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

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20190311183929.12796-5-saeedm@mellanox.com>
Date:   Mon, 11 Mar 2019 11:39:23 -0700
From:   Saeed Mahameed <saeedm@...lanox.com>
To:     "David S. Miller" <davem@...emloft.net>
Cc:     netdev@...r.kernel.org, Eli Britstein <elibr@...lanox.com>,
        Maor Gottlieb <maorg@...lanox.com>,
        Mark Bloch <markb@...lanox.com>,
        Saeed Mahameed <saeedm@...lanox.com>
Subject: [net 04/10] net/mlx5: Consolidate update FTE for all removal changes

From: Eli Britstein <elibr@...lanox.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@...lanox.com>
Reviewed-by: Maor Gottlieb <maorg@...lanox.com>
Reviewed-by: Mark Bloch <markb@...lanox.com>
Signed-off-by: Saeed Mahameed <saeedm@...lanox.com>
---
 .../net/ethernet/mellanox/mlx5/core/fs_core.c | 31 +++++++++++++++----
 1 file changed, 25 insertions(+), 6 deletions(-)

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);
-- 
2.20.1