From: Dariusz Sosnowski <dsosnowski@nvidia.com>
To: Viacheslav Ovsiienko <viacheslavo@nvidia.com>,
Bing Zhao <bingz@nvidia.com>, Ori Kam <orika@nvidia.com>,
Suanming Mou <suanmingm@nvidia.com>,
Matan Azrad <matan@nvidia.com>
Cc: <dev@dpdk.org>, Raslan Darawsheh <rasland@nvidia.com>
Subject: [PATCH 5/5] net/mlx5: support count and age on root group
Date: Tue, 4 Nov 2025 18:46:12 +0100 [thread overview]
Message-ID: <20251104174612.1341962-6-dsosnowski@nvidia.com> (raw)
In-Reply-To: <20251104174612.1341962-1-dsosnowski@nvidia.com>
This patch enables support for count and age flow actions
in root table i.e. flow group 0, when working
with HW Steering flow engine (dv_flow_en = 2).
It depends on support in HWS layer introduced in previous patches.
Specifically this patch:
- Adjusts action validation during flow actions translation
to allow count and age on root group.
- Adds creation and destruction of mlx5dr count action
for use on root group.
- Adjusts count action construction (done during rule create)
to select correct action variant based on the group
where the rule is created.
Signed-off-by: Dariusz Sosnowski <dsosnowski@nvidia.com>
---
doc/guides/nics/mlx5.rst | 34 ++++++++-----
doc/guides/rel_notes/release_25_11.rst | 4 ++
drivers/net/mlx5/mlx5_flow_hw.c | 66 ++++++++++++++++++--------
drivers/net/mlx5/mlx5_hws_cnt.c | 43 +++++++++++++----
drivers/net/mlx5/mlx5_hws_cnt.h | 17 +++++--
5 files changed, 121 insertions(+), 43 deletions(-)
diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst
index fde98ae993..1ddd82b5d6 100644
--- a/doc/guides/nics/mlx5.rst
+++ b/doc/guides/nics/mlx5.rst
@@ -2698,19 +2698,26 @@ DPDK 19.05 19.02 21.05 21.05
Limitations
^^^^^^^^^^^
-Because freeing a counter (by destroying a flow rule or destroying indirect action)
-does not immediately make it available for the application,
-the PMD might return:
+With :ref:`HW steering <mlx5_hws>`:
-- ``ENOENT`` if no counter is available in ``free``, ``reuse``
- or ``wait_reset`` rings.
- No counter will be available until the application releases some of them.
-- ``EAGAIN`` if no counter is available in ``free`` and ``reuse`` rings,
- but there are counters in ``wait_reset`` ring.
- This means that after the next service thread cycle new counters will be available.
+#. Because freeing a counter (by destroying a flow rule or destroying indirect action)
+ does not immediately make it available for the application,
+ the PMD might return:
-The application has to be aware that flow rule create or indirect action create
-might need be retried.
+ - ``ENOENT`` if no counter is available in ``free``, ``reuse``
+ or ``wait_reset`` rings.
+ No counter will be available until the application releases some of them.
+ - ``EAGAIN`` if no counter is available in ``free`` and ``reuse`` rings,
+ but there are counters in ``wait_reset`` ring.
+ This means that after the next service thread cycle new counters will be available.
+
+ The application has to be aware that flow rule create or indirect action create
+ might need be retried.
+
+#. Using count action on root tables requires:
+
+ - Linux kernel >= v6.4
+ - rdma-core >= v60.0
.. _mlx5_age:
@@ -2754,6 +2761,11 @@ With :ref:`HW steering <mlx5_hws>`,
(``RTE_FLOW_PORT_FLAG_STRICT_QUEUE`` passed to ``rte_flow_configure()``),
indirect age actions can be created only through asynchronous flow API.
+#. Using age action on root tables requires:
+
+ - Linux kernel >= v6.4
+ - rdma-core >= v60.0
+
.. _mlx5_quota:
diff --git a/doc/guides/rel_notes/release_25_11.rst b/doc/guides/rel_notes/release_25_11.rst
index c5ba335cfc..40c6f30ae2 100644
--- a/doc/guides/rel_notes/release_25_11.rst
+++ b/doc/guides/rel_notes/release_25_11.rst
@@ -147,6 +147,10 @@ New Features
* Added multi-process per port.
* Optimized code.
+* **Updated NVIDIA mlx5 driver.**
+
+ * Support count and age flow actions on root tables with HW Steering flow engine.
+
* **Added PQC ML algorithms in cryptodev.**
* Added PQC ML-KEM support with reference to FIPS203.
diff --git a/drivers/net/mlx5/mlx5_flow_hw.c b/drivers/net/mlx5/mlx5_flow_hw.c
index 4aa0a01b5e..cca8fe6f46 100644
--- a/drivers/net/mlx5/mlx5_flow_hw.c
+++ b/drivers/net/mlx5/mlx5_flow_hw.c
@@ -1801,7 +1801,7 @@ flow_hw_represented_port_compile(struct rte_eth_dev *dev,
static __rte_always_inline int
flow_hw_cnt_compile(struct rte_eth_dev *dev, uint32_t start_pos,
- struct mlx5_hw_actions *acts)
+ struct mlx5_hw_actions *acts, bool is_root)
{
struct mlx5_priv *priv = dev->data->dev_private;
uint32_t pos = start_pos;
@@ -1815,7 +1815,8 @@ flow_hw_cnt_compile(struct rte_eth_dev *dev, uint32_t start_pos,
(priv->hws_cpool,
cnt_id,
&acts->rule_acts[pos].action,
- &acts->rule_acts[pos].counter.offset);
+ &acts->rule_acts[pos].counter.offset,
+ is_root);
if (ret != 0)
return ret;
acts->cnt_id = cnt_id;
@@ -2461,6 +2462,28 @@ mlx5_create_ipv6_ext_reformat(struct rte_eth_dev *dev,
return -EINVAL;
}
+static bool
+is_counter_supported_root(struct rte_eth_dev *dev)
+{
+ struct mlx5_priv *priv = dev->data->dev_private;
+ struct mlx5_hws_cnt_pool *pool = mlx5_hws_cnt_host_pool(priv->hws_cpool);
+
+ return pool->is_root_supported;
+}
+
+static bool
+is_indirect_action_type_supported_root(struct rte_eth_dev *dev,
+ const enum rte_flow_action_type type)
+{
+ switch (type) {
+ case RTE_FLOW_ACTION_TYPE_COUNT:
+ case RTE_FLOW_ACTION_TYPE_AGE:
+ return is_counter_supported_root(dev);
+ default:
+ return false;
+ }
+}
+
/**
* Translate rte_flow actions to DR action.
*
@@ -2545,8 +2568,10 @@ __flow_hw_translate_actions_template(struct rte_eth_dev *dev,
goto err;
break;
case RTE_FLOW_ACTION_TYPE_INDIRECT:
- if (is_root) {
- DRV_LOG(ERR, "Indirect action is not supported in root table.");
+ if (is_root && !is_indirect_action_type_supported_root(dev, masks->type)) {
+ DRV_LOG(ERR, "Indirect action type (%d) is not supported on "
+ "root table.",
+ masks->type);
goto err;
}
if (actions->conf && masks->conf) {
@@ -2781,12 +2806,10 @@ __flow_hw_translate_actions_template(struct rte_eth_dev *dev,
goto err;
break;
case RTE_FLOW_ACTION_TYPE_AGE:
- if (is_root) {
- __flow_hw_action_template_destroy(dev, acts);
+ if (is_root && !is_counter_supported_root(dev)) {
rte_flow_error_set(&sub_error, ENOTSUP,
- RTE_FLOW_ERROR_TYPE_ACTION,
- NULL,
- "Age action on root table is not supported in HW steering mode");
+ RTE_FLOW_ERROR_TYPE_ACTION, NULL,
+ "Age action is not supported on group 0");
goto err;
}
if (__flow_hw_act_data_general_append(priv, acts,
@@ -2796,12 +2819,10 @@ __flow_hw_translate_actions_template(struct rte_eth_dev *dev,
goto err;
break;
case RTE_FLOW_ACTION_TYPE_COUNT:
- if (is_root) {
- __flow_hw_action_template_destroy(dev, acts);
+ if (is_root && !is_counter_supported_root(dev)) {
rte_flow_error_set(&sub_error, ENOTSUP,
- RTE_FLOW_ERROR_TYPE_ACTION,
- NULL,
- "Counter action on root table is not supported in HW steering mode");
+ RTE_FLOW_ERROR_TYPE_ACTION, NULL,
+ "Count action is not supported on group 0");
goto err;
}
if ((at->action_flags & MLX5_FLOW_ACTION_AGE) ||
@@ -2815,7 +2836,7 @@ __flow_hw_translate_actions_template(struct rte_eth_dev *dev,
if (masks->conf &&
((const struct rte_flow_action_count *)
masks->conf)->id) {
- err = flow_hw_cnt_compile(dev, dr_pos, acts);
+ err = flow_hw_cnt_compile(dev, dr_pos, acts, is_root);
if (err)
goto err;
} else if (__flow_hw_act_data_general_append
@@ -3157,7 +3178,7 @@ flow_hw_construct_quota(struct mlx5_priv *priv,
static __rte_always_inline int
flow_hw_shared_action_construct(struct rte_eth_dev *dev, uint32_t queue,
const struct rte_flow_action *action,
- struct rte_flow_template_table *table __rte_unused,
+ struct rte_flow_template_table *table,
const uint64_t item_flags, uint64_t action_flags,
struct rte_flow_hw *flow,
struct mlx5dr_rule_action *rule_act)
@@ -3176,6 +3197,7 @@ flow_hw_shared_action_construct(struct rte_eth_dev *dev, uint32_t queue,
((1u << MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1);
uint32_t *cnt_queue;
cnt_id_t age_cnt;
+ bool is_root = mlx5_group_id_is_root(table->grp->group_id);
memset(&act_data, 0, sizeof(act_data));
switch (type) {
@@ -3201,7 +3223,8 @@ flow_hw_shared_action_construct(struct rte_eth_dev *dev, uint32_t queue,
if (mlx5_hws_cnt_pool_get_action_offset(priv->hws_cpool,
act_idx,
&rule_act->action,
- &rule_act->counter.offset))
+ &rule_act->counter.offset,
+ is_root))
return -1;
flow->flags |= MLX5_FLOW_HW_FLOW_FLAG_CNT_ID;
flow->cnt_id = act_idx;
@@ -3242,7 +3265,7 @@ flow_hw_shared_action_construct(struct rte_eth_dev *dev, uint32_t queue,
}
if (mlx5_hws_cnt_pool_get_action_offset(priv->hws_cpool,
age_cnt, &rule_act->action,
- &rule_act->counter.offset))
+ &rule_act->counter.offset, is_root))
return -1;
break;
case MLX5_INDIRECT_ACTION_TYPE_CT:
@@ -3476,6 +3499,7 @@ flow_hw_actions_construct(struct rte_eth_dev *dev,
struct mlx5_aso_mtr *aso_mtr;
struct mlx5_multi_pattern_segment *mp_segment = NULL;
struct rte_flow_hw_aux *aux;
+ bool is_root = mlx5_group_id_is_root(table->grp->group_id);
attr.group = table->grp->group_id;
ft_flag = mlx5_hw_act_flag[!!table->grp->group_id][table->type];
@@ -3673,7 +3697,8 @@ flow_hw_actions_construct(struct rte_eth_dev *dev,
(priv->hws_cpool,
cnt_id,
&rule_acts[act_data->action_dst].action,
- &rule_acts[act_data->action_dst].counter.offset
+ &rule_acts[act_data->action_dst].counter.offset,
+ is_root
);
if (ret != 0)
goto error;
@@ -3685,7 +3710,8 @@ flow_hw_actions_construct(struct rte_eth_dev *dev,
(priv->hws_cpool,
act_data->shared_counter.id,
&rule_acts[act_data->action_dst].action,
- &rule_acts[act_data->action_dst].counter.offset
+ &rule_acts[act_data->action_dst].counter.offset,
+ is_root
);
if (ret != 0)
goto error;
diff --git a/drivers/net/mlx5/mlx5_hws_cnt.c b/drivers/net/mlx5/mlx5_hws_cnt.c
index fb01fce4e5..7a2bb6fec4 100644
--- a/drivers/net/mlx5/mlx5_hws_cnt.c
+++ b/drivers/net/mlx5/mlx5_hws_cnt.c
@@ -439,6 +439,10 @@ mlx5_hws_cnt_pool_init(struct mlx5_dev_ctx_shared *sh,
}
cntp->cfg = *pcfg;
+ cntp->is_root_supported = mlx5dr_action_counter_root_is_supported();
+ DRV_LOG(DEBUG, "ibdev %s counter and age action %s supported on group 0",
+ sh->ibdev_name,
+ cntp->is_root_supported ? "is" : "is not");
if (cntp->cfg.host_cpool)
return cntp;
if (pcfg->request_num > sh->hws_max_nb_counters) {
@@ -659,9 +663,13 @@ mlx5_hws_cnt_pool_action_destroy(struct mlx5_hws_cnt_pool *cpool)
for (idx = 0; idx < cpool->dcs_mng.batch_total; idx++) {
struct mlx5_hws_cnt_dcs *dcs = &cpool->dcs_mng.dcs[idx];
- if (dcs->dr_action != NULL) {
- mlx5dr_action_destroy(dcs->dr_action);
- dcs->dr_action = NULL;
+ if (dcs->root_action != NULL) {
+ mlx5dr_action_destroy(dcs->root_action);
+ dcs->root_action = NULL;
+ }
+ if (dcs->hws_action != NULL) {
+ mlx5dr_action_destroy(dcs->hws_action);
+ dcs->hws_action = NULL;
}
}
}
@@ -673,11 +681,14 @@ mlx5_hws_cnt_pool_action_create(struct mlx5_priv *priv,
struct mlx5_hws_cnt_pool *hpool = mlx5_hws_cnt_host_pool(cpool);
uint32_t idx;
int ret = 0;
- uint32_t flags;
+ uint32_t root_flags;
+ uint32_t hws_flags;
- flags = MLX5DR_ACTION_FLAG_HWS_RX | MLX5DR_ACTION_FLAG_HWS_TX;
+ root_flags = MLX5DR_ACTION_FLAG_ROOT_RX | MLX5DR_ACTION_FLAG_ROOT_TX;
+ hws_flags = MLX5DR_ACTION_FLAG_HWS_RX | MLX5DR_ACTION_FLAG_HWS_TX;
if (priv->sh->config.dv_esw_en && priv->master) {
- flags |= (is_unified_fdb(priv) ?
+ root_flags |= MLX5DR_ACTION_FLAG_ROOT_FDB;
+ hws_flags |= (is_unified_fdb(priv) ?
(MLX5DR_ACTION_FLAG_HWS_FDB_RX |
MLX5DR_ACTION_FLAG_HWS_FDB_TX |
MLX5DR_ACTION_FLAG_HWS_FDB_UNIFIED) :
@@ -687,10 +698,24 @@ mlx5_hws_cnt_pool_action_create(struct mlx5_priv *priv,
struct mlx5_hws_cnt_dcs *hdcs = &hpool->dcs_mng.dcs[idx];
struct mlx5_hws_cnt_dcs *dcs = &cpool->dcs_mng.dcs[idx];
- dcs->dr_action = mlx5dr_action_create_counter(priv->dr_ctx,
+ dcs->hws_action = mlx5dr_action_create_counter(priv->dr_ctx,
+ (struct mlx5dr_devx_obj *)hdcs->obj,
+ hws_flags);
+ if (dcs->hws_action == NULL) {
+ mlx5_hws_cnt_pool_action_destroy(cpool);
+ ret = -ENOSYS;
+ break;
+ }
+
+ if (!cpool->is_root_supported) {
+ dcs->root_action = NULL;
+ continue;
+ }
+
+ dcs->root_action = mlx5dr_action_create_counter(priv->dr_ctx,
(struct mlx5dr_devx_obj *)hdcs->obj,
- flags);
- if (dcs->dr_action == NULL) {
+ root_flags);
+ if (dcs->root_action == NULL) {
mlx5_hws_cnt_pool_action_destroy(cpool);
ret = -ENOSYS;
break;
diff --git a/drivers/net/mlx5/mlx5_hws_cnt.h b/drivers/net/mlx5/mlx5_hws_cnt.h
index 8d7a9d34df..7d65ae82a0 100644
--- a/drivers/net/mlx5/mlx5_hws_cnt.h
+++ b/drivers/net/mlx5/mlx5_hws_cnt.h
@@ -31,7 +31,8 @@
#define MLX5_HWS_AGE_IDX_MASK (RTE_BIT32(MLX5_INDIRECT_ACTION_TYPE_OFFSET) - 1)
struct mlx5_hws_cnt_dcs {
- void *dr_action;
+ struct mlx5dr_action *root_action; /* mlx5dr action used on root groups. */
+ struct mlx5dr_action *hws_action; /* mlx5dr action used on non-root groups. */
uint32_t batch_sz;
uint32_t iidx; /* internal index of first counter in this bulk. */
struct mlx5_devx_obj *obj;
@@ -113,6 +114,7 @@ struct __rte_cache_aligned mlx5_hws_cnt_pool {
struct mlx5_hws_cnt_pool_caches *cache;
uint64_t time_of_last_age_check;
struct mlx5_priv *priv;
+ uint32_t is_root_supported:1;
};
/* HWS AGE status. */
@@ -632,12 +634,21 @@ mlx5_hws_cnt_pool_get_size(struct mlx5_hws_cnt_pool *cpool)
static __rte_always_inline int
mlx5_hws_cnt_pool_get_action_offset(struct mlx5_hws_cnt_pool *cpool,
cnt_id_t cnt_id, struct mlx5dr_action **action,
- uint32_t *offset)
+ uint32_t *offset, bool is_root)
{
uint8_t idx = cnt_id >> MLX5_HWS_CNT_DCS_IDX_OFFSET;
idx &= MLX5_HWS_CNT_DCS_IDX_MASK;
- *action = cpool->dcs_mng.dcs[idx].dr_action;
+ if (likely(!is_root)) {
+ *action = cpool->dcs_mng.dcs[idx].hws_action;
+ } else {
+ /*
+ * Any table using counter on root group should be rejected on validation
+ * when counter on root is not supported.
+ */
+ MLX5_ASSERT(cpool->dcs_mng.dcs[idx].root_action != NULL);
+ *action = cpool->dcs_mng.dcs[idx].root_action;
+ }
*offset = cnt_id & MLX5_HWS_CNT_IDX_MASK;
return 0;
}
--
2.39.5
prev parent reply other threads:[~2025-11-04 17:48 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-11-04 17:46 [PATCH 0/5] net/mlx5: support count and age actions " Dariusz Sosnowski
2025-11-04 17:46 ` [PATCH 1/5] common/mlx5: detect DevX counters support in rdma-core Dariusz Sosnowski
2025-11-04 17:46 ` [PATCH 2/5] net/mlx5/hws: support counter from DevX bulk on root Dariusz Sosnowski
2025-11-04 17:46 ` [PATCH 3/5] net/mlx5: rework root group checks in table create Dariusz Sosnowski
2025-11-04 17:46 ` [PATCH 4/5] net/mlx5: improve error reporting on masked indirect actions Dariusz Sosnowski
2025-11-04 17:46 ` Dariusz Sosnowski [this message]
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=20251104174612.1341962-6-dsosnowski@nvidia.com \
--to=dsosnowski@nvidia.com \
--cc=bingz@nvidia.com \
--cc=dev@dpdk.org \
--cc=matan@nvidia.com \
--cc=orika@nvidia.com \
--cc=rasland@nvidia.com \
--cc=suanmingm@nvidia.com \
--cc=viacheslavo@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).