-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <lnet/lib-lnet.h>
int
-lnet_create_peer_table(void)
+lnet_peer_tables_create(void)
{
- struct list_head *hash;
- int i;
-
- LASSERT (the_lnet.ln_peer_hash == NULL);
- LIBCFS_ALLOC(hash, LNET_PEER_HASHSIZE * sizeof(struct list_head));
-
- if (hash == NULL) {
- CERROR("Can't allocate peer hash table\n");
+ struct lnet_peer_table *ptable;
+ cfs_list_t *hash;
+ int i;
+ int j;
+
+ the_lnet.ln_peer_tables = cfs_percpt_alloc(lnet_cpt_table(),
+ sizeof(*ptable));
+ if (the_lnet.ln_peer_tables == NULL) {
+ CERROR("Failed to allocate cpu-partition peer tables\n");
return -ENOMEM;
}
- for (i = 0; i < LNET_PEER_HASHSIZE; i++)
- CFS_INIT_LIST_HEAD(&hash[i]);
+ cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
+ CFS_INIT_LIST_HEAD(&ptable->pt_deathrow);
+
+ LIBCFS_CPT_ALLOC(hash, lnet_cpt_table(), i,
+ LNET_PEER_HASH_SIZE * sizeof(*hash));
+ if (hash == NULL) {
+ CERROR("Failed to create peer hash table\n");
+ lnet_peer_tables_destroy();
+ return -ENOMEM;
+ }
+
+ for (j = 0; j < LNET_PEER_HASH_SIZE; j++)
+ CFS_INIT_LIST_HEAD(&hash[j]);
+ ptable->pt_hash = hash; /* sign of initialization */
+ }
- the_lnet.ln_peer_hash = hash;
return 0;
}
void
-lnet_destroy_peer_table(void)
+lnet_peer_tables_destroy(void)
{
- int i;
+ struct lnet_peer_table *ptable;
+ cfs_list_t *hash;
+ int i;
+ int j;
- if (the_lnet.ln_peer_hash == NULL)
- return;
+ if (the_lnet.ln_peer_tables == NULL)
+ return;
+
+ cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
+ hash = ptable->pt_hash;
+ if (hash == NULL) /* not intialized */
+ break;
+
+ LASSERT(cfs_list_empty(&ptable->pt_deathrow));
+
+ ptable->pt_hash = NULL;
+ for (j = 0; j < LNET_PEER_HASH_SIZE; j++)
+ LASSERT(cfs_list_empty(&hash[j]));
- for (i = 0; i < LNET_PEER_HASHSIZE; i++)
- LASSERT (list_empty(&the_lnet.ln_peer_hash[i]));
-
- LIBCFS_FREE(the_lnet.ln_peer_hash,
- LNET_PEER_HASHSIZE * sizeof (struct list_head));
- the_lnet.ln_peer_hash = NULL;
+ LIBCFS_FREE(hash, LNET_PEER_HASH_SIZE * sizeof(*hash));
+ }
+
+ cfs_percpt_free(the_lnet.ln_peer_tables);
+ the_lnet.ln_peer_tables = NULL;
}
void
-lnet_clear_peer_table(void)
+lnet_peer_tables_cleanup(void)
{
- int i;
-
- LASSERT (the_lnet.ln_shutdown); /* i.e. no new peers */
-
- for (i = 0; i < LNET_PEER_HASHSIZE; i++) {
- struct list_head *peers = &the_lnet.ln_peer_hash[i];
-
- LNET_LOCK();
- while (!list_empty(peers)) {
- lnet_peer_t *lp = list_entry(peers->next,
- lnet_peer_t, lp_hashlist);
-
- list_del(&lp->lp_hashlist);
- lnet_peer_decref_locked(lp); /* lose hash table's ref */
+ struct lnet_peer_table *ptable;
+ int i;
+ int j;
+
+ LASSERT(the_lnet.ln_shutdown); /* i.e. no new peers */
+
+ cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
+ lnet_net_lock(i);
+
+ for (j = 0; j < LNET_PEER_HASH_SIZE; j++) {
+ cfs_list_t *peers = &ptable->pt_hash[j];
+
+ while (!cfs_list_empty(peers)) {
+ lnet_peer_t *lp = cfs_list_entry(peers->next,
+ lnet_peer_t,
+ lp_hashlist);
+ cfs_list_del_init(&lp->lp_hashlist);
+ /* lose hash table's ref */
+ lnet_peer_decref_locked(lp);
+ }
}
- LNET_UNLOCK();
+
+ lnet_net_unlock(i);
}
- LNET_LOCK();
- for (i = 3; the_lnet.ln_npeers != 0;i++) {
- LNET_UNLOCK();
+ cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
+ CFS_LIST_HEAD (deathrow);
+ lnet_peer_t *lp;
- if ((i & (i-1)) == 0)
- CDEBUG(D_WARNING,"Waiting for %d peers\n",
- the_lnet.ln_npeers);
- cfs_pause(cfs_time_seconds(1));
+ lnet_net_lock(i);
- LNET_LOCK();
- }
- LNET_UNLOCK();
+ for (j = 3; ptable->pt_number != 0; j++) {
+ lnet_net_unlock(i);
+
+ if ((j & (j - 1)) == 0) {
+ CDEBUG(D_WARNING,
+ "Waiting for %d peers on peer table\n",
+ ptable->pt_number);
+ }
+ cfs_pause(cfs_time_seconds(1) / 2);
+ lnet_net_lock(i);
+ }
+ cfs_list_splice_init(&ptable->pt_deathrow, &deathrow);
+
+ lnet_net_unlock(i);
+
+ while (!cfs_list_empty(&deathrow)) {
+ lp = cfs_list_entry(deathrow.next,
+ lnet_peer_t, lp_hashlist);
+ cfs_list_del(&lp->lp_hashlist);
+ LIBCFS_FREE(lp, sizeof(*lp));
+ }
+ }
}
void
-lnet_destroy_peer_locked (lnet_peer_t *lp)
+lnet_destroy_peer_locked(lnet_peer_t *lp)
{
- lnet_ni_decref_locked(lp->lp_ni);
- LNET_UNLOCK();
+ struct lnet_peer_table *ptable;
- LASSERT (lp->lp_refcount == 0);
- LASSERT (lp->lp_rtr_refcount == 0);
- LASSERT (list_empty(&lp->lp_txq));
- LASSERT (lp->lp_txqnob == 0);
+ LASSERT(lp->lp_refcount == 0);
+ LASSERT(lp->lp_rtr_refcount == 0);
+ LASSERT(cfs_list_empty(&lp->lp_txq));
+ LASSERT(cfs_list_empty(&lp->lp_hashlist));
+ LASSERT(lp->lp_txqnob == 0);
- LIBCFS_FREE(lp, sizeof(*lp));
+ ptable = the_lnet.ln_peer_tables[lp->lp_cpt];
+ LASSERT(ptable->pt_number > 0);
+ ptable->pt_number--;
- LNET_LOCK();
+ lnet_ni_decref_locked(lp->lp_ni, lp->lp_cpt);
+ lp->lp_ni = NULL;
- LASSERT(the_lnet.ln_npeers > 0);
- the_lnet.ln_npeers--;
+ cfs_list_add(&lp->lp_hashlist, &ptable->pt_deathrow);
}
lnet_peer_t *
-lnet_find_peer_locked (lnet_nid_t nid)
+lnet_find_peer_locked(struct lnet_peer_table *ptable, lnet_nid_t nid)
{
- unsigned int idx = LNET_NIDADDR(nid) % LNET_PEER_HASHSIZE;
- struct list_head *peers = &the_lnet.ln_peer_hash[idx];
- struct list_head *tmp;
- lnet_peer_t *lp;
+ cfs_list_t *peers;
+ lnet_peer_t *lp;
- if (the_lnet.ln_shutdown)
- return NULL;
+ LASSERT(!the_lnet.ln_shutdown);
- list_for_each (tmp, peers) {
- lp = list_entry(tmp, lnet_peer_t, lp_hashlist);
-
+ peers = &ptable->pt_hash[lnet_nid2peerhash(nid)];
+ cfs_list_for_each_entry(lp, peers, lp_hashlist) {
if (lp->lp_nid == nid) {
- lnet_peer_addref_locked(lp);
+ lnet_peer_addref_locked(lp);
return lp;
- }
+ }
}
-
+
return NULL;
}
int
-lnet_nid2peer_locked(lnet_peer_t **lpp, lnet_nid_t nid)
+lnet_nid2peer_locked(lnet_peer_t **lpp, lnet_nid_t nid, int cpt)
{
- lnet_peer_t *lp;
- lnet_peer_t *lp2;
+ struct lnet_peer_table *ptable;
+ lnet_peer_t *lp = NULL;
+ lnet_peer_t *lp2;
+ int cpt2;
+ int rc = 0;
+
+ *lpp = NULL;
+ if (the_lnet.ln_shutdown) /* it's shutting down */
+ return -ESHUTDOWN;
+
+ /* cpt can be LNET_LOCK_EX if it's called from router functions */
+ cpt2 = cpt != LNET_LOCK_EX ? cpt : lnet_cpt_of_nid_locked(nid);
+
+ ptable = the_lnet.ln_peer_tables[cpt2];
+ lp = lnet_find_peer_locked(ptable, nid);
+ if (lp != NULL) {
+ *lpp = lp;
+ return 0;
+ }
+
+ if (!cfs_list_empty(&ptable->pt_deathrow)) {
+ lp = cfs_list_entry(ptable->pt_deathrow.next,
+ lnet_peer_t, lp_hashlist);
+ cfs_list_del(&lp->lp_hashlist);
+ }
+
+ /*
+ * take extra refcount in case another thread has shutdown LNet
+ * and destroyed locks and peer-table before I finish the allocation
+ */
+ ptable->pt_number++;
+ lnet_net_unlock(cpt);
+
+ if (lp != NULL)
+ memset(lp, 0, sizeof(*lp));
+ else
+ LIBCFS_CPT_ALLOC(lp, lnet_cpt_table(), cpt2, sizeof(*lp));
- lp = lnet_find_peer_locked(nid);
- if (lp != NULL) {
- *lpp = lp;
- return 0;
- }
-
- LNET_UNLOCK();
-
- LIBCFS_ALLOC(lp, sizeof(*lp));
if (lp == NULL) {
- *lpp = NULL;
- LNET_LOCK();
- return -ENOMEM;
- }
+ rc = -ENOMEM;
+ lnet_net_lock(cpt);
+ goto out;
+ }
- memset(lp, 0, sizeof(*lp)); /* zero counters etc */
-
CFS_INIT_LIST_HEAD(&lp->lp_txq);
- CFS_INIT_LIST_HEAD(&lp->lp_rtrq);
-
+ CFS_INIT_LIST_HEAD(&lp->lp_rtrq);
+ CFS_INIT_LIST_HEAD(&lp->lp_routes);
+
lp->lp_notify = 0;
lp->lp_notifylnd = 0;
lp->lp_notifying = 0;
lp->lp_alive_count = 0;
lp->lp_timestamp = 0;
lp->lp_alive = !lnet_peers_start_down(); /* 1 bit!! */
- lp->lp_last_alive = cfs_time_current_sec(); /* assumes alive */
- lp->lp_last_query = 0; /* didn't ask LND yet */
+ lp->lp_last_alive = cfs_time_current(); /* assumes alive */
+ lp->lp_last_query = 0; /* haven't asked NI yet */
lp->lp_ping_timestamp = 0;
- lp->lp_nid = nid;
- lp->lp_refcount = 2; /* 1 for caller; 1 for hash */
- lp->lp_rtr_refcount = 0;
-
- LNET_LOCK();
-
- lp2 = lnet_find_peer_locked(nid);
- if (lp2 != NULL) {
- LNET_UNLOCK();
- LIBCFS_FREE(lp, sizeof(*lp));
- LNET_LOCK();
-
- if (the_lnet.ln_shutdown) {
- lnet_peer_decref_locked(lp2);
- *lpp = NULL;
- return -ESHUTDOWN;
- }
-
- *lpp = lp2;
- return 0;
- }
-
- lp->lp_ni = lnet_net2ni_locked(LNET_NIDNET(nid));
- if (lp->lp_ni == NULL) {
- LNET_UNLOCK();
- LIBCFS_FREE(lp, sizeof(*lp));
- LNET_LOCK();
-
- *lpp = NULL;
- return the_lnet.ln_shutdown ? -ESHUTDOWN : -EHOSTUNREACH;
- }
+ lp->lp_ping_feats = LNET_PING_FEAT_INVAL;
+ lp->lp_nid = nid;
+ lp->lp_cpt = cpt2;
+ lp->lp_refcount = 2; /* 1 for caller; 1 for hash */
+ lp->lp_rtr_refcount = 0;
+
+ lnet_net_lock(cpt);
+
+ if (the_lnet.ln_shutdown) {
+ rc = -ESHUTDOWN;
+ goto out;
+ }
- lp->lp_txcredits =
- lp->lp_mintxcredits = lp->lp_ni->ni_peertxcredits;
- lp->lp_rtrcredits =
- lp->lp_minrtrcredits = lnet_peer_buffer_credits(lp->lp_ni);
+ lp2 = lnet_find_peer_locked(ptable, nid);
+ if (lp2 != NULL) {
+ *lpp = lp2;
+ goto out;
+ }
+
+ lp->lp_ni = lnet_net2ni_locked(LNET_NIDNET(nid), cpt2);
+ if (lp->lp_ni == NULL) {
+ rc = -EHOSTUNREACH;
+ goto out;
+ }
- LASSERT (!the_lnet.ln_shutdown);
- /* can't add peers after shutdown starts */
+ lp->lp_txcredits =
+ lp->lp_mintxcredits = lp->lp_ni->ni_peertxcredits;
+ lp->lp_rtrcredits =
+ lp->lp_minrtrcredits = lnet_peer_buffer_credits(lp->lp_ni);
- list_add_tail(&lp->lp_hashlist, lnet_nid2peerhash(nid));
- the_lnet.ln_npeers++;
- the_lnet.ln_peertable_version++;
- *lpp = lp;
- return 0;
+ cfs_list_add_tail(&lp->lp_hashlist,
+ &ptable->pt_hash[lnet_nid2peerhash(nid)]);
+ ptable->pt_version++;
+ *lpp = lp;
+
+ return 0;
+out:
+ if (lp != NULL)
+ cfs_list_add(&lp->lp_hashlist, &ptable->pt_deathrow);
+ ptable->pt_number--;
+ return rc;
}
void
lnet_debug_peer(lnet_nid_t nid)
{
- char *aliveness = "NA";
- int rc;
- lnet_peer_t *lp;
+ char *aliveness = "NA";
+ lnet_peer_t *lp;
+ int rc;
+ int cpt;
- LNET_LOCK();
+ cpt = lnet_cpt_of_nid(nid);
+ lnet_net_lock(cpt);
- rc = lnet_nid2peer_locked(&lp, nid);
- if (rc != 0) {
- LNET_UNLOCK();
+ rc = lnet_nid2peer_locked(&lp, nid, cpt);
+ if (rc != 0) {
+ lnet_net_unlock(cpt);
CDEBUG(D_WARNING, "No peer %s\n", libcfs_nid2str(nid));
return;
}
- if (lnet_isrouter(lp) || lp->lp_ni->ni_peertimeout > 0)
+ if (lnet_isrouter(lp) || lnet_peer_aliveness_enabled(lp))
aliveness = lp->lp_alive ? "up" : "down";
CDEBUG(D_WARNING, "%-24s %4d %5s %5d %5d %5d %5d %5d %ld\n",
lnet_peer_decref_locked(lp);
- LNET_UNLOCK();
+ lnet_net_unlock(cpt);
}