staging: lustre: do less intense allocating retry for ko2iblnd
authorLiang Zhen <liang.zhen@intel.com>
Wed, 2 Mar 2016 23:53:30 +0000 (18:53 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 3 Mar 2016 00:01:38 +0000 (16:01 -0800)
ko2iblnd may retry too frequent for growing pools, all schedulers
are spinning if another thread is in progress of allocating a new
pool and can't finish right away because of high system load.

Signed-off-by: Liang Zhen <liang.zhen@intel.com>
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-7054
Reviewed-on: http://review.whamcloud.com/16470
Reviewed-by: Doug Oucharek <doug.s.oucharek@intel.com>
Reviewed-by: James Simmons <uja.ornl@yahoo.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c

index 135ccf18917b52b31017f0d95b73eb0b46727c3c..0d32e6541a3f7207fdea3bd9b023619247f87caa 100644 (file)
@@ -1218,6 +1218,7 @@ static kib_hca_dev_t *kiblnd_current_hdev(kib_dev_t *dev)
                if (!(i++ % 50))
                        CDEBUG(D_NET, "%s: Wait for failover\n",
                               dev->ibd_ifname);
+               set_current_state(TASK_INTERRUPTIBLE);
                schedule_timeout(cfs_time_seconds(1) / 100);
 
                read_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
@@ -1684,6 +1685,9 @@ struct list_head *kiblnd_pool_alloc_node(kib_poolset_t *ps)
 {
        struct list_head *node;
        kib_pool_t *pool;
+       unsigned int interval = 1;
+       unsigned long time_before;
+       unsigned int trips = 0;
        int rc;
 
  again:
@@ -1709,9 +1713,15 @@ struct list_head *kiblnd_pool_alloc_node(kib_poolset_t *ps)
        if (ps->ps_increasing) {
                /* another thread is allocating a new pool */
                spin_unlock(&ps->ps_lock);
-               CDEBUG(D_NET, "Another thread is allocating new %s pool, waiting for her to complete\n",
-                      ps->ps_name);
-               schedule();
+               trips++;
+               CDEBUG(D_NET, "Another thread is allocating new %s pool, waiting %d HZs for her to complete. trips = %d\n",
+                      ps->ps_name, interval, trips);
+
+               set_current_state(TASK_INTERRUPTIBLE);
+               schedule_timeout(interval);
+               if (interval < cfs_time_seconds(1))
+                       interval *= 2;
+
                goto again;
        }
 
@@ -1725,8 +1735,10 @@ struct list_head *kiblnd_pool_alloc_node(kib_poolset_t *ps)
        spin_unlock(&ps->ps_lock);
 
        CDEBUG(D_NET, "%s pool exhausted, allocate new pool\n", ps->ps_name);
-
+       time_before = cfs_time_current();
        rc = ps->ps_pool_create(ps, ps->ps_pool_size, &pool);
+       CDEBUG(D_NET, "ps_pool_create took %lu HZ to complete",
+              cfs_time_current() - time_before);
 
        spin_lock(&ps->ps_lock);
        ps->ps_increasing = 0;