Whamcloud - gitweb
LU-12930 various: use schedule_timeout_*interruptible
[fs/lustre-release.git] / lnet / klnds / gnilnd / gnilnd.c
index 4b79eeb..14fb5d6 100644 (file)
@@ -1,7 +1,7 @@
 /*
  * Copyright (C) 2012 Cray, Inc.
  *
- * Copyright (c) 2013, 2015, Intel Corporation.
+ * Copyright (c) 2013, 2017, Intel Corporation.
  *
  *   Author: Nic Henke <nic@cray.com>
  *   Author: James Shimek <jshimek@cray.com>
@@ -25,7 +25,7 @@
 #include "gnilnd.h"
 
 /* Primary entry points from LNET.  There are no guarantees against reentrance. */
-lnd_t the_kgnilnd = {
+const struct lnet_lnd the_kgnilnd = {
        .lnd_type       = GNILND,
        .lnd_startup    = kgnilnd_startup,
        .lnd_shutdown   = kgnilnd_shutdown,
@@ -33,7 +33,6 @@ lnd_t the_kgnilnd = {
        .lnd_send       = kgnilnd_send,
        .lnd_recv       = kgnilnd_recv,
        .lnd_eager_recv = kgnilnd_eager_recv,
-       .lnd_query      = kgnilnd_query,
 };
 
 kgn_data_t      kgnilnd_data;
@@ -272,7 +271,7 @@ kgnilnd_create_conn(kgn_conn_t **connp, kgn_device_t *dev)
        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);
@@ -292,8 +291,8 @@ kgnilnd_create_conn(kgn_conn_t **connp, kgn_device_t *dev)
         * 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 */
@@ -505,7 +504,9 @@ kgnilnd_destroy_conn(kgn_conn_t *conn)
 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
@@ -601,11 +602,12 @@ kgnilnd_peer_notify(kgn_peer_t *peer, int error, int alive)
                        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);
@@ -877,7 +879,7 @@ kgnilnd_set_conn_params(kgn_dgram_t *dgram)
        /* 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
@@ -1259,8 +1261,8 @@ kgnilnd_peer_increase_reconnect_locked(kgn_peer_t *peer)
                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",
@@ -1596,8 +1598,7 @@ kgnilnd_del_conn_or_peer(kgn_net_t *net, lnet_nid_t nid, int command,
               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));
+               schedule_timeout_uninterruptible(cfs_time_seconds(1));
                i++;
 
                CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, "Waiting on %d peers %d closes %d detaches\n",
@@ -1800,7 +1801,7 @@ kgnilnd_report_node_state(lnet_nid_t nid, int down)
 }
 
 int
-kgnilnd_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
+kgnilnd_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg)
 {
        struct libcfs_ioctl_data *data = arg;
        kgn_net_t                *net = ni->ni_data;
@@ -1919,77 +1920,6 @@ kgnilnd_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
        return rc;
 }
 
-void
-kgnilnd_query(lnet_ni_t *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;
-       lnet_process_id_t       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)
 {
@@ -2301,7 +2231,8 @@ int kgnilnd_base_startup(void)
 
        /* 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);
 
@@ -2532,8 +2463,7 @@ kgnilnd_base_shutdown(void)
 
                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
@@ -2552,8 +2482,7 @@ kgnilnd_base_shutdown(void)
                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 */
@@ -2585,8 +2514,7 @@ kgnilnd_base_shutdown(void)
                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,
@@ -2677,7 +2605,7 @@ kgnilnd_base_shutdown(void)
 }
 
 int
-kgnilnd_startup(lnet_ni_t *ni)
+kgnilnd_startup(struct lnet_ni *ni)
 {
        int               rc, devno;
        kgn_net_t        *net;
@@ -2775,7 +2703,7 @@ kgnilnd_startup(lnet_ni_t *ni)
 }
 
 void
-kgnilnd_shutdown(lnet_ni_t *ni)
+kgnilnd_shutdown(struct lnet_ni *ni)
 {
        kgn_net_t     *net = ni->ni_data;
        int           i;
@@ -2837,8 +2765,7 @@ kgnilnd_shutdown(lnet_ni_t *ni)
                                "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 */
@@ -2891,7 +2818,7 @@ static int __init kgnilnd_init(void)
        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();
@@ -2903,7 +2830,7 @@ static int __init kgnilnd_init(void)
 
 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);