From: Tariq Toukan <tariqt@nvidia.com>
To: "David S. Miller" <davem@davemloft.net>,
Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
Eric Dumazet <edumazet@google.com>,
"Andrew Lunn" <andrew+netdev@lunn.ch>
Cc: Saeed Mahameed <saeedm@nvidia.com>,
Leon Romanovsky <leon@kernel.org>,
Tariq Toukan <tariqt@nvidia.com>, <netdev@vger.kernel.org>,
<linux-rdma@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
Moshe Shemesh <moshe@nvidia.com>, Mark Bloch <mbloch@nvidia.com>,
Vlad Dogaru <vdogaru@nvidia.com>,
Yevgeny Kliteynik <kliteyn@nvidia.com>,
Gal Pressman <gal@nvidia.com>
Subject: [PATCH net-next 4/4] net/mlx5: HWS, handle modify header actions dependency
Date: Tue, 20 May 2025 21:46:42 +0300 [thread overview]
Message-ID: <1747766802-958178-5-git-send-email-tariqt@nvidia.com> (raw)
In-Reply-To: <1747766802-958178-1-git-send-email-tariqt@nvidia.com>
From: Yevgeny Kliteynik <kliteyn@nvidia.com>
Having adjacent accelerated modify header actions (so-called
pattern-argument actions) may result in inconsistent outcome.
These inconsistencies can take the form of writes to the same
field or a read coupled with a write to the same field. The
solution is to detect such dependencies and insert nops between
the offending actions.
The existing implementation had a few issues, which pretty much
required a complete rewrite of the code that handles these
dependencies.
In the new implementation we're doing the following:
* Checking any two adjacent actions for conflicts (not just
odd-even pairs).
* Marking 'set' and 'add' action fields as destination, rather
than source, for the purposes of checking for conflicts.
* Checking all types of actions ('add', 'set', 'copy') for
dependencies.
* Managing offsets of the args in the buffer - copy the action
args to the right place in the buffer.
* Checking that after inserting nops we're still within the number
of supported actions - return an error otherwise.
Signed-off-by: Vlad Dogaru <vdogaru@nvidia.com>
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
.../mellanox/mlx5/core/steering/hws/action.c | 21 ++++--
.../mellanox/mlx5/core/steering/hws/pat_arg.c | 74 ++++++++++---------
.../mellanox/mlx5/core/steering/hws/pat_arg.h | 6 +-
3 files changed, 55 insertions(+), 46 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/action.c b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/action.c
index 64d115feef2c..fb62f3bc4bd4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/action.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/action.c
@@ -1190,14 +1190,15 @@ hws_action_create_modify_header_hws(struct mlx5hws_action *action,
struct mlx5hws_action_mh_pattern *pattern,
u32 log_bulk_size)
{
+ u16 num_actions, max_mh_actions = 0, hw_max_actions;
struct mlx5hws_context *ctx = action->ctx;
- u16 num_actions, max_mh_actions = 0;
int i, ret, size_in_bytes;
u32 pat_id, arg_id = 0;
__be64 *new_pattern;
size_t pat_max_sz;
pat_max_sz = MLX5HWS_ARG_CHUNK_SIZE_MAX * MLX5HWS_ARG_DATA_SIZE;
+ hw_max_actions = pat_max_sz / MLX5HWS_MODIFY_ACTION_SIZE;
size_in_bytes = pat_max_sz * sizeof(__be64);
new_pattern = kcalloc(num_of_patterns, size_in_bytes, GFP_KERNEL);
if (!new_pattern)
@@ -1211,10 +1212,14 @@ hws_action_create_modify_header_hws(struct mlx5hws_action *action,
cur_num_actions = pattern[i].sz / MLX5HWS_MODIFY_ACTION_SIZE;
- mlx5hws_pat_calc_nop(pattern[i].data, cur_num_actions,
- pat_max_sz / MLX5HWS_MODIFY_ACTION_SIZE,
- &new_num_actions, &nop_locations,
- &new_pattern[i * pat_max_sz]);
+ ret = mlx5hws_pat_calc_nop(pattern[i].data, cur_num_actions,
+ hw_max_actions, &new_num_actions,
+ &nop_locations,
+ &new_pattern[i * pat_max_sz]);
+ if (ret) {
+ mlx5hws_err(ctx, "Too many actions after nop insertion\n");
+ goto free_new_pat;
+ }
action[i].modify_header.nop_locations = nop_locations;
action[i].modify_header.num_of_actions = new_num_actions;
@@ -2116,10 +2121,12 @@ static void hws_action_modify_write(struct mlx5hws_send_engine *queue,
if (unlikely(!new_arg_data))
return;
- for (i = 0, j = 0; i < num_of_actions; i++, j++) {
- memcpy(&new_arg_data[j], arg_data, MLX5HWS_MODIFY_ACTION_SIZE);
+ for (i = 0, j = 0; j < num_of_actions; i++, j++) {
if (BIT(i) & nop_locations)
j++;
+ memcpy(&new_arg_data[j * MLX5HWS_MODIFY_ACTION_SIZE],
+ &arg_data[i * MLX5HWS_MODIFY_ACTION_SIZE],
+ MLX5HWS_MODIFY_ACTION_SIZE);
}
}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/pat_arg.c b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/pat_arg.c
index 78de19c074a7..51e4c551e0ef 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/pat_arg.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/pat_arg.c
@@ -490,8 +490,8 @@ hws_action_modify_get_target_fields(u8 action_type, __be64 *pattern,
switch (action_type) {
case MLX5_ACTION_TYPE_SET:
case MLX5_ACTION_TYPE_ADD:
- *src_field = MLX5_GET(set_action_in, pattern, field);
- *dst_field = INVALID_FIELD;
+ *src_field = INVALID_FIELD;
+ *dst_field = MLX5_GET(set_action_in, pattern, field);
break;
case MLX5_ACTION_TYPE_COPY:
*src_field = MLX5_GET(copy_action_in, pattern, src_field);
@@ -522,57 +522,59 @@ bool mlx5hws_pat_verify_actions(struct mlx5hws_context *ctx, __be64 pattern[], s
return true;
}
-void mlx5hws_pat_calc_nop(__be64 *pattern, size_t num_actions,
- size_t max_actions, size_t *new_size,
- u32 *nop_locations, __be64 *new_pat)
+int mlx5hws_pat_calc_nop(__be64 *pattern, size_t num_actions,
+ size_t max_actions, size_t *new_size,
+ u32 *nop_locations, __be64 *new_pat)
{
- u16 prev_src_field = 0, prev_dst_field = 0;
+ u16 prev_src_field = INVALID_FIELD, prev_dst_field = INVALID_FIELD;
u16 src_field, dst_field;
u8 action_type;
+ bool dependent;
size_t i, j;
*new_size = num_actions;
*nop_locations = 0;
if (num_actions == 1)
- return;
+ return 0;
for (i = 0, j = 0; i < num_actions; i++, j++) {
- action_type = MLX5_GET(set_action_in, &pattern[i], action_type);
+ if (j >= max_actions)
+ return -EINVAL;
+ action_type = MLX5_GET(set_action_in, &pattern[i], action_type);
hws_action_modify_get_target_fields(action_type, &pattern[i],
&src_field, &dst_field);
- if (i % 2) {
- if (action_type == MLX5_ACTION_TYPE_COPY &&
- (prev_src_field == src_field ||
- prev_dst_field == dst_field)) {
- /* need Nop */
- *new_size += 1;
- *nop_locations |= BIT(i);
- memset(&new_pat[j], 0, MLX5HWS_MODIFY_ACTION_SIZE);
- MLX5_SET(set_action_in, &new_pat[j],
- action_type,
- MLX5_MODIFICATION_TYPE_NOP);
- j++;
- } else if (prev_src_field == src_field) {
- /* need Nop */
- *new_size += 1;
- *nop_locations |= BIT(i);
- MLX5_SET(set_action_in, &new_pat[j],
- action_type,
- MLX5_MODIFICATION_TYPE_NOP);
- j++;
- }
- }
- memcpy(&new_pat[j], &pattern[i], MLX5HWS_MODIFY_ACTION_SIZE);
- /* check if no more space */
- if (j > max_actions) {
- *new_size = num_actions;
- *nop_locations = 0;
- return;
+
+ /* For every action, look at it and the previous one. The two
+ * actions are dependent if:
+ */
+ dependent =
+ (i > 0) &&
+ /* At least one of the actions is a write and */
+ (dst_field != INVALID_FIELD ||
+ prev_dst_field != INVALID_FIELD) &&
+ /* One reads from the other's source */
+ (dst_field == prev_src_field ||
+ src_field == prev_dst_field ||
+ /* Or both write to the same destination */
+ dst_field == prev_dst_field);
+
+ if (dependent) {
+ *new_size += 1;
+ *nop_locations |= BIT(i);
+ memset(&new_pat[j], 0, MLX5HWS_MODIFY_ACTION_SIZE);
+ MLX5_SET(set_action_in, &new_pat[j], action_type,
+ MLX5_MODIFICATION_TYPE_NOP);
+ j++;
+ if (j >= max_actions)
+ return -EINVAL;
}
+ memcpy(&new_pat[j], &pattern[i], MLX5HWS_MODIFY_ACTION_SIZE);
prev_src_field = src_field;
prev_dst_field = dst_field;
}
+
+ return 0;
}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/pat_arg.h b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/pat_arg.h
index 91bd2572a341..7fbd8dc7aa18 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/pat_arg.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/steering/hws/pat_arg.h
@@ -96,7 +96,7 @@ int mlx5hws_arg_write_inline_arg_data(struct mlx5hws_context *ctx,
u8 *arg_data,
size_t data_size);
-void mlx5hws_pat_calc_nop(__be64 *pattern, size_t num_actions,
- size_t max_actions, size_t *new_size,
- u32 *nop_locations, __be64 *new_pat);
+int mlx5hws_pat_calc_nop(__be64 *pattern, size_t num_actions,
+ size_t max_actions, size_t *new_size,
+ u32 *nop_locations, __be64 *new_pat);
#endif /* MLX5HWS_PAT_ARG_H_ */
--
2.31.1
next prev parent reply other threads:[~2025-05-20 18:47 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-05-20 18:46 [PATCH net-next 0/4] net/mlx5: HWS, set of fixes and adjustments Tariq Toukan
2025-05-20 18:46 ` [PATCH net-next 1/4] net/mlx5: SWS, fix reformat id error handling Tariq Toukan
2025-05-20 18:46 ` [PATCH net-next 2/4] net/mlx5: HWS, register reformat actions with fw Tariq Toukan
2025-05-20 18:46 ` [PATCH net-next 3/4] net/mlx5: HWS, fix typo - 'nope' to 'nop' Tariq Toukan
2025-05-20 18:46 ` Tariq Toukan [this message]
2025-05-22 3:50 ` [PATCH net-next 0/4] net/mlx5: HWS, set of fixes and adjustments patchwork-bot+netdevbpf
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1747766802-958178-5-git-send-email-tariqt@nvidia.com \
--to=tariqt@nvidia.com \
--cc=andrew+netdev@lunn.ch \
--cc=davem@davemloft.net \
--cc=edumazet@google.com \
--cc=gal@nvidia.com \
--cc=kliteyn@nvidia.com \
--cc=kuba@kernel.org \
--cc=leon@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-rdma@vger.kernel.org \
--cc=mbloch@nvidia.com \
--cc=moshe@nvidia.com \
--cc=netdev@vger.kernel.org \
--cc=pabeni@redhat.com \
--cc=saeedm@nvidia.com \
--cc=vdogaru@nvidia.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).