#include "gnilnd.h"
/* Primary entry points from LNET. There are no guarantees against reentrance. */
-struct lnet_lnd the_kgnilnd = {
+const struct lnet_lnd the_kgnilnd = {
.lnd_type = GNILND,
.lnd_startup = kgnilnd_startup,
.lnd_shutdown = kgnilnd_shutdown,
.lnd_send = kgnilnd_send,
.lnd_recv = kgnilnd_recv,
.lnd_eager_recv = kgnilnd_eager_recv,
- .lnd_query = kgnilnd_query,
};
kgn_data_t kgnilnd_data;
conn->gnc_next_tx = (int) GNILND_MAX_MSG_ID - 10;
/* if this fails, we have conflicts and MAX_TX is too large */
- CLASSERT(GNILND_MAX_MSG_ID < GNILND_MSGID_CLOSE);
+ BUILD_BUG_ON(GNILND_MAX_MSG_ID >= GNILND_MSGID_CLOSE);
/* get a new unique CQ id for this conn */
write_lock(&kgnilnd_data.kgn_peer_conn_lock);
* check context */
conn->gnc_device = dev;
- conn->gnc_timeout = MAX(*kgnilnd_tunables.kgn_timeout,
- GNILND_MIN_TIMEOUT);
+ conn->gnc_timeout = max(*kgnilnd_tunables.kgn_timeout,
+ GNILND_MIN_TIMEOUT);
kgnilnd_update_reaper_timeout(conn->gnc_timeout);
/* this is the ep_handle for doing SMSG & BTE */
void
kgnilnd_peer_alive(kgn_peer_t *peer)
{
- set_mb(peer->gnp_last_alive, jiffies);
+ time64_t now = ktime_get_seconds();
+
+ set_mb(peer->gnp_last_alive, now);
}
void
return;
}
- LIBCFS_ALLOC(nets, nnets * sizeof(*nets));
+ CFS_ALLOC_PTR_ARRAY(nets, nnets);
if (nets == NULL) {
up_read(&kgnilnd_data.kgn_net_rw_sem);
peer_nid = kgnilnd_lnd2lnetnid(net->gnn_ni->ni_nid,
peer->gnp_nid);
- CDEBUG(D_NET, "peer 0x%p->%s last_alive %lu (%lus ago)\n",
+ CDEBUG(D_NET, "peer 0x%p->%s last_alive %lld (%llds ago)\n",
peer, libcfs_nid2str(peer_nid), peer->gnp_last_alive,
- cfs_duration_sec(jiffies - peer->gnp_last_alive));
+ ktime_get_seconds() - peer->gnp_last_alive);
lnet_notify(net->gnn_ni, peer_nid, alive,
+ (alive) ? true : false,
peer->gnp_last_alive);
kgnilnd_net_decref(net);
}
- LIBCFS_FREE(nets, nnets * sizeof(*nets));
+ CFS_FREE_PTR_ARRAY(nets, nnets);
}
}
/* set timeout vals in conn early so we can use them for the NAK */
/* use max of the requested and our timeout, peer will do the same */
- conn->gnc_timeout = MAX(conn->gnc_timeout, connreq->gncr_timeout);
+ conn->gnc_timeout = max(conn->gnc_timeout, connreq->gncr_timeout);
/* only ep_bind really mucks around with the CQ */
/* only ep bind if we are not connecting to ourself and the dstnid is not a wildcard. this check
current_to += *kgnilnd_tunables.kgn_min_reconnect_interval / 2;
}
- current_to = MIN(current_to,
- *kgnilnd_tunables.kgn_max_reconnect_interval);
+ current_to = min(current_to,
+ *kgnilnd_tunables.kgn_max_reconnect_interval);
peer->gnp_reconnect_interval = current_to;
CDEBUG(D_NET, "peer %s can reconnect at %lu interval %lu\n",
return rc;
}
- i = 4;
- while (atomic_read(&kgnilnd_data.kgn_npending_conns) ||
- atomic_read(&kgnilnd_data.kgn_npending_detach) ||
- atomic_read(&kgnilnd_data.kgn_npending_unlink)) {
-
- set_current_state(TASK_UNINTERRUPTIBLE);
- schedule_timeout(cfs_time_seconds(1));
- i++;
-
- CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, "Waiting on %d peers %d closes %d detaches\n",
+ wait_var_event_warning(&kgnilnd_data,
+ !atomic_read(&kgnilnd_data.kgn_npending_conns) &&
+ !atomic_read(&kgnilnd_data.kgn_npending_detach) &&
+ !atomic_read(&kgnilnd_data.kgn_npending_unlink),
+ "Waiting on %d peers %d closes %d detaches\n",
atomic_read(&kgnilnd_data.kgn_npending_unlink),
atomic_read(&kgnilnd_data.kgn_npending_conns),
atomic_read(&kgnilnd_data.kgn_npending_detach));
- }
return rc;
}
return rc;
}
-void
-kgnilnd_query(struct lnet_ni *ni, lnet_nid_t nid, cfs_time_t *when)
-{
- kgn_net_t *net = ni->ni_data;
- kgn_tx_t *tx;
- kgn_peer_t *peer = NULL;
- kgn_conn_t *conn = NULL;
- struct lnet_process_id id = {
- .nid = nid,
- .pid = LNET_PID_LUSTRE,
- };
- ENTRY;
-
- /* I expect to find him, so only take a read lock */
- read_lock(&kgnilnd_data.kgn_peer_conn_lock);
- peer = kgnilnd_find_peer_locked(nid);
- if (peer != NULL) {
- /* LIE if in a quiesce - we will update the timeouts after,
- * but we don't want sends failing during it */
- if (kgnilnd_data.kgn_quiesce_trigger) {
- *when = jiffies;
- read_unlock(&kgnilnd_data.kgn_peer_conn_lock);
- GOTO(out, 0);
- }
-
- /* Update to best guess, might refine on later checks */
- *when = peer->gnp_last_alive;
-
- /* we have a peer, how about a conn? */
- conn = kgnilnd_find_conn_locked(peer);
-
- if (conn == NULL) {
- /* if there is no conn, check peer last errno to see if clean disconnect
- * - if it was, we lie to LNet because we believe a TX would complete
- * on reconnect */
- if (kgnilnd_conn_clean_errno(peer->gnp_last_errno)) {
- *when = jiffies;
- }
- /* we still want to fire a TX and new conn in this case */
- } else {
- /* gnp_last_alive is valid, run for the hills */
- read_unlock(&kgnilnd_data.kgn_peer_conn_lock);
- GOTO(out, 0);
- }
- }
- /* if we get here, either we have no peer or no conn for him, so fire off
- * new TX to trigger conn setup */
- read_unlock(&kgnilnd_data.kgn_peer_conn_lock);
-
- /* if we couldn't find him, we'll fire up a TX and get connected -
- * if we don't do this, after ni_peer_timeout, LNet will declare him dead.
- * So really we treat kgnilnd_query as a bit of a 'connect now' type
- * event because it'll only do this when it wants to send
- *
- * Use a real TX for this to get the proper gnp_tx_queue behavior, etc
- * normally we'd use kgnilnd_send_ctlmsg for this, but we don't really
- * care that this goes out quickly since we already know we need a new conn
- * formed */
- if (CFS_FAIL_CHECK(CFS_FAIL_GNI_NOOP_SEND))
- return;
-
- tx = kgnilnd_new_tx_msg(GNILND_MSG_NOOP, ni->ni_nid);
- if (tx != NULL) {
- kgnilnd_launch_tx(tx, net, &id);
- }
-out:
- CDEBUG(D_NETTRACE, "peer 0x%p->%s when %lu\n", peer,
- libcfs_nid2str(nid), *when);
- EXIT;
-}
-
int
kgnilnd_dev_init(kgn_device_t *dev)
{
"%d physical mappings of %d pages still mapped\n",
dev->gnd_map_nphys, dev->gnd_map_physnop);
- LASSERTF(dev->gnd_map_nvirt == 0 && dev->gnd_map_virtnob == 0,
- "%d virtual mappings of %llu bytes still mapped\n",
- dev->gnd_map_nvirt, dev->gnd_map_virtnob);
-
LASSERTF(atomic_read(&dev->gnd_n_mdd) == 0 &&
atomic_read(&dev->gnd_n_mdd_held) == 0 &&
atomic64_read(&dev->gnd_nbytes_map) == 0,
init_rwsem(&dev->gnd_conn_sem);
/* alloc & setup nid based dgram table */
- LIBCFS_ALLOC(dev->gnd_dgrams,
- sizeof(struct list_head) * *kgnilnd_tunables.kgn_peer_hash_size);
+ CFS_ALLOC_PTR_ARRAY(dev->gnd_dgrams,
+ *kgnilnd_tunables.kgn_peer_hash_size);
if (dev->gnd_dgrams == NULL)
GOTO(failed, rc = -ENOMEM);
/* OK to call kgnilnd_api_shutdown() to cleanup now */
kgnilnd_data.kgn_init = GNILND_INIT_DATA;
- try_module_get(THIS_MODULE);
+ if (!try_module_get(THIS_MODULE))
+ GOTO(failed, rc = -ENOENT);
rwlock_init(&kgnilnd_data.kgn_peer_conn_lock);
- LIBCFS_ALLOC(kgnilnd_data.kgn_peers,
- sizeof(struct list_head) * *kgnilnd_tunables.kgn_peer_hash_size);
+ CFS_ALLOC_PTR_ARRAY(kgnilnd_data.kgn_peers,
+ *kgnilnd_tunables.kgn_peer_hash_size);
if (kgnilnd_data.kgn_peers == NULL)
GOTO(failed, rc = -ENOMEM);
INIT_LIST_HEAD(&kgnilnd_data.kgn_peers[i]);
}
- LIBCFS_ALLOC(kgnilnd_data.kgn_conns,
- sizeof(struct list_head) * *kgnilnd_tunables.kgn_peer_hash_size);
+ CFS_ALLOC_PTR_ARRAY(kgnilnd_data.kgn_conns,
+ *kgnilnd_tunables.kgn_peer_hash_size);
if (kgnilnd_data.kgn_conns == NULL)
GOTO(failed, rc = -ENOMEM);
INIT_LIST_HEAD(&kgnilnd_data.kgn_conns[i]);
}
- LIBCFS_ALLOC(kgnilnd_data.kgn_nets,
- sizeof(struct list_head) * *kgnilnd_tunables.kgn_net_hash_size);
+ CFS_ALLOC_PTR_ARRAY(kgnilnd_data.kgn_nets,
+ *kgnilnd_tunables.kgn_net_hash_size);
if (kgnilnd_data.kgn_nets == NULL)
GOTO(failed, rc = -ENOMEM);
CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET,
"Waiting for conns to be cleaned up %d\n",atomic_read(&kgnilnd_data.kgn_nconns));
- set_current_state(TASK_UNINTERRUPTIBLE);
- schedule_timeout(cfs_time_seconds(1));
+ schedule_timeout_uninterruptible(cfs_time_seconds(1));
}
/* Peer state all cleaned up BEFORE setting shutdown, so threads don't
* have to worry about shutdown races. NB connections may be created
i++;
CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET,
"Waiting for ruhroh thread to terminate\n");
- set_current_state(TASK_UNINTERRUPTIBLE);
- schedule_timeout(cfs_time_seconds(1));
+ schedule_timeout_uninterruptible(cfs_time_seconds(1));
}
/* Flag threads to terminate */
CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, /* power of 2? */
"Waiting for %d threads to terminate\n",
atomic_read(&kgnilnd_data.kgn_nthreads));
- set_current_state(TASK_UNINTERRUPTIBLE);
- schedule_timeout(cfs_time_seconds(1));
+ schedule_timeout_uninterruptible(cfs_time_seconds(1));
}
LASSERTF(atomic_read(&kgnilnd_data.kgn_npeers) == 0,
for (i = 0; i < *kgnilnd_tunables.kgn_peer_hash_size; i++)
LASSERT(list_empty(&kgnilnd_data.kgn_peers[i]));
- LIBCFS_FREE(kgnilnd_data.kgn_peers,
- sizeof (struct list_head) *
- *kgnilnd_tunables.kgn_peer_hash_size);
+ CFS_FREE_PTR_ARRAY(kgnilnd_data.kgn_peers,
+ *kgnilnd_tunables.kgn_peer_hash_size);
}
down_write(&kgnilnd_data.kgn_net_rw_sem);
for (i = 0; i < *kgnilnd_tunables.kgn_net_hash_size; i++)
LASSERT(list_empty(&kgnilnd_data.kgn_nets[i]));
- LIBCFS_FREE(kgnilnd_data.kgn_nets,
- sizeof (struct list_head) *
- *kgnilnd_tunables.kgn_net_hash_size);
+ CFS_FREE_PTR_ARRAY(kgnilnd_data.kgn_nets,
+ *kgnilnd_tunables.kgn_net_hash_size);
}
up_write(&kgnilnd_data.kgn_net_rw_sem);
for (i = 0; i < *kgnilnd_tunables.kgn_peer_hash_size; i++)
LASSERT(list_empty(&kgnilnd_data.kgn_conns[i]));
- LIBCFS_FREE(kgnilnd_data.kgn_conns,
- sizeof (struct list_head) *
- *kgnilnd_tunables.kgn_peer_hash_size);
+ CFS_FREE_PTR_ARRAY(kgnilnd_data.kgn_conns,
+ *kgnilnd_tunables.kgn_peer_hash_size);
}
for (i = 0; i < kgnilnd_data.kgn_ndevs; i++) {
"dgrams left %d\n", atomic_read(&dev->gnd_ndgrams));
if (dev->gnd_dgrams != NULL) {
- for (i = 0; i < *kgnilnd_tunables.kgn_peer_hash_size; i++)
+ for (i = 0; i < *kgnilnd_tunables.kgn_peer_hash_size;
+ i++)
LASSERT(list_empty(&dev->gnd_dgrams[i]));
- LIBCFS_FREE(dev->gnd_dgrams,
- sizeof (struct list_head) *
- *kgnilnd_tunables.kgn_peer_hash_size);
+ CFS_FREE_PTR_ARRAY(dev->gnd_dgrams,
+ *kgnilnd_tunables.kgn_peer_hash_size);
}
kgnilnd_free_phys_fmablk(dev);
"Waiting for %d references to clear on net %d\n",
atomic_read(&net->gnn_refcount),
net->gnn_netnum);
- set_current_state(TASK_UNINTERRUPTIBLE);
- schedule_timeout(cfs_time_seconds(1));
+ schedule_timeout_uninterruptible(cfs_time_seconds(1));
}
/* release ref from kgnilnd_startup */
if (rc != 0)
return rc;
- printk(KERN_INFO "Lustre: kgnilnd build version: "KGNILND_BUILD_REV"\n");
+ LCONSOLE_INFO("Lustre: kgnilnd build version: "LUSTRE_VERSION_STRING"\n");
kgnilnd_insert_sysctl();
kgnilnd_proc_init();
MODULE_AUTHOR("Cray, Inc. <nic@cray.com>");
MODULE_DESCRIPTION("Gemini LNet Network Driver");
-MODULE_VERSION(KGNILND_BUILD_REV);
+MODULE_VERSION(LUSTRE_VERSION_STRING);
MODULE_LICENSE("GPL");
module_init(kgnilnd_init);