Skip to contentSkip to conversation

net/mlx5: flow fast path validation

PATCH Superseded Dariusz Sosnowski 3 participants 32 messages Jun 5, 2024 – Jun 30, 2024 Network drivers

lore
Share & tools
lore
#0 lore raw

cover net/mlx5: flow fast path validation

This patchset adds validations for flow items and actions supplied by the user to fast path async flow API functions implemented in mlx5 PMD.

- Patch 1: Adds the ability to generically calculate flow item's mask size using rte_flow_conv().
This change is used in follow up patches to check for non-zero mask in pattern template.
- Patches 2-7: Prepares mlx5 PMD for adding fast path validations. Specifically:

- Required functionality for validating queue indexes and target represented ports are extracted for reuse.
- Additional info, required for validation, is stored in pattern and actions templates.
- Introduces mlx5_fp_debug_enabled() function for checking if mlx5 PMD was compiled in debug mode.

- Patch 8: Introduce validations to:

- rte_flow_async_create()
- rte_flow_async_create_by_index()
- rte_flow_async_actions_update()
- rte_flow_async_destroy()

These validations are enabled if and only if RTE_LIBRTE_MLX5_DEBUG macro is defined during compilation.

Depends-on: series-32087 ("net/mlx5: add match with Tx queue item")
Depends-on: series-32080 ("net/mlx5: validate HWS template items")

Dariusz Sosnowski (9):
ethdev: support duplicating only item mask net/mlx5: extract target port validation net/mlx5: extract queue index validation net/mlx5: store pattern template items net/mlx5: store original actions in template net/mlx5: store expected type on indirect action net/mlx5: store modify field action common/mlx5: add debug mode indicator net/mlx5: add async flow operation validation

 doc/guides/rel_notes/release_24_07.rst |   1 +
 drivers/common/mlx5/mlx5_common.h      |  13 +
 drivers/net/mlx5/mlx5_flow.c           |  59 ++-
 drivers/net/mlx5/mlx5_flow.h           |  12 +
 drivers/net/mlx5/mlx5_flow_hw.c        | 620 +++++++++++++++++++++++--
 lib/ethdev/rte_flow.c                  |  15 +-
 lib/ethdev/rte_flow.h                  |  12 +
 7 files changed, 682 insertions(+), 50 deletions(-)

--
2.39.2

#1 lore raw

PATCH 1/9 ethdev: support duplicating only item mask

Extend rte_flow_conv() to support working only on item's mask. This allows drivers to get only the mask's size when working on pattern templates and duplicate items having only the mask in a generic way.

---
 lib/ethdev/rte_flow.c | 15 +++++++++++++--
 lib/ethdev/rte_flow.h | 12 ++++++++++++
 2 files changed, 25 insertions(+), 2 deletions(-)
lib/ethdev/rte_flow.c +13 -2
index 7ab1100ea0..ca2f85c3fa 100644
@@ -60,9 +60,10 @@ rte_flow_conv_copy(void *buf, const void *data, const size_t size,
 	bool rte_type = type >= 0;
 
 	size_t sz = rte_type ? desc[type].size : sizeof(void *);
-	if (buf == NULL || data == NULL)
+	if (data == NULL)
 		return 0;
-	rte_memcpy(buf, data, (size > sz ? sz : size));
+	if (buf != NULL)
+		rte_memcpy(buf, data, (size > sz ? sz : size));
 	if (rte_type && desc[type].desc_fn)
 		sz += desc[type].desc_fn(size > 0 ? buf : NULL, data);
 	return sz;
@@ -1090,6 +1091,7 @@ rte_flow_conv(enum rte_flow_conv_op op,
 
 	switch (op) {
 		const struct rte_flow_attr *attr;
+		const struct rte_flow_item *item;
 
 	case RTE_FLOW_CONV_OP_NONE:
 		ret = 0;
@@ -1104,6 +1106,15 @@ rte_flow_conv(enum rte_flow_conv_op op,
 	case RTE_FLOW_CONV_OP_ITEM:
 		ret = rte_flow_conv_pattern(dst, size, src, 1, error);
 		break;
+	case RTE_FLOW_CONV_OP_ITEM_MASK:
+		item = src;
+		if (item->mask == NULL) {
+			ret = rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM_MASK,
+						 item, "Mask not provided");
+			break;
+		}
+		ret = rte_flow_conv_item_spec(dst, size, src, RTE_FLOW_CONV_ITEM_MASK);
+		break;
 	case RTE_FLOW_CONV_OP_ACTION:
 		ret = rte_flow_conv_actions(dst, size, src, 1, error);
 		break;
lib/ethdev/rte_flow.h +12 -0
index 6e8ab1d4c7..35f180e735 100644
@@ -4409,6 +4409,18 @@ enum rte_flow_conv_op {
 	 */
 	RTE_FLOW_CONV_OP_ITEM,
 
+	/**
+	 * Convert a single item mask.
+	 *
+	 * Duplicates only @p mask.
+	 *
+	 * - @p src type:
+	 *   @code const struct rte_flow_item * @endcode
+	 * - @p dst type:
+	 *   @code struct rte_flow_item * @endcode
+	 */
+	RTE_FLOW_CONV_OP_ITEM_MASK,
+
 	/**
 	 * Convert a single action.
 	 *
Signature · 1 line
2.39.2
#2 lore raw

PATCH 2/9 net/mlx5: extract target port validation

In reply to Dariusz Sosnowski

Extract code responsible for validation if port specified in configuration of RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT action is correct. Allow for reuse of this logic for both RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT actions and RTE_FLOW_ITEM_TYPE_REPRESENTED_PORT items.

---
 drivers/net/mlx5/mlx5_flow_hw.c | 63 +++++++++++++++++++++------------
 1 file changed, 40 insertions(+), 23 deletions(-)
drivers/net/mlx5/mlx5_flow_hw.c +40 -23
index e8562660dd..24b86535b1 100644
@@ -5716,6 +5716,42 @@ flow_hw_validate_action_port_representor(struct rte_eth_dev *dev __rte_unused,
 	return 0;
 }
 
+static int
+flow_hw_validate_target_port_id(struct rte_eth_dev *dev,
+				uint16_t target_port_id)
+{
+	struct mlx5_priv *port_priv;
+	struct mlx5_priv *dev_priv;
+
+	if (target_port_id == MLX5_REPRESENTED_PORT_ESW_MGR)
+		return 0;
+
+	port_priv = mlx5_port_to_eswitch_info(target_port_id, false);
+	if (!port_priv) {
+		rte_errno = EINVAL;
+		DRV_LOG(ERR, "Port %u Failed to obtain E-Switch info for port %u",
+			dev->data->port_id, target_port_id);
+		return -rte_errno;
+	}
+
+	dev_priv = mlx5_dev_to_eswitch_info(dev);
+	if (!dev_priv) {
+		rte_errno = EINVAL;
+		DRV_LOG(ERR, "Port %u Failed to obtain E-Switch info for transfer proxy",
+			dev->data->port_id);
+		return -rte_errno;
+	}
+
+	if (port_priv->domain_id != dev_priv->domain_id) {
+		rte_errno = EINVAL;
+		DRV_LOG(ERR, "Port %u Failed to obtain E-Switch info for transfer proxy",
+			dev->data->port_id);
+		return -rte_errno;
+	}
+
+	return 0;
+}
+
 static int
 flow_hw_validate_action_represented_port(struct rte_eth_dev *dev,
 					 const struct rte_flow_action *action,
@@ -5732,32 +5768,13 @@ flow_hw_validate_action_represented_port(struct rte_eth_dev *dev,
 					  "cannot use represented_port actions"
 					  " without an E-Switch");
 	if (mask_conf && mask_conf->port_id) {
-		struct mlx5_priv *port_priv;
-		struct mlx5_priv *dev_priv;
-
 		if (!action_conf)
 			return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
 						  action, "port index was not provided");
-		port_priv = mlx5_port_to_eswitch_info(action_conf->port_id, false);
-		if (!port_priv)
-			return rte_flow_error_set(error, rte_errno,
-						  RTE_FLOW_ERROR_TYPE_ACTION,
-						  action,
-						  "failed to obtain E-Switch"
-						  " info for port");
-		dev_priv = mlx5_dev_to_eswitch_info(dev);
-		if (!dev_priv)
-			return rte_flow_error_set(error, rte_errno,
-						  RTE_FLOW_ERROR_TYPE_ACTION,
-						  action,
-						  "failed to obtain E-Switch"
-						  " info for transfer proxy");
-		if (port_priv->domain_id != dev_priv->domain_id)
-			return rte_flow_error_set(error, rte_errno,
-						  RTE_FLOW_ERROR_TYPE_ACTION,
-						  action,
-						  "cannot forward to port from"
-						  " a different E-Switch");
+
+		if (flow_hw_validate_target_port_id(dev, action_conf->port_id))
+			return rte_flow_error_set(error, rte_errno, RTE_FLOW_ERROR_TYPE_ACTION,
+						  action, "port index is invalid");
 	}
 	return 0;
 }
Signature · 1 line
2.39.2
#3 lore raw

PATCH 3/9 net/mlx5: extract queue index validation

In reply to Dariusz Sosnowski

Extract validation of queue index out of validation of RTE_FLOW_ACTION_TYPE_QUEUE action to allow reuse in template API flow rule creation implementation.

---
 drivers/net/mlx5/mlx5_flow.c | 59 +++++++++++++++++++++++++-----------
 drivers/net/mlx5/mlx5_flow.h |  3 ++
 2 files changed, 44 insertions(+), 18 deletions(-)
drivers/net/mlx5/mlx5_flow.c +41 -18
index 18c8890e4a..9643fa424f 100644
@@ -2020,6 +2020,46 @@ mlx5_flow_validate_action_drop(struct rte_eth_dev *dev,
 	return 0;
 }
 
+/*
+ * Check if a queue specified in the queue action is valid.
+ *
+ * @param[in] dev
+ *   Pointer to the Ethernet device structure.
+ * @param[in] action
+ *   Pointer to the queue action.
+ * @param[out] error
+ *   Pointer to error structure.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_flow_validate_target_queue(struct rte_eth_dev *dev,
+				const struct rte_flow_action *action,
+				struct rte_flow_error *error)
+{
+	const struct rte_flow_action_queue *queue = action->conf;
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	if (mlx5_is_external_rxq(dev, queue->index))
+		return 0;
+	if (!priv->rxqs_n)
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  NULL, "No Rx queues configured");
+	if (queue->index >= priv->rxqs_n)
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &queue->index,
+					  "queue index out of range");
+	if (mlx5_rxq_get(dev, queue->index) == NULL)
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &queue->index,
+					  "queue is not configured");
+	return 0;
+}
+
 /*
  * Validate the queue action.
  *
@@ -2044,7 +2084,6 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 				const struct rte_flow_attr *attr,
 				struct rte_flow_error *error)
 {
-	struct mlx5_priv *priv = dev->data->dev_private;
 	const struct rte_flow_action_queue *queue = action->conf;
 
 	if (!queue)
@@ -2060,23 +2099,7 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 		return rte_flow_error_set(error, ENOTSUP,
 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 					  "queue action not supported for egress.");
-	if (mlx5_is_external_rxq(dev, queue->index))
-		return 0;
-	if (!priv->rxqs_n)
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  NULL, "No Rx queues configured");
-	if (queue->index >= priv->rxqs_n)
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  &queue->index,
-					  "queue index out of range");
-	if (mlx5_rxq_get(dev, queue->index) == NULL)
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  &queue->index,
-					  "queue is not configured");
-	return 0;
+	return mlx5_flow_validate_target_queue(dev, action, error);
 }
 
 /**
drivers/net/mlx5/mlx5_flow.h +3 -0
index 5f51f7257e..7be95a7d18 100644
@@ -2948,6 +2948,9 @@ int mlx5_flow_validate_action_mark(struct rte_eth_dev *dev,
 				   uint64_t action_flags,
 				   const struct rte_flow_attr *attr,
 				   struct rte_flow_error *error);
+int mlx5_flow_validate_target_queue(struct rte_eth_dev *dev,
+				    const struct rte_flow_action *action,
+				    struct rte_flow_error *error);
 int mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 				    uint64_t action_flags,
 				    struct rte_eth_dev *dev,
Signature · 1 line
2.39.2
#4 lore raw

PATCH 4/9 net/mlx5: store pattern template items

In reply to Dariusz Sosnowski

Store items which define the pattern template on creation. This allows validation of items, provided by the user during flow rule creation, against pattern template.

---
 drivers/net/mlx5/mlx5_flow.h    |  2 ++
 drivers/net/mlx5/mlx5_flow_hw.c | 29 +++++++++++++++++++++++++++++
 2 files changed, 31 insertions(+)
drivers/net/mlx5/mlx5_flow.h +2 -0
index 7be95a7d18..5997ffb5e0 100644
@@ -1489,6 +1489,8 @@ struct rte_flow_pattern_template {
 	/* Manages all GENEVE TLV options used by this pattern template. */
 	struct mlx5_geneve_tlv_options_mng geneve_opt_mng;
 	uint8_t flex_item; /* flex item index. */
+	/* Items on which this pattern template is based on. */
+	struct rte_flow_item *items;
 };
 
 /* Flow action template struct. */
drivers/net/mlx5/mlx5_flow_hw.c +29 -0
index 24b86535b1..132f6ca264 100644
@@ -8401,6 +8401,7 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 		.mask = &tag_m,
 		.last = NULL
 	};
+	int it_items_size;
 	unsigned int i = 0;
 	int rc;
 
@@ -8444,6 +8445,31 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 	it->attr = *attr;
 	it->item_flags = item_flags;
 	it->orig_item_nb = orig_item_nb;
+	it_items_size = rte_flow_conv(RTE_FLOW_CONV_OP_PATTERN, NULL, 0, tmpl_items, error);
+	if (it_items_size <= 0) {
+		rte_flow_error_set(error, ENOMEM,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+				   NULL,
+				   "Failed to determine buffer size for pattern");
+		goto error;
+	}
+	it_items_size = RTE_ALIGN(it_items_size, 16);
+	it->items = mlx5_malloc(MLX5_MEM_ZERO, it_items_size, 0, rte_dev_numa_node(dev->device));
+	if (it->items == NULL) {
+		rte_flow_error_set(error, ENOMEM,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+				   NULL,
+				   "Cannot allocate memory for pattern");
+		goto error;
+	}
+	rc = rte_flow_conv(RTE_FLOW_CONV_OP_PATTERN, it->items, it_items_size, tmpl_items, error);
+	if (rc <= 0) {
+		rte_flow_error_set(error, ENOMEM,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+				   NULL,
+				   "Failed to store pattern");
+		goto error;
+	}
 	it->mt = mlx5dr_match_template_create(tmpl_items, attr->relaxed_matching);
 	if (!it->mt) {
 		rte_flow_error_set(error, rte_errno,
@@ -8458,6 +8484,7 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 		else if (attr->egress)
 			it->implicit_tag = true;
 		mlx5_free(copied_items);
+		copied_items = NULL;
 	}
 	/* Either inner or outer, can't both. */
 	if (it->item_flags & (MLX5_FLOW_ITEM_OUTER_IPV6_ROUTING_EXT |
@@ -8518,6 +8545,7 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 			mlx5_geneve_tlv_options_unregister(priv, &it->geneve_opt_mng);
 		if (it->mt)
 			claim_zero(mlx5dr_match_template_destroy(it->mt));
+		mlx5_free(it->items);
 		mlx5_free(it);
 	}
 	if (copied_items)
@@ -8560,6 +8588,7 @@ flow_hw_pattern_template_destroy(struct rte_eth_dev *dev,
 	flow_hw_flex_item_release(dev, &template->flex_item);
 	mlx5_geneve_tlv_options_unregister(priv, &template->geneve_opt_mng);
 	claim_zero(mlx5dr_match_template_destroy(template->mt));
+	mlx5_free(template->items);
 	mlx5_free(template);
 	return 0;
 }
Signature · 1 line
2.39.2
#5 lore raw

PATCH 5/9 net/mlx5: store original actions in template

In reply to Dariusz Sosnowski

When actions template is created in mlx5 PMD, some actions will be replaced with another or some actions will be added to implement required template semantics on NVIDIA NICs. For example:

- RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID actions is replaced with modify field action.
- Modify field action is added for preserving metadata between FDB and NIC domains.
- Modify field action is added when quota action is used to amend ASO syndrome.

Types of actions and their order in flow rules based on some actions template must match the original, not modified one.

This patch adds preserving of the original actions for actions template to allow for easier validation of ordering and types on fast path.

---
 drivers/net/mlx5/mlx5_flow.h    |  1 +
 drivers/net/mlx5/mlx5_flow_hw.c | 11 +++++++++++
 2 files changed, 12 insertions(+)
drivers/net/mlx5/mlx5_flow.h +1 -0
index 5997ffb5e0..a40e8385b3 100644
@@ -1499,6 +1499,7 @@ struct rte_flow_actions_template {
 	/* Template attributes. */
 	struct rte_flow_actions_template_attr attr;
 	struct rte_flow_action *actions; /* Cached flow actions. */
+	struct rte_flow_action *orig_actions; /* Original flow actions. */
 	struct rte_flow_action *masks; /* Cached action masks.*/
 	struct mlx5dr_action_template *tmpl; /* mlx5dr action template. */
 	uint64_t action_flags; /* Bit-map of all valid action in template. */
drivers/net/mlx5/mlx5_flow_hw.c +11 -0
index 132f6ca264..52ad419c95 100644
@@ -7415,6 +7415,7 @@ flow_hw_actions_template_create(struct rte_eth_dev *dev,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	int len, act_len, mask_len;
+	int orig_act_len;
 	unsigned int act_num;
 	unsigned int i;
 	struct rte_flow_actions_template *at = NULL;
@@ -7529,6 +7530,10 @@ flow_hw_actions_template_create(struct rte_eth_dev *dev,
 	len += RTE_ALIGN(mask_len, 16);
 	len += RTE_ALIGN(act_num * sizeof(*at->dr_off), 16);
 	len += RTE_ALIGN(act_num * sizeof(*at->src_off), 16);
+	orig_act_len = rte_flow_conv(RTE_FLOW_CONV_OP_ACTIONS, NULL, 0, actions, error);
+	if (orig_act_len <= 0)
+		return NULL;
+	len += RTE_ALIGN(orig_act_len, 16);
 	at = mlx5_malloc(MLX5_MEM_ZERO, len + sizeof(*at),
 			 RTE_CACHE_LINE_SIZE, rte_socket_id());
 	if (!at) {
@@ -7556,6 +7561,12 @@ flow_hw_actions_template_create(struct rte_eth_dev *dev,
 	at->src_off = RTE_PTR_ADD(at->dr_off,
 				  RTE_ALIGN(act_num * sizeof(*at->dr_off), 16));
 	memcpy(at->src_off, src_off, act_num * sizeof(at->src_off[0]));
+	at->orig_actions = RTE_PTR_ADD(at->src_off,
+				       RTE_ALIGN(act_num * sizeof(*at->src_off), 16));
+	orig_act_len = rte_flow_conv(RTE_FLOW_CONV_OP_ACTIONS, at->orig_actions, orig_act_len,
+				     actions, error);
+	if (orig_act_len <= 0)
+		goto error;
 	at->actions_num = act_num;
 	for (i = 0; i < at->actions_num; ++i)
 		at->dr_off[i] = UINT16_MAX;
Signature · 1 line
2.39.2
#6 lore raw

PATCH 6/9 net/mlx5: store expected type on indirect action

In reply to Dariusz Sosnowski

When template table is created, list of unmasked actions is recorded for future flow rule insertions.
This patch expands entries for RTE_FLOW_ACTION_TYPE_INDIRECT actions in this list with information about expected indirect action type. This will be used in follow up commits which add flow rule operation validation. Specifically, this will be used to verify if indirect action provided by the user references an action of a correct type.

---
 drivers/net/mlx5/mlx5_flow.h    |  4 ++++
 drivers/net/mlx5/mlx5_flow_hw.c | 22 ++++++++++++++++++++--
 2 files changed, 24 insertions(+), 2 deletions(-)
drivers/net/mlx5/mlx5_flow.h +4 -0
index a40e8385b3..cd6fed4723 100644
@@ -1409,6 +1409,10 @@ struct mlx5_action_construct_data {
 	uint16_t action_dst; /* mlx5dr_rule_action dst offset. */
 	indirect_list_callback_t indirect_list_cb;
 	union {
+		struct {
+			/* Expected type of indirection action. */
+			enum rte_flow_action_type expected_type;
+		} indirect;
 		struct {
 			/* encap data len. */
 			uint16_t len;
drivers/net/mlx5/mlx5_flow_hw.c +20 -2
index 52ad419c95..5e9d43ab22 100644
@@ -768,6 +768,24 @@ __flow_hw_act_data_general_append(struct mlx5_priv *priv,
 	return 0;
 }
 
+static __rte_always_inline int
+__flow_hw_act_data_indirect_append(struct mlx5_priv *priv,
+				   struct mlx5_hw_actions *acts,
+				   enum rte_flow_action_type type,
+				   enum rte_flow_action_type mask_type,
+				   uint16_t action_src,
+				   uint16_t action_dst)
+{
+	struct mlx5_action_construct_data *act_data;
+
+	act_data = __flow_hw_act_data_alloc(priv, type, action_src, action_dst);
+	if (!act_data)
+		return -1;
+	act_data->indirect.expected_type = mask_type;
+	LIST_INSERT_HEAD(&acts->act_list, act_data, next);
+	return 0;
+}
+
 static __rte_always_inline int
 flow_hw_act_data_indirect_list_append(struct mlx5_priv *priv,
 				      struct mlx5_hw_actions *acts,
@@ -2215,9 +2233,9 @@ __flow_hw_actions_translate(struct rte_eth_dev *dev,
 				if (flow_hw_shared_action_translate
 				(dev, actions, acts, src_pos, dr_pos))
 					goto err;
-			} else if (__flow_hw_act_data_general_append
+			} else if (__flow_hw_act_data_indirect_append
 					(priv, acts, RTE_FLOW_ACTION_TYPE_INDIRECT,
-					 src_pos, dr_pos)){
+					 masks->type, src_pos, dr_pos)){
 				goto err;
 			}
 			break;
Signature · 1 line
2.39.2
#7 lore raw

PATCH 7/9 net/mlx5: store modify field action

In reply to Dariusz Sosnowski

When template table is created, list of unmasked actions is recorded for future flow rule insertions.
This patch expands entries for RTE_FLOW_ACTION_TYPE_MODIFY_FIELD actions in this list with a copy of the action from the template. This will be used in follow up commits which add flow rule operation validation. Specifically, to validate that RTE_FLOW_ACTION_TYPE_MODIFY_FIELD action passed by the user is correctly configured.

---
 drivers/net/mlx5/mlx5_flow.h    | 2 ++
 drivers/net/mlx5/mlx5_flow_hw.c | 4 +++-
 2 files changed, 5 insertions(+), 1 deletion(-)
drivers/net/mlx5/mlx5_flow.h +2 -0
index cd6fed4723..5ea3f22a2e 100644
@@ -1439,6 +1439,8 @@ struct mlx5_action_construct_data {
 			 * PRM actions.
 			 */
 			uint32_t mask[MLX5_ACT_MAX_MOD_FIELDS];
+			/* Copy of action passed to the action template. */
+			struct rte_flow_action_modify_field action;
 		} modify_header;
 		struct {
 			bool symmetric_hash_function; /* Symmetric RSS hash */
drivers/net/mlx5/mlx5_flow_hw.c +3 -1
index 5e9d43ab22..a04e8647cb 100644
@@ -883,6 +883,7 @@ __flow_hw_act_data_hdr_modify_append(struct mlx5_priv *priv,
 				     enum rte_flow_action_type type,
 				     uint16_t action_src,
 				     uint16_t action_dst,
+				     const struct rte_flow_action_modify_field *mf,
 				     uint16_t mhdr_cmds_off,
 				     uint16_t mhdr_cmds_end,
 				     bool shared,
@@ -895,6 +896,7 @@ __flow_hw_act_data_hdr_modify_append(struct mlx5_priv *priv,
 	act_data = __flow_hw_act_data_alloc(priv, type, action_src, action_dst);
 	if (!act_data)
 		return -1;
+	act_data->modify_header.action = *mf;
 	act_data->modify_header.mhdr_cmds_off = mhdr_cmds_off;
 	act_data->modify_header.mhdr_cmds_end = mhdr_cmds_end;
 	act_data->modify_header.shared = shared;
@@ -1372,7 +1374,7 @@ flow_hw_modify_field_compile(struct rte_eth_dev *dev,
 	if (shared)
 		return 0;
 	ret = __flow_hw_act_data_hdr_modify_append(priv, acts, RTE_FLOW_ACTION_TYPE_MODIFY_FIELD,
-						   src_pos, mhdr->pos,
+						   src_pos, mhdr->pos, conf,
 						   cmds_start, cmds_end, shared,
 						   field, dcopy, mask);
 	if (ret)
Signature · 1 line
2.39.2
#8 lore raw

PATCH 8/9 common/mlx5: add debug mode indicator

In reply to Dariusz Sosnowski

Add mlx5_fp_debug_enabled() function which:

- returns true if RTE_LIBRTE_MLX5_DEBUG is defined,
- returns false otherwise.

This allows for conditional execution of code meant to be executed only when mlx5 debug mode is enabled, without adding conditional compilation guards inside source code.
When mlx5 debug mode is disabled, any code running if mlx5_fp_debug_enabled() returns true will be removed by optimizing compiler due to dead code elimination.

---
 drivers/common/mlx5/mlx5_common.h | 13 +++++++++++++
 1 file changed, 13 insertions(+)
drivers/common/mlx5/mlx5_common.h +13 -0
index 14c70edbef..1abd1e8239 100644
@@ -109,6 +109,19 @@ pmd_drv_log_basename(const char *s)
 
 #endif /* RTE_LIBRTE_MLX5_DEBUG */
 
+/**
+ * Returns true if debug mode is enabled for fast path operations.
+ */
+static inline bool
+mlx5_fp_debug_enabled(void)
+{
+#ifdef RTE_LIBRTE_MLX5_DEBUG
+	return true;
+#else
+	return false;
+#endif
+}
+
 /* Allocate a buffer on the stack and fill it with a printf format string. */
 #define MKSTR(name, ...) \
 	int mkstr_size_##name = snprintf(NULL, 0, "" __VA_ARGS__); \
Signature · 1 line
2.39.2
#9 lore raw

PATCH 9/9 net/mlx5: add async flow operation validation

In reply to Dariusz Sosnowski

This patch adds validation to implementations of the following API functions:

- rte_flow_async_create()
- rte_flow_async_create_by_index()
- rte_flow_async_update()
- rte_flow_async_destroy()

These validations are enabled if and only if RTE_LIBRTE_MLX5_DEBUG macro is defined.

---
 doc/guides/rel_notes/release_24_07.rst |   1 +
 drivers/net/mlx5/mlx5_flow_hw.c        | 491 ++++++++++++++++++++++++-
 2 files changed, 488 insertions(+), 4 deletions(-)
doc/guides/rel_notes/release_24_07.rst +1 -0
index 3a3257fcd5..39b074c63c 100644
@@ -85,6 +85,7 @@ New Features
 
   * Added match with Tx queue.
   * Added match with external Tx queue.
+  * Added flow item and actions validation to async flow API.
 
 
 Removed Items
drivers/net/mlx5/mlx5_flow_hw.c +487 -4

527 lines not on this page · raw message

Signature · 1 line
2.39.2
#10 lore raw
In reply to Dariusz Sosnowski
Quoted 1 line

- Patches 2-7: Prepares mlx5 PMD for adding fast path validations.

Correction: Patches 2-8

Quoted 1 line

- Patch 8: Introduce validations to:

Correction: Patch 9

Best regards,
Dariusz Sosnowski

#11 lore raw

PATCH v2 1/1 ethdev: support duplicating only item mask

In reply to Dariusz Sosnowski

Extend rte_flow_conv() to support working only on item's mask. This allows drivers to get only the mask's size when working on pattern templates and duplicate items having only the mask in a generic way.

---
v2:
- Split series 32095 into separate ethdev and mlx5 series.
- Rebased on latest commit in next-net-main.

---
 lib/ethdev/rte_flow.c | 15 +++++++++++++--
 lib/ethdev/rte_flow.h | 12 ++++++++++++
 2 files changed, 25 insertions(+), 2 deletions(-)
lib/ethdev/rte_flow.c +13 -2
index 7ab1100ea0..ca2f85c3fa 100644
@@ -60,9 +60,10 @@ rte_flow_conv_copy(void *buf, const void *data, const size_t size,
 	bool rte_type = type >= 0;
 
 	size_t sz = rte_type ? desc[type].size : sizeof(void *);
-	if (buf == NULL || data == NULL)
+	if (data == NULL)
 		return 0;
-	rte_memcpy(buf, data, (size > sz ? sz : size));
+	if (buf != NULL)
+		rte_memcpy(buf, data, (size > sz ? sz : size));
 	if (rte_type && desc[type].desc_fn)
 		sz += desc[type].desc_fn(size > 0 ? buf : NULL, data);
 	return sz;
@@ -1090,6 +1091,7 @@ rte_flow_conv(enum rte_flow_conv_op op,
 
 	switch (op) {
 		const struct rte_flow_attr *attr;
+		const struct rte_flow_item *item;
 
 	case RTE_FLOW_CONV_OP_NONE:
 		ret = 0;
@@ -1104,6 +1106,15 @@ rte_flow_conv(enum rte_flow_conv_op op,
 	case RTE_FLOW_CONV_OP_ITEM:
 		ret = rte_flow_conv_pattern(dst, size, src, 1, error);
 		break;
+	case RTE_FLOW_CONV_OP_ITEM_MASK:
+		item = src;
+		if (item->mask == NULL) {
+			ret = rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM_MASK,
+						 item, "Mask not provided");
+			break;
+		}
+		ret = rte_flow_conv_item_spec(dst, size, src, RTE_FLOW_CONV_ITEM_MASK);
+		break;
 	case RTE_FLOW_CONV_OP_ACTION:
 		ret = rte_flow_conv_actions(dst, size, src, 1, error);
 		break;
lib/ethdev/rte_flow.h +12 -0
index fb84992b61..f864578f80 100644
@@ -4426,6 +4426,18 @@ enum rte_flow_conv_op {
 	 */
 	RTE_FLOW_CONV_OP_ITEM,
 
+	/**
+	 * Convert a single item mask.
+	 *
+	 * Duplicates only @p mask.
+	 *
+	 * - @p src type:
+	 *   @code const struct rte_flow_item * @endcode
+	 * - @p dst type:
+	 *   @code struct rte_flow_item * @endcode
+	 */
+	RTE_FLOW_CONV_OP_ITEM_MASK,
+
 	/**
 	 * Convert a single action.
 	 *

--
2.39.2

Applied
#12 lore raw

Re: [PATCH v2] ethdev: support duplicating only item mask

Quoted 7 lines · Dariusz Sosnowski

Extend rte_flow_conv() to support working only on item's mask. This allows drivers to get only the mask's size when working on pattern templates and duplicate items having only the mask in a generic way.

Applied to dpdk-next-net/main, thanks.

#13 lore raw

v2 cover net/mlx5: flow fast path validation

In reply to Dariusz Sosnowski

This patchset adds validations for flow items and actions supplied by the user to fast path async flow API functions implemented in mlx5 PMD.

- Patches 1-7: Prepares mlx5 PMD for adding fast path validations. Specifically:

- Required functionality for validating queue indexes and target represented ports are extracted for reuse.
- Additional info, required for validation, is stored in pattern and actions templates.
- Introduces mlx5_fp_debug_enabled() function for checking if mlx5 PMD was compiled in debug mode.

- Patch 8: Introduce validations to:

- rte_flow_async_create()
- rte_flow_async_create_by_index()
- rte_flow_async_actions_update()
- rte_flow_async_destroy()

These validations are enabled if and only if RTE_LIBRTE_MLX5_DEBUG macro is defined during compilation.

Depends-on: series-32150 ("ethdev: support duplicating only item mask")

v2:
- Split series 32095 into separate ethdev and mlx5 series.
- Rebased on latest commit in next-net-mlx-main.
- Removed Depends-on tags for already merged mlx5 patch series.

Dariusz Sosnowski (8):
net/mlx5: extract target port validation net/mlx5: extract queue index validation net/mlx5: store pattern template items net/mlx5: store original actions in template net/mlx5: store expected type on indirect action net/mlx5: store modify field action common/mlx5: add debug mode indicator net/mlx5: add async flow operation validation

 doc/guides/rel_notes/release_24_07.rst |   1 +
 drivers/common/mlx5/mlx5_common.h      |  13 +
 drivers/net/mlx5/mlx5_flow.c           |  59 ++-
 drivers/net/mlx5/mlx5_flow.h           |  12 +
 drivers/net/mlx5/mlx5_flow_hw.c        | 620 +++++++++++++++++++++++--
 5 files changed, 657 insertions(+), 48 deletions(-)

--
2.39.2

#14 lore raw

PATCH v2 1/8 net/mlx5: extract target port validation

Extract code responsible for validation if port specified in configuration of RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT action is correct. Allow for reuse of this logic for both RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT actions and RTE_FLOW_ITEM_TYPE_REPRESENTED_PORT items.

---
 drivers/net/mlx5/mlx5_flow_hw.c | 63 +++++++++++++++++++++------------
 1 file changed, 40 insertions(+), 23 deletions(-)
drivers/net/mlx5/mlx5_flow_hw.c +40 -23
index eb89dcf454..aee0201cbc 100644
@@ -6046,6 +6046,42 @@ flow_hw_validate_action_port_representor(struct rte_eth_dev *dev __rte_unused,
 	return 0;
 }
 
+static int
+flow_hw_validate_target_port_id(struct rte_eth_dev *dev,
+				uint16_t target_port_id)
+{
+	struct mlx5_priv *port_priv;
+	struct mlx5_priv *dev_priv;
+
+	if (target_port_id == MLX5_REPRESENTED_PORT_ESW_MGR)
+		return 0;
+
+	port_priv = mlx5_port_to_eswitch_info(target_port_id, false);
+	if (!port_priv) {
+		rte_errno = EINVAL;
+		DRV_LOG(ERR, "Port %u Failed to obtain E-Switch info for port %u",
+			dev->data->port_id, target_port_id);
+		return -rte_errno;
+	}
+
+	dev_priv = mlx5_dev_to_eswitch_info(dev);
+	if (!dev_priv) {
+		rte_errno = EINVAL;
+		DRV_LOG(ERR, "Port %u Failed to obtain E-Switch info for transfer proxy",
+			dev->data->port_id);
+		return -rte_errno;
+	}
+
+	if (port_priv->domain_id != dev_priv->domain_id) {
+		rte_errno = EINVAL;
+		DRV_LOG(ERR, "Port %u Failed to obtain E-Switch info for transfer proxy",
+			dev->data->port_id);
+		return -rte_errno;
+	}
+
+	return 0;
+}
+
 static int
 flow_hw_validate_action_represented_port(struct rte_eth_dev *dev,
 					 const struct rte_flow_action *action,
@@ -6062,32 +6098,13 @@ flow_hw_validate_action_represented_port(struct rte_eth_dev *dev,
 					  "cannot use represented_port actions"
 					  " without an E-Switch");
 	if (mask_conf && mask_conf->port_id) {
-		struct mlx5_priv *port_priv;
-		struct mlx5_priv *dev_priv;
-
 		if (!action_conf)
 			return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
 						  action, "port index was not provided");
-		port_priv = mlx5_port_to_eswitch_info(action_conf->port_id, false);
-		if (!port_priv)
-			return rte_flow_error_set(error, rte_errno,
-						  RTE_FLOW_ERROR_TYPE_ACTION,
-						  action,
-						  "failed to obtain E-Switch"
-						  " info for port");
-		dev_priv = mlx5_dev_to_eswitch_info(dev);
-		if (!dev_priv)
-			return rte_flow_error_set(error, rte_errno,
-						  RTE_FLOW_ERROR_TYPE_ACTION,
-						  action,
-						  "failed to obtain E-Switch"
-						  " info for transfer proxy");
-		if (port_priv->domain_id != dev_priv->domain_id)
-			return rte_flow_error_set(error, rte_errno,
-						  RTE_FLOW_ERROR_TYPE_ACTION,
-						  action,
-						  "cannot forward to port from"
-						  " a different E-Switch");
+
+		if (flow_hw_validate_target_port_id(dev, action_conf->port_id))
+			return rte_flow_error_set(error, rte_errno, RTE_FLOW_ERROR_TYPE_ACTION,
+						  action, "port index is invalid");
 	}
 	return 0;
 }
Signature · 1 line
2.39.2
#15 lore raw

PATCH v2 2/8 net/mlx5: extract queue index validation

In reply to Dariusz Sosnowski

Extract validation of queue index out of validation of RTE_FLOW_ACTION_TYPE_QUEUE action to allow reuse in template API flow rule creation implementation.

---
 drivers/net/mlx5/mlx5_flow.c | 59 +++++++++++++++++++++++++-----------
 drivers/net/mlx5/mlx5_flow.h |  3 ++
 2 files changed, 44 insertions(+), 18 deletions(-)
drivers/net/mlx5/mlx5_flow.c +41 -18
index 7bcbbc74b5..5374c95954 100644
@@ -2020,6 +2020,46 @@ mlx5_flow_validate_action_drop(struct rte_eth_dev *dev,
 	return 0;
 }
 
+/*
+ * Check if a queue specified in the queue action is valid.
+ *
+ * @param[in] dev
+ *   Pointer to the Ethernet device structure.
+ * @param[in] action
+ *   Pointer to the queue action.
+ * @param[out] error
+ *   Pointer to error structure.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_flow_validate_target_queue(struct rte_eth_dev *dev,
+				const struct rte_flow_action *action,
+				struct rte_flow_error *error)
+{
+	const struct rte_flow_action_queue *queue = action->conf;
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	if (mlx5_is_external_rxq(dev, queue->index))
+		return 0;
+	if (!priv->rxqs_n)
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  NULL, "No Rx queues configured");
+	if (queue->index >= priv->rxqs_n)
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &queue->index,
+					  "queue index out of range");
+	if (mlx5_rxq_get(dev, queue->index) == NULL)
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &queue->index,
+					  "queue is not configured");
+	return 0;
+}
+
 /*
  * Validate the queue action.
  *
@@ -2044,7 +2084,6 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 				const struct rte_flow_attr *attr,
 				struct rte_flow_error *error)
 {
-	struct mlx5_priv *priv = dev->data->dev_private;
 	const struct rte_flow_action_queue *queue = action->conf;
 
 	if (!queue)
@@ -2060,23 +2099,7 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 		return rte_flow_error_set(error, ENOTSUP,
 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 					  "queue action not supported for egress.");
-	if (mlx5_is_external_rxq(dev, queue->index))
-		return 0;
-	if (!priv->rxqs_n)
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  NULL, "No Rx queues configured");
-	if (queue->index >= priv->rxqs_n)
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  &queue->index,
-					  "queue index out of range");
-	if (mlx5_rxq_get(dev, queue->index) == NULL)
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  &queue->index,
-					  "queue is not configured");
-	return 0;
+	return mlx5_flow_validate_target_queue(dev, action, error);
 }
 
 /**
drivers/net/mlx5/mlx5_flow.h +3 -0
index 92e2ecedb3..da7d06d033 100644
@@ -3018,6 +3018,9 @@ int mlx5_flow_validate_action_mark(struct rte_eth_dev *dev,
 				   uint64_t action_flags,
 				   const struct rte_flow_attr *attr,
 				   struct rte_flow_error *error);
+int mlx5_flow_validate_target_queue(struct rte_eth_dev *dev,
+				    const struct rte_flow_action *action,
+				    struct rte_flow_error *error);
 int mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 				    uint64_t action_flags,
 				    struct rte_eth_dev *dev,
Signature · 1 line
2.39.2
#16 lore raw

PATCH v2 3/8 net/mlx5: store pattern template items

In reply to Dariusz Sosnowski

Store items which define the pattern template on creation. This allows validation of items, provided by the user during flow rule creation, against pattern template.

---
 drivers/net/mlx5/mlx5_flow.h    |  2 ++
 drivers/net/mlx5/mlx5_flow_hw.c | 29 +++++++++++++++++++++++++++++
 2 files changed, 31 insertions(+)
drivers/net/mlx5/mlx5_flow.h +2 -0
index da7d06d033..5c2cc2b7c1 100644
@@ -1522,6 +1522,8 @@ struct rte_flow_pattern_template {
 	/* Manages all GENEVE TLV options used by this pattern template. */
 	struct mlx5_geneve_tlv_options_mng geneve_opt_mng;
 	uint8_t flex_item; /* flex item index. */
+	/* Items on which this pattern template is based on. */
+	struct rte_flow_item *items;
 };
 
 /* Flow action template struct. */
drivers/net/mlx5/mlx5_flow_hw.c +29 -0
index aee0201cbc..0ddae54ed2 100644
@@ -8767,6 +8767,7 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 		.mask = &tag_m,
 		.last = NULL
 	};
+	int it_items_size;
 	unsigned int i = 0;
 	int rc;
 
@@ -8810,6 +8811,31 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 	it->attr = *attr;
 	it->item_flags = item_flags;
 	it->orig_item_nb = orig_item_nb;
+	it_items_size = rte_flow_conv(RTE_FLOW_CONV_OP_PATTERN, NULL, 0, tmpl_items, error);
+	if (it_items_size <= 0) {
+		rte_flow_error_set(error, ENOMEM,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+				   NULL,
+				   "Failed to determine buffer size for pattern");
+		goto error;
+	}
+	it_items_size = RTE_ALIGN(it_items_size, 16);
+	it->items = mlx5_malloc(MLX5_MEM_ZERO, it_items_size, 0, rte_dev_numa_node(dev->device));
+	if (it->items == NULL) {
+		rte_flow_error_set(error, ENOMEM,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+				   NULL,
+				   "Cannot allocate memory for pattern");
+		goto error;
+	}
+	rc = rte_flow_conv(RTE_FLOW_CONV_OP_PATTERN, it->items, it_items_size, tmpl_items, error);
+	if (rc <= 0) {
+		rte_flow_error_set(error, ENOMEM,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+				   NULL,
+				   "Failed to store pattern");
+		goto error;
+	}
 	it->mt = mlx5dr_match_template_create(tmpl_items, attr->relaxed_matching);
 	if (!it->mt) {
 		rte_flow_error_set(error, rte_errno,
@@ -8824,6 +8850,7 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 		else if (attr->egress)
 			it->implicit_tag = true;
 		mlx5_free(copied_items);
+		copied_items = NULL;
 	}
 	/* Either inner or outer, can't both. */
 	if (it->item_flags & (MLX5_FLOW_ITEM_OUTER_IPV6_ROUTING_EXT |
@@ -8884,6 +8911,7 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 			mlx5_geneve_tlv_options_unregister(priv, &it->geneve_opt_mng);
 		if (it->mt)
 			claim_zero(mlx5dr_match_template_destroy(it->mt));
+		mlx5_free(it->items);
 		mlx5_free(it);
 	}
 	if (copied_items)
@@ -8926,6 +8954,7 @@ flow_hw_pattern_template_destroy(struct rte_eth_dev *dev,
 	flow_hw_flex_item_release(dev, &template->flex_item);
 	mlx5_geneve_tlv_options_unregister(priv, &template->geneve_opt_mng);
 	claim_zero(mlx5dr_match_template_destroy(template->mt));
+	mlx5_free(template->items);
 	mlx5_free(template);
 	return 0;
 }
Signature · 1 line
2.39.2
#17 lore raw

PATCH v2 4/8 net/mlx5: store original actions in template

In reply to Dariusz Sosnowski

When actions template is created in mlx5 PMD, some actions will be replaced with another or some actions will be added to implement required template semantics on NVIDIA NICs. For example:

- RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID actions is replaced with modify field action.
- Modify field action is added for preserving metadata between FDB and NIC domains.
- Modify field action is added when quota action is used to amend ASO syndrome.

Types of actions and their order in flow rules based on some actions template must match the original, not modified one.

This patch adds preserving of the original actions for actions template to allow for easier validation of ordering and types on fast path.

---
 drivers/net/mlx5/mlx5_flow.h    |  1 +
 drivers/net/mlx5/mlx5_flow_hw.c | 11 +++++++++++
 2 files changed, 12 insertions(+)
drivers/net/mlx5/mlx5_flow.h +1 -0
index 5c2cc2b7c1..8e99e76e5d 100644
@@ -1532,6 +1532,7 @@ struct rte_flow_actions_template {
 	/* Template attributes. */
 	struct rte_flow_actions_template_attr attr;
 	struct rte_flow_action *actions; /* Cached flow actions. */
+	struct rte_flow_action *orig_actions; /* Original flow actions. */
 	struct rte_flow_action *masks; /* Cached action masks.*/
 	struct mlx5dr_action_template *tmpl; /* mlx5dr action template. */
 	uint64_t action_flags; /* Bit-map of all valid action in template. */
drivers/net/mlx5/mlx5_flow_hw.c +11 -0
index 0ddae54ed2..ddc9cf24d6 100644
@@ -7745,6 +7745,7 @@ __flow_hw_actions_template_create(struct rte_eth_dev *dev,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	int len, act_len, mask_len;
+	int orig_act_len;
 	unsigned int act_num;
 	unsigned int i;
 	struct rte_flow_actions_template *at = NULL;
@@ -7862,6 +7863,10 @@ __flow_hw_actions_template_create(struct rte_eth_dev *dev,
 	len += RTE_ALIGN(mask_len, 16);
 	len += RTE_ALIGN(act_num * sizeof(*at->dr_off), 16);
 	len += RTE_ALIGN(act_num * sizeof(*at->src_off), 16);
+	orig_act_len = rte_flow_conv(RTE_FLOW_CONV_OP_ACTIONS, NULL, 0, actions, error);
+	if (orig_act_len <= 0)
+		return NULL;
+	len += RTE_ALIGN(orig_act_len, 16);
 	at = mlx5_malloc(MLX5_MEM_ZERO, len + sizeof(*at),
 			 RTE_CACHE_LINE_SIZE, rte_socket_id());
 	if (!at) {
@@ -7889,6 +7894,12 @@ __flow_hw_actions_template_create(struct rte_eth_dev *dev,
 	at->src_off = RTE_PTR_ADD(at->dr_off,
 				  RTE_ALIGN(act_num * sizeof(*at->dr_off), 16));
 	memcpy(at->src_off, src_off, act_num * sizeof(at->src_off[0]));
+	at->orig_actions = RTE_PTR_ADD(at->src_off,
+				       RTE_ALIGN(act_num * sizeof(*at->src_off), 16));
+	orig_act_len = rte_flow_conv(RTE_FLOW_CONV_OP_ACTIONS, at->orig_actions, orig_act_len,
+				     actions, error);
+	if (orig_act_len <= 0)
+		goto error;
 	at->actions_num = act_num;
 	for (i = 0; i < at->actions_num; ++i)
 		at->dr_off[i] = UINT16_MAX;
Signature · 1 line
2.39.2
#18 lore raw

PATCH v2 5/8 net/mlx5: store expected type on indirect action

In reply to Dariusz Sosnowski

When template table is created, list of unmasked actions is recorded for future flow rule insertions.
This patch expands entries for RTE_FLOW_ACTION_TYPE_INDIRECT actions in this list with information about expected indirect action type. This will be used in follow up commits which add flow rule operation validation. Specifically, this will be used to verify if indirect action provided by the user references an action of a correct type.

---
 drivers/net/mlx5/mlx5_flow.h    |  4 ++++
 drivers/net/mlx5/mlx5_flow_hw.c | 22 ++++++++++++++++++++--
 2 files changed, 24 insertions(+), 2 deletions(-)
drivers/net/mlx5/mlx5_flow.h +4 -0
index 8e99e76e5d..33847e2272 100644
@@ -1442,6 +1442,10 @@ struct mlx5_action_construct_data {
 	uint16_t action_dst; /* mlx5dr_rule_action dst offset. */
 	indirect_list_callback_t indirect_list_cb;
 	union {
+		struct {
+			/* Expected type of indirection action. */
+			enum rte_flow_action_type expected_type;
+		} indirect;
 		struct {
 			/* encap data len. */
 			uint16_t len;
drivers/net/mlx5/mlx5_flow_hw.c +20 -2
index ddc9cf24d6..21a885517a 100644
@@ -997,6 +997,24 @@ __flow_hw_act_data_general_append(struct mlx5_priv *priv,
 	return 0;
 }
 
+static __rte_always_inline int
+__flow_hw_act_data_indirect_append(struct mlx5_priv *priv,
+				   struct mlx5_hw_actions *acts,
+				   enum rte_flow_action_type type,
+				   enum rte_flow_action_type mask_type,
+				   uint16_t action_src,
+				   uint16_t action_dst)
+{
+	struct mlx5_action_construct_data *act_data;
+
+	act_data = __flow_hw_act_data_alloc(priv, type, action_src, action_dst);
+	if (!act_data)
+		return -1;
+	act_data->indirect.expected_type = mask_type;
+	LIST_INSERT_HEAD(&acts->act_list, act_data, next);
+	return 0;
+}
+
 static __rte_always_inline int
 flow_hw_act_data_indirect_list_append(struct mlx5_priv *priv,
 				      struct mlx5_hw_actions *acts,
@@ -2482,9 +2500,9 @@ __flow_hw_translate_actions_template(struct rte_eth_dev *dev,
 				if (flow_hw_shared_action_translate
 				(dev, actions, acts, src_pos, dr_pos))
 					goto err;
-			} else if (__flow_hw_act_data_general_append
+			} else if (__flow_hw_act_data_indirect_append
 					(priv, acts, RTE_FLOW_ACTION_TYPE_INDIRECT,
-					 src_pos, dr_pos)){
+					 masks->type, src_pos, dr_pos)){
 				goto err;
 			}
 			break;
Signature · 1 line
2.39.2
#19 lore raw

PATCH v2 6/8 net/mlx5: store modify field action

In reply to Dariusz Sosnowski

When template table is created, list of unmasked actions is recorded for future flow rule insertions.
This patch expands entries for RTE_FLOW_ACTION_TYPE_MODIFY_FIELD actions in this list with a copy of the action from the template. This will be used in follow up commits which add flow rule operation validation. Specifically, to validate that RTE_FLOW_ACTION_TYPE_MODIFY_FIELD action passed by the user is correctly configured.

---
 drivers/net/mlx5/mlx5_flow.h    | 2 ++
 drivers/net/mlx5/mlx5_flow_hw.c | 4 +++-
 2 files changed, 5 insertions(+), 1 deletion(-)
drivers/net/mlx5/mlx5_flow.h +2 -0
index 33847e2272..6974d4e075 100644
@@ -1472,6 +1472,8 @@ struct mlx5_action_construct_data {
 			 * PRM actions.
 			 */
 			uint32_t mask[MLX5_ACT_MAX_MOD_FIELDS];
+			/* Copy of action passed to the action template. */
+			struct rte_flow_action_modify_field action;
 		} modify_header;
 		struct {
 			bool symmetric_hash_function; /* Symmetric RSS hash */
drivers/net/mlx5/mlx5_flow_hw.c +3 -1
index 21a885517a..19d6105be8 100644
@@ -1112,6 +1112,7 @@ __flow_hw_act_data_hdr_modify_append(struct mlx5_priv *priv,
 				     enum rte_flow_action_type type,
 				     uint16_t action_src,
 				     uint16_t action_dst,
+				     const struct rte_flow_action_modify_field *mf,
 				     uint16_t mhdr_cmds_off,
 				     uint16_t mhdr_cmds_end,
 				     bool shared,
@@ -1124,6 +1125,7 @@ __flow_hw_act_data_hdr_modify_append(struct mlx5_priv *priv,
 	act_data = __flow_hw_act_data_alloc(priv, type, action_src, action_dst);
 	if (!act_data)
 		return -1;
+	act_data->modify_header.action = *mf;
 	act_data->modify_header.mhdr_cmds_off = mhdr_cmds_off;
 	act_data->modify_header.mhdr_cmds_end = mhdr_cmds_end;
 	act_data->modify_header.shared = shared;
@@ -1601,7 +1603,7 @@ flow_hw_modify_field_compile(struct rte_eth_dev *dev,
 	if (shared)
 		return 0;
 	ret = __flow_hw_act_data_hdr_modify_append(priv, acts, RTE_FLOW_ACTION_TYPE_MODIFY_FIELD,
-						   src_pos, mhdr->pos,
+						   src_pos, mhdr->pos, conf,
 						   cmds_start, cmds_end, shared,
 						   field, dcopy, mask);
 	if (ret)
Signature · 1 line
2.39.2
#20 lore raw

PATCH v2 7/8 common/mlx5: add debug mode indicator

In reply to Dariusz Sosnowski

Add mlx5_fp_debug_enabled() function which:

- returns true if RTE_LIBRTE_MLX5_DEBUG is defined,
- returns false otherwise.

This allows for conditional execution of code meant to be executed only when mlx5 debug mode is enabled, without adding conditional compilation guards inside source code.
When mlx5 debug mode is disabled, any code running if mlx5_fp_debug_enabled() returns true will be removed by optimizing compiler due to dead code elimination.

---
 drivers/common/mlx5/mlx5_common.h | 13 +++++++++++++
 1 file changed, 13 insertions(+)
drivers/common/mlx5/mlx5_common.h +13 -0
index 14c70edbef..1abd1e8239 100644
@@ -109,6 +109,19 @@ pmd_drv_log_basename(const char *s)
 
 #endif /* RTE_LIBRTE_MLX5_DEBUG */
 
+/**
+ * Returns true if debug mode is enabled for fast path operations.
+ */
+static inline bool
+mlx5_fp_debug_enabled(void)
+{
+#ifdef RTE_LIBRTE_MLX5_DEBUG
+	return true;
+#else
+	return false;
+#endif
+}
+
 /* Allocate a buffer on the stack and fill it with a printf format string. */
 #define MKSTR(name, ...) \
 	int mkstr_size_##name = snprintf(NULL, 0, "" __VA_ARGS__); \
Signature · 1 line
2.39.2
#21 lore raw

PATCH v2 8/8 net/mlx5: add async flow operation validation

In reply to Dariusz Sosnowski

This patch adds validation to implementations of the following API functions:

- rte_flow_async_create()
- rte_flow_async_create_by_index()
- rte_flow_async_update()
- rte_flow_async_destroy()

These validations are enabled if and only if RTE_LIBRTE_MLX5_DEBUG macro is defined.

---
 doc/guides/rel_notes/release_24_07.rst |   1 +
 drivers/net/mlx5/mlx5_flow_hw.c        | 491 ++++++++++++++++++++++++-
 2 files changed, 488 insertions(+), 4 deletions(-)
doc/guides/rel_notes/release_24_07.rst +1 -0
index 7688ed2764..5f37e2283c 100644
@@ -86,6 +86,7 @@ New Features
   * Added match with Tx queue.
   * Added match with external Tx queue.
   * Added match with E-Switch manager.
+  * Added flow item and actions validation to async flow API.
 
 
 Removed Items
drivers/net/mlx5/mlx5_flow_hw.c +487 -4

527 lines not on this page · raw message

Signature · 1 line
2.39.2
#22 lore raw

v3 cover net/mlx5: flow fast path validation

In reply to Dariusz Sosnowski

This patchset adds validations for flow items and actions supplied by the user to fast path async flow API functions implemented in mlx5 PMD.

- Patches 1-7: Prepares mlx5 PMD for adding fast path validations. Specifically:

- Required functionality for validating queue indexes and target represented ports are extracted for reuse.
- Additional info, required for validation, is stored in pattern and actions templates.
- Introduces mlx5_fp_debug_enabled() function for checking if mlx5 PMD was compiled in debug mode.

- Patch 8: Introduce validations to:

- rte_flow_async_create()
- rte_flow_async_create_by_index()
- rte_flow_async_actions_update()
- rte_flow_async_destroy()

These validations are enabled if and only if RTE_LIBRTE_MLX5_DEBUG macro is defined during compilation.

v3:
- Rebased on 24.07-rc1.
- Removed Depends-on tag for already marged ethdev patch.

v2:
- Split series 32095 into separate ethdev and mlx5 series.
- Rebased on latest commit in next-net-mlx-main.
- Removed Depends-on tags for already merged mlx5 patch series.

Dariusz Sosnowski (8):
net/mlx5: extract target port validation net/mlx5: extract queue index validation net/mlx5: store pattern template items net/mlx5: store original actions in template net/mlx5: store expected type on indirect action net/mlx5: store modify field action common/mlx5: add debug mode indicator net/mlx5: add async flow operation validation

 doc/guides/rel_notes/release_24_07.rst |   1 +
 drivers/common/mlx5/mlx5_common.h      |  13 +
 drivers/net/mlx5/mlx5_flow.c           |  59 ++-
 drivers/net/mlx5/mlx5_flow.h           |  12 +
 drivers/net/mlx5/mlx5_flow_hw.c        | 620 +++++++++++++++++++++++--
 5 files changed, 657 insertions(+), 48 deletions(-)

--
2.39.2

#23 lore raw

PATCH v3 1/8 net/mlx5: extract target port validation

Extract code responsible for validation if port specified in configuration of RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT action is correct. Allow for reuse of this logic for both RTE_FLOW_ACTION_TYPE_REPRESENTED_PORT actions and RTE_FLOW_ITEM_TYPE_REPRESENTED_PORT items.

---
 drivers/net/mlx5/mlx5_flow_hw.c | 63 +++++++++++++++++++++------------
 1 file changed, 40 insertions(+), 23 deletions(-)
drivers/net/mlx5/mlx5_flow_hw.c +40 -23
index b8d71b145d..c0d763198d 100644
@@ -6051,6 +6051,42 @@ flow_hw_validate_action_port_representor(struct rte_eth_dev *dev __rte_unused,
 	return 0;
 }
 
+static int
+flow_hw_validate_target_port_id(struct rte_eth_dev *dev,
+				uint16_t target_port_id)
+{
+	struct mlx5_priv *port_priv;
+	struct mlx5_priv *dev_priv;
+
+	if (target_port_id == MLX5_REPRESENTED_PORT_ESW_MGR)
+		return 0;
+
+	port_priv = mlx5_port_to_eswitch_info(target_port_id, false);
+	if (!port_priv) {
+		rte_errno = EINVAL;
+		DRV_LOG(ERR, "Port %u Failed to obtain E-Switch info for port %u",
+			dev->data->port_id, target_port_id);
+		return -rte_errno;
+	}
+
+	dev_priv = mlx5_dev_to_eswitch_info(dev);
+	if (!dev_priv) {
+		rte_errno = EINVAL;
+		DRV_LOG(ERR, "Port %u Failed to obtain E-Switch info for transfer proxy",
+			dev->data->port_id);
+		return -rte_errno;
+	}
+
+	if (port_priv->domain_id != dev_priv->domain_id) {
+		rte_errno = EINVAL;
+		DRV_LOG(ERR, "Port %u Failed to obtain E-Switch info for transfer proxy",
+			dev->data->port_id);
+		return -rte_errno;
+	}
+
+	return 0;
+}
+
 static int
 flow_hw_validate_action_represented_port(struct rte_eth_dev *dev,
 					 const struct rte_flow_action *action,
@@ -6067,32 +6103,13 @@ flow_hw_validate_action_represented_port(struct rte_eth_dev *dev,
 					  "cannot use represented_port actions"
 					  " without an E-Switch");
 	if (mask_conf && mask_conf->port_id) {
-		struct mlx5_priv *port_priv;
-		struct mlx5_priv *dev_priv;
-
 		if (!action_conf)
 			return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION,
 						  action, "port index was not provided");
-		port_priv = mlx5_port_to_eswitch_info(action_conf->port_id, false);
-		if (!port_priv)
-			return rte_flow_error_set(error, rte_errno,
-						  RTE_FLOW_ERROR_TYPE_ACTION,
-						  action,
-						  "failed to obtain E-Switch"
-						  " info for port");
-		dev_priv = mlx5_dev_to_eswitch_info(dev);
-		if (!dev_priv)
-			return rte_flow_error_set(error, rte_errno,
-						  RTE_FLOW_ERROR_TYPE_ACTION,
-						  action,
-						  "failed to obtain E-Switch"
-						  " info for transfer proxy");
-		if (port_priv->domain_id != dev_priv->domain_id)
-			return rte_flow_error_set(error, rte_errno,
-						  RTE_FLOW_ERROR_TYPE_ACTION,
-						  action,
-						  "cannot forward to port from"
-						  " a different E-Switch");
+
+		if (flow_hw_validate_target_port_id(dev, action_conf->port_id))
+			return rte_flow_error_set(error, rte_errno, RTE_FLOW_ERROR_TYPE_ACTION,
+						  action, "port index is invalid");
 	}
 	return 0;
 }
Signature · 1 line
2.39.2
#24 lore raw

PATCH v3 2/8 net/mlx5: extract queue index validation

In reply to Dariusz Sosnowski

Extract validation of queue index out of validation of RTE_FLOW_ACTION_TYPE_QUEUE action to allow reuse in template API flow rule creation implementation.

---
 drivers/net/mlx5/mlx5_flow.c | 59 +++++++++++++++++++++++++-----------
 drivers/net/mlx5/mlx5_flow.h |  3 ++
 2 files changed, 44 insertions(+), 18 deletions(-)
drivers/net/mlx5/mlx5_flow.c +41 -18
index 3a59fb525a..daf0e41c6f 100644
@@ -2020,6 +2020,46 @@ mlx5_flow_validate_action_drop(struct rte_eth_dev *dev,
 	return 0;
 }
 
+/*
+ * Check if a queue specified in the queue action is valid.
+ *
+ * @param[in] dev
+ *   Pointer to the Ethernet device structure.
+ * @param[in] action
+ *   Pointer to the queue action.
+ * @param[out] error
+ *   Pointer to error structure.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+int
+mlx5_flow_validate_target_queue(struct rte_eth_dev *dev,
+				const struct rte_flow_action *action,
+				struct rte_flow_error *error)
+{
+	const struct rte_flow_action_queue *queue = action->conf;
+	struct mlx5_priv *priv = dev->data->dev_private;
+
+	if (mlx5_is_external_rxq(dev, queue->index))
+		return 0;
+	if (!priv->rxqs_n)
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  NULL, "No Rx queues configured");
+	if (queue->index >= priv->rxqs_n)
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &queue->index,
+					  "queue index out of range");
+	if (mlx5_rxq_get(dev, queue->index) == NULL)
+		return rte_flow_error_set(error, EINVAL,
+					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+					  &queue->index,
+					  "queue is not configured");
+	return 0;
+}
+
 /*
  * Validate the queue action.
  *
@@ -2044,7 +2084,6 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 				const struct rte_flow_attr *attr,
 				struct rte_flow_error *error)
 {
-	struct mlx5_priv *priv = dev->data->dev_private;
 	const struct rte_flow_action_queue *queue = action->conf;
 
 	if (!queue)
@@ -2060,23 +2099,7 @@ mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 		return rte_flow_error_set(error, ENOTSUP,
 					  RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
 					  "queue action not supported for egress.");
-	if (mlx5_is_external_rxq(dev, queue->index))
-		return 0;
-	if (!priv->rxqs_n)
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  NULL, "No Rx queues configured");
-	if (queue->index >= priv->rxqs_n)
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  &queue->index,
-					  "queue index out of range");
-	if (mlx5_rxq_get(dev, queue->index) == NULL)
-		return rte_flow_error_set(error, EINVAL,
-					  RTE_FLOW_ERROR_TYPE_ACTION_CONF,
-					  &queue->index,
-					  "queue is not configured");
-	return 0;
+	return mlx5_flow_validate_target_queue(dev, action, error);
 }
 
 /**
drivers/net/mlx5/mlx5_flow.h +3 -0
index 92e2ecedb3..da7d06d033 100644
@@ -3018,6 +3018,9 @@ int mlx5_flow_validate_action_mark(struct rte_eth_dev *dev,
 				   uint64_t action_flags,
 				   const struct rte_flow_attr *attr,
 				   struct rte_flow_error *error);
+int mlx5_flow_validate_target_queue(struct rte_eth_dev *dev,
+				    const struct rte_flow_action *action,
+				    struct rte_flow_error *error);
 int mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
 				    uint64_t action_flags,
 				    struct rte_eth_dev *dev,
Signature · 1 line
2.39.2
#25 lore raw

PATCH v3 3/8 net/mlx5: store pattern template items

In reply to Dariusz Sosnowski

Store items which define the pattern template on creation. This allows validation of items, provided by the user during flow rule creation, against pattern template.

---
 drivers/net/mlx5/mlx5_flow.h    |  2 ++
 drivers/net/mlx5/mlx5_flow_hw.c | 29 +++++++++++++++++++++++++++++
 2 files changed, 31 insertions(+)
drivers/net/mlx5/mlx5_flow.h +2 -0
index da7d06d033..5c2cc2b7c1 100644
@@ -1522,6 +1522,8 @@ struct rte_flow_pattern_template {
 	/* Manages all GENEVE TLV options used by this pattern template. */
 	struct mlx5_geneve_tlv_options_mng geneve_opt_mng;
 	uint8_t flex_item; /* flex item index. */
+	/* Items on which this pattern template is based on. */
+	struct rte_flow_item *items;
 };
 
 /* Flow action template struct. */
drivers/net/mlx5/mlx5_flow_hw.c +29 -0
index c0d763198d..1ad261d529 100644
@@ -8772,6 +8772,7 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 		.mask = &tag_m,
 		.last = NULL
 	};
+	int it_items_size;
 	unsigned int i = 0;
 	int rc;
 
@@ -8815,6 +8816,31 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 	it->attr = *attr;
 	it->item_flags = item_flags;
 	it->orig_item_nb = orig_item_nb;
+	it_items_size = rte_flow_conv(RTE_FLOW_CONV_OP_PATTERN, NULL, 0, tmpl_items, error);
+	if (it_items_size <= 0) {
+		rte_flow_error_set(error, ENOMEM,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+				   NULL,
+				   "Failed to determine buffer size for pattern");
+		goto error;
+	}
+	it_items_size = RTE_ALIGN(it_items_size, 16);
+	it->items = mlx5_malloc(MLX5_MEM_ZERO, it_items_size, 0, rte_dev_numa_node(dev->device));
+	if (it->items == NULL) {
+		rte_flow_error_set(error, ENOMEM,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+				   NULL,
+				   "Cannot allocate memory for pattern");
+		goto error;
+	}
+	rc = rte_flow_conv(RTE_FLOW_CONV_OP_PATTERN, it->items, it_items_size, tmpl_items, error);
+	if (rc <= 0) {
+		rte_flow_error_set(error, ENOMEM,
+				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
+				   NULL,
+				   "Failed to store pattern");
+		goto error;
+	}
 	it->mt = mlx5dr_match_template_create(tmpl_items, attr->relaxed_matching);
 	if (!it->mt) {
 		rte_flow_error_set(error, rte_errno,
@@ -8829,6 +8855,7 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 		else if (attr->egress)
 			it->implicit_tag = true;
 		mlx5_free(copied_items);
+		copied_items = NULL;
 	}
 	/* Either inner or outer, can't both. */
 	if (it->item_flags & (MLX5_FLOW_ITEM_OUTER_IPV6_ROUTING_EXT |
@@ -8889,6 +8916,7 @@ flow_hw_pattern_template_create(struct rte_eth_dev *dev,
 			mlx5_geneve_tlv_options_unregister(priv, &it->geneve_opt_mng);
 		if (it->mt)
 			claim_zero(mlx5dr_match_template_destroy(it->mt));
+		mlx5_free(it->items);
 		mlx5_free(it);
 	}
 	if (copied_items)
@@ -8931,6 +8959,7 @@ flow_hw_pattern_template_destroy(struct rte_eth_dev *dev,
 	flow_hw_flex_item_release(dev, &template->flex_item);
 	mlx5_geneve_tlv_options_unregister(priv, &template->geneve_opt_mng);
 	claim_zero(mlx5dr_match_template_destroy(template->mt));
+	mlx5_free(template->items);
 	mlx5_free(template);
 	return 0;
 }
Signature · 1 line
2.39.2
#26 lore raw

PATCH v3 4/8 net/mlx5: store original actions in template

In reply to Dariusz Sosnowski

When actions template is created in mlx5 PMD, some actions will be replaced with another or some actions will be added to implement required template semantics on NVIDIA NICs. For example:

- RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID actions is replaced with modify field action.
- Modify field action is added for preserving metadata between FDB and NIC domains.
- Modify field action is added when quota action is used to amend ASO syndrome.

Types of actions and their order in flow rules based on some actions template must match the original, not modified one.

This patch adds preserving of the original actions for actions template to allow for easier validation of ordering and types on fast path.

---
 drivers/net/mlx5/mlx5_flow.h    |  1 +
 drivers/net/mlx5/mlx5_flow_hw.c | 11 +++++++++++
 2 files changed, 12 insertions(+)
drivers/net/mlx5/mlx5_flow.h +1 -0
index 5c2cc2b7c1..8e99e76e5d 100644
@@ -1532,6 +1532,7 @@ struct rte_flow_actions_template {
 	/* Template attributes. */
 	struct rte_flow_actions_template_attr attr;
 	struct rte_flow_action *actions; /* Cached flow actions. */
+	struct rte_flow_action *orig_actions; /* Original flow actions. */
 	struct rte_flow_action *masks; /* Cached action masks.*/
 	struct mlx5dr_action_template *tmpl; /* mlx5dr action template. */
 	uint64_t action_flags; /* Bit-map of all valid action in template. */
drivers/net/mlx5/mlx5_flow_hw.c +11 -0
index 1ad261d529..1ccaf844b8 100644
@@ -7750,6 +7750,7 @@ __flow_hw_actions_template_create(struct rte_eth_dev *dev,
 {
 	struct mlx5_priv *priv = dev->data->dev_private;
 	int len, act_len, mask_len;
+	int orig_act_len;
 	unsigned int act_num;
 	unsigned int i;
 	struct rte_flow_actions_template *at = NULL;
@@ -7867,6 +7868,10 @@ __flow_hw_actions_template_create(struct rte_eth_dev *dev,
 	len += RTE_ALIGN(mask_len, 16);
 	len += RTE_ALIGN(act_num * sizeof(*at->dr_off), 16);
 	len += RTE_ALIGN(act_num * sizeof(*at->src_off), 16);
+	orig_act_len = rte_flow_conv(RTE_FLOW_CONV_OP_ACTIONS, NULL, 0, actions, error);
+	if (orig_act_len <= 0)
+		return NULL;
+	len += RTE_ALIGN(orig_act_len, 16);
 	at = mlx5_malloc(MLX5_MEM_ZERO, len + sizeof(*at),
 			 RTE_CACHE_LINE_SIZE, rte_socket_id());
 	if (!at) {
@@ -7894,6 +7899,12 @@ __flow_hw_actions_template_create(struct rte_eth_dev *dev,
 	at->src_off = RTE_PTR_ADD(at->dr_off,
 				  RTE_ALIGN(act_num * sizeof(*at->dr_off), 16));
 	memcpy(at->src_off, src_off, act_num * sizeof(at->src_off[0]));
+	at->orig_actions = RTE_PTR_ADD(at->src_off,
+				       RTE_ALIGN(act_num * sizeof(*at->src_off), 16));
+	orig_act_len = rte_flow_conv(RTE_FLOW_CONV_OP_ACTIONS, at->orig_actions, orig_act_len,
+				     actions, error);
+	if (orig_act_len <= 0)
+		goto error;
 	at->actions_num = act_num;
 	for (i = 0; i < at->actions_num; ++i)
 		at->dr_off[i] = UINT16_MAX;
Signature · 1 line
2.39.2
#27 lore raw

PATCH v3 5/8 net/mlx5: store expected type on indirect action

In reply to Dariusz Sosnowski

When template table is created, list of unmasked actions is recorded for future flow rule insertions.
This patch expands entries for RTE_FLOW_ACTION_TYPE_INDIRECT actions in this list with information about expected indirect action type. This will be used in follow up commits which add flow rule operation validation. Specifically, this will be used to verify if indirect action provided by the user references an action of a correct type.

---
 drivers/net/mlx5/mlx5_flow.h    |  4 ++++
 drivers/net/mlx5/mlx5_flow_hw.c | 22 ++++++++++++++++++++--
 2 files changed, 24 insertions(+), 2 deletions(-)
drivers/net/mlx5/mlx5_flow.h +4 -0
index 8e99e76e5d..33847e2272 100644
@@ -1442,6 +1442,10 @@ struct mlx5_action_construct_data {
 	uint16_t action_dst; /* mlx5dr_rule_action dst offset. */
 	indirect_list_callback_t indirect_list_cb;
 	union {
+		struct {
+			/* Expected type of indirection action. */
+			enum rte_flow_action_type expected_type;
+		} indirect;
 		struct {
 			/* encap data len. */
 			uint16_t len;
drivers/net/mlx5/mlx5_flow_hw.c +20 -2
index 1ccaf844b8..f337f76450 100644
@@ -997,6 +997,24 @@ __flow_hw_act_data_general_append(struct mlx5_priv *priv,
 	return 0;
 }
 
+static __rte_always_inline int
+__flow_hw_act_data_indirect_append(struct mlx5_priv *priv,
+				   struct mlx5_hw_actions *acts,
+				   enum rte_flow_action_type type,
+				   enum rte_flow_action_type mask_type,
+				   uint16_t action_src,
+				   uint16_t action_dst)
+{
+	struct mlx5_action_construct_data *act_data;
+
+	act_data = __flow_hw_act_data_alloc(priv, type, action_src, action_dst);
+	if (!act_data)
+		return -1;
+	act_data->indirect.expected_type = mask_type;
+	LIST_INSERT_HEAD(&acts->act_list, act_data, next);
+	return 0;
+}
+
 static __rte_always_inline int
 flow_hw_act_data_indirect_list_append(struct mlx5_priv *priv,
 				      struct mlx5_hw_actions *acts,
@@ -2486,9 +2504,9 @@ __flow_hw_translate_actions_template(struct rte_eth_dev *dev,
 				if (flow_hw_shared_action_translate
 				(dev, actions, acts, src_pos, dr_pos))
 					goto err;
-			} else if (__flow_hw_act_data_general_append
+			} else if (__flow_hw_act_data_indirect_append
 					(priv, acts, RTE_FLOW_ACTION_TYPE_INDIRECT,
-					 src_pos, dr_pos)){
+					 masks->type, src_pos, dr_pos)){
 				goto err;
 			}
 			break;
Signature · 1 line
2.39.2
#28 lore raw

PATCH v3 6/8 net/mlx5: store modify field action

In reply to Dariusz Sosnowski

When template table is created, list of unmasked actions is recorded for future flow rule insertions.
This patch expands entries for RTE_FLOW_ACTION_TYPE_MODIFY_FIELD actions in this list with a copy of the action from the template. This will be used in follow up commits which add flow rule operation validation. Specifically, to validate that RTE_FLOW_ACTION_TYPE_MODIFY_FIELD action passed by the user is correctly configured.

---
 drivers/net/mlx5/mlx5_flow.h    | 2 ++
 drivers/net/mlx5/mlx5_flow_hw.c | 4 +++-
 2 files changed, 5 insertions(+), 1 deletion(-)
drivers/net/mlx5/mlx5_flow.h +2 -0
index 33847e2272..6974d4e075 100644
@@ -1472,6 +1472,8 @@ struct mlx5_action_construct_data {
 			 * PRM actions.
 			 */
 			uint32_t mask[MLX5_ACT_MAX_MOD_FIELDS];
+			/* Copy of action passed to the action template. */
+			struct rte_flow_action_modify_field action;
 		} modify_header;
 		struct {
 			bool symmetric_hash_function; /* Symmetric RSS hash */
drivers/net/mlx5/mlx5_flow_hw.c +3 -1
index f337f76450..bc084af87e 100644
@@ -1112,6 +1112,7 @@ __flow_hw_act_data_hdr_modify_append(struct mlx5_priv *priv,
 				     enum rte_flow_action_type type,
 				     uint16_t action_src,
 				     uint16_t action_dst,
+				     const struct rte_flow_action_modify_field *mf,
 				     uint16_t mhdr_cmds_off,
 				     uint16_t mhdr_cmds_end,
 				     bool shared,
@@ -1124,6 +1125,7 @@ __flow_hw_act_data_hdr_modify_append(struct mlx5_priv *priv,
 	act_data = __flow_hw_act_data_alloc(priv, type, action_src, action_dst);
 	if (!act_data)
 		return -1;
+	act_data->modify_header.action = *mf;
 	act_data->modify_header.mhdr_cmds_off = mhdr_cmds_off;
 	act_data->modify_header.mhdr_cmds_end = mhdr_cmds_end;
 	act_data->modify_header.shared = shared;
@@ -1605,7 +1607,7 @@ flow_hw_modify_field_compile(struct rte_eth_dev *dev,
 	if (shared)
 		return 0;
 	ret = __flow_hw_act_data_hdr_modify_append(priv, acts, RTE_FLOW_ACTION_TYPE_MODIFY_FIELD,
-						   src_pos, mhdr->pos,
+						   src_pos, mhdr->pos, conf,
 						   cmds_start, cmds_end, shared,
 						   field, dcopy, mask);
 	if (ret)
Signature · 1 line
2.39.2
#29 lore raw

PATCH v3 7/8 common/mlx5: add debug mode indicator

In reply to Dariusz Sosnowski

Add mlx5_fp_debug_enabled() function which:

- returns true if RTE_LIBRTE_MLX5_DEBUG is defined,
- returns false otherwise.

This allows for conditional execution of code meant to be executed only when mlx5 debug mode is enabled, without adding conditional compilation guards inside source code.
When mlx5 debug mode is disabled, any code running if mlx5_fp_debug_enabled() returns true will be removed by optimizing compiler due to dead code elimination.

---
 drivers/common/mlx5/mlx5_common.h | 13 +++++++++++++
 1 file changed, 13 insertions(+)
drivers/common/mlx5/mlx5_common.h +13 -0
index 14c70edbef..1abd1e8239 100644
@@ -109,6 +109,19 @@ pmd_drv_log_basename(const char *s)
 
 #endif /* RTE_LIBRTE_MLX5_DEBUG */
 
+/**
+ * Returns true if debug mode is enabled for fast path operations.
+ */
+static inline bool
+mlx5_fp_debug_enabled(void)
+{
+#ifdef RTE_LIBRTE_MLX5_DEBUG
+	return true;
+#else
+	return false;
+#endif
+}
+
 /* Allocate a buffer on the stack and fill it with a printf format string. */
 #define MKSTR(name, ...) \
 	int mkstr_size_##name = snprintf(NULL, 0, "" __VA_ARGS__); \
Signature · 1 line
2.39.2
#30 lore raw

PATCH v3 8/8 net/mlx5: add async flow operation validation

In reply to Dariusz Sosnowski

This patch adds validation to implementations of the following API functions:

- rte_flow_async_create()
- rte_flow_async_create_by_index()
- rte_flow_async_update()
- rte_flow_async_destroy()

These validations are enabled if and only if RTE_LIBRTE_MLX5_DEBUG macro is defined.

---
 doc/guides/rel_notes/release_24_07.rst |   1 +
 drivers/net/mlx5/mlx5_flow_hw.c        | 491 ++++++++++++++++++++++++-
 2 files changed, 488 insertions(+), 4 deletions(-)
doc/guides/rel_notes/release_24_07.rst +1 -0
index 7c88de381b..cccb3ca834 100644
@@ -97,6 +97,7 @@ New Features
   * Added match with Tx queue.
   * Added match with external Tx queue.
   * Added match with E-Switch manager.
+  * Added flow item and actions validation to async flow API.
 
 * **Updated TAP driver.**
 
drivers/net/mlx5/mlx5_flow_hw.c +487 -4

527 lines not on this page · raw message

Signature · 1 line
2.39.2
#31 lore raw
In reply to Dariusz Sosnowski

Hi,
From: Dariusz Sosnowski <[email protected]>
Sent: Wednesday, June 26, 2024 9:14 PM
To: Slava Ovsiienko; Bing Zhao; Ori Kam; Suanming Mou; Matan Azrad
Cc: [email protected]
Subject: [PATCH v3 0/8] net/mlx5: flow fast path validation

This patchset adds validations for flow items and actions supplied by the user to fast path async flow API functions implemented in mlx5 PMD.

- Patches 1-7: Prepares mlx5 PMD for adding fast path validations. Specifically:

- Required functionality for validating queue indexes and target represented ports are extracted for reuse.
- Additional info, required for validation, is stored in pattern and actions templates.
- Introduces mlx5_fp_debug_enabled() function for checking if mlx5 PMD was compiled in debug mode.

- Patch 8: Introduce validations to:

- rte_flow_async_create()
- rte_flow_async_create_by_index()
- rte_flow_async_actions_update()
- rte_flow_async_destroy()

These validations are enabled if and only if RTE_LIBRTE_MLX5_DEBUG macro is defined during compilation.

v3:
- Rebased on 24.07-rc1.
- Removed Depends-on tag for already marged ethdev patch.

v2:
- Split series 32095 into separate ethdev and mlx5 series.
- Rebased on latest commit in next-net-mlx-main.
- Removed Depends-on tags for already merged mlx5 patch series.

Dariusz Sosnowski (8):
net/mlx5: extract target port validation net/mlx5: extract queue index validation net/mlx5: store pattern template items net/mlx5: store original actions in template net/mlx5: store expected type on indirect action net/mlx5: store modify field action common/mlx5: add debug mode indicator net/mlx5: add async flow operation validation

 doc/guides/rel_notes/release_24_07.rst |   1 +
 drivers/common/mlx5/mlx5_common.h      |  13 +
 drivers/net/mlx5/mlx5_flow.c           |  59 ++-
 drivers/net/mlx5/mlx5_flow.h           |  12 +
 drivers/net/mlx5/mlx5_flow_hw.c        | 620 +++++++++++++++++++++++--
 5 files changed, 657 insertions(+), 48 deletions(-)

--
2.39.2

Series applied to next-net-mlx,

Kindest regards,
Raslan Darawsheh