotenet_t, lrn_list);
- route2 = NULL;
- cfs_list_for_each (e2, &rnet->lrn_routes) {
- lnet_nid_t nid1;
- lnet_nid_t nid2;
- int net;
+ route2 = NULL;
+ cfs_list_for_each(e2, &rnet->lrn_routes) {
+ lnet_nid_t nid1;
+ lnet_nid_t nid2;
+ int net;
- route = cfs_list_entry(e2, lnet_route_t, lr_list);
+ route = cfs_list_entry(e2, lnet_route_t,
+ lr_list);
- if (route2 == NULL) {
- route2 = route;
- continue;
- }
+ if (route2 == NULL) {
+ route2 = route;
+ continue;
+ }
- if (route->lr_gateway->lp_ni ==
- route2->lr_gateway->lp_ni)
- continue;
+ if (route->lr_gateway->lp_ni ==
+ route2->lr_gateway->lp_ni)
+ continue;
- nid1 = route->lr_gateway->lp_nid;
- nid2 = route2->lr_gateway->lp_nid;
- net = rnet->lrn_net;
+ nid1 = route->lr_gateway->lp_nid;
+ nid2 = route2->lr_gateway->lp_nid;
+ net = rnet->lrn_net;
- LNET_UNLOCK();
+ lnet_net_unlock(cpt);
- CERROR("Routes to %s via %s and %s not supported\n",
- libcfs_net2str(net), libcfs_nid2str(nid1),
- libcfs_nid2str(nid2));
- return -EINVAL;
- }
- }
+ CERROR("Routes to %s via %s and %s not "
+ "supported\n",
+ libcfs_net2str(net),
+ libcfs_nid2str(nid1),
+ libcfs_nid2str(nid2));
+ return -EINVAL;
+ }
+ }
+ }
- LNET_UNLOCK();
- return 0;
+ lnet_net_unlock(cpt);
+ return 0;
}
int
-lnet_del_route (__u32 net, lnet_nid_t gw_nid)
+lnet_del_route(__u32 net, lnet_nid_t gw_nid)
{
struct lnet_peer *gateway;
- lnet_remotenet_t *rnet;
- lnet_route_t *route;
- cfs_list_t *e1;
- cfs_list_t *e2;
- int rc = -ENOENT;
-
- CDEBUG(D_NET, "Del route: net %s : gw %s\n",
- libcfs_net2str(net), libcfs_nid2str(gw_nid));
-
- /* NB Caller may specify either all routes via the given gateway
- * or a specific route entry actual NIDs) */
+ lnet_remotenet_t *rnet;
+ lnet_route_t *route;
+ cfs_list_t *e1;
+ cfs_list_t *e2;
+ int rc = -ENOENT;
+ cfs_list_t *rn_list;
+ int idx = 0;
+
+ CDEBUG(D_NET, "Del route: net %s : gw %s\n",
+ libcfs_net2str(net), libcfs_nid2str(gw_nid));
+
+ /* NB Caller may specify either all routes via the given gateway
+ * or a specific route entry actual NIDs) */
+
+ lnet_net_lock(LNET_LOCK_EX);
+ if (net == LNET_NIDNET(LNET_NID_ANY))
+ rn_list = &the_lnet.ln_remote_nets_hash[0];
+ else
+ rn_list = lnet_net2rnethash(net);
again:
- LNET_LOCK();
+ cfs_list_for_each(e1, rn_list) {
+ rnet = cfs_list_entry(e1, lnet_remotenet_t, lrn_list);
- cfs_list_for_each (e1, &the_lnet.ln_remote_nets) {
- rnet = cfs_list_entry(e1, lnet_remotenet_t, lrn_list);
-
- if (!(net == LNET_NIDNET(LNET_NID_ANY) ||
- net == rnet->lrn_net))
- continue;
+ if (!(net == LNET_NIDNET(LNET_NID_ANY) ||
+ net == rnet->lrn_net))
+ continue;
- cfs_list_for_each (e2, &rnet->lrn_routes) {
- route = cfs_list_entry(e2, lnet_route_t, lr_list);
+ cfs_list_for_each(e2, &rnet->lrn_routes) {
+ route = cfs_list_entry(e2, lnet_route_t, lr_list);
gateway = route->lr_gateway;
if (!(gw_nid == LNET_NID_ANY ||
cfs_list_del(&route->lr_list);
cfs_list_del(&route->lr_gwlist);
- the_lnet.ln_remote_nets_version++;
+ the_lnet.ln_remote_nets_version++;
- if (cfs_list_empty(&rnet->lrn_routes))
- cfs_list_del(&rnet->lrn_list);
- else
- rnet = NULL;
+ if (cfs_list_empty(&rnet->lrn_routes))
+ cfs_list_del(&rnet->lrn_list);
+ else
+ rnet = NULL;
lnet_rtr_decref_locked(gateway);
lnet_peer_decref_locked(gateway);
- LNET_UNLOCK();
- LIBCFS_FREE(route, sizeof (*route));
+ lnet_net_unlock(LNET_LOCK_EX);
- if (rnet != NULL)
- LIBCFS_FREE(rnet, sizeof(*rnet));
+ LIBCFS_FREE(route, sizeof(*route));
- rc = 0;
- goto again;
- }
- }
+ if (rnet != NULL)
+ LIBCFS_FREE(rnet, sizeof(*rnet));
+
+ rc = 0;
+ lnet_net_lock(LNET_LOCK_EX);
+ goto again;
+ }
+ }
+
+ if (net == LNET_NIDNET(LNET_NID_ANY) &&
+ ++idx < LNET_REMOTE_NETS_HASH_SIZE) {
+ rn_list = &the_lnet.ln_remote_nets_hash[idx];
+ goto again;
+ }
+ lnet_net_unlock(LNET_LOCK_EX);
- LNET_UNLOCK();
- return rc;
+ return rc;
}
void
}
int
-lnet_get_route (int idx, __u32 *net, __u32 *hops,
- lnet_nid_t *gateway, __u32 *alive)
+lnet_get_route(int idx, __u32 *net, __u32 *hops,
+ lnet_nid_t *gateway, __u32 *alive)
{
- cfs_list_t *e1;
- cfs_list_t *e2;
- lnet_remotenet_t *rnet;
- lnet_route_t *route;
-
- LNET_LOCK();
-
- cfs_list_for_each (e1, &the_lnet.ln_remote_nets) {
- rnet = cfs_list_entry(e1, lnet_remotenet_t, lrn_list);
-
- cfs_list_for_each (e2, &rnet->lrn_routes) {
- route = cfs_list_entry(e2, lnet_route_t, lr_list);
-
- if (idx-- == 0) {
- *net = rnet->lrn_net;
- *hops = route->lr_hops;
- *gateway = route->lr_gateway->lp_nid;
- *alive = route->lr_gateway->lp_alive;
- LNET_UNLOCK();
- return 0;
- }
- }
- }
+ cfs_list_t *e1;
+ cfs_list_t *e2;
+ lnet_remotenet_t *rnet;
+ lnet_route_t *route;
+ int cpt;
+ int i;
+ cfs_list_t *rn_list;
+
+ cpt = lnet_net_lock_current();
+
+ for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
+ rn_list = &the_lnet.ln_remote_nets_hash[i];
+ cfs_list_for_each(e1, rn_list) {
+ rnet = cfs_list_entry(e1, lnet_remotenet_t, lrn_list);
+
+ cfs_list_for_each(e2, &rnet->lrn_routes) {
+ route = cfs_list_entry(e2, lnet_route_t,
+ lr_list);
+
+ if (idx-- == 0) {
+ *net = rnet->lrn_net;
+ *hops = route->lr_hops;
+ *gateway = route->lr_gateway->lp_nid;
+ *alive = route->lr_gateway->lp_alive;
+ lnet_net_unlock(cpt);
+ return 0;
+ }
+ }
+ }
+ }
- LNET_UNLOCK();
- return -ENOENT;
+ lnet_net_unlock(cpt);
+ return -ENOENT;
}
void
lnet_swap_pinginfo(lnet_ping_info_t *info)
{
- int i;
- lnet_ni_status_t *stat;
+ int i;
+ lnet_ni_status_t *stat;
__swab32s(&info->pi_magic);
- __swab32s(&info->pi_version);
+ __swab32s(&info->pi_features);
__swab32s(&info->pi_pid);
__swab32s(&info->pi_nnis);
for (i = 0; i < info->pi_nnis && i < LNET_MAX_RTR_NIS; i++) {
if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
CDEBUG(D_NET, "%s: Unexpected magic %08x\n",
libcfs_nid2str(gw->lp_nid), info->pi_magic);
- gw->lp_ping_version = LNET_PROTO_PING_UNKNOWN;
+ gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
return;
}
- gw->lp_ping_version = info->pi_version;
- if (gw->lp_ping_version == LNET_PROTO_PING_VERSION_1)
- return; /* v1 doesn't carry NI status info */
-
- if (gw->lp_ping_version != LNET_PROTO_PING_VERSION) {
- CDEBUG(D_NET, "%s: Unexpected version 0x%x\n",
- libcfs_nid2str(gw->lp_nid), gw->lp_ping_version);
- gw->lp_ping_version = LNET_PROTO_PING_UNKNOWN;
- return;
+ gw->lp_ping_feats = info->pi_features;
+ if ((gw->lp_ping_feats & LNET_PING_FEAT_MASK) == 0) {
+ CDEBUG(D_NET, "%s: Unexpected features 0x%x\n",
+ libcfs_nid2str(gw->lp_nid), gw->lp_ping_feats);
+ return; /* nothing I can understand */
}
+ if ((gw->lp_ping_feats & LNET_PING_FEAT_NI_STATUS) == 0)
+ return; /* can't carry NI status info */
+
cfs_list_for_each_entry(rtr, &gw->lp_routes, lr_gwlist) {
int ptl_status = LNET_NI_STATUS_INVALID;
int down = 0;
if (nid == LNET_NID_ANY) {
CDEBUG(D_NET, "%s: unexpected LNET_NID_ANY\n",
libcfs_nid2str(gw->lp_nid));
- gw->lp_ping_version = LNET_PROTO_PING_UNKNOWN;
+ gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
return;
}
CDEBUG(D_NET, "%s: Unexpected status 0x%x\n",
libcfs_nid2str(gw->lp_nid), stat->ns_status);
- gw->lp_ping_version = LNET_PROTO_PING_UNKNOWN;
+ gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
return;
}
if (event->unlinked) {
LNetInvalidateHandle(&rcd->rcd_mdh);
return;
- }
+ }
LASSERT(event->type == LNET_EVENT_SEND ||
event->type == LNET_EVENT_REPLY);
lp = rcd->rcd_gateway;
LASSERT(lp != NULL);
- LNET_LOCK();
+ /* NB: it's called with holding lnet_res_lock, we have a few
+ * places need to hold both locks at the same time, please take
+ * care of lock ordering */
+ lnet_net_lock(lp->lp_cpt);
if (!lnet_isrouter(lp) || lp->lp_rcd != rcd) {
/* ignore if no longer a router or rcd is replaced */
goto out;
lnet_parse_rc_info(rcd);
out:
- LNET_UNLOCK();
+ lnet_net_unlock(lp->lp_cpt);
}
void
LASSERT (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);
for (;;) {
- LNET_LOCK();
+ int cpt = lnet_net_lock_current();
all_known = 1;
cfs_list_for_each (entry, &the_lnet.ln_routers) {
}
}
- LNET_UNLOCK();
+ lnet_net_unlock(cpt);
if (all_known)
return;
if (now < ni->ni_last_alive + timeout)
continue;
+ lnet_ni_lock(ni);
+ /* re-check with lock */
+ if (now < ni->ni_last_alive + timeout) {
+ lnet_ni_unlock(ni);
+ continue;
+ }
+
LASSERT(ni->ni_status != NULL);
if (ni->ni_status->ns_status != LNET_NI_STATUS_DOWN) {
* NI status to "down" */
ni->ni_status->ns_status = LNET_NI_STATUS_DOWN;
}
+ lnet_ni_unlock(ni);
}
}
void
-lnet_destroy_rc_data (lnet_rc_data_t *rcd)
+lnet_destroy_rc_data(lnet_rc_data_t *rcd)
{
LASSERT(cfs_list_empty(&rcd->rcd_list));
/* detached from network */
LASSERT(LNetHandleIsInvalid(rcd->rcd_mdh));
if (rcd->rcd_gateway != NULL) {
- LNET_LOCK();
+ int cpt = rcd->rcd_gateway->lp_cpt;
+
+ lnet_net_lock(cpt);
lnet_peer_decref_locked(rcd->rcd_gateway);
- LNET_UNLOCK();
+ lnet_net_unlock(cpt);
}
if (rcd->rcd_pinginfo != NULL)
int rc;
int i;
- LNET_UNLOCK();
+ lnet_net_unlock(gateway->lp_cpt);
LIBCFS_ALLOC(rcd, sizeof(*rcd));
if (rcd == NULL)
}
LASSERT(rc == 0);
- LNET_LOCK();
+ lnet_net_lock(gateway->lp_cpt);
/* router table changed or someone has created rcd for this gateway */
if (!lnet_isrouter(gateway) || gateway->lp_rcd != NULL) {
- LNET_UNLOCK();
+ lnet_net_unlock(gateway->lp_cpt);
goto out;
}
lnet_destroy_rc_data(rcd);
}
- LNET_LOCK();
+ lnet_net_lock(gateway->lp_cpt);
return gateway->lp_rcd;
}
mdh = rcd->rcd_mdh;
if (rtr->lp_ping_deadline == 0) {
- rtr->lp_ping_deadline = \
+ rtr->lp_ping_deadline =
cfs_time_shift(router_ping_timeout);
}
- LNET_UNLOCK();
+ lnet_net_unlock(rtr->lp_cpt);
- rc = LNetGet(LNET_NID_ANY, mdh, id, LNET_RESERVED_PORTAL,
- LNET_PROTO_PING_MATCHBITS, 0);
+ rc = LNetGet(LNET_NID_ANY, mdh, id, LNET_RESERVED_PORTAL,
+ LNET_PROTO_PING_MATCHBITS, 0);
- LNET_LOCK();
+ lnet_net_lock(rtr->lp_cpt);
if (rc != 0)
rtr->lp_ping_notsent = 0; /* no event pending */
}
LASSERT (check_routers_before_use);
LASSERT (dead_router_check_interval > 0);
- LNET_LOCK();
+ lnet_net_lock(0);
/* As an approximation, allow each router the same number of
* outstanding events as it is allowed outstanding sends */
id.nid = rtr->lp_nid;
id.pid = LUSTRE_SRV_LNET_PID;
- LNET_UNLOCK();
+ lnet_net_unlock(0);
rc = LNetSetAsync(id, 1);
if (rc != 0) {
return rc;
}
- LNET_LOCK();
- /* NB router list doesn't change in userspace */
- LASSERT (version == the_lnet.ln_routers_version);
- }
+ lnet_net_lock(0);
+ /* NB router list doesn't change in userspace */
+ LASSERT(version == the_lnet.ln_routers_version);
+ }
- LNET_UNLOCK();
+ lnet_net_unlock(0);
if (nrtr == 0) {
CDEBUG(D_NET,
return 0;
#ifdef __KERNEL__
- cfs_sema_init(&the_lnet.ln_rc_signal, 0);
+ sema_init(&the_lnet.ln_rc_signal, 0);
/* EQ size doesn't matter; the callback is guaranteed to get every
* event */
eqsz = 0;
if (rc < 0) {
CERROR("Can't start router checker thread: %d\n", rc);
/* block until event callback signals exit */
- cfs_down(&the_lnet.ln_rc_signal);
+ down(&the_lnet.ln_rc_signal);
rc = LNetEQFree(the_lnet.ln_rc_eqh);
LASSERT (rc == 0);
the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN;
#ifdef __KERNEL__
/* block until event callback signals exit */
- cfs_down(&the_lnet.ln_rc_signal);
+ down(&the_lnet.ln_rc_signal);
#else
lnet_router_checker();
#endif
CFS_INIT_LIST_HEAD(&head);
- LNET_LOCK();
+ lnet_net_lock(LNET_LOCK_EX);
if (the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING) {
/* router checker is stopping, prune all */
cfs_list_splice_init(&the_lnet.ln_rcd_deathrow, &head);
if (!cfs_list_empty(&head)) {
- LNET_UNLOCK();
+ lnet_net_unlock(LNET_LOCK_EX);
cfs_list_for_each_entry(rcd, &head, rcd_list)
LNetMDUnlink(rcd->rcd_mdh);
- LNET_LOCK();
- }
+ lnet_net_lock(LNET_LOCK_EX);
+ }
cfs_list_splice_init(&head, &the_lnet.ln_rcd_zombie);
while (!cfs_list_empty(&the_lnet.ln_rcd_zombie)) {
cfs_list_for_each_entry_safe(rcd, tmp, &the_lnet.ln_rcd_zombie,
rcd_list) {
- if (!LNetHandleIsInvalid(rcd->rcd_mdh))
+ if (LNetHandleIsInvalid(rcd->rcd_mdh))
cfs_list_move(&rcd->rcd_list, &head);
- }
+ }
wait_unlink = wait_unlink &&
!cfs_list_empty(&the_lnet.ln_rcd_zombie);
- LNET_UNLOCK();
+ lnet_net_unlock(LNET_LOCK_EX);
while (!cfs_list_empty(&head)) {
rcd = cfs_list_entry(head.next,
}
if (!wait_unlink)
- break;
+ return;
i++;
CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET,
"Waiting for rc buffers to unlink\n");
cfs_pause(cfs_time_seconds(1) / 4);
- LNET_LOCK();
+ lnet_net_lock(LNET_LOCK_EX);
}
+
+ lnet_net_unlock(LNET_LOCK_EX);
}
LASSERT (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);
while (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING) {
- __u64 version;
+ __u64 version;
+ int cpt;
+ int cpt2;
- LNET_LOCK();
+ cpt = lnet_net_lock_current();
rescan:
- version = the_lnet.ln_routers_version;
+ version = the_lnet.ln_routers_version;
+
+ cfs_list_for_each(entry, &the_lnet.ln_routers) {
+ rtr = cfs_list_entry(entry, lnet_peer_t, lp_rtr_list);
+
+ cpt2 = lnet_cpt_of_nid_locked(rtr->lp_nid);
+ if (cpt != cpt2) {
+ lnet_net_unlock(cpt);
+ cpt = cpt2;
+ lnet_net_lock(cpt);
+ /* the routers list has changed */
+ if (version != the_lnet.ln_routers_version)
+ goto rescan;
+ }
- cfs_list_for_each (entry, &the_lnet.ln_routers) {
- rtr = cfs_list_entry(entry, lnet_peer_t, lp_rtr_list);
lnet_ping_router_locked(rtr);
/* NB dropped lock */
if (the_lnet.ln_routing)
lnet_update_ni_status_locked();
- LNET_UNLOCK();
+ lnet_net_unlock(cpt);
lnet_prune_rc_data(0); /* don't wait for UNLINK */
lnet_prune_rc_data(1); /* wait for UNLINK */
the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN;
- cfs_up(&the_lnet.ln_rc_signal);
- /* The unlink event callback will signal final completion */
- return 0;
+ up(&the_lnet.ln_rc_signal);
+ /* The unlink event callback will signal final completion */
+ return 0;
}
void
}
lnet_rtrbuf_t *
-lnet_new_rtrbuf(lnet_rtrbufpool_t *rbp)
+lnet_new_rtrbuf(lnet_rtrbufpool_t *rbp, int cpt)
{
int npages = rbp->rbp_npages;
int sz = offsetof(lnet_rtrbuf_t, rb_kiov[npages]);
lnet_rtrbuf_t *rb;
int i;
- LIBCFS_ALLOC(rb, sz);
- if (rb == NULL)
- return NULL;
+ LIBCFS_CPT_ALLOC(rb, lnet_cpt_table(), cpt, sz);
+ if (rb == NULL)
+ return NULL;
- rb->rb_pool = rbp;
+ rb->rb_pool = rbp;
- for (i = 0; i < npages; i++) {
- page = cfs_alloc_page(CFS_ALLOC_ZERO | CFS_ALLOC_STD);
+ for (i = 0; i < npages; i++) {
+ page = cfs_page_cpt_alloc(lnet_cpt_table(), cpt,
+ CFS_ALLOC_ZERO | CFS_ALLOC_STD);
if (page == NULL) {
while (--i >= 0)
cfs_free_page(rb->rb_kiov[i].kiov_page);
}
int
-lnet_rtrpool_alloc_bufs(lnet_rtrbufpool_t *rbp, int nbufs)
+lnet_rtrpool_alloc_bufs(lnet_rtrbufpool_t *rbp, int nbufs, int cpt)
{
lnet_rtrbuf_t *rb;
int i;
}
for (i = 0; i < nbufs; i++) {
- rb = lnet_new_rtrbuf(rbp);
+ rb = lnet_new_rtrbuf(rbp, cpt);
if (rb == NULL) {
CERROR("Failed to allocate %d router bufs of %d pages\n",
void
lnet_rtrpools_free(void)
{
+ lnet_rtrbufpool_t *rtrp;
+ int i;
+
if (the_lnet.ln_rtrpools == NULL) /* uninitialized or freed */
return;
- lnet_rtrpool_free_bufs(&the_lnet.ln_rtrpools[0]);
- lnet_rtrpool_free_bufs(&the_lnet.ln_rtrpools[1]);
- lnet_rtrpool_free_bufs(&the_lnet.ln_rtrpools[2]);
+ cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
+ lnet_rtrpool_free_bufs(&rtrp[0]);
+ lnet_rtrpool_free_bufs(&rtrp[1]);
+ lnet_rtrpool_free_bufs(&rtrp[2]);
+ }
- LIBCFS_FREE(the_lnet.ln_rtrpools,
- sizeof(lnet_rtrbufpool_t) * LNET_NRBPOOLS);
+ cfs_percpt_free(the_lnet.ln_rtrpools);
the_lnet.ln_rtrpools = NULL;
}
static int
lnet_nrb_tiny_calculate(int npages)
{
- if (tiny_router_buffers > 0)
- return tiny_router_buffers;
+ int nrbs = LNET_NRB_TINY;
- if (tiny_router_buffers == 0)
- return LNET_NRB_TINY;
+ if (tiny_router_buffers < 0) {
+ LCONSOLE_ERROR_MSG(0x10c,
+ "tiny_router_buffers=%d invalid when "
+ "routing enabled\n", tiny_router_buffers);
+ return -1;
+ }
+
+ if (tiny_router_buffers > 0)
+ nrbs = tiny_router_buffers;
- LCONSOLE_ERROR_MSG(0x10c, "tiny_router_buffers=%d invalid when "
- "routing enabled\n", tiny_router_buffers);
- return -1;
+ nrbs /= LNET_CPT_NUMBER;
+ return max(nrbs, LNET_NRB_TINY_MIN);
}
static int
lnet_nrb_small_calculate(int npages)
{
- if (small_router_buffers > 0)
- return tiny_router_buffers;
+ int nrbs = LNET_NRB_SMALL;
+
+ if (small_router_buffers < 0) {
+ LCONSOLE_ERROR_MSG(0x10c,
+ "small_router_buffers=%d invalid when "
+ "routing enabled\n", small_router_buffers);
+ return -1;
+ }
- if (small_router_buffers == 0)
- return LNET_NRB_SMALL;
+ if (small_router_buffers > 0)
+ nrbs = small_router_buffers;
- LCONSOLE_ERROR_MSG(0x10d, "small_router_buffers=%d invalid when "
- "routing enabled\n", small_router_buffers);
- return -1;
+ nrbs /= LNET_CPT_NUMBER;
+ return max(nrbs, LNET_NRB_SMALL_MIN);
}
static int
lnet_nrb_large_calculate(int npages)
{
- if (large_router_buffers > 0)
- return large_router_buffers;
+ int nrbs = LNET_NRB_LARGE;
- if (large_router_buffers == 0)
- return LNET_NRB_LARGE;
+ if (large_router_buffers < 0) {
+ LCONSOLE_ERROR_MSG(0x10c,
+ "large_router_buffers=%d invalid when "
+ "routing enabled\n", large_router_buffers);
+ return -1;
+ }
+
+ if (large_router_buffers > 0)
+ nrbs = large_router_buffers;
- LCONSOLE_ERROR_MSG(0x10e, "large_router_buffers=%d invalid when"
- " routing enabled\n", large_router_buffers);
- return -1;
+ nrbs /= LNET_CPT_NUMBER;
+ return max(nrbs, LNET_NRB_LARGE_MIN);
}
int
int nrb_small;
int nrb_large;
int rc;
+ int i;
if (!strcmp(forwarding, "")) {
/* not set either way */
if (nrb_large < 0)
return -EINVAL;
- LIBCFS_ALLOC(the_lnet.ln_rtrpools,
- sizeof(lnet_rtrbufpool_t) * LNET_NRBPOOLS);
+ the_lnet.ln_rtrpools = cfs_percpt_alloc(lnet_cpt_table(),
+ LNET_NRBPOOLS *
+ sizeof(lnet_rtrbufpool_t));
if (the_lnet.ln_rtrpools == NULL) {
LCONSOLE_ERROR_MSG(0x10c,
"Failed to initialize router buffe pool\n");
return -ENOMEM;
}
- do { /* iterate over rtrpools on all CPTs in upcoming patches */
- rtrp = the_lnet.ln_rtrpools;
-
+ cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
lnet_rtrpool_init(&rtrp[0], 0);
- rc = lnet_rtrpool_alloc_bufs(&rtrp[0], nrb_tiny);
+ rc = lnet_rtrpool_alloc_bufs(&rtrp[0], nrb_tiny, i);
if (rc != 0)
goto failed;
lnet_rtrpool_init(&rtrp[1], small_pages);
- rc = lnet_rtrpool_alloc_bufs(&rtrp[1], nrb_small);
+ rc = lnet_rtrpool_alloc_bufs(&rtrp[1], nrb_small, i);
if (rc != 0)
goto failed;
lnet_rtrpool_init(&rtrp[2], large_pages);
- rc = lnet_rtrpool_alloc_bufs(&rtrp[2], nrb_large);
+ rc = lnet_rtrpool_alloc_bufs(&rtrp[2], nrb_large, i);
if (rc != 0)
goto failed;
- } while (0);
+ }
- LNET_LOCK();
+ lnet_net_lock(LNET_LOCK_EX);
the_lnet.ln_routing = 1;
- LNET_UNLOCK();
+ lnet_net_unlock(LNET_LOCK_EX);
return 0;
}
int
-lnet_notify (lnet_ni_t *ni, lnet_nid_t nid, int alive, cfs_time_t when)
+lnet_notify(lnet_ni_t *ni, lnet_nid_t nid, int alive, cfs_time_t when)
{
- lnet_peer_t *lp = NULL;
- cfs_time_t now = cfs_time_current();
+ struct lnet_peer *lp = NULL;
+ cfs_time_t now = cfs_time_current();
+ int cpt = lnet_cpt_of_nid(nid);
LASSERT (!cfs_in_interrupt ());
return 0;
}
- LNET_LOCK();
+ lnet_net_lock(cpt);
- lp = lnet_find_peer_locked(nid);
- if (lp == NULL) {
- /* nid not found */
- LNET_UNLOCK();
+ if (the_lnet.ln_shutdown) {
+ lnet_net_unlock(cpt);
+ return -ESHUTDOWN;
+ }
+
+ lp = lnet_find_peer_locked(the_lnet.ln_peer_tables[cpt], nid);
+ if (lp == NULL) {
+ /* nid not found */
+ lnet_net_unlock(cpt);
CDEBUG(D_NET, "%s not found\n", libcfs_nid2str(nid));
return 0;
}
lnet_ni_notify_locked(ni, lp);
- lnet_peer_decref_locked(lp);
+ lnet_peer_decref_locked(lp);
- LNET_UNLOCK();
- return 0;
+ lnet_net_unlock(cpt);
+ return 0;
}
EXPORT_SYMBOL(lnet_notify);
live_router_check_interval, dead_router_check_interval,
interval);
- LNET_LOCK();
- LASSERT (!running); /* recursion check */
- running = 1;
- LNET_UNLOCK();
+ LASSERT(LNET_CPT_NUMBER == 1);
+
+ lnet_net_lock(0);
+ LASSERT(!running); /* recursion check */
+ running = 1;
+ lnet_net_unlock(0);
- last = now;
+ last = now;
if (the_lnet.ln_rc_state == LNET_RC_STATE_STOPPING)
lnet_prune_rc_data(0); /* unlink all rcd and nowait */
LASSERT (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);
- LNET_LOCK();
+ lnet_net_lock(0);
version = the_lnet.ln_routers_version;
cfs_list_for_each_entry (rtr, &the_lnet.ln_routers, lp_rtr_list) {
LASSERT (version == the_lnet.ln_routers_version);
}
- LNET_UNLOCK();
+ lnet_net_unlock(0);
- running = 0; /* lock only needed for the recursion check */
- return;
+ running = 0; /* lock only needed for the recursion check */
+ return;
}
/* NB lnet_peers_start_down depends on me,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff