net/mlx5: add metadata register copy table
authorViacheslav Ovsiienko <viacheslavo@mellanox.com>
Thu, 7 Nov 2019 17:10:04 +0000 (17:10 +0000)
committerFerruh Yigit <ferruh.yigit@intel.com>
Mon, 11 Nov 2019 13:23:02 +0000 (14:23 +0100)
While reg_c[meta] can be copied to reg_b simply by modify-header
action (it is supported by hardware), it is not possible to copy
reg_c[mark] to the STE flow_tag as flow_tag is not a metadata
register and this is not supported by hardware. Instead, it
should be manually set by a flow per each unique MARK ID. For
this purpose, there should be a dedicated flow table -
RX_CP_TBL and all the Rx flow should pass by the table
to properly copy values from the register to flow tag field.

And for each MARK action, a copy flow should be added
to RX_CP_TBL according to the MARK ID like:
  (if reg_c[mark] == mark_id),
    flow_tag := mark_id / reg_b := reg_c[meta] / jump to RX_ACT_TBL

For SET_META action, there can be only one default flow like:
  reg_b := reg_c[meta] / jump to RX_ACT_TBL

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Acked-by: Matan Azrad <matan@mellanox.com>
drivers/net/mlx5/mlx5.c
drivers/net/mlx5/mlx5.h
drivers/net/mlx5/mlx5_defs.h
drivers/net/mlx5/mlx5_flow.c
drivers/net/mlx5/mlx5_flow.h
drivers/net/mlx5/mlx5_flow_dv.c

index ff62ab2..91aaa9b 100644 (file)
@@ -1039,6 +1039,8 @@ mlx5_dev_close(struct rte_eth_dev *dev)
                priv->txqs = NULL;
        }
        mlx5_proc_priv_uninit(dev);
+       if (priv->mreg_cp_tbl)
+               mlx5_hlist_destroy(priv->mreg_cp_tbl, NULL, NULL);
        mlx5_mprq_free_mp(dev);
        mlx5_free_shared_dr(priv);
        if (priv->rss_conf.rss_key != NULL)
@@ -2458,9 +2460,22 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
                        goto error;
                }
        }
+       if (priv->config.dv_flow_en &&
+           priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
+           mlx5_flow_ext_mreg_supported(eth_dev) &&
+           priv->sh->dv_regc0_mask) {
+               priv->mreg_cp_tbl = mlx5_hlist_create(MLX5_FLOW_MREG_HNAME,
+                                                     MLX5_FLOW_MREG_HTABLE_SZ);
+               if (!priv->mreg_cp_tbl) {
+                       err = ENOMEM;
+                       goto error;
+               }
+       }
        return eth_dev;
 error:
        if (priv) {
+               if (priv->mreg_cp_tbl)
+                       mlx5_hlist_destroy(priv->mreg_cp_tbl, NULL, NULL);
                if (priv->sh)
                        mlx5_free_shared_dr(priv);
                if (priv->nl_socket_route >= 0)
index ddfb593..fab58c9 100644 (file)
@@ -569,8 +569,9 @@ struct mlx5_flow_tbl_resource {
 #define MLX5_HAIRPIN_TX_TABLE (UINT16_MAX - 1)
 /* Reserve the last two tables for metadata register copy. */
 #define MLX5_FLOW_MREG_ACT_TABLE_GROUP (MLX5_MAX_TABLES - 1)
-#define MLX5_FLOW_MREG_CP_TABLE_GROUP \
-       (MLX5_FLOW_MREG_ACT_TABLE_GROUP - 1)
+#define MLX5_FLOW_MREG_CP_TABLE_GROUP (MLX5_MAX_TABLES - 2)
+/* Tables for metering splits should be added here. */
+#define MLX5_MAX_TABLES_EXTERNAL (MLX5_MAX_TABLES - 3)
 #define MLX5_MAX_TABLES_FDB UINT16_MAX
 
 #define MLX5_DBR_PAGE_SIZE 4096 /* Must be >= 512. */
@@ -736,6 +737,8 @@ struct mlx5_priv {
        LIST_HEAD(dbrpage, mlx5_devx_dbr_page) dbrpgs; /* Door-bell pages. */
        struct mlx5_vlan_vmwa_context *vmwa_context; /* VLAN WA context. */
        struct mlx5_flow_id_pool *qrss_id_pool;
+       struct mlx5_hlist *mreg_cp_tbl;
+       /* Hash table of Rx metadata register copy table. */
 #ifndef RTE_ARCH_64
        rte_spinlock_t uar_lock_cq; /* CQs share a common distinct UAR */
        rte_spinlock_t uar_lock[MLX5_UAR_PAGE_NUM_MAX];
index a77c430..0ef532f 100644 (file)
 #define MLX5_XMETA_MODE_META16 1
 #define MLX5_XMETA_MODE_META32 2
 
+/* Size of the simple hash table for metadata register table. */
+#define MLX5_FLOW_MREG_HTABLE_SZ 4096
+#define MLX5_FLOW_MREG_HNAME "MARK_COPY_TABLE"
+
 /* Definition of static_assert found in /usr/include/assert.h */
 #ifndef HAVE_STATIC_ASSERT
 #define static_assert _Static_assert
index 06ff8f8..61afc48 100644 (file)
@@ -675,7 +675,17 @@ flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
                        container_of((*priv->rxqs)[idx],
                                     struct mlx5_rxq_ctrl, rxq);
 
-               if (mark) {
+               /*
+                * To support metadata register copy on Tx loopback,
+                * this must be always enabled (metadata may arive
+                * from other port - not from local flows only.
+                */
+               if (priv->config.dv_flow_en &&
+                   priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
+                   mlx5_flow_ext_mreg_supported(dev)) {
+                       rxq_ctrl->rxq.mark = 1;
+                       rxq_ctrl->flow_mark_n = 1;
+               } else if (mark) {
                        rxq_ctrl->rxq.mark = 1;
                        rxq_ctrl->flow_mark_n++;
                }
@@ -739,7 +749,12 @@ flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
                        container_of((*priv->rxqs)[idx],
                                     struct mlx5_rxq_ctrl, rxq);
 
-               if (mark) {
+               if (priv->config.dv_flow_en &&
+                   priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
+                   mlx5_flow_ext_mreg_supported(dev)) {
+                       rxq_ctrl->rxq.mark = 1;
+                       rxq_ctrl->flow_mark_n = 1;
+               } else if (mark) {
                        rxq_ctrl->flow_mark_n--;
                        rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
                }
@@ -2749,6 +2764,398 @@ flow_check_hairpin_split(struct rte_eth_dev *dev,
        return 0;
 }
 
+/* Declare flow create/destroy prototype in advance. */
+static struct rte_flow *
+flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
+                const struct rte_flow_attr *attr,
+                const struct rte_flow_item items[],
+                const struct rte_flow_action actions[],
+                bool external, struct rte_flow_error *error);
+
+static void
+flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
+                 struct rte_flow *flow);
+
+/**
+ * Add a flow of copying flow metadata registers in RX_CP_TBL.
+ *
+ * As mark_id is unique, if there's already a registered flow for the mark_id,
+ * return by increasing the reference counter of the resource. Otherwise, create
+ * the resource (mcp_res) and flow.
+ *
+ * Flow looks like,
+ *   - If ingress port is ANY and reg_c[1] is mark_id,
+ *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
+ *
+ * For default flow (zero mark_id), flow is like,
+ *   - If ingress port is ANY,
+ *     reg_b := reg_c[0] and jump to RX_ACT_TBL.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param mark_id
+ *   ID of MARK action, zero means default flow for META.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL.
+ *
+ * @return
+ *   Associated resource on success, NULL otherwise and rte_errno is set.
+ */
+static struct mlx5_flow_mreg_copy_resource *
+flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
+                         struct rte_flow_error *error)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       struct rte_flow_attr attr = {
+               .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
+               .ingress = 1,
+       };
+       struct mlx5_rte_flow_item_tag tag_spec = {
+               .data = mark_id,
+       };
+       struct rte_flow_item items[] = {
+               [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
+       };
+       struct rte_flow_action_mark ftag = {
+               .id = mark_id,
+       };
+       struct mlx5_flow_action_copy_mreg cp_mreg = {
+               .dst = REG_B,
+               .src = 0,
+       };
+       struct rte_flow_action_jump jump = {
+               .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
+       };
+       struct rte_flow_action actions[] = {
+               [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
+       };
+       struct mlx5_flow_mreg_copy_resource *mcp_res;
+       int ret;
+
+       /* Fill the register fileds in the flow. */
+       ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
+       if (ret < 0)
+               return NULL;
+       tag_spec.id = ret;
+       ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
+       if (ret < 0)
+               return NULL;
+       cp_mreg.src = ret;
+       /* Check if already registered. */
+       assert(priv->mreg_cp_tbl);
+       mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
+       if (mcp_res) {
+               /* For non-default rule. */
+               if (mark_id)
+                       mcp_res->refcnt++;
+               assert(mark_id || mcp_res->refcnt == 1);
+               return mcp_res;
+       }
+       /* Provide the full width of FLAG specific value. */
+       if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
+               tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
+       /* Build a new flow. */
+       if (mark_id) {
+               items[0] = (struct rte_flow_item){
+                       .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
+                       .spec = &tag_spec,
+               };
+               items[1] = (struct rte_flow_item){
+                       .type = RTE_FLOW_ITEM_TYPE_END,
+               };
+               actions[0] = (struct rte_flow_action){
+                       .type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
+                       .conf = &ftag,
+               };
+               actions[1] = (struct rte_flow_action){
+                       .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
+                       .conf = &cp_mreg,
+               };
+               actions[2] = (struct rte_flow_action){
+                       .type = RTE_FLOW_ACTION_TYPE_JUMP,
+                       .conf = &jump,
+               };
+               actions[3] = (struct rte_flow_action){
+                       .type = RTE_FLOW_ACTION_TYPE_END,
+               };
+       } else {
+               /* Default rule, wildcard match. */
+               attr.priority = MLX5_FLOW_PRIO_RSVD;
+               items[0] = (struct rte_flow_item){
+                       .type = RTE_FLOW_ITEM_TYPE_END,
+               };
+               actions[0] = (struct rte_flow_action){
+                       .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
+                       .conf = &cp_mreg,
+               };
+               actions[1] = (struct rte_flow_action){
+                       .type = RTE_FLOW_ACTION_TYPE_JUMP,
+                       .conf = &jump,
+               };
+               actions[2] = (struct rte_flow_action){
+                       .type = RTE_FLOW_ACTION_TYPE_END,
+               };
+       }
+       /* Build a new entry. */
+       mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
+       if (!mcp_res) {
+               rte_errno = ENOMEM;
+               return NULL;
+       }
+       /*
+        * The copy Flows are not included in any list. There
+        * ones are referenced from other Flows and can not
+        * be applied, removed, deleted in ardbitrary order
+        * by list traversing.
+        */
+       mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
+                                        actions, false, error);
+       if (!mcp_res->flow)
+               goto error;
+       mcp_res->refcnt++;
+       mcp_res->hlist_ent.key = mark_id;
+       ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
+                               &mcp_res->hlist_ent);
+       assert(!ret);
+       if (ret)
+               goto error;
+       return mcp_res;
+error:
+       if (mcp_res->flow)
+               flow_list_destroy(dev, NULL, mcp_res->flow);
+       rte_free(mcp_res);
+       return NULL;
+}
+
+/**
+ * Release flow in RX_CP_TBL.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @flow
+ *   Parent flow for wich copying is provided.
+ */
+static void
+flow_mreg_del_copy_action(struct rte_eth_dev *dev,
+                         struct rte_flow *flow)
+{
+       struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
+       struct mlx5_priv *priv = dev->data->dev_private;
+
+       if (!mcp_res || !priv->mreg_cp_tbl)
+               return;
+       if (flow->copy_applied) {
+               assert(mcp_res->appcnt);
+               flow->copy_applied = 0;
+               --mcp_res->appcnt;
+               if (!mcp_res->appcnt)
+                       flow_drv_remove(dev, mcp_res->flow);
+       }
+       /*
+        * We do not check availability of metadata registers here,
+        * because copy resources are allocated in this case.
+        */
+       if (--mcp_res->refcnt)
+               return;
+       assert(mcp_res->flow);
+       flow_list_destroy(dev, NULL, mcp_res->flow);
+       mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
+       rte_free(mcp_res);
+       flow->mreg_copy = NULL;
+}
+
+/**
+ * Start flow in RX_CP_TBL.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @flow
+ *   Parent flow for wich copying is provided.
+ *
+ * @return
+ *   0 on success, a negative errno value otherwise and rte_errno is set.
+ */
+static int
+flow_mreg_start_copy_action(struct rte_eth_dev *dev,
+                           struct rte_flow *flow)
+{
+       struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
+       int ret;
+
+       if (!mcp_res || flow->copy_applied)
+               return 0;
+       if (!mcp_res->appcnt) {
+               ret = flow_drv_apply(dev, mcp_res->flow, NULL);
+               if (ret)
+                       return ret;
+       }
+       ++mcp_res->appcnt;
+       flow->copy_applied = 1;
+       return 0;
+}
+
+/**
+ * Stop flow in RX_CP_TBL.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @flow
+ *   Parent flow for wich copying is provided.
+ */
+static void
+flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
+                          struct rte_flow *flow)
+{
+       struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
+
+       if (!mcp_res || !flow->copy_applied)
+               return;
+       assert(mcp_res->appcnt);
+       --mcp_res->appcnt;
+       flow->copy_applied = 0;
+       if (!mcp_res->appcnt)
+               flow_drv_remove(dev, mcp_res->flow);
+}
+
+/**
+ * Remove the default copy action from RX_CP_TBL.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ */
+static void
+flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
+{
+       struct mlx5_flow_mreg_copy_resource *mcp_res;
+       struct mlx5_priv *priv = dev->data->dev_private;
+
+       /* Check if default flow is registered. */
+       if (!priv->mreg_cp_tbl)
+               return;
+       mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, 0ULL);
+       if (!mcp_res)
+               return;
+       assert(mcp_res->flow);
+       flow_list_destroy(dev, NULL, mcp_res->flow);
+       mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
+       rte_free(mcp_res);
+}
+
+/**
+ * Add the default copy action in in RX_CP_TBL.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL.
+ *
+ * @return
+ *   0 for success, negative value otherwise and rte_errno is set.
+ */
+static int
+flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
+                                 struct rte_flow_error *error)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       struct mlx5_flow_mreg_copy_resource *mcp_res;
+
+       /* Check whether extensive metadata feature is engaged. */
+       if (!priv->config.dv_flow_en ||
+           priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
+           !mlx5_flow_ext_mreg_supported(dev) ||
+           !priv->sh->dv_regc0_mask)
+               return 0;
+       mcp_res = flow_mreg_add_copy_action(dev, 0, error);
+       if (!mcp_res)
+               return -rte_errno;
+       return 0;
+}
+
+/**
+ * Add a flow of copying flow metadata registers in RX_CP_TBL.
+ *
+ * All the flow having Q/RSS action should be split by
+ * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
+ * performs the following,
+ *   - CQE->flow_tag := reg_c[1] (MARK)
+ *   - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
+ * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
+ * but there should be a flow per each MARK ID set by MARK action.
+ *
+ * For the aforementioned reason, if there's a MARK action in flow's action
+ * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
+ * the MARK ID to CQE's flow_tag like,
+ *   - If reg_c[1] is mark_id,
+ *     flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
+ *
+ * For SET_META action which stores value in reg_c[0], as the destination is
+ * also a flow metadata register (reg_b), adding a default flow is enough. Zero
+ * MARK ID means the default flow. The default flow looks like,
+ *   - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
+ *
+ * @param dev
+ *   Pointer to Ethernet device.
+ * @param flow
+ *   Pointer to flow structure.
+ * @param[in] actions
+ *   Pointer to the list of actions.
+ * @param[out] error
+ *   Perform verbose error reporting if not NULL.
+ *
+ * @return
+ *   0 on success, negative value otherwise and rte_errno is set.
+ */
+static int
+flow_mreg_update_copy_table(struct rte_eth_dev *dev,
+                           struct rte_flow *flow,
+                           const struct rte_flow_action *actions,
+                           struct rte_flow_error *error)
+{
+       struct mlx5_priv *priv = dev->data->dev_private;
+       struct mlx5_dev_config *config = &priv->config;
+       struct mlx5_flow_mreg_copy_resource *mcp_res;
+       const struct rte_flow_action_mark *mark;
+
+       /* Check whether extensive metadata feature is engaged. */
+       if (!config->dv_flow_en ||
+           config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
+           !mlx5_flow_ext_mreg_supported(dev) ||
+           !priv->sh->dv_regc0_mask)
+               return 0;
+       /* Find MARK action. */
+       for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
+               switch (actions->type) {
+               case RTE_FLOW_ACTION_TYPE_FLAG:
+                       mcp_res = flow_mreg_add_copy_action
+                               (dev, MLX5_FLOW_MARK_DEFAULT, error);
+                       if (!mcp_res)
+                               return -rte_errno;
+                       flow->mreg_copy = mcp_res;
+                       if (dev->data->dev_started) {
+                               mcp_res->appcnt++;
+                               flow->copy_applied = 1;
+                       }
+                       return 0;
+               case RTE_FLOW_ACTION_TYPE_MARK:
+                       mark = (const struct rte_flow_action_mark *)
+                               actions->conf;
+                       mcp_res =
+                               flow_mreg_add_copy_action(dev, mark->id, error);
+                       if (!mcp_res)
+                               return -rte_errno;
+                       flow->mreg_copy = mcp_res;
+                       if (dev->data->dev_started) {
+                               mcp_res->appcnt++;
+                               flow->copy_applied = 1;
+                       }
+                       return 0;
+               default:
+                       break;
+               }
+       }
+       return 0;
+}
+
 #define MLX5_MAX_SPLIT_ACTIONS 24
 #define MLX5_MAX_SPLIT_ITEMS 24
 
@@ -3472,6 +3879,22 @@ flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
                if (ret < 0)
                        goto error;
        }
+       /*
+        * Update the metadata register copy table. If extensive
+        * metadata feature is enabled and registers are supported
+        * we might create the extra rte_flow for each unique
+        * MARK/FLAG action ID.
+        *
+        * The table is updated for ingress Flows only, because
+        * the egress Flows belong to the different device and
+        * copy table should be updated in peer NIC Rx domain.
+        */
+       if (attr->ingress &&
+           (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
+               ret = flow_mreg_update_copy_table(dev, flow, actions, error);
+               if (ret)
+                       goto error;
+       }
        if (dev->data->dev_started) {
                ret = flow_drv_apply(dev, flow, error);
                if (ret < 0)
@@ -3487,6 +3910,8 @@ error_before_flow:
                                     hairpin_id);
        return NULL;
 error:
+       assert(flow);
+       flow_mreg_del_copy_action(dev, flow);
        ret = rte_errno; /* Save rte_errno before cleanup. */
        if (flow->hairpin_flow_id)
                mlx5_flow_id_release(priv->sh->flow_id_pool,
@@ -3595,6 +4020,7 @@ flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
        flow_drv_destroy(dev, flow);
        if (list)
                TAILQ_REMOVE(list, flow, next);
+       flow_mreg_del_copy_action(dev, flow);
        rte_free(flow->fdir);
        rte_free(flow);
 }
@@ -3631,8 +4057,11 @@ mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
 {
        struct rte_flow *flow;
 
-       TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
+       TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
                flow_drv_remove(dev, flow);
+               flow_mreg_stop_copy_action(dev, flow);
+       }
+       flow_mreg_del_default_copy_action(dev);
        flow_rxq_flags_clear(dev);
 }
 
@@ -3654,7 +4083,15 @@ mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
        struct rte_flow_error error;
        int ret = 0;
 
+       /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
+       ret = flow_mreg_add_default_copy_action(dev, &error);
+       if (ret < 0)
+               return -rte_errno;
+       /* Apply Flows created by application. */
        TAILQ_FOREACH(flow, list, next) {
+               ret = flow_mreg_start_copy_action(dev, flow);
+               if (ret < 0)
+                       goto error;
                ret = flow_drv_apply(dev, flow, &error);
                if (ret < 0)
                        goto error;
index 8795f54..f84ea9d 100644 (file)
@@ -38,6 +38,7 @@ enum mlx5_rte_flow_item_type {
 enum mlx5_rte_flow_action_type {
        MLX5_RTE_FLOW_ACTION_TYPE_END = INT_MIN,
        MLX5_RTE_FLOW_ACTION_TYPE_TAG,
+       MLX5_RTE_FLOW_ACTION_TYPE_MARK,
        MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
 };
 
@@ -417,6 +418,21 @@ struct mlx5_flow_dv_push_vlan_action_resource {
        rte_be32_t vlan_tag; /**< VLAN tag value. */
 };
 
+/* Metadata register copy table entry. */
+struct mlx5_flow_mreg_copy_resource {
+       /*
+        * Hash list entry for copy table.
+        *  - Key is 32/64-bit MARK action ID.
+        *  - MUST be the first entry.
+        */
+       struct mlx5_hlist_entry hlist_ent;
+       LIST_ENTRY(mlx5_flow_mreg_copy_resource) next;
+       /* List entry for device flows. */
+       uint32_t refcnt; /* Reference counter. */
+       uint32_t appcnt; /* Apply/Remove counter. */
+       struct rte_flow *flow; /* Built flow for copy. */
+};
+
 /*
  * Max number of actions per DV flow.
  * See CREATE_FLOW_MAX_FLOW_ACTIONS_SUPPORTED
@@ -510,10 +526,13 @@ struct rte_flow {
        enum mlx5_flow_drv_type drv_type; /**< Driver type. */
        struct mlx5_flow_rss rss; /**< RSS context. */
        struct mlx5_flow_counter *counter; /**< Holds flow counter. */
+       struct mlx5_flow_mreg_copy_resource *mreg_copy;
+       /**< pointer to metadata register copy table resource. */
        LIST_HEAD(dev_flows, mlx5_flow) dev_flows;
        /**< Device flows that are part of the flow. */
        struct mlx5_fdir *fdir; /**< Pointer to associated FDIR if any. */
        uint32_t hairpin_flow_id; /**< The flow id used for hairpin. */
+       uint32_t copy_applied:1; /**< The MARK copy Flow os applied. */
 };
 
 typedef int (*mlx5_flow_validate_t)(struct rte_eth_dev *dev,
index 204e15e..9ca0ffe 100644 (file)
@@ -4086,8 +4086,11 @@ flow_dv_validate_attributes(struct rte_eth_dev *dev,
                                          NULL,
                                          "groups are not supported");
 #else
-       uint32_t max_group = attributes->transfer ? MLX5_MAX_TABLES_FDB :
-                                                   MLX5_MAX_TABLES;
+       uint32_t max_group = attributes->transfer ?
+                            MLX5_MAX_TABLES_FDB :
+                               external ?
+                               MLX5_MAX_TABLES_EXTERNAL :
+                               MLX5_MAX_TABLES;
        uint32_t table;
        int ret;
 
@@ -4721,6 +4724,7 @@ flow_dv_validate(struct rte_eth_dev *dev, const struct rte_flow_attr *attr,
                                                MLX5_FLOW_ACTION_DEC_TCP_ACK;
                        break;
                case MLX5_RTE_FLOW_ACTION_TYPE_TAG:
+               case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
                case MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG:
                        break;
                default:
@@ -6557,6 +6561,8 @@ __flow_dv_translate(struct rte_eth_dev *dev,
                                action_flags |= MLX5_FLOW_ACTION_MARK_EXT;
                                break;
                        }
+                       /* Fall-through */
+               case MLX5_RTE_FLOW_ACTION_TYPE_MARK:
                        /* Legacy (non-extensive) MARK action. */
                        tag_resource.tag = mlx5_flow_mark_set
                              (((const struct rte_flow_action_mark *)