]> git.baikalelectronics.ru Git - kernel.git/commitdiff
net/mlx5: Fix possible deadlock from lockdep when adding fte to fg
authorRoi Dayan <roid@mellanox.com>
Mon, 20 Aug 2018 08:43:03 +0000 (11:43 +0300)
committerSaeed Mahameed <saeedm@mellanox.com>
Thu, 6 Sep 2018 00:08:34 +0000 (17:08 -0700)
This is a false positive report due to incorrect nested lock
annotations as we lock multiple fgs with the same subclass.
Instead of locking all fgs only lock the one being used as was
done before.

Fixes: c3b2221ab081 ("net/mlx5: Support multiple updates of steering rules in parallel")
Signed-off-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
drivers/net/ethernet/mellanox/mlx5/core/fs_core.c

index 384b560f2a9328116168e87b66605ddf65cd2151..37d114c668b7ba70ca968f76c88c42af84967f25 100644 (file)
@@ -1578,6 +1578,33 @@ static u64 matched_fgs_get_version(struct list_head *match_head)
        return version;
 }
 
+static struct fs_fte *
+lookup_fte_locked(struct mlx5_flow_group *g,
+                 u32 *match_value,
+                 bool take_write)
+{
+       struct fs_fte *fte_tmp;
+
+       if (take_write)
+               nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
+       else
+               nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
+       fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
+                                        rhash_fte);
+       if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
+               fte_tmp = NULL;
+               goto out;
+       }
+
+       nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
+out:
+       if (take_write)
+               up_write_ref_node(&g->node);
+       else
+               up_read_ref_node(&g->node);
+       return fte_tmp;
+}
+
 static struct mlx5_flow_handle *
 try_add_to_existing_fg(struct mlx5_flow_table *ft,
                       struct list_head *match_head,
@@ -1600,10 +1627,6 @@ try_add_to_existing_fg(struct mlx5_flow_table *ft,
        if (IS_ERR(fte))
                return  ERR_PTR(-ENOMEM);
 
-       list_for_each_entry(iter, match_head, list) {
-               nested_down_read_ref_node(&iter->g->node, FS_LOCK_PARENT);
-       }
-
 search_again_locked:
        version = matched_fgs_get_version(match_head);
        /* Try to find a fg that already contains a matching fte */
@@ -1611,20 +1634,9 @@ search_again_locked:
                struct fs_fte *fte_tmp;
 
                g = iter->g;
-               fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, spec->match_value,
-                                                rhash_fte);
-               if (!fte_tmp || !tree_get_node(&fte_tmp->node))
+               fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
+               if (!fte_tmp)
                        continue;
-
-               nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
-               if (!take_write) {
-                       list_for_each_entry(iter, match_head, list)
-                               up_read_ref_node(&iter->g->node);
-               } else {
-                       list_for_each_entry(iter, match_head, list)
-                               up_write_ref_node(&iter->g->node);
-               }
-
                rule = add_rule_fg(g, spec->match_value,
                                   flow_act, dest, dest_num, fte_tmp);
                up_write_ref_node(&fte_tmp->node);
@@ -1633,19 +1645,6 @@ search_again_locked:
                return rule;
        }
 
-       /* No group with matching fte found. Try to add a new fte to any
-        * matching fg.
-        */
-
-       if (!take_write) {
-               list_for_each_entry(iter, match_head, list)
-                       up_read_ref_node(&iter->g->node);
-               list_for_each_entry(iter, match_head, list)
-                       nested_down_write_ref_node(&iter->g->node,
-                                                  FS_LOCK_PARENT);
-               take_write = true;
-       }
-
        /* Check the ft version, for case that new flow group
         * was added while the fgs weren't locked
         */
@@ -1657,27 +1656,30 @@ search_again_locked:
        /* Check the fgs version, for case the new FTE with the
         * same values was added while the fgs weren't locked
         */
-       if (version != matched_fgs_get_version(match_head))
+       if (version != matched_fgs_get_version(match_head)) {
+               take_write = true;
                goto search_again_locked;
+       }
 
        list_for_each_entry(iter, match_head, list) {
                g = iter->g;
 
                if (!g->node.active)
                        continue;
+
+               nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
+
                err = insert_fte(g, fte);
                if (err) {
+                       up_write_ref_node(&g->node);
                        if (err == -ENOSPC)
                                continue;
-                       list_for_each_entry(iter, match_head, list)
-                               up_write_ref_node(&iter->g->node);
                        kmem_cache_free(steering->ftes_cache, fte);
                        return ERR_PTR(err);
                }
 
                nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
-               list_for_each_entry(iter, match_head, list)
-                       up_write_ref_node(&iter->g->node);
+               up_write_ref_node(&g->node);
                rule = add_rule_fg(g, spec->match_value,
                                   flow_act, dest, dest_num, fte);
                up_write_ref_node(&fte->node);
@@ -1686,8 +1688,6 @@ search_again_locked:
        }
        rule = ERR_PTR(-ENOENT);
 out:
-       list_for_each_entry(iter, match_head, list)
-               up_write_ref_node(&iter->g->node);
        kmem_cache_free(steering->ftes_cache, fte);
        return rule;
 }