rpms/kernel
9
4
Fork 6
Code Issues Pull Requests Releases Activity
a9
kernel/SOURCES/1373-net-mlx5-hws-rework-rehash-loop.patch

210 lines
6.9 KiB
Diff
Raw Permalink Blame History

From a380ec59fea7f1801a94b324bd7f688f2d3be0dd Mon Sep 17 00:00:00 2001
From: Kamal Heib <kheib@redhat.com>
Date: Thu, 16 Apr 2026 17:42:05 -0400
Subject: [PATCH] net/mlx5: HWS, rework rehash loop
JIRA: https://redhat.atlassian.net/browse/RHEL-169055
commit ef94799a87415790d4297cf06075f99b70c420cd
Author: Yevgeny Kliteynik <kliteyn@nvidia.com>
Date: Sun May 11 22:38:09 2025 +0300
net/mlx5: HWS, rework rehash loop
Reworking the rehash loop - simplifying the code and making it less
error prone:
- Instead of doing round-robin on all the queues with batch of rules in
each cycle, just go over all the queues and move all the rules that
belong to this queue.
- If at some stage of moving the rule we get a failure (which should
not happen), this can't be rolled back. So instead of aborting
rehash and leaving the matcher in a broken state, allow the loop
to continue: attempt to move the rest of the rules and delete the
old matcher. A rule that failed to move to a new matcher will loose
its match STE once the rehash is completed and the old matcher is
deleted, so the rule won't match any traffic any more. This rule's
packets will fall back to the steering pipeline w/o HW offload.
Rehash procedure will return an error, which will cause the rule
insertion to fail for the rule that started this whole rehash.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Vlad Dogaru <vdogaru@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1746992290-568936-10-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Kamal Heib <kheib@redhat.com>
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c
index 456fac895f5e..9e057f808ea5 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/bwc.c
@@ -610,95 +610,69 @@ hws_bwc_matcher_find_at(struct mlx5hws_bwc_matcher *bwc_matcher,
static int hws_bwc_matcher_move_all_simple(struct mlx5hws_bwc_matcher *bwc_matcher)
{
+ bool move_error = false, poll_error = false, drain_error = false;
struct mlx5hws_context *ctx = bwc_matcher->matcher->tbl->ctx;
+ struct mlx5hws_matcher *matcher = bwc_matcher->matcher;
u16 bwc_queues = mlx5hws_bwc_queues(ctx);
- struct mlx5hws_bwc_rule **bwc_rules;
struct mlx5hws_rule_attr rule_attr;
- u32 *pending_rules;
- int i, j, ret = 0;
- bool all_done;
- u16 burst_th;
+ struct mlx5hws_bwc_rule *bwc_rule;
+ struct mlx5hws_send_engine *queue;
+ struct list_head *rules_list;
+ u32 pending_rules;
+ int i, ret = 0;
mlx5hws_bwc_rule_fill_attr(bwc_matcher, 0, 0, &rule_attr);
- pending_rules = kcalloc(bwc_queues, sizeof(*pending_rules), GFP_KERNEL);
- if (!pending_rules)
- return -ENOMEM;
-
- bwc_rules = kcalloc(bwc_queues, sizeof(*bwc_rules), GFP_KERNEL);
- if (!bwc_rules) {
- ret = -ENOMEM;
- goto free_pending_rules;
- }
-
for (i = 0; i < bwc_queues; i++) {
if (list_empty(&bwc_matcher->rules[i]))
- bwc_rules[i] = NULL;
- else
- bwc_rules[i] = list_first_entry(&bwc_matcher->rules[i],
- struct mlx5hws_bwc_rule,
- list_node);
- }
+ continue;
- do {
- all_done = true;
+ pending_rules = 0;
+ rule_attr.queue_id = mlx5hws_bwc_get_queue_id(ctx, i);
+ rules_list = &bwc_matcher->rules[i];
- for (i = 0; i < bwc_queues; i++) {
- rule_attr.queue_id = mlx5hws_bwc_get_queue_id(ctx, i);
- burst_th = hws_bwc_get_burst_th(ctx, rule_attr.queue_id);
-
- for (j = 0; j < burst_th && bwc_rules[i]; j++) {
- rule_attr.burst = !!((j + 1) % burst_th);
- ret = mlx5hws_matcher_resize_rule_move(bwc_matcher->matcher,
- bwc_rules[i]->rule,
- &rule_attr);
- if (unlikely(ret)) {
- mlx5hws_err(ctx,
- "Moving BWC rule failed during rehash (%d)\n",
- ret);
- goto free_bwc_rules;
- }
+ list_for_each_entry(bwc_rule, rules_list, list_node) {
+ ret = mlx5hws_matcher_resize_rule_move(matcher,
+ bwc_rule->rule,
+ &rule_attr);
+ if (unlikely(ret && !move_error)) {
+ mlx5hws_err(ctx,
+ "Moving BWC rule: move failed (%d), attempting to move rest of the rules\n",
+ ret);
+ move_error = true;
+ }
- all_done = false;
- pending_rules[i]++;
- bwc_rules[i] = list_is_last(&bwc_rules[i]->list_node,
- &bwc_matcher->rules[i]) ?
- NULL : list_next_entry(bwc_rules[i], list_node);
-
- ret = mlx5hws_bwc_queue_poll(ctx,
- rule_attr.queue_id,
- &pending_rules[i],
- false);
- if (unlikely(ret)) {
- mlx5hws_err(ctx,
- "Moving BWC rule failed during rehash (%d)\n",
- ret);
- goto free_bwc_rules;
- }
+ pending_rules++;
+ ret = mlx5hws_bwc_queue_poll(ctx,
+ rule_attr.queue_id,
+ &pending_rules,
+ false);
+ if (unlikely(ret && !poll_error)) {
+ mlx5hws_err(ctx,
+ "Moving BWC rule: poll failed (%d), attempting to move rest of the rules\n",
+ ret);
+ poll_error = true;
}
}
- } while (!all_done);
-
- /* drain all the bwc queues */
- for (i = 0; i < bwc_queues; i++) {
- if (pending_rules[i]) {
- u16 queue_id = mlx5hws_bwc_get_queue_id(ctx, i);
- mlx5hws_send_engine_flush_queue(&ctx->send_queue[queue_id]);
- ret = mlx5hws_bwc_queue_poll(ctx, queue_id,
- &pending_rules[i], true);
- if (unlikely(ret)) {
+ if (pending_rules) {
+ queue = &ctx->send_queue[rule_attr.queue_id];
+ mlx5hws_send_engine_flush_queue(queue);
+ ret = mlx5hws_bwc_queue_poll(ctx,
+ rule_attr.queue_id,
+ &pending_rules,
+ true);
+ if (unlikely(ret && !drain_error)) {
mlx5hws_err(ctx,
- "Moving BWC rule failed during rehash (%d)\n", ret);
- goto free_bwc_rules;
+ "Moving BWC rule: drain failed (%d), attempting to move rest of the rules\n",
+ ret);
+ drain_error = true;
}
}
}
-free_bwc_rules:
- kfree(bwc_rules);
-free_pending_rules:
- kfree(pending_rules);
+ if (move_error || poll_error || drain_error)
+ ret = -EINVAL;
return ret;
}
@@ -742,15 +716,18 @@ static int hws_bwc_matcher_move(struct mlx5hws_bwc_matcher *bwc_matcher)
}
ret = hws_bwc_matcher_move_all(bwc_matcher);
- if (ret) {
- mlx5hws_err(ctx, "Rehash error: moving rules failed\n");
- return -ENOMEM;
- }
+ if (ret)
+ mlx5hws_err(ctx, "Rehash error: moving rules failed, attempting to remove the old matcher\n");
+
+ /* Error during rehash can't be rolled back.
+ * The best option here is to allow the rehash to complete and remove
+ * the old matcher - can't leave the matcher in the 'in_resize' state.
+ */
bwc_matcher->matcher = new_matcher;
mlx5hws_matcher_destroy(old_matcher);
- return 0;
+ return ret;
}
static int
--
2.50.1 (Apple Git-155)
Reference in New Issue View Git Blame Copy Permalink