* [PATCH net-next 0/2] net/mlx5: less memory per offloaded tc rule
@ 2026-10-09 5:20 Alexandre Cassen
2026-10-09 5:21 ` [PATCH net-next 1/2] net/mlx5e: TC, free NIC flow parse attributes after offload Alexandre Cassen
2026-10-09 5:21 ` [PATCH net-next 2/2] net/mlx5: HWS, shrink FTE action array to fit Alexandre Cassen
0 siblings, 2 replies; 5+ messages in thread
From: Alexandre Cassen @ 2026-10-09 5:20 UTC (permalink / raw)
To: Saeed Mahameed, Leon Romanovsky, Tariq Toukan, Mark Bloch
Cc: Andrew Lunn, David S. Miller, Eric Dumazet, Jakub Kicinski,
Paolo Abeni, Moshe Shemesh, Yevgeny Kliteynik, netdev, linux-rdma
Two mlx5 allocations hold more memory than an offloaded tc rule needs.
A NIC flow keeps a 4 KB allocation for parse attributes that nothing
reads after offload. An FTE under hardware steering keeps the 1 KB
action array it was built in while a rule rarely needs more than three
entries.
Freeing the first after offload saves 4 KB per NIC flow. Shrinking the
second saves about 1 KB per FTE under hardware steering.
Alexandre Cassen (2):
net/mlx5e: TC, free NIC flow parse attributes after offload
net/mlx5: HWS, shrink FTE action array to fit
.../net/ethernet/mellanox/mlx5/core/en_tc.c | 18 +++++++++++++-----
.../mellanox/mlx5/core/steering/hws/fs_hws.c | 10 +++++++++-
2 files changed, 22 insertions(+), 6 deletions(-)
--
2.43.0
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH net-next 1/2] net/mlx5e: TC, free NIC flow parse attributes after offload
2026-10-09 5:20 [PATCH net-next 0/2] net/mlx5: less memory per offloaded tc rule Alexandre Cassen
@ 2026-10-09 5:21 ` Alexandre Cassen
2026-10-10 15:42 ` Simon Horman
2026-10-09 5:21 ` [PATCH net-next 2/2] net/mlx5: HWS, shrink FTE action array to fit Alexandre Cassen
1 sibling, 1 reply; 5+ messages in thread
From: Alexandre Cassen @ 2026-10-09 5:21 UTC (permalink / raw)
To: Saeed Mahameed, Leon Romanovsky, Tariq Toukan, Mark Bloch
Cc: Andrew Lunn, David S. Miller, Eric Dumazet, Jakub Kicinski,
Paolo Abeni, Moshe Shemesh, Yevgeny Kliteynik, netdev, linux-rdma
A NIC flow holds a 4 KB allocation for its parse attributes until
deletion although nothing reads them after offload. Free them once the
rule is in.
Signed-off-by: Alexandre Cassen <acassen@corp.free.fr>
---
.../net/ethernet/mellanox/mlx5/core/en_tc.c | 18 +++++++++++++-----
1 file changed, 13 insertions(+), 5 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
index c99e824b4c6d..708b7987ea4c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
@@ -1414,6 +1414,16 @@ void mlx5e_del_offloaded_nic_rule(struct mlx5e_priv *priv,
MLX5E_TC_FT_LEVEL);
}
+static void mlx5e_tc_free_parse_attr(struct mlx5_flow_attr *attr)
+{
+ if (!attr->parse_attr)
+ return;
+
+ mlx5e_mod_hdr_dealloc(&attr->parse_attr->mod_hdr_acts);
+ kvfree(attr->parse_attr);
+ attr->parse_attr = NULL;
+}
+
static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
struct mlx5e_tc_flow *flow)
{
@@ -1436,10 +1446,8 @@ static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
}
mutex_unlock(&tc->t_lock);
- if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) {
- mlx5e_mod_hdr_dealloc(&attr->parse_attr->mod_hdr_acts);
+ if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR)
mlx5e_tc_detach_mod_hdr(priv, flow, attr);
- }
if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT)
mlx5_fc_destroy(priv->mdev, attr->counter);
@@ -1450,7 +1458,7 @@ static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
free_flow_post_acts(flow);
mlx5_tc_ct_delete_flow(get_ct_priv(flow->priv), attr);
- kvfree(attr->parse_attr);
+ mlx5e_tc_free_parse_attr(attr);
kfree(flow->attr);
}
@@ -4759,6 +4767,7 @@ mlx5e_add_nic_flow(struct mlx5e_priv *priv,
if (err)
goto err_free;
+ mlx5e_tc_free_parse_attr(flow->attr);
flow_flag_set(flow, OFFLOADED);
*__flow = flow;
@@ -4766,7 +4775,6 @@ mlx5e_add_nic_flow(struct mlx5e_priv *priv,
err_free:
flow_flag_set(flow, FAILED);
- mlx5e_mod_hdr_dealloc(&parse_attr->mod_hdr_acts);
mlx5e_flow_put(priv, flow);
out:
return err;
--
2.43.0
^ permalink raw reply related [flat|nested] 5+ messages in thread
* [PATCH net-next 2/2] net/mlx5: HWS, shrink FTE action array to fit
2026-10-09 5:20 [PATCH net-next 0/2] net/mlx5: less memory per offloaded tc rule Alexandre Cassen
2026-10-09 5:21 ` [PATCH net-next 1/2] net/mlx5e: TC, free NIC flow parse attributes after offload Alexandre Cassen
@ 2026-10-09 5:21 ` Alexandre Cassen
2026-10-10 15:42 ` Simon Horman
1 sibling, 1 reply; 5+ messages in thread
From: Alexandre Cassen @ 2026-10-09 5:21 UTC (permalink / raw)
To: Saeed Mahameed, Leon Romanovsky, Tariq Toukan, Mark Bloch
Cc: Andrew Lunn, David S. Miller, Eric Dumazet, Jakub Kicinski,
Paolo Abeni, Moshe Shemesh, Yevgeny Kliteynik, netdev, linux-rdma
An FTE keeps the actions it must release in the array used to build
them. This is a 1 KB allocation sized for MLX5_FLOW_CONTEXT_ACTION_MAX
entries while a rule rarely needs more than three. Keep a copy sized to
the actions in use.
Signed-off-by: Alexandre Cassen <acassen@corp.free.fr>
---
.../ethernet/mellanox/mlx5/core/steering/hws/fs_hws.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/fs_hws.c b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/fs_hws.c
index 5a172c572a68..da6e57c103d6 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/fs_hws.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/fs_hws.c
@@ -746,6 +746,7 @@ static int mlx5_fs_fte_get_hws_actions(struct mlx5_flow_root_namespace *ns,
struct mlx5_fs_hws_context *fs_ctx = &ns->fs_hws_context;
struct mlx5hws_action_dest_attr *dest_actions;
struct mlx5hws_context *ctx = fs_ctx->hws_ctx;
+ struct mlx5_fs_hws_rule_action *rule_actions;
struct mlx5_fs_hws_rule_action *fs_actions;
struct mlx5_core_dev *dev = ns->dev;
struct mlx5hws_action *dest_action;
@@ -1041,13 +1042,20 @@ static int mlx5_fs_fte_get_hws_actions(struct mlx5_flow_root_namespace *ns,
fs_actions[num_fs_actions++].action = tmp_action;
(*ractions)[num_actions++].action = tmp_action;
+ rule_actions = kmemdup_array(fs_actions, num_fs_actions,
+ sizeof(*fs_actions), GFP_KERNEL);
+ if (!rule_actions) {
+ err = -ENOMEM;
+ goto free_actions;
+ }
+ kfree(fs_actions);
kfree(dest_actions);
/* Actions created specifically for this rule will be destroyed
* once rule is deleted.
*/
fte->fs_hws_rule.num_fs_actions = num_fs_actions;
- fte->fs_hws_rule.hws_fs_actions = fs_actions;
+ fte->fs_hws_rule.hws_fs_actions = rule_actions;
return 0;
--
2.43.0
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH net-next 1/2] net/mlx5e: TC, free NIC flow parse attributes after offload
2026-10-09 5:21 ` [PATCH net-next 1/2] net/mlx5e: TC, free NIC flow parse attributes after offload Alexandre Cassen
@ 2026-10-10 15:42 ` Simon Horman
0 siblings, 0 replies; 5+ messages in thread
From: Simon Horman @ 2026-10-10 15:42 UTC (permalink / raw)
To: Alexandre Cassen
Cc: Saeed Mahameed, Leon Romanovsky, Tariq Toukan, Mark Bloch,
Andrew Lunn, David S. Miller, Eric Dumazet, Jakub Kicinski,
Paolo Abeni, Moshe Shemesh, Yevgeny Kliteynik, netdev, linux-rdma
On Fri, Oct 09, 2026 at 07:21:00AM +0200, Alexandre Cassen wrote:
> A NIC flow holds a 4 KB allocation for its parse attributes until
> deletion although nothing reads them after offload. Free them once the
> rule is in.
>
> Signed-off-by: Alexandre Cassen <acassen@corp.free.fr>
Reviewed-by: Simon Horman <horms@kernel.org>
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH net-next 2/2] net/mlx5: HWS, shrink FTE action array to fit
2026-10-09 5:21 ` [PATCH net-next 2/2] net/mlx5: HWS, shrink FTE action array to fit Alexandre Cassen
@ 2026-10-10 15:42 ` Simon Horman
0 siblings, 0 replies; 5+ messages in thread
From: Simon Horman @ 2026-10-10 15:42 UTC (permalink / raw)
To: Alexandre Cassen
Cc: Saeed Mahameed, Leon Romanovsky, Tariq Toukan, Mark Bloch,
Andrew Lunn, David S. Miller, Eric Dumazet, Jakub Kicinski,
Paolo Abeni, Moshe Shemesh, Yevgeny Kliteynik, netdev, linux-rdma
On Fri, Oct 09, 2026 at 07:21:01AM +0200, Alexandre Cassen wrote:
> An FTE keeps the actions it must release in the array used to build
> them. This is a 1 KB allocation sized for MLX5_FLOW_CONTEXT_ACTION_MAX
> entries while a rule rarely needs more than three. Keep a copy sized to
> the actions in use.
>
> Signed-off-by: Alexandre Cassen <acassen@corp.free.fr>
Reviewed-by: Simon Horman <horms@kernel.org>
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2026-10-10 15:42 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-10-09 5:20 [PATCH net-next 0/2] net/mlx5: less memory per offloaded tc rule Alexandre Cassen
2026-10-09 5:21 ` [PATCH net-next 1/2] net/mlx5e: TC, free NIC flow parse attributes after offload Alexandre Cassen
2026-10-10 15:42 ` Simon Horman
2026-10-09 5:21 ` [PATCH net-next 2/2] net/mlx5: HWS, shrink FTE action array to fit Alexandre Cassen
2026-10-10 15:42 ` Simon Horman
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox