{
struct mlx5_priv *priv = dev->data->dev_private;
struct mlx5_dev_config *config = &priv->config;
+ enum modify_reg start_reg;
switch (feature) {
case MLX5_HAIRPIN_RX:
case MLX5_METADATA_TX:
return REG_A;
case MLX5_METADATA_FDB:
- return REG_C_0;
+ switch (config->dv_xmeta_en) {
+ case MLX5_XMETA_MODE_LEGACY:
+ return REG_NONE;
+ case MLX5_XMETA_MODE_META16:
+ return REG_C_0;
+ case MLX5_XMETA_MODE_META32:
+ return REG_C_1;
+ }
+ break;
case MLX5_FLOW_MARK:
switch (config->dv_xmeta_en) {
case MLX5_XMETA_MODE_LEGACY:
}
break;
case MLX5_COPY_MARK:
- return REG_C_3;
+ case MLX5_MTR_SFX:
+ /*
+ * Metadata COPY_MARK register using is in meter suffix sub
+ * flow while with meter. It's safe to share the same register.
+ */
+ return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
+ case MLX5_MTR_COLOR:
+ RTE_ASSERT(priv->mtr_color_reg != REG_NONE);
+ return priv->mtr_color_reg;
case MLX5_APP_TAG:
/*
- * Suppose engaging reg_c_2 .. reg_c_7 registers.
- * reg_c_2 is reserved for coloring by meters.
- * reg_c_3 is reserved for split flows TAG.
+ * If meter is enable, it will engage two registers for color
+ * match and flow match. If meter color match is not using the
+ * REG_C_2, need to skip the REG_C_x be used by meter color
+ * match.
+ * If meter is disable, free to use all available registers.
*/
- if (id > (REG_C_7 - REG_C_4))
- return rte_flow_error_set
- (error, EINVAL,
- RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
- NULL, "invalid tag id");
- if (config->flow_mreg_c[id + REG_C_4 - REG_C_0] == REG_NONE)
- return rte_flow_error_set
- (error, ENOTSUP,
- RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
- NULL, "unsupported tag id");
- return config->flow_mreg_c[id + REG_C_4 - REG_C_0];
+ if (priv->mtr_color_reg != REG_NONE)
+ start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_3 :
+ REG_C_4;
+ else
+ start_reg = REG_C_2;
+ if (id > (REG_C_7 - start_reg))
+ return rte_flow_error_set(error, EINVAL,
+ RTE_FLOW_ERROR_TYPE_ITEM,
+ NULL, "invalid tag id");
+ if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
+ return rte_flow_error_set(error, ENOTSUP,
+ RTE_FLOW_ERROR_TYPE_ITEM,
+ NULL, "unsupported tag id");
+ /*
+ * This case means meter is using the REG_C_x great than 2.
+ * Take care not to conflict with meter color REG_C_x.
+ * If the available index REG_C_y >= REG_C_x, skip the
+ * color register.
+ */
+ if (start_reg == REG_C_3 && config->flow_mreg_c
+ [id + REG_C_3 - REG_C_0] >= priv->mtr_color_reg) {
+ if (config->flow_mreg_c[id + 1 + REG_C_3 - REG_C_0] !=
+ REG_NONE)
+ return config->flow_mreg_c
+ [id + 1 + REG_C_3 - REG_C_0];
+ return rte_flow_error_set(error, ENOTSUP,
+ RTE_FLOW_ERROR_TYPE_ITEM,
+ NULL, "unsupported tag id");
+ }
+ return config->flow_mreg_c[id + start_reg - REG_C_0];
}
assert(false);
return rte_flow_error_set(error, EINVAL,
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++;
}
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;
}
RTE_FLOW_ERROR_TYPE_ACTION_CONF,
NULL, "No queues configured");
for (i = 0; i != rss->queue_num; ++i) {
+ if (rss->queue[i] >= priv->rxqs_n)
+ return rte_flow_error_set
+ (error, EINVAL,
+ RTE_FLOW_ERROR_TYPE_ACTION_CONF,
+ &rss->queue[i], "queue index out of range");
if (!(*priv->rxqs)[rss->queue[i]])
return rte_flow_error_set
(error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
return 0;
}
+/* Allocate unique ID for the split Q/RSS subflows. */
+static uint32_t
+flow_qrss_get_id(struct rte_eth_dev *dev)
+{
+ struct mlx5_priv *priv = dev->data->dev_private;
+ uint32_t qrss_id, ret;
+
+ ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
+ if (ret)
+ return 0;
+ assert(qrss_id);
+ return qrss_id;
+}
+
+/* Free unique ID for the split Q/RSS subflows. */
+static void
+flow_qrss_free_id(struct rte_eth_dev *dev, uint32_t qrss_id)
+{
+ struct mlx5_priv *priv = dev->data->dev_private;
+
+ if (qrss_id)
+ mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
+}
+
+/**
+ * Release resource related QUEUE/RSS action split.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param flow
+ * Flow to release id's from.
+ */
+static void
+flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
+ struct rte_flow *flow)
+{
+ struct mlx5_flow *dev_flow;
+
+ LIST_FOREACH(dev_flow, &flow->dev_flows, next)
+ if (dev_flow->qrss_id)
+ flow_qrss_free_id(dev, dev_flow->qrss_id);
+}
+
static int
flow_null_validate(struct rte_eth_dev *dev __rte_unused,
const struct rte_flow_attr *attr __rte_unused,
const struct mlx5_flow_driver_ops *fops;
enum mlx5_flow_drv_type type = flow->drv_type;
+ flow_mreg_split_qrss_release(dev, flow);
assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
fops = flow_get_drv_ops(type);
fops->destroy(dev, flow);
return 0;
}
+/**
+ * Get port id item from the item list.
+ *
+ * @param[in] item
+ * Pointer to the list of items.
+ *
+ * @return
+ * Pointer to the port id item if exist, else return NULL.
+ */
+static const struct rte_flow_item *
+find_port_id_item(const struct rte_flow_item *item)
+{
+ assert(item);
+ for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
+ if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
+ return item;
+ }
+ return NULL;
+}
+
/**
* Get RSS action from the action list.
*
MLX5_EXPANSION_ROOT_OUTER;
}
+/**
+ * Get QUEUE/RSS action from the action list.
+ *
+ * @param[in] actions
+ * Pointer to the list of actions.
+ * @param[out] qrss
+ * Pointer to the return pointer.
+ * @param[out] qrss_type
+ * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
+ * if no QUEUE/RSS is found.
+ *
+ * @return
+ * Total number of actions.
+ */
+static int
+flow_parse_qrss_action(const struct rte_flow_action actions[],
+ const struct rte_flow_action **qrss)
+{
+ int actions_n = 0;
+
+ for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
+ switch (actions->type) {
+ case RTE_FLOW_ACTION_TYPE_QUEUE:
+ case RTE_FLOW_ACTION_TYPE_RSS:
+ *qrss = actions;
+ break;
+ default:
+ break;
+ }
+ actions_n++;
+ }
+ /* Count RTE_FLOW_ACTION_TYPE_END. */
+ return actions_n + 1;
+}
+
+/**
+ * Check meter action from the action list.
+ *
+ * @param[in] actions
+ * Pointer to the list of actions.
+ * @param[out] mtr
+ * Pointer to the meter exist flag.
+ *
+ * @return
+ * Total number of actions.
+ */
+static int
+flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
+{
+ int actions_n = 0;
+
+ assert(mtr);
+ *mtr = 0;
+ for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
+ switch (actions->type) {
+ case RTE_FLOW_ACTION_TYPE_METER:
+ *mtr = 1;
+ break;
+ default:
+ break;
+ }
+ actions_n++;
+ }
+ /* Count RTE_FLOW_ACTION_TYPE_END. */
+ return actions_n + 1;
+}
+
/**
* Check if the flow should be splited due to hairpin.
* The reason for the split is that in current HW we can't
switch (actions->type) {
case RTE_FLOW_ACTION_TYPE_QUEUE:
queue = actions->conf;
+ if (queue == NULL)
+ return 0;
if (mlx5_rxq_get_type(dev, queue->index) !=
MLX5_RXQ_TYPE_HAIRPIN)
return 0;
break;
case RTE_FLOW_ACTION_TYPE_RSS:
rss = actions->conf;
+ if (rss == NULL || rss->queue_num == 0)
+ return 0;
if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
MLX5_RXQ_TYPE_HAIRPIN)
return 0;
return 0;
}
-#define MLX5_MAX_SPLIT_ACTIONS 24
-#define MLX5_MAX_SPLIT_ITEMS 24
+/* 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);
/**
- * Split the hairpin flow.
- * Since HW can't support encap on Rx we move the encap to Tx.
- * If the count action is after the encap then we also
- * move the count action. in this case the count will also measure
- * the outer bytes.
+ * 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[in] actions
- * Associated actions (list terminated by the END action).
- * @param[out] actions_rx
- * Rx flow actions.
- * @param[out] actions_tx
- * Tx flow actions..
- * @param[out] pattern_tx
- * The pattern items for the Tx flow.
- * @param[out] flow_id
- * The flow ID connected to this flow.
+ * @param mark_id
+ * ID of MARK action, zero means default flow for META.
+ * @param[out] error
+ * Perform verbose error reporting if not NULL.
*
* @return
- * 0 on success.
+ * Associated resource on success, NULL otherwise and rte_errno is set.
*/
-static int
-flow_hairpin_split(struct rte_eth_dev *dev,
- const struct rte_flow_action actions[],
- struct rte_flow_action actions_rx[],
- struct rte_flow_action actions_tx[],
- struct rte_flow_item pattern_tx[],
- uint32_t *flow_id)
+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;
- const struct rte_flow_action_raw_encap *raw_encap;
- const struct rte_flow_action_raw_decap *raw_decap;
- struct mlx5_rte_flow_action_set_tag *set_tag;
- struct rte_flow_action *tag_action;
- struct mlx5_rte_flow_item_tag *tag_item;
- struct rte_flow_item *item;
- char *addr;
- int encap = 0;
+ 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;
- mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
- for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
- switch (actions->type) {
- case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
- case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
- rte_memcpy(actions_tx, actions,
- sizeof(struct rte_flow_action));
- actions_tx++;
- break;
- case RTE_FLOW_ACTION_TYPE_COUNT:
- if (encap) {
- rte_memcpy(actions_tx, actions,
- sizeof(struct rte_flow_action));
- actions_tx++;
- } else {
- rte_memcpy(actions_rx, actions,
- sizeof(struct rte_flow_action));
- actions_rx++;
- }
- break;
- case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
- raw_encap = actions->conf;
- if (raw_encap->size >
- (sizeof(struct rte_flow_item_eth) +
- sizeof(struct rte_flow_item_ipv4))) {
- memcpy(actions_tx, actions,
- sizeof(struct rte_flow_action));
- actions_tx++;
- encap = 1;
- } else {
- rte_memcpy(actions_rx, actions,
- sizeof(struct rte_flow_action));
- actions_rx++;
- }
- break;
- case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
- raw_decap = actions->conf;
- if (raw_decap->size <
- (sizeof(struct rte_flow_item_eth) +
- sizeof(struct rte_flow_item_ipv4))) {
- memcpy(actions_tx, actions,
- sizeof(struct rte_flow_action));
- actions_tx++;
- } else {
- rte_memcpy(actions_rx, actions,
- sizeof(struct rte_flow_action));
- actions_rx++;
- }
- break;
- default:
- rte_memcpy(actions_rx, actions,
- sizeof(struct rte_flow_action));
- actions_rx++;
- break;
- }
+ /* 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;
}
- /* Add set meta action and end action for the Rx flow. */
- tag_action = actions_rx;
- tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
- actions_rx++;
- rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
- actions_rx++;
- set_tag = (void *)actions_rx;
- set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
- assert(set_tag->id > REG_NONE);
- set_tag->data = *flow_id;
- tag_action->conf = set_tag;
- /* Create Tx item list. */
- rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
- addr = (void *)&pattern_tx[2];
- item = pattern_tx;
- item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
- tag_item = (void *)addr;
- tag_item->data = *flow_id;
- tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
- assert(set_tag->id > REG_NONE);
- item->spec = tag_item;
- addr += sizeof(struct mlx5_rte_flow_item_tag);
- tag_item = (void *)addr;
- tag_item->data = UINT32_MAX;
- tag_item->id = UINT16_MAX;
- item->mask = tag_item;
- addr += sizeof(struct mlx5_rte_flow_item_tag);
- item->last = NULL;
- item++;
- item->type = RTE_FLOW_ITEM_TYPE_END;
- return 0;
+ /* 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
+
+/**
+ * Split the hairpin flow.
+ * Since HW can't support encap on Rx we move the encap to Tx.
+ * If the count action is after the encap then we also
+ * move the count action. in this case the count will also measure
+ * the outer bytes.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param[in] actions
+ * Associated actions (list terminated by the END action).
+ * @param[out] actions_rx
+ * Rx flow actions.
+ * @param[out] actions_tx
+ * Tx flow actions..
+ * @param[out] pattern_tx
+ * The pattern items for the Tx flow.
+ * @param[out] flow_id
+ * The flow ID connected to this flow.
+ *
+ * @return
+ * 0 on success.
+ */
+static int
+flow_hairpin_split(struct rte_eth_dev *dev,
+ const struct rte_flow_action actions[],
+ struct rte_flow_action actions_rx[],
+ struct rte_flow_action actions_tx[],
+ struct rte_flow_item pattern_tx[],
+ uint32_t *flow_id)
+{
+ struct mlx5_priv *priv = dev->data->dev_private;
+ const struct rte_flow_action_raw_encap *raw_encap;
+ const struct rte_flow_action_raw_decap *raw_decap;
+ struct mlx5_rte_flow_action_set_tag *set_tag;
+ struct rte_flow_action *tag_action;
+ struct mlx5_rte_flow_item_tag *tag_item;
+ struct rte_flow_item *item;
+ char *addr;
+ int encap = 0;
+
+ mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
+ for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
+ switch (actions->type) {
+ case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
+ case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
+ rte_memcpy(actions_tx, actions,
+ sizeof(struct rte_flow_action));
+ actions_tx++;
+ break;
+ case RTE_FLOW_ACTION_TYPE_COUNT:
+ if (encap) {
+ rte_memcpy(actions_tx, actions,
+ sizeof(struct rte_flow_action));
+ actions_tx++;
+ } else {
+ rte_memcpy(actions_rx, actions,
+ sizeof(struct rte_flow_action));
+ actions_rx++;
+ }
+ break;
+ case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
+ raw_encap = actions->conf;
+ if (raw_encap->size >
+ (sizeof(struct rte_flow_item_eth) +
+ sizeof(struct rte_flow_item_ipv4))) {
+ memcpy(actions_tx, actions,
+ sizeof(struct rte_flow_action));
+ actions_tx++;
+ encap = 1;
+ } else {
+ rte_memcpy(actions_rx, actions,
+ sizeof(struct rte_flow_action));
+ actions_rx++;
+ }
+ break;
+ case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
+ raw_decap = actions->conf;
+ if (raw_decap->size <
+ (sizeof(struct rte_flow_item_eth) +
+ sizeof(struct rte_flow_item_ipv4))) {
+ memcpy(actions_tx, actions,
+ sizeof(struct rte_flow_action));
+ actions_tx++;
+ } else {
+ rte_memcpy(actions_rx, actions,
+ sizeof(struct rte_flow_action));
+ actions_rx++;
+ }
+ break;
+ default:
+ rte_memcpy(actions_rx, actions,
+ sizeof(struct rte_flow_action));
+ actions_rx++;
+ break;
+ }
+ }
+ /* Add set meta action and end action for the Rx flow. */
+ tag_action = actions_rx;
+ tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
+ actions_rx++;
+ rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
+ actions_rx++;
+ set_tag = (void *)actions_rx;
+ set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
+ assert(set_tag->id > REG_NONE);
+ set_tag->data = *flow_id;
+ tag_action->conf = set_tag;
+ /* Create Tx item list. */
+ rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
+ addr = (void *)&pattern_tx[2];
+ item = pattern_tx;
+ item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
+ tag_item = (void *)addr;
+ tag_item->data = *flow_id;
+ tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
+ assert(set_tag->id > REG_NONE);
+ item->spec = tag_item;
+ addr += sizeof(struct mlx5_rte_flow_item_tag);
+ tag_item = (void *)addr;
+ tag_item->data = UINT32_MAX;
+ tag_item->id = UINT16_MAX;
+ item->mask = tag_item;
+ addr += sizeof(struct mlx5_rte_flow_item_tag);
+ item->last = NULL;
+ item++;
+ item->type = RTE_FLOW_ITEM_TYPE_END;
+ return 0;
+}
+
+/**
+ * The last stage of splitting chain, just creates the subflow
+ * without any modification.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param[in] flow
+ * Parent flow structure pointer.
+ * @param[in, out] sub_flow
+ * Pointer to return the created subflow, may be NULL.
+ * @param[in] attr
+ * Flow rule attributes.
+ * @param[in] items
+ * Pattern specification (list terminated by the END pattern item).
+ * @param[in] actions
+ * Associated actions (list terminated by the END action).
+ * @param[in] external
+ * This flow rule is created by request external to PMD.
+ * @param[out] error
+ * Perform verbose error reporting if not NULL.
+ * @return
+ * 0 on success, negative value otherwise
+ */
+static int
+flow_create_split_inner(struct rte_eth_dev *dev,
+ struct rte_flow *flow,
+ struct mlx5_flow **sub_flow,
+ 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)
+{
+ struct mlx5_flow *dev_flow;
+
+ dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
+ if (!dev_flow)
+ return -rte_errno;
+ dev_flow->flow = flow;
+ dev_flow->external = external;
+ /* Subflow object was created, we must include one in the list. */
+ LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
+ if (sub_flow)
+ *sub_flow = dev_flow;
+ return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
+}
+
+/**
+ * Split the meter flow.
+ *
+ * As meter flow will split to three sub flow, other than meter
+ * action, the other actions make sense to only meter accepts
+ * the packet. If it need to be dropped, no other additional
+ * actions should be take.
+ *
+ * One kind of special action which decapsulates the L3 tunnel
+ * header will be in the prefix sub flow, as not to take the
+ * L3 tunnel header into account.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param[in] actions
+ * Associated actions (list terminated by the END action).
+ * @param[out] actions_sfx
+ * Suffix flow actions.
+ * @param[out] actions_pre
+ * Prefix flow actions.
+ * @param[out] pattern_sfx
+ * The pattern items for the suffix flow.
+ * @param[out] tag_sfx
+ * Pointer to suffix flow tag.
+ *
+ * @return
+ * 0 on success.
+ */
+static int
+flow_meter_split_prep(struct rte_eth_dev *dev,
+ const struct rte_flow_action actions[],
+ struct rte_flow_action actions_sfx[],
+ struct rte_flow_action actions_pre[])
+{
+ struct rte_flow_action *tag_action;
+ struct mlx5_rte_flow_action_set_tag *set_tag;
+ struct rte_flow_error error;
+ const struct rte_flow_action_raw_encap *raw_encap;
+ const struct rte_flow_action_raw_decap *raw_decap;
+ uint32_t tag_id;
+
+ /* Add the extra tag action first. */
+ tag_action = actions_pre;
+ tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
+ actions_pre++;
+ /* Prepare the actions for prefix and suffix flow. */
+ for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
+ switch (actions->type) {
+ case RTE_FLOW_ACTION_TYPE_METER:
+ case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
+ case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
+ memcpy(actions_pre, actions,
+ sizeof(struct rte_flow_action));
+ actions_pre++;
+ break;
+ case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
+ raw_encap = actions->conf;
+ if (raw_encap->size >
+ (sizeof(struct rte_flow_item_eth) +
+ sizeof(struct rte_flow_item_ipv4))) {
+ memcpy(actions_sfx, actions,
+ sizeof(struct rte_flow_action));
+ actions_sfx++;
+ } else {
+ rte_memcpy(actions_pre, actions,
+ sizeof(struct rte_flow_action));
+ actions_pre++;
+ }
+ break;
+ case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
+ raw_decap = actions->conf;
+ /* Size 0 decap means 50 bytes as vxlan decap. */
+ if (raw_decap->size && (raw_decap->size <
+ (sizeof(struct rte_flow_item_eth) +
+ sizeof(struct rte_flow_item_ipv4)))) {
+ memcpy(actions_sfx, actions,
+ sizeof(struct rte_flow_action));
+ actions_sfx++;
+ } else {
+ rte_memcpy(actions_pre, actions,
+ sizeof(struct rte_flow_action));
+ actions_pre++;
+ }
+ break;
+ default:
+ memcpy(actions_sfx, actions,
+ sizeof(struct rte_flow_action));
+ actions_sfx++;
+ break;
+ }
+ }
+ /* Add end action to the actions. */
+ actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
+ actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
+ actions_pre++;
+ /* Set the tag. */
+ set_tag = (void *)actions_pre;
+ set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
+ /*
+ * Get the id from the qrss_pool to make qrss share the id with meter.
+ */
+ tag_id = flow_qrss_get_id(dev);
+ set_tag->data = rte_cpu_to_be_32(tag_id);
+ tag_action->conf = set_tag;
+ return tag_id;
+}
+
+/**
+ * Split action list having QUEUE/RSS for metadata register copy.
+ *
+ * Once Q/RSS action is detected in user's action list, the flow action
+ * should be split in order to copy metadata registers, which will happen in
+ * RX_CP_TBL like,
+ * - CQE->flow_tag := reg_c[1] (MARK)
+ * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
+ * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
+ * This is because the last action of each flow must be a terminal action
+ * (QUEUE, RSS or DROP).
+ *
+ * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
+ * stored and kept in the mlx5_flow structure per each sub_flow.
+ *
+ * The Q/RSS action is replaced with,
+ * - SET_TAG, setting the allocated flow ID to reg_c[2].
+ * And the following JUMP action is added at the end,
+ * - JUMP, to RX_CP_TBL.
+ *
+ * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
+ * flow_create_split_metadata() routine. The flow will look like,
+ * - If flow ID matches (reg_c[2]), perform Q/RSS.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param[out] split_actions
+ * Pointer to store split actions to jump to CP_TBL.
+ * @param[in] actions
+ * Pointer to the list of original flow actions.
+ * @param[in] qrss
+ * Pointer to the Q/RSS action.
+ * @param[in] actions_n
+ * Number of original actions.
+ * @param[out] error
+ * Perform verbose error reporting if not NULL.
+ *
+ * @return
+ * non-zero unique flow_id on success, otherwise 0 and
+ * error/rte_error are set.
+ */
+static uint32_t
+flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
+ struct rte_flow_action *split_actions,
+ const struct rte_flow_action *actions,
+ const struct rte_flow_action *qrss,
+ int actions_n, struct rte_flow_error *error)
+{
+ struct mlx5_rte_flow_action_set_tag *set_tag;
+ struct rte_flow_action_jump *jump;
+ const int qrss_idx = qrss - actions;
+ uint32_t flow_id = 0;
+ int ret = 0;
+
+ /*
+ * Given actions will be split
+ * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
+ * - Add jump to mreg CP_TBL.
+ * As a result, there will be one more action.
+ */
+ ++actions_n;
+ memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
+ set_tag = (void *)(split_actions + actions_n);
+ /*
+ * If tag action is not set to void(it means we are not the meter
+ * suffix flow), add the tag action. Since meter suffix flow already
+ * has the tag added.
+ */
+ if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
+ /*
+ * Allocate the new subflow ID. This one is unique within
+ * device and not shared with representors. Otherwise,
+ * we would have to resolve multi-thread access synch
+ * issue. Each flow on the shared device is appended
+ * with source vport identifier, so the resulting
+ * flows will be unique in the shared (by master and
+ * representors) domain even if they have coinciding
+ * IDs.
+ */
+ flow_id = flow_qrss_get_id(dev);
+ if (!flow_id)
+ return rte_flow_error_set(error, ENOMEM,
+ RTE_FLOW_ERROR_TYPE_ACTION,
+ NULL, "can't allocate id "
+ "for split Q/RSS subflow");
+ /* Internal SET_TAG action to set flow ID. */
+ *set_tag = (struct mlx5_rte_flow_action_set_tag){
+ .data = flow_id,
+ };
+ ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
+ if (ret < 0)
+ return ret;
+ set_tag->id = ret;
+ /* Construct new actions array. */
+ /* Replace QUEUE/RSS action. */
+ split_actions[qrss_idx] = (struct rte_flow_action){
+ .type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
+ .conf = set_tag,
+ };
+ }
+ /* JUMP action to jump to mreg copy table (CP_TBL). */
+ jump = (void *)(set_tag + 1);
+ *jump = (struct rte_flow_action_jump){
+ .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
+ };
+ split_actions[actions_n - 2] = (struct rte_flow_action){
+ .type = RTE_FLOW_ACTION_TYPE_JUMP,
+ .conf = jump,
+ };
+ split_actions[actions_n - 1] = (struct rte_flow_action){
+ .type = RTE_FLOW_ACTION_TYPE_END,
+ };
+ return flow_id;
}
/**
- * The last stage of splitting chain, just creates the subflow
- * without any modification.
+ * Extend the given action list for Tx metadata copy.
+ *
+ * Copy the given action list to the ext_actions and add flow metadata register
+ * copy action in order to copy reg_a set by WQE to reg_c[0].
+ *
+ * @param[out] ext_actions
+ * Pointer to the extended action list.
+ * @param[in] actions
+ * Pointer to the list of actions.
+ * @param[in] actions_n
+ * Number of actions in the list.
+ * @param[out] error
+ * Perform verbose error reporting if not NULL.
+ *
+ * @return
+ * 0 on success, negative value otherwise
+ */
+static int
+flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
+ struct rte_flow_action *ext_actions,
+ const struct rte_flow_action *actions,
+ int actions_n, struct rte_flow_error *error)
+{
+ struct mlx5_flow_action_copy_mreg *cp_mreg =
+ (struct mlx5_flow_action_copy_mreg *)
+ (ext_actions + actions_n + 1);
+ int ret;
+
+ ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
+ if (ret < 0)
+ return ret;
+ cp_mreg->dst = ret;
+ ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
+ if (ret < 0)
+ return ret;
+ cp_mreg->src = ret;
+ memcpy(ext_actions, actions,
+ sizeof(*ext_actions) * actions_n);
+ ext_actions[actions_n - 1] = (struct rte_flow_action){
+ .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
+ .conf = cp_mreg,
+ };
+ ext_actions[actions_n] = (struct rte_flow_action){
+ .type = RTE_FLOW_ACTION_TYPE_END,
+ };
+ return 0;
+}
+
+/**
+ * The splitting for metadata feature.
+ *
+ * - Q/RSS action on NIC Rx should be split in order to pass by
+ * the mreg copy table (RX_CP_TBL) and then it jumps to the
+ * action table (RX_ACT_TBL) which has the split Q/RSS action.
+ *
+ * - All the actions on NIC Tx should have a mreg copy action to
+ * copy reg_a from WQE to reg_c[0].
*
* @param dev
* Pointer to Ethernet device.
* @param[in] flow
* Parent flow structure pointer.
- * @param[in, out] sub_flow
- * Pointer to return the created subflow, may be NULL.
* @param[in] attr
* Flow rule attributes.
* @param[in] items
* 0 on success, negative value otherwise
*/
static int
-flow_create_split_inner(struct rte_eth_dev *dev,
- struct rte_flow *flow,
- struct mlx5_flow **sub_flow,
- 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)
+flow_create_split_metadata(struct rte_eth_dev *dev,
+ struct rte_flow *flow,
+ 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)
{
- struct mlx5_flow *dev_flow;
+ struct mlx5_priv *priv = dev->data->dev_private;
+ struct mlx5_dev_config *config = &priv->config;
+ const struct rte_flow_action *qrss = NULL;
+ struct rte_flow_action *ext_actions = NULL;
+ struct mlx5_flow *dev_flow = NULL;
+ uint32_t qrss_id = 0;
+ int mtr_sfx = 0;
+ size_t act_size;
+ int actions_n;
+ int ret;
- dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
- if (!dev_flow)
- return -rte_errno;
- dev_flow->flow = flow;
- dev_flow->external = external;
- /* Subflow object was created, we must include one in the list. */
- LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
- if (sub_flow)
- *sub_flow = dev_flow;
- return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
+ /* 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))
+ return flow_create_split_inner(dev, flow, NULL, attr, items,
+ actions, external, error);
+ actions_n = flow_parse_qrss_action(actions, &qrss);
+ if (qrss) {
+ /* Exclude hairpin flows from splitting. */
+ if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
+ const struct rte_flow_action_queue *queue;
+
+ queue = qrss->conf;
+ if (mlx5_rxq_get_type(dev, queue->index) ==
+ MLX5_RXQ_TYPE_HAIRPIN)
+ qrss = NULL;
+ } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
+ const struct rte_flow_action_rss *rss;
+
+ rss = qrss->conf;
+ if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
+ MLX5_RXQ_TYPE_HAIRPIN)
+ qrss = NULL;
+ }
+ }
+ if (qrss) {
+ /* Check if it is in meter suffix table. */
+ mtr_sfx = attr->group == (attr->transfer ?
+ (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
+ MLX5_FLOW_TABLE_LEVEL_SUFFIX);
+ /*
+ * Q/RSS action on NIC Rx should be split in order to pass by
+ * the mreg copy table (RX_CP_TBL) and then it jumps to the
+ * action table (RX_ACT_TBL) which has the split Q/RSS action.
+ */
+ act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
+ sizeof(struct rte_flow_action_set_tag) +
+ sizeof(struct rte_flow_action_jump);
+ ext_actions = rte_zmalloc(__func__, act_size, 0);
+ if (!ext_actions)
+ return rte_flow_error_set(error, ENOMEM,
+ RTE_FLOW_ERROR_TYPE_ACTION,
+ NULL, "no memory to split "
+ "metadata flow");
+ /*
+ * If we are the suffix flow of meter, tag already exist.
+ * Set the tag action to void.
+ */
+ if (mtr_sfx)
+ ext_actions[qrss - actions].type =
+ RTE_FLOW_ACTION_TYPE_VOID;
+ else
+ ext_actions[qrss - actions].type =
+ MLX5_RTE_FLOW_ACTION_TYPE_TAG;
+ /*
+ * Create the new actions list with removed Q/RSS action
+ * and appended set tag and jump to register copy table
+ * (RX_CP_TBL). We should preallocate unique tag ID here
+ * in advance, because it is needed for set tag action.
+ */
+ qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
+ qrss, actions_n, error);
+ if (!mtr_sfx && !qrss_id) {
+ ret = -rte_errno;
+ goto exit;
+ }
+ } else if (attr->egress && !attr->transfer) {
+ /*
+ * All the actions on NIC Tx should have a metadata register
+ * copy action to copy reg_a from WQE to reg_c[meta]
+ */
+ act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
+ sizeof(struct mlx5_flow_action_copy_mreg);
+ ext_actions = rte_zmalloc(__func__, act_size, 0);
+ if (!ext_actions)
+ return rte_flow_error_set(error, ENOMEM,
+ RTE_FLOW_ERROR_TYPE_ACTION,
+ NULL, "no memory to split "
+ "metadata flow");
+ /* Create the action list appended with copy register. */
+ ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
+ actions_n, error);
+ if (ret < 0)
+ goto exit;
+ }
+ /* Add the unmodified original or prefix subflow. */
+ ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
+ ext_actions ? ext_actions : actions,
+ external, error);
+ if (ret < 0)
+ goto exit;
+ assert(dev_flow);
+ if (qrss) {
+ const struct rte_flow_attr q_attr = {
+ .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
+ .ingress = 1,
+ };
+ /* Internal PMD action to set register. */
+ struct mlx5_rte_flow_item_tag q_tag_spec = {
+ .data = qrss_id,
+ .id = 0,
+ };
+ struct rte_flow_item q_items[] = {
+ {
+ .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
+ .spec = &q_tag_spec,
+ .last = NULL,
+ .mask = NULL,
+ },
+ {
+ .type = RTE_FLOW_ITEM_TYPE_END,
+ },
+ };
+ struct rte_flow_action q_actions[] = {
+ {
+ .type = qrss->type,
+ .conf = qrss->conf,
+ },
+ {
+ .type = RTE_FLOW_ACTION_TYPE_END,
+ },
+ };
+ uint64_t hash_fields = dev_flow->hash_fields;
+
+ /*
+ * Configure the tag item only if there is no meter subflow.
+ * Since tag is already marked in the meter suffix subflow
+ * we can just use the meter suffix items as is.
+ */
+ if (qrss_id) {
+ /* Not meter subflow. */
+ assert(!mtr_sfx);
+ /*
+ * Put unique id in prefix flow due to it is destroyed
+ * after suffix flow and id will be freed after there
+ * is no actual flows with this id and identifier
+ * reallocation becomes possible (for example, for
+ * other flows in other threads).
+ */
+ dev_flow->qrss_id = qrss_id;
+ qrss_id = 0;
+ ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
+ error);
+ if (ret < 0)
+ goto exit;
+ q_tag_spec.id = ret;
+ }
+ dev_flow = NULL;
+ /* Add suffix subflow to execute Q/RSS. */
+ ret = flow_create_split_inner(dev, flow, &dev_flow,
+ &q_attr, mtr_sfx ? items :
+ q_items, q_actions,
+ external, error);
+ if (ret < 0)
+ goto exit;
+ assert(dev_flow);
+ dev_flow->hash_fields = hash_fields;
+ }
+
+exit:
+ /*
+ * We do not destroy the partially created sub_flows in case of error.
+ * These ones are included into parent flow list and will be destroyed
+ * by flow_drv_destroy.
+ */
+ flow_qrss_free_id(dev, qrss_id);
+ rte_free(ext_actions);
+ return ret;
+}
+
+/**
+ * The splitting for meter feature.
+ *
+ * - The meter flow will be split to two flows as prefix and
+ * suffix flow. The packets make sense only it pass the prefix
+ * meter action.
+ *
+ * - Reg_C_5 is used for the packet to match betweend prefix and
+ * suffix flow.
+ *
+ * @param dev
+ * Pointer to Ethernet device.
+ * @param[in] flow
+ * Parent flow structure pointer.
+ * @param[in] attr
+ * Flow rule attributes.
+ * @param[in] items
+ * Pattern specification (list terminated by the END pattern item).
+ * @param[in] actions
+ * Associated actions (list terminated by the END action).
+ * @param[in] external
+ * This flow rule is created by request external to PMD.
+ * @param[out] error
+ * Perform verbose error reporting if not NULL.
+ * @return
+ * 0 on success, negative value otherwise
+ */
+static int
+flow_create_split_meter(struct rte_eth_dev *dev,
+ struct rte_flow *flow,
+ 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)
+{
+ struct mlx5_priv *priv = dev->data->dev_private;
+ struct rte_flow_action *sfx_actions = NULL;
+ struct rte_flow_action *pre_actions = NULL;
+ struct rte_flow_item *sfx_items = NULL;
+ const struct rte_flow_item *sfx_port_id_item;
+ struct mlx5_flow *dev_flow = NULL;
+ struct rte_flow_attr sfx_attr = *attr;
+ uint32_t mtr = 0;
+ uint32_t mtr_tag_id = 0;
+ size_t act_size;
+ size_t item_size;
+ int actions_n = 0;
+ int ret;
+
+ if (priv->mtr_en)
+ actions_n = flow_check_meter_action(actions, &mtr);
+ if (mtr) {
+ struct mlx5_rte_flow_item_tag *tag_spec;
+ /* The five prefix actions: meter, decap, encap, tag, end. */
+ act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
+ sizeof(struct rte_flow_action_set_tag);
+ /* tag, end. */
+#define METER_SUFFIX_ITEM 3
+ item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
+ sizeof(struct mlx5_rte_flow_item_tag);
+ sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
+ if (!sfx_actions)
+ return rte_flow_error_set(error, ENOMEM,
+ RTE_FLOW_ERROR_TYPE_ACTION,
+ NULL, "no memory to split "
+ "meter flow");
+ pre_actions = sfx_actions + actions_n;
+ mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
+ pre_actions);
+ if (!mtr_tag_id) {
+ ret = -rte_errno;
+ goto exit;
+ }
+ /* Add the prefix subflow. */
+ ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
+ pre_actions, external, error);
+ if (ret) {
+ ret = -rte_errno;
+ goto exit;
+ }
+ dev_flow->mtr_flow_id = mtr_tag_id;
+ /* Prepare the suffix flow match pattern. */
+ sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
+ act_size);
+ tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
+ METER_SUFFIX_ITEM);
+ tag_spec->data = rte_cpu_to_be_32(dev_flow->mtr_flow_id);
+ tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
+ error);
+ sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
+ sfx_items->spec = tag_spec;
+ sfx_items->last = NULL;
+ sfx_items->mask = NULL;
+ sfx_items++;
+ sfx_port_id_item = find_port_id_item(items);
+ if (sfx_port_id_item) {
+ memcpy(sfx_items, sfx_port_id_item,
+ sizeof(*sfx_items));
+ sfx_items++;
+ }
+ sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
+ sfx_items -= METER_SUFFIX_ITEM;
+ /* Setting the sfx group atrr. */
+ sfx_attr.group = sfx_attr.transfer ?
+ (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
+ MLX5_FLOW_TABLE_LEVEL_SUFFIX;
+ }
+ /* Add the prefix subflow. */
+ ret = flow_create_split_metadata(dev, flow, &sfx_attr,
+ sfx_items ? sfx_items : items,
+ sfx_actions ? sfx_actions : actions,
+ external, error);
+exit:
+ if (sfx_actions)
+ rte_free(sfx_actions);
+ return ret;
}
/**
{
int ret;
- ret = flow_create_split_inner(dev, flow, NULL, attr, items,
- actions, external, error);
+ ret = flow_create_split_meter(dev, flow, attr, items,
+ actions, external, error);
assert(ret <= 0);
return ret;
}
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)
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,
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);
}
{
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);
}
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;
return 0;
}
+/**
+ * Create the needed meter and suffix tables.
+ *
+ * @param[in] dev
+ * Pointer to Ethernet device.
+ * @param[in] fm
+ * Pointer to the flow meter.
+ *
+ * @return
+ * Pointer to table set on success, NULL otherwise.
+ */
+struct mlx5_meter_domains_infos *
+mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
+ const struct mlx5_flow_meter *fm)
+{
+ const struct mlx5_flow_driver_ops *fops;
+
+ fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
+ return fops->create_mtr_tbls(dev, fm);
+}
+
+/**
+ * Destroy the meter table set.
+ *
+ * @param[in] dev
+ * Pointer to Ethernet device.
+ * @param[in] tbl
+ * Pointer to the meter table set.
+ *
+ * @return
+ * 0 on success.
+ */
+int
+mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
+ struct mlx5_meter_domains_infos *tbls)
+{
+ const struct mlx5_flow_driver_ops *fops;
+
+ fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
+ return fops->destroy_mtr_tbls(dev, tbls);
+}
+
+/**
+ * Create policer rules.
+ *
+ * @param[in] dev
+ * Pointer to Ethernet device.
+ * @param[in] fm
+ * Pointer to flow meter structure.
+ * @param[in] attr
+ * Pointer to flow attributes.
+ *
+ * @return
+ * 0 on success, -1 otherwise.
+ */
+int
+mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
+ struct mlx5_flow_meter *fm,
+ const struct rte_flow_attr *attr)
+{
+ const struct mlx5_flow_driver_ops *fops;
+
+ fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
+ return fops->create_policer_rules(dev, fm, attr);
+}
+
+/**
+ * Destroy policer rules.
+ *
+ * @param[in] fm
+ * Pointer to flow meter structure.
+ * @param[in] attr
+ * Pointer to flow attributes.
+ *
+ * @return
+ * 0 on success, -1 otherwise.
+ */
+int
+mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
+ struct mlx5_flow_meter *fm,
+ const struct rte_flow_attr *attr)
+{
+ const struct mlx5_flow_driver_ops *fops;
+
+ fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
+ return fops->destroy_policer_rules(dev, fm, attr);
+}
+
+/**
+ * Allocate a counter.
+ *
+ * @param[in] dev
+ * Pointer to Ethernet device structure.
+ *
+ * @return
+ * Pointer to allocated counter on success, NULL otherwise.
+ */
+struct mlx5_flow_counter *
+mlx5_counter_alloc(struct rte_eth_dev *dev)
+{
+ const struct mlx5_flow_driver_ops *fops;
+ struct rte_flow_attr attr = { .transfer = 0 };
+
+ if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
+ fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
+ return fops->counter_alloc(dev);
+ }
+ DRV_LOG(ERR,
+ "port %u counter allocate is not supported.",
+ dev->data->port_id);
+ return NULL;
+}
+
+/**
+ * Free a counter.
+ *
+ * @param[in] dev
+ * Pointer to Ethernet device structure.
+ * @param[in] cnt
+ * Pointer to counter to be free.
+ */
+void
+mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
+{
+ const struct mlx5_flow_driver_ops *fops;
+ struct rte_flow_attr attr = { .transfer = 0 };
+
+ if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
+ fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
+ fops->counter_free(dev, cnt);
+ return;
+ }
+ DRV_LOG(ERR,
+ "port %u counter free is not supported.",
+ dev->data->port_id);
+}
+
+/**
+ * Query counter statistics.
+ *
+ * @param[in] dev
+ * Pointer to Ethernet device structure.
+ * @param[in] cnt
+ * Pointer to counter to query.
+ * @param[in] clear
+ * Set to clear counter statistics.
+ * @param[out] pkts
+ * The counter hits packets number to save.
+ * @param[out] bytes
+ * The counter hits bytes number to save.
+ *
+ * @return
+ * 0 on success, a negative errno value otherwise.
+ */
+int
+mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
+ bool clear, uint64_t *pkts, uint64_t *bytes)
+{
+ const struct mlx5_flow_driver_ops *fops;
+ struct rte_flow_attr attr = { .transfer = 0 };
+
+ if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
+ fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
+ return fops->counter_query(dev, cnt, clear, pkts, bytes);
+ }
+ DRV_LOG(ERR,
+ "port %u counter query is not supported.",
+ dev->data->port_id);
+ return -ENOTSUP;
+}
+
#define MLX5_POOL_QUERY_FREQ_US 1000000
/**