diff mbox series

[net,04/10] net/mlx5: Consolidate update FTE for all removal changes

Message ID 20190311183929.12796-5-saeedm@mellanox.com
State Accepted
Delegated to: David Miller
Headers show
Series [net,01/10] net/mlx5: Fix multiple updates of steering rules in parallel | expand

Commit Message

Saeed Mahameed March 11, 2019, 6:39 p.m. UTC
From: Eli Britstein <elibr@mellanox.com>

With commit a18e879d4e45 ("net/mlx5e: Annul encap action ordering
requirement") and a use-case of e-switch remote mirroring, the
incremental/stepped FTE removal process done by the fs core got us to
illegal transient states and FW errors:

SET_FLOW_TABLE_ENTRY(0x936) op_mod(0x0) failed, status bad
parameter(0x3), syndrome (0x9c2e40)

To avoid that and improve FTE removal performance, aggregate the FTE's
updates that should be applied. Remove the FTE if it is empty, or apply
one FW update command with the aggregated updates.

Fixes: a18e879d4e45 ("net/mlx5e: Annul encap action ordering requirement")
Signed-off-by: Eli Britstein <elibr@mellanox.com>
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 .../net/ethernet/mellanox/mlx5/core/fs_core.c | 31 +++++++++++++++----
 1 file changed, 25 insertions(+), 6 deletions(-)
diff mbox series

Patch

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c b/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c
index 42a6952ff9d0..0be3eb86dd84 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c
@@ -430,9 +430,6 @@  static void modify_fte(struct fs_fte *fte)
 	struct mlx5_core_dev *dev;
 	int err;
 
-	if (!fte->modify_mask)
-		return;
-
 	fs_get_obj(fg, fte->node.parent);
 	fs_get_obj(ft, fg->node.parent);
 	dev = get_dev(&fte->node);
@@ -475,7 +472,6 @@  static void del_sw_hw_rule(struct fs_node *node)
 			BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 	}
 out:
-	modify_fte(fte);
 	kfree(rule);
 }
 
@@ -602,7 +598,7 @@  static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 	fte->node.type =  FS_TYPE_FLOW_ENTRY;
 	fte->action = *flow_act;
 
-	tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
+	tree_init_node(&fte->node, NULL, del_sw_fte);
 
 	return fte;
 }
@@ -1882,10 +1878,33 @@  EXPORT_SYMBOL(mlx5_add_flow_rules);
 
 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
 {
+	struct fs_fte *fte;
 	int i;
 
+	/* In order to consolidate the HW changes we lock the FTE for other
+	 * changes, and increase its refcount, in order not to perform the
+	 * "del" functions of the FTE. Will handle them here.
+	 * The removal of the rules is done under locked FTE.
+	 * After removing all the handle's rules, if there are remaining
+	 * rules, it means we just need to modify the FTE in FW, and
+	 * unlock/decrease the refcount we increased before.
+	 * Otherwise, it means the FTE should be deleted. First delete the
+	 * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
+	 * the FTE, which will handle the last decrease of the refcount, as
+	 * well as required handling of its parent.
+	 */
+	fs_get_obj(fte, handle->rule[0]->node.parent);
+	down_write_ref_node(&fte->node, false);
 	for (i = handle->num_rules - 1; i >= 0; i--)
-		tree_remove_node(&handle->rule[i]->node, false);
+		tree_remove_node(&handle->rule[i]->node, true);
+	if (fte->modify_mask && fte->dests_size) {
+		modify_fte(fte);
+		up_write_ref_node(&fte->node, false);
+	} else {
+		del_hw_fte(&fte->node);
+		up_write(&fte->node.lock);
+		tree_put_node(&fte->node, false);
+	}
 	kfree(handle);
 }
 EXPORT_SYMBOL(mlx5_del_flow_rules);