dpif-netdev: report numa node number on pmd thread create failure
authorPanu Matilainen <pmatilai@redhat.com>
Wed, 6 Apr 2016 12:52:30 +0000 (15:52 +0300)
committerDaniele Di Proietto <diproiettod@vmware.com>
Wed, 6 Apr 2016 18:38:32 +0000 (11:38 -0700)
Since PMD threads are placed on the NUMA node of the port regardless
of a possible pmd-cpu-mask setting, this can lead to a somewhat
confusing "out of unpinned cores" message - there might be plenty
of available cores in the mask but they cannot be used if the port
is on different NUMA node than the cores. Report the NUMA node
number to help diagnosing the issue.

Reported-at: https://bugzilla.redhat.com/show_bug.cgi?id=1295952
Signed-off-by: Panu Matilainen <pmatilai@redhat.com>
Acked-by: Daniele Di Proietto <diproiettod@vmware.com>
lib/dpif-netdev.c

index 119dc2d..d7d9704 100644 (file)
@@ -3145,7 +3145,7 @@ dp_netdev_set_pmds_on_numa(struct dp_netdev *dp, int numa_id)
         n_unpinned = ovs_numa_get_n_unpinned_cores_on_numa(numa_id);
         if (!n_unpinned) {
             VLOG_ERR("Cannot create pmd threads due to out of unpinned "
-                     "cores on numa node");
+                     "cores on numa node %d", numa_id);
             return;
         }