Both checksum and IPv6 next_hdr needs to be updated when
adding or removing srv6 header into/from IPv6 packets.
1. Add srv6
ste1 (push buffer with next_hdr 0) --> ste2 (IPv6 next_hdr to 0x2b)
--> ste3 (load next hop IPv6 address, and srv6 next_hdr restore)
2. Remove srv6
ste1 (set srv6 next_hdr 0 and save original) --> ste2 (load final
IPv6 destination, restore srv6 next_hdr) --> ste3 (remove srv6 and
copy srv6 next_hdr to ipv6 next_hdr)
Add helpers to generate the 2 modify header resources for add/remove
actions.
Remove srv6 should be shared globally and add srv6 can be shared or
unique per each flow rules.
Signed-off-by: Rongwei Liu <rongw...@nvidia.com>
---
drivers/net/mlx5/mlx5.h | 29 +++
drivers/net/mlx5/mlx5_flow_dv.c | 386 ++++++++++++++++++++++++++++++++
2 files changed, 415 insertions(+)
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 3fbec4db9e..2cb6364957 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -2314,4 +2314,33 @@ void mlx5_flex_parser_clone_free_cb(void *tool_ctx,
int mlx5_alloc_srh_flex_parser(struct rte_eth_dev *dev);
void mlx5_free_srh_flex_parser(struct rte_eth_dev *dev);
+
+int
+flow_dv_generate_ipv6_routing_pop_mhdr1(struct rte_eth_dev *dev,
+ const struct rte_flow_attr *attr,
+ struct mlx5_modification_cmd *cmd,
+ uint32_t cmd_num);
+
+int
+flow_dv_generate_ipv6_routing_pop_mhdr2(struct rte_eth_dev *dev,
+ const struct rte_flow_attr *attr,
+ struct mlx5_modification_cmd *cmd,
+ uint32_t cmd_num);
+
+int
+flow_dv_generate_ipv6_routing_push_mhdr1(struct rte_eth_dev *dev,
+ const struct rte_flow_attr *attr,
+ struct mlx5_modification_cmd *cmd,
+ uint32_t cmd_num);
+
+int
+flow_dv_generate_ipv6_routing_push_mhdr2(struct rte_eth_dev *dev,
+ const struct rte_flow_attr *attr,
+ struct mlx5_modification_cmd *cmd,
+ uint32_t cmd_num, uint8_t *buf);
+
+int
+flow_dv_ipv6_routing_pop_mhdr_cmd(struct rte_eth_dev *dev, uint8_t *mh_data,
+ uint8_t *anchor_id);
+
#endif /* RTE_PMD_MLX5_H_ */
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index f136f43b0a..4a1f61eeb7 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -2128,6 +2128,392 @@ flow_dv_convert_action_modify_field
field, dcopy, resource, type, error);
}
+/**
+ * Generate the 1st modify header data for IPv6 routing pop.
+ *
+ * @param[in] dev
+ * Pointer to the rte_eth_dev structure.
+ * @param[in] attr
+ * Pointer to the rte_flow table attribute.
+ * @param[in,out] cmd
+ * Pointer to modify header command buffer.
+ * @param[in] cmd_num
+ * Modify header command number.
+ *
+ * @return
+ * Positive on success, a negative value otherwise.
+ */
+int
+flow_dv_generate_ipv6_routing_pop_mhdr1(struct rte_eth_dev *dev,
+ const struct rte_flow_attr *attr,
+ struct mlx5_modification_cmd *cmd,
+ uint32_t cmd_num)
+{
+ struct mlx5_priv *priv = dev->data->dev_private;
+ struct rte_flow_action_modify_data data;
+ struct field_modify_info field[MLX5_ACT_MAX_MOD_FIELDS] = {
+ {0, 0, MLX5_MODI_OUT_NONE} };
+ struct field_modify_info dcopy[MLX5_ACT_MAX_MOD_FIELDS] = {
+ {0, 0, MLX5_MODI_OUT_NONE} };
+ uint32_t mask[MLX5_ACT_MAX_MOD_FIELDS] = { 0 };
+ struct rte_flow_item item = {
+ .spec = NULL,
+ .mask = NULL
+ };
+ union {
+ struct mlx5_flow_dv_modify_hdr_resource resource;
+ uint8_t data[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
+ sizeof(struct mlx5_modification_cmd) *
MLX5_MHDR_MAX_CMD];
+ } dummy;
+ struct mlx5_flow_dv_modify_hdr_resource *resource;
+ uint32_t value = 0;
+ struct rte_flow_error error;
+
+#define IPV6_ROUTING_POP_MHDR_NUM1 3
+ if (cmd_num < IPV6_ROUTING_POP_MHDR_NUM1) {
+ DRV_LOG(ERR, "Not enough modify header buffer");
+ return -1;
+ }
+ memset(&data, 0, sizeof(data));
+ memset(&dummy, 0, sizeof(dummy));
+ /* save next_hdr to seg_left. */
+ data.field = RTE_FLOW_FIELD_FLEX_ITEM;
+ data.flex_handle = (struct rte_flow_item_flex_handle *)
+ (uintptr_t)&priv->sh->srh_flex_parser.flex;
+ data.offset = offsetof(struct rte_ipv6_routing_ext, segments_left) *
CHAR_BIT;
+ /* For COPY fill the destination field (dcopy) without mask. */
+ mlx5_flow_field_id_to_modify_info(&data, dcopy, NULL, 8, dev, attr,
&error);
+ /* Then construct the source field (field) with mask. */
+ data.offset = offsetof(struct rte_ipv6_routing_ext, next_hdr) *
CHAR_BIT;
+ mlx5_flow_field_id_to_modify_info(&data, field, mask, 8, dev, attr,
&error);
+ item.mask = &mask;
+ resource = &dummy.resource;
+ if (flow_dv_convert_modify_action(&item, field, dcopy, resource,
+ MLX5_MODIFICATION_TYPE_COPY, &error))
{
+ DRV_LOG(ERR, "Generate save srv6 next header modify header
failed");
+ return -1;
+ }
+ MLX5_ASSERT(resource->actions_num == 1);
+ /* add nop. */
+ resource->actions[1].data0 = 0;
+ resource->actions[1].action_type = MLX5_MODIFICATION_TYPE_NOP;
+ resource->actions[1].data0 = RTE_BE32(resource->actions[1].data0);
+ resource->actions[1].data1 = 0;
+ resource->actions_num += 1;
+ /* clear srv6 next_hdr. */
+ memset(&field, 0, sizeof(field));
+ memset(&dcopy, 0, sizeof(dcopy));
+ memset(&mask, 0, sizeof(mask));
+ mlx5_flow_field_id_to_modify_info(&data, field, mask, 8, dev, attr,
&error);
+ item.spec = (void *)(uintptr_t)&value;
+ if (flow_dv_convert_modify_action(&item, field, dcopy, resource,
+ MLX5_MODIFICATION_TYPE_SET, &error)) {
+ DRV_LOG(ERR, "Generate clear srv6 next header modify header
failed");
+ return -1;
+ }
+ MLX5_ASSERT(resource->actions_num == IPV6_ROUTING_POP_MHDR_NUM1);
+#undef IPV6_ROUTING_POP_MHDR_NUM1
+ memcpy(cmd, resource->actions,
+ resource->actions_num * sizeof(struct mlx5_modification_cmd));
+ return resource->actions_num;
+}
+
+/**
+ * Generate the 2nd modify header data for IPv6 routing pop.
+ *
+ * @param[in] dev
+ * Pointer to the rte_eth_dev structure.
+ * @param[in] attr
+ * Pointer to the rte_flow table attribute.
+ * @param[in,out] cmd
+ * Pointer to modify header command buffer.
+ * @param[in] cmd_num
+ * Modify header command number.
+ *
+ * @return
+ * Positive on success, a negative value otherwise.
+ */
+int
+flow_dv_generate_ipv6_routing_pop_mhdr2(struct rte_eth_dev *dev,
+ const struct rte_flow_attr *attr,
+ struct mlx5_modification_cmd *cmd,
+ uint32_t cmd_num)
+{
+ struct mlx5_priv *priv = dev->data->dev_private;
+ struct rte_flow_action_modify_data data;
+ struct field_modify_info field[MLX5_ACT_MAX_MOD_FIELDS] = {
+ {0, 0, MLX5_MODI_OUT_NONE} };
+ struct field_modify_info dcopy[MLX5_ACT_MAX_MOD_FIELDS] = {
+ {0, 0, MLX5_MODI_OUT_NONE} };
+ uint32_t mask[MLX5_ACT_MAX_MOD_FIELDS] = { 0 };
+ struct rte_flow_item item = {
+ .spec = NULL,
+ .mask = NULL
+ };
+ union {
+ struct mlx5_flow_dv_modify_hdr_resource resource;
+ uint8_t data[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
+ sizeof(struct mlx5_modification_cmd) *
MLX5_MHDR_MAX_CMD];
+ } dummy;
+ struct mlx5_flow_dv_modify_hdr_resource *resource;
+ struct rte_flow_error error;
+
+#define IPV6_ROUTING_POP_MHDR_NUM2 5
+ if (cmd_num < IPV6_ROUTING_POP_MHDR_NUM2) {
+ DRV_LOG(ERR, "Note enough modify header buffer");
+ return -1;
+ }
+ memset(&data, 0, sizeof(data));
+ memset(&dummy, 0, sizeof(dummy));
+ resource = &dummy.resource;
+ item.mask = &mask;
+ data.field = RTE_FLOW_FIELD_IPV6_DST;
+ data.level = 0;
+ data.offset = 0;
+ mlx5_flow_field_id_to_modify_info(&data, dcopy, NULL, 128, dev, attr,
&error);
+ data.field = RTE_FLOW_FIELD_FLEX_ITEM;
+ data.offset = 32;
+ data.flex_handle = (struct rte_flow_item_flex_handle *)
+ (uintptr_t)&priv->sh->srh_flex_parser.flex;
+ mlx5_flow_field_id_to_modify_info(&data, field, mask, 128, dev, attr,
&error);
+ if (flow_dv_convert_modify_action(&item, field, dcopy, resource,
+ MLX5_MODIFICATION_TYPE_COPY, &error))
{
+ DRV_LOG(ERR, "Generate load final IPv6 address modify header
failed");
+ return -1;
+ }
+ MLX5_ASSERT(resource->actions_num == 4);
+ memset(&field, 0, sizeof(field));
+ memset(&dcopy, 0, sizeof(dcopy));
+ memset(&mask, 0, sizeof(mask));
+ /* copy seg_left to srv6.next_hdr */
+ data.offset = offsetof(struct rte_ipv6_routing_ext, next_hdr) *
CHAR_BIT;
+ mlx5_flow_field_id_to_modify_info(&data, dcopy, NULL, 8, dev, attr,
&error);
+ data.offset = offsetof(struct rte_ipv6_routing_ext, segments_left) *
CHAR_BIT;
+ mlx5_flow_field_id_to_modify_info(&data, field, mask, 8, dev, attr,
&error);
+ if (flow_dv_convert_modify_action(&item, field, dcopy, resource,
+ MLX5_MODIFICATION_TYPE_COPY, &error))
{
+ DRV_LOG(ERR, "Generate restore srv6 next header modify header
failed");
+ return -1;
+ }
+ MLX5_ASSERT(resource->actions_num == IPV6_ROUTING_POP_MHDR_NUM2);
+#undef IPV6_ROUTING_POP_MHDR_NUM2
+ memcpy(cmd, resource->actions,
+ resource->actions_num * sizeof(struct mlx5_modification_cmd));
+ return resource->actions_num;
+}
+
+/**
+ * Generate the 2nd modify header data for IPv6 routing push.
+ *
+ * @param[in] dev
+ * Pointer to the rte_eth_dev structure.
+ * @param[in] attr
+ * Pointer to the rte_flow table attribute.
+ * @param[in,out] cmd
+ * Pointer to modify header command buffer.
+ * @param[in] cmd_num
+ * Modify header command number.
+ *
+ * @return
+ * Positive on success, a negative value otherwise.
+ */
+int
+flow_dv_generate_ipv6_routing_push_mhdr1(struct rte_eth_dev *dev,
+ const struct rte_flow_attr *attr,
+ struct mlx5_modification_cmd *cmd,
+ uint32_t cmd_num)
+{
+ struct rte_flow_action_modify_data data;
+ struct field_modify_info field[MLX5_ACT_MAX_MOD_FIELDS] = {
+ {0, 0, MLX5_MODI_OUT_NONE} };
+ struct field_modify_info dcopy[MLX5_ACT_MAX_MOD_FIELDS] = {
+ {0, 0, MLX5_MODI_OUT_NONE} };
+ uint32_t mask[MLX5_ACT_MAX_MOD_FIELDS] = { 0 };
+ struct rte_flow_item item = {
+ .spec = NULL,
+ .mask = NULL
+ };
+ union {
+ struct mlx5_flow_dv_modify_hdr_resource resource;
+ uint8_t data[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
+ sizeof(struct mlx5_modification_cmd) *
MLX5_MHDR_MAX_CMD];
+ } dummy;
+ struct mlx5_flow_dv_modify_hdr_resource *resource;
+ struct rte_flow_error error;
+ uint8_t value;
+
+#define IPV6_ROUTING_PUSH_MHDR_NUM1 1
+ if (cmd_num < IPV6_ROUTING_PUSH_MHDR_NUM1) {
+ DRV_LOG(ERR, "Not enough modify header buffer");
+ return -1;
+ }
+ memset(&data, 0, sizeof(data));
+ memset(&dummy, 0, sizeof(dummy));
+ resource = &dummy.resource;
+ /* Set IPv6 proto to 0x2b. */
+ data.field = RTE_FLOW_FIELD_IPV6_PROTO;
+ mlx5_flow_field_id_to_modify_info(&data, field, mask, 8, dev, attr,
&error);
+ resource = &dummy.resource;
+ item.mask = &mask;
+ value = IPPROTO_ROUTING;
+ item.spec = (void *)(uintptr_t)&value;
+ if (flow_dv_convert_modify_action(&item, field, dcopy, resource,
+ MLX5_MODIFICATION_TYPE_SET, &error)) {
+ DRV_LOG(ERR, "Generate modify IPv6 protocol to 0x2b failed");
+ return -1;
+ }
+ MLX5_ASSERT(resource->actions_num == IPV6_ROUTING_PUSH_MHDR_NUM1);
+#undef IPV6_ROUTING_PUSH_MHDR_NUM1
+ memcpy(cmd, resource->actions,
+ resource->actions_num * sizeof(struct mlx5_modification_cmd));
+ return resource->actions_num;
+}
+
+/**
+ * Generate the 2nd modify header data for IPv6 routing push.
+ *
+ * @param[in] dev
+ * Pointer to the rte_eth_dev structure.
+ * @param[in] attr
+ * Pointer to the rte_flow table attribute.
+ * @param[in,out] cmd
+ * Pointer to modify header command buffer.
+ * @param[in] cmd_num
+ * Modify header command number.
+ *
+ * @return
+ * Positive on success, a negative value otherwise.
+ */
+int
+flow_dv_generate_ipv6_routing_push_mhdr2(struct rte_eth_dev *dev,
+ const struct rte_flow_attr *attr,
+ struct mlx5_modification_cmd *cmd,
+ uint32_t cmd_num, uint8_t *buf)
+{
+ struct mlx5_priv *priv = dev->data->dev_private;
+ struct rte_flow_action_modify_data data;
+ struct field_modify_info field[MLX5_ACT_MAX_MOD_FIELDS] = {
+ {0, 0, MLX5_MODI_OUT_NONE} };
+ struct field_modify_info dcopy[MLX5_ACT_MAX_MOD_FIELDS] = {
+ {0, 0, MLX5_MODI_OUT_NONE} };
+ uint32_t mask[MLX5_ACT_MAX_MOD_FIELDS] = { 0 };
+ struct rte_flow_item item = {
+ .spec = NULL,
+ .mask = NULL
+ };
+ union {
+ struct mlx5_flow_dv_modify_hdr_resource resource;
+ uint8_t data[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
+ sizeof(struct mlx5_modification_cmd) *
MLX5_MHDR_MAX_CMD];
+ } dummy;
+ struct mlx5_flow_dv_modify_hdr_resource *resource;
+ struct rte_flow_error error;
+ uint8_t next_hdr = *buf;
+
+#define IPV6_ROUTING_PUSH_MHDR_NUM2 5
+ if (cmd_num < IPV6_ROUTING_PUSH_MHDR_NUM2) {
+ DRV_LOG(ERR, "Not enough modify header buffer");
+ return -1;
+ }
+ memset(&data, 0, sizeof(data));
+ memset(&dummy, 0, sizeof(dummy));
+ resource = &dummy.resource;
+ item.mask = &mask;
+ item.spec = buf + sizeof(struct rte_ipv6_routing_ext) +
+ (*(buf + 3) - 1) * 16; /* seg_left-1 IPv6 address */
+ data.field = RTE_FLOW_FIELD_IPV6_DST;
+ mlx5_flow_field_id_to_modify_info(&data, field, mask, 128, dev, attr,
&error);
+ if (flow_dv_convert_modify_action(&item, field, dcopy, resource,
+ MLX5_MODIFICATION_TYPE_SET, &error)) {
+ DRV_LOG(ERR, "Generate load srv6 next hop modify header
failed");
+ return -1;
+ }
+ MLX5_ASSERT(resource->actions_num == 4);
+ memset(&field, 0, sizeof(field));
+ memset(&mask, 0, sizeof(mask));
+ data.field = RTE_FLOW_FIELD_FLEX_ITEM;
+ data.flex_handle = (struct rte_flow_item_flex_handle *)
+ (uintptr_t)&priv->sh->srh_flex_parser.flex;
+ data.offset = offsetof(struct rte_ipv6_routing_ext, next_hdr) *
CHAR_BIT;
+ item.spec = (void *)(uintptr_t)&next_hdr;
+ mlx5_flow_field_id_to_modify_info(&data, field, mask, 8, dev, attr,
&error);
+ if (flow_dv_convert_modify_action(&item, field, dcopy, resource,
+ MLX5_MODIFICATION_TYPE_SET, &error)) {
+ DRV_LOG(ERR, "Generate srv6 next header restore modify header
failed");
+ return -1;
+ }
+ MLX5_ASSERT(resource->actions_num == IPV6_ROUTING_PUSH_MHDR_NUM2);
+#undef IPV6_ROUTING_PUSH_MHDR_NUM2
+ memcpy(cmd, resource->actions,
+ resource->actions_num * sizeof(struct mlx5_modification_cmd));
+ return resource->actions_num;
+}
+
+/**
+ * Generate IPv6 routing pop modification_cmd.
+ *
+ * @param[in] dev
+ * Pointer to the rte_eth_dev structure.
+ * @param[in,out] mh_data
+ * Pointer to modify header data buffer.
+ * @param[in,out] anchor_id
+ * Anchor ID for REMOVE command.
+ *
+ * @return
+ * Positive on success, a negative value otherwise.
+ */
+int
+flow_dv_ipv6_routing_pop_mhdr_cmd(struct rte_eth_dev *dev, uint8_t *mh_data,
+ uint8_t *anchor_id)
+{
+ struct rte_flow_action_modify_data data;
+ struct field_modify_info field[MLX5_ACT_MAX_MOD_FIELDS] = {
+ {0, 0, MLX5_MODI_OUT_NONE} };
+ struct field_modify_info dcopy[MLX5_ACT_MAX_MOD_FIELDS] = {
+ {0, 0, MLX5_MODI_OUT_NONE} };
+ uint32_t mask[MLX5_ACT_MAX_MOD_FIELDS] = { 0 };
+ struct rte_flow_item item = {
+ .spec = NULL,
+ .mask = NULL
+ };
+ union {
+ struct mlx5_flow_dv_modify_hdr_resource resource;
+ uint8_t data[sizeof(struct mlx5_flow_dv_modify_hdr_resource) +
+ sizeof(struct mlx5_modification_cmd) *
MLX5_MHDR_MAX_CMD];
+ } dummy;
+ struct mlx5_flow_dv_modify_hdr_resource *resource;
+ struct rte_flow_error error;
+ struct mlx5_priv *priv = dev->data->dev_private;
+
+ if (!priv || !priv->sh->cdev->config.hca_attr.flex.parse_graph_anchor) {
+ DRV_LOG(ERR, "Doesn't support srv6 as reformat anchor");
+ return -1;
+ }
+ /* Restore IPv6 protocol from flex parser. */
+ memset(&data, 0, sizeof(data));
+ memset(&dummy, 0, sizeof(dummy));
+ data.field = RTE_FLOW_FIELD_IPV6_PROTO;
+ mlx5_flow_field_id_to_modify_info(&data, dcopy, NULL, 8, dev, NULL,
&error);
+ /* Then construct the source field (field) with mask. */
+ data.field = RTE_FLOW_FIELD_FLEX_ITEM;
+ data.flex_handle = (struct rte_flow_item_flex_handle *)
+ (uintptr_t)&priv->sh->srh_flex_parser.flex;
+ data.offset = offsetof(struct rte_ipv6_routing_ext, next_hdr) *
CHAR_BIT;
+ mlx5_flow_field_id_to_modify_info(&data, field, mask, 8, dev, NULL,
&error);
+ item.mask = &mask;
+ resource = &dummy.resource;
+ if (flow_dv_convert_modify_action(&item, field, dcopy, resource,
+ MLX5_MODIFICATION_TYPE_COPY,
+ &error)) {
+ DRV_LOG(ERR, "Generate copy IPv6 protocol from srv6 next header
failed");
+ return -1;
+ }
+ MLX5_ASSERT(resource->actions_num == 1);
+ memcpy(mh_data, resource->actions, sizeof(struct
mlx5_modification_cmd));
+ *anchor_id = priv->sh->srh_flex_parser.flex.devx_fp->anchor_id;
+ return 1;
+}
+
/**
* Validate MARK item.
*
--
2.27.0
