X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fmlx5%2Fmlx5_flow.c;h=b8abe6a12e2a3ae2be4150f4a401029136d0164d;hb=27efd5dead05f5e11da0b535593c2320b422cfbf;hp=1148db018081776da37cbf41da92c8b38f74312f;hpb=3c84f34eea712483b34b57262ab25efd4def2ff9;p=dpdk.git diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c index 1148db0180..b8abe6a12e 100644 --- a/drivers/net/mlx5/mlx5_flow.c +++ b/drivers/net/mlx5/mlx5_flow.c @@ -286,6 +286,10 @@ static struct mlx5_flow_tunnel_info tunnels_info[] = { .tunnel = MLX5_FLOW_LAYER_VXLAN, .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP, }, + { + .tunnel = MLX5_FLOW_LAYER_GENEVE, + .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP, + }, { .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE, .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP, @@ -316,12 +320,6 @@ static struct mlx5_flow_tunnel_info tunnels_info[] = { }, }; -enum mlx5_feature_name { - MLX5_HAIRPIN_RX, - MLX5_HAIRPIN_TX, - MLX5_APPLICATION, -}; - /** * Translate tag ID to register. * @@ -338,36 +336,126 @@ enum mlx5_feature_name { * The request register on success, a negative errno * value otherwise and rte_errno is set. */ -__rte_unused -static enum modify_reg flow_get_reg_id(struct rte_eth_dev *dev, - enum mlx5_feature_name feature, - uint32_t id, - struct rte_flow_error *error) +enum modify_reg +mlx5_flow_get_reg_id(struct rte_eth_dev *dev, + enum mlx5_feature_name feature, + uint32_t id, + struct rte_flow_error *error) { - static enum modify_reg id2reg[] = { - [0] = REG_A, - [1] = REG_C_2, - [2] = REG_C_3, - [3] = REG_C_4, - [4] = REG_B,}; + struct mlx5_priv *priv = dev->data->dev_private; + struct mlx5_dev_config *config = &priv->config; + enum modify_reg start_reg; - dev = (void *)dev; switch (feature) { case MLX5_HAIRPIN_RX: return REG_B; case MLX5_HAIRPIN_TX: return REG_A; - case MLX5_APPLICATION: - if (id > 4) + case MLX5_METADATA_RX: + switch (config->dv_xmeta_en) { + case MLX5_XMETA_MODE_LEGACY: + return REG_B; + case MLX5_XMETA_MODE_META16: + return REG_C_0; + case MLX5_XMETA_MODE_META32: + return REG_C_1; + } + break; + case MLX5_METADATA_TX: + return REG_A; + case MLX5_METADATA_FDB: + return REG_C_0; + case MLX5_FLOW_MARK: + switch (config->dv_xmeta_en) { + case MLX5_XMETA_MODE_LEGACY: + return REG_NONE; + case MLX5_XMETA_MODE_META16: + return REG_C_1; + case MLX5_XMETA_MODE_META32: + return REG_C_0; + } + break; + case MLX5_COPY_MARK: + 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: + /* + * 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 (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"); - return id2reg[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]; } - return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM, + assert(false); + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, "invalid feature name"); } +/** + * Check extensive flow metadata register support. + * + * @param dev + * Pointer to rte_eth_dev structure. + * + * @return + * True if device supports extensive flow metadata register, otherwise false. + */ +bool +mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev) +{ + struct mlx5_priv *priv = dev->data->dev_private; + struct mlx5_dev_config *config = &priv->config; + + /* + * Having available reg_c can be regarded inclusively as supporting + * extensive flow metadata register, which could mean, + * - metadata register copy action by modify header. + * - 16 modify header actions is supported. + * - reg_c's are preserved across different domain (FDB and NIC) on + * packet loopback by flow lookup miss. + */ + return config->flow_mreg_c[2] != REG_NONE; +} + /** * Discover the maximum number of priority available. * @@ -606,18 +694,28 @@ flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow) { struct mlx5_priv *priv = dev->data->dev_private; struct rte_flow *flow = dev_flow->flow; - const int mark = !!(flow->actions & + const int mark = !!(dev_flow->actions & (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK)); const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL); unsigned int i; for (i = 0; i != flow->rss.queue_num; ++i) { - int idx = (*flow->queue)[i]; + int idx = (*flow->rss.queue)[i]; struct mlx5_rxq_ctrl *rxq_ctrl = 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++; } @@ -669,19 +767,24 @@ flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow) { struct mlx5_priv *priv = dev->data->dev_private; struct rte_flow *flow = dev_flow->flow; - const int mark = !!(flow->actions & + const int mark = !!(dev_flow->actions & (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK)); const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL); unsigned int i; assert(dev->data->dev_started); for (i = 0; i != flow->rss.queue_num; ++i) { - int idx = (*flow->queue)[i]; + int idx = (*flow->rss.queue)[i]; struct mlx5_rxq_ctrl *rxq_ctrl = 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; } @@ -1276,11 +1379,17 @@ mlx5_flow_validate_item_eth(const struct rte_flow_item *item, return rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM, item, "multiple L2 layers not supported"); - if (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)) + if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) || + (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3))) + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ITEM, item, + "L2 layer should not follow " + "L3 layers"); + if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) || + (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN))) return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM, item, - "inner L2 layer should not " - "follow inner L3 layers"); + "L2 layer should not follow VLAN"); if (!mask) mask = &rte_flow_item_eth_mask; ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, @@ -1327,8 +1436,6 @@ mlx5_flow_validate_item_vlan(const struct rte_flow_item *item, const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN : MLX5_FLOW_LAYER_OUTER_VLAN; - const uint64_t l2m = tunnel ? MLX5_FLOW_LAYER_INNER_L2 : - MLX5_FLOW_LAYER_OUTER_L2; if (item_flags & vlanm) return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM, item, @@ -1336,11 +1443,7 @@ mlx5_flow_validate_item_vlan(const struct rte_flow_item *item, else if ((item_flags & l34m) != 0) return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM, item, - "L2 layer cannot follow L3/L4 layer"); - else if ((item_flags & l2m) == 0) - return rte_flow_error_set(error, EINVAL, - RTE_FLOW_ERROR_TYPE_ITEM, item, - "no L2 layer before VLAN"); + "VLAN cannot follow L3/L4 layer"); if (!mask) mask = &rte_flow_item_vlan_mask; ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, @@ -1403,6 +1506,8 @@ mlx5_flow_validate_item_vlan(const struct rte_flow_item *item, int mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item, uint64_t item_flags, + uint64_t last_item, + uint16_t ether_type, const struct rte_flow_item_ipv4 *acc_mask, struct rte_flow_error *error) { @@ -1423,7 +1528,16 @@ mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item, MLX5_FLOW_LAYER_OUTER_L4; int ret; uint8_t next_proto = 0xFF; + const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 | + MLX5_FLOW_LAYER_OUTER_VLAN | + MLX5_FLOW_LAYER_INNER_VLAN); + if ((last_item & l2_vlan) && ether_type && + ether_type != RTE_ETHER_TYPE_IPV4) + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ITEM, item, + "IPv4 cannot follow L2/VLAN layer " + "which ether type is not IPv4"); if (item_flags & MLX5_FLOW_LAYER_IPIP) { if (mask && spec) next_proto = mask->hdr.next_proto_id & @@ -1453,10 +1567,6 @@ mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item, return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM, item, "L3 cannot follow an NVGRE layer."); - else if (!tunnel && !(item_flags & MLX5_FLOW_LAYER_OUTER_L2)) - return rte_flow_error_set(error, EINVAL, - RTE_FLOW_ERROR_TYPE_ITEM, item, - "no L2 layer before IPV4"); if (!mask) mask = &rte_flow_item_ipv4_mask; else if (mask->hdr.next_proto_id != 0 && @@ -1494,6 +1604,8 @@ mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item, int mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item, uint64_t item_flags, + uint64_t last_item, + uint16_t ether_type, const struct rte_flow_item_ipv6 *acc_mask, struct rte_flow_error *error) { @@ -1519,7 +1631,16 @@ mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item, MLX5_FLOW_LAYER_OUTER_L4; int ret; uint8_t next_proto = 0xFF; + const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 | + MLX5_FLOW_LAYER_OUTER_VLAN | + MLX5_FLOW_LAYER_INNER_VLAN); + if ((last_item & l2_vlan) && ether_type && + ether_type != RTE_ETHER_TYPE_IPV6) + return rte_flow_error_set(error, EINVAL, + RTE_FLOW_ERROR_TYPE_ITEM, item, + "IPv6 cannot follow L2/VLAN layer " + "which ether type is not IPv6"); if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) { if (mask && spec) next_proto = mask->hdr.proto & spec->hdr.proto; @@ -1548,10 +1669,6 @@ mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item, return rte_flow_error_set(error, EINVAL, RTE_FLOW_ERROR_TYPE_ITEM, item, "L3 cannot follow an NVGRE layer."); - else if (!tunnel && !(item_flags & MLX5_FLOW_LAYER_OUTER_L2)) - return rte_flow_error_set(error, EINVAL, - RTE_FLOW_ERROR_TYPE_ITEM, item, - "no L2 layer before IPV6"); if (!mask) mask = &rte_flow_item_ipv6_mask; ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask, @@ -2168,6 +2285,49 @@ mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item, 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, @@ -2457,6 +2617,7 @@ flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow) 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); @@ -2527,136 +2688,1295 @@ find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level) } /** - * Create a flow and add it to @p list. + * 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 if the flow should be splited due to hairpin. + * The reason for the split is that in current HW we can't + * support encap on Rx, so if a flow have encap we move it + * to Tx. * * @param dev * Pointer to Ethernet device. - * @param list - * Pointer to a TAILQ flow list. * @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 - * A flow on success, NULL otherwise and rte_errno is set. + * > 0 the number of actions and the flow should be split, + * 0 when no split required. */ +static int +flow_check_hairpin_split(struct rte_eth_dev *dev, + const struct rte_flow_attr *attr, + const struct rte_flow_action actions[]) +{ + int queue_action = 0; + int action_n = 0; + int encap = 0; + const struct rte_flow_action_queue *queue; + const struct rte_flow_action_rss *rss; + const struct rte_flow_action_raw_encap *raw_encap; + + if (!attr->ingress) + return 0; + for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) { + switch (actions->type) { + case RTE_FLOW_ACTION_TYPE_QUEUE: + queue = actions->conf; + if (mlx5_rxq_get_type(dev, queue->index) != + MLX5_RXQ_TYPE_HAIRPIN) + return 0; + queue_action = 1; + action_n++; + break; + case RTE_FLOW_ACTION_TYPE_RSS: + rss = actions->conf; + if (mlx5_rxq_get_type(dev, rss->queue[0]) != + MLX5_RXQ_TYPE_HAIRPIN) + return 0; + queue_action = 1; + action_n++; + break; + case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP: + case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP: + encap = 1; + action_n++; + 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))) + encap = 1; + action_n++; + break; + default: + action_n++; + break; + } + } + if (encap == 1 && queue_action) + return action_n; + 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) + 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 rte_flow *flow = NULL; - struct mlx5_flow *dev_flow; - const struct rte_flow_action_rss *rss; - union { - struct rte_flow_expand_rss buf; - uint8_t buffer[2048]; - } expand_buffer; - struct rte_flow_expand_rss *buf = &expand_buffer.buf; + 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; - uint32_t i; - uint32_t flow_size; - ret = flow_drv_validate(dev, attr, items, actions, external, error); + /* Fill the register fileds in the flow. */ + ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error); if (ret < 0) return NULL; - flow_size = sizeof(struct rte_flow); - rss = flow_get_rss_action(actions); - if (rss) - flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t), - sizeof(void *)); - else - flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *)); - flow = rte_calloc(__func__, 1, flow_size, 0); - if (!flow) { - rte_errno = ENOMEM; + 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; } - flow->drv_type = flow_get_drv_type(dev, attr); - flow->ingress = attr->ingress; - flow->transfer = attr->transfer; - assert(flow->drv_type > MLX5_FLOW_TYPE_MIN && - flow->drv_type < MLX5_FLOW_TYPE_MAX); - flow->queue = (void *)(flow + 1); - LIST_INIT(&flow->dev_flows); - if (rss && rss->types) { - unsigned int graph_root; - - graph_root = find_graph_root(items, rss->level); - ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer), - items, rss->types, - mlx5_support_expansion, - graph_root); - assert(ret > 0 && - (unsigned int)ret < sizeof(expand_buffer.buffer)); + /* 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 { - buf->entries = 1; - buf->entry[0].pattern = (void *)(uintptr_t)items; - } - for (i = 0; i < buf->entries; ++i) { - dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern, - actions, error); - if (!dev_flow) - goto error; - dev_flow->flow = flow; - dev_flow->external = external; - LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next); - ret = flow_drv_translate(dev, dev_flow, attr, - buf->entry[i].pattern, - actions, error); - if (ret < 0) - goto error; + /* 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, + }; } - if (dev->data->dev_started) { - ret = flow_drv_apply(dev, flow, error); - if (ret < 0) - goto error; + /* Build a new entry. */ + mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0); + if (!mcp_res) { + rte_errno = ENOMEM; + return NULL; } - TAILQ_INSERT_TAIL(list, flow, next); - flow_rxq_flags_set(dev, flow); - return flow; + /* + * 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: - ret = rte_errno; /* Save rte_errno before cleanup. */ - assert(flow); - flow_drv_destroy(dev, flow); - rte_free(flow); - rte_errno = ret; /* Restore rte_errno. */ + if (mcp_res->flow) + flow_list_destroy(dev, NULL, mcp_res->flow); + rte_free(mcp_res); return NULL; } /** - * Create a dedicated flow rule on e-switch table 0 (root table), to direct all - * incoming packets to table 1. - * - * Other flow rules, requested for group n, will be created in - * e-switch table n+1. - * Jump action to e-switch group n will be created to group n+1. + * Release flow in RX_CP_TBL. * - * Used when working in switchdev mode, to utilise advantages of table 1 - * and above. + * @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 - * Pointer to flow on success, NULL otherwise and rte_errno is set. + * 0 on success, a negative errno value otherwise and rte_errno is set. */ -struct rte_flow * -mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev) +static int +flow_mreg_start_copy_action(struct rte_eth_dev *dev, + struct rte_flow *flow) { - const struct rte_flow_attr attr = { - .group = 0, - .priority = 0, - .ingress = 1, + 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 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; + 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; + /* + * 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 = (void *)(split_actions + actions_n); + *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; + /* 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, + }; + /* Construct new actions array. */ + memcpy(split_actions, actions, sizeof(*split_actions) * actions_n); + /* Replace QUEUE/RSS action. */ + split_actions[qrss_idx] = (struct rte_flow_action){ + .type = MLX5_RTE_FLOW_ACTION_TYPE_TAG, + .conf = set_tag, + }; + 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; +} + +/** + * 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] 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_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_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; + size_t act_size; + int actions_n; + int ret; + + /* 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) { + /* + * 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"); + /* + * 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 (!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_id) { + 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; + /* + * Put unique id in prefix flow due to it is destroyed after + * prefix 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; + dev_flow = NULL; + ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error); + if (ret < 0) + goto exit; + q_tag_spec.id = ret; + /* Add suffix subflow to execute Q/RSS. */ + ret = flow_create_split_inner(dev, flow, &dev_flow, + &q_attr, 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; +} + +/** + * Split the flow to subflow set. The splitters might be linked + * in the chain, like this: + * flow_create_split_outer() calls: + * flow_create_split_meter() calls: + * flow_create_split_metadata(meter_subflow_0) calls: + * flow_create_split_inner(metadata_subflow_0) + * flow_create_split_inner(metadata_subflow_1) + * flow_create_split_inner(metadata_subflow_2) + * flow_create_split_metadata(meter_subflow_1) calls: + * flow_create_split_inner(metadata_subflow_0) + * flow_create_split_inner(metadata_subflow_1) + * flow_create_split_inner(metadata_subflow_2) + * + * This provide flexible way to add new levels of flow splitting. + * The all of successfully created subflows are included to the + * parent flow dev_flow list. + * + * @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_outer(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) +{ + int ret; + + ret = flow_create_split_metadata(dev, flow, attr, items, + actions, external, error); + assert(ret <= 0); + return ret; +} + +/** + * Create a flow and add it to @p list. + * + * @param dev + * Pointer to Ethernet device. + * @param list + * Pointer to a TAILQ flow list. If this parameter NULL, + * no list insertion occurred, flow is just created, + * this is caller's responsibility to track the + * created flow. + * @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 + * A flow on success, NULL otherwise and rte_errno is set. + */ +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) +{ + struct mlx5_priv *priv = dev->data->dev_private; + struct rte_flow *flow = NULL; + struct mlx5_flow *dev_flow; + const struct rte_flow_action_rss *rss; + union { + struct rte_flow_expand_rss buf; + uint8_t buffer[2048]; + } expand_buffer; + union { + struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS]; + uint8_t buffer[2048]; + } actions_rx; + union { + struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS]; + uint8_t buffer[2048]; + } actions_hairpin_tx; + union { + struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS]; + uint8_t buffer[2048]; + } items_tx; + struct rte_flow_expand_rss *buf = &expand_buffer.buf; + const struct rte_flow_action *p_actions_rx = actions; + int ret; + uint32_t i; + uint32_t flow_size; + int hairpin_flow = 0; + uint32_t hairpin_id = 0; + struct rte_flow_attr attr_tx = { .priority = 0 }; + + hairpin_flow = flow_check_hairpin_split(dev, attr, actions); + if (hairpin_flow > 0) { + if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) { + rte_errno = EINVAL; + return NULL; + } + flow_hairpin_split(dev, actions, actions_rx.actions, + actions_hairpin_tx.actions, items_tx.items, + &hairpin_id); + p_actions_rx = actions_rx.actions; + } + ret = flow_drv_validate(dev, attr, items, p_actions_rx, external, + error); + if (ret < 0) + goto error_before_flow; + flow_size = sizeof(struct rte_flow); + rss = flow_get_rss_action(p_actions_rx); + if (rss) + flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t), + sizeof(void *)); + else + flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *)); + flow = rte_calloc(__func__, 1, flow_size, 0); + if (!flow) { + rte_errno = ENOMEM; + goto error_before_flow; + } + flow->drv_type = flow_get_drv_type(dev, attr); + if (hairpin_id != 0) + flow->hairpin_flow_id = hairpin_id; + assert(flow->drv_type > MLX5_FLOW_TYPE_MIN && + flow->drv_type < MLX5_FLOW_TYPE_MAX); + flow->rss.queue = (void *)(flow + 1); + if (rss) { + /* + * The following information is required by + * mlx5_flow_hashfields_adjust() in advance. + */ + flow->rss.level = rss->level; + /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */ + flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types; + } + LIST_INIT(&flow->dev_flows); + if (rss && rss->types) { + unsigned int graph_root; + + graph_root = find_graph_root(items, rss->level); + ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer), + items, rss->types, + mlx5_support_expansion, + graph_root); + assert(ret > 0 && + (unsigned int)ret < sizeof(expand_buffer.buffer)); + } else { + buf->entries = 1; + buf->entry[0].pattern = (void *)(uintptr_t)items; + } + for (i = 0; i < buf->entries; ++i) { + /* + * The splitter may create multiple dev_flows, + * depending on configuration. In the simplest + * case it just creates unmodified original flow. + */ + ret = flow_create_split_outer(dev, flow, attr, + buf->entry[i].pattern, + p_actions_rx, external, + error); + if (ret < 0) + goto error; + } + /* Create the tx flow. */ + if (hairpin_flow) { + attr_tx.group = MLX5_HAIRPIN_TX_TABLE; + attr_tx.ingress = 0; + attr_tx.egress = 1; + dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items, + actions_hairpin_tx.actions, error); + if (!dev_flow) + goto error; + dev_flow->flow = flow; + dev_flow->external = 0; + LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next); + ret = flow_drv_translate(dev, dev_flow, &attr_tx, + items_tx.items, + actions_hairpin_tx.actions, error); + 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) + goto error; + } + if (list) + TAILQ_INSERT_TAIL(list, flow, next); + flow_rxq_flags_set(dev, flow); + return flow; +error_before_flow: + if (hairpin_id) + mlx5_flow_id_release(priv->sh->flow_id_pool, + 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->hairpin_flow_id); + assert(flow); + flow_drv_destroy(dev, flow); + rte_free(flow); + rte_errno = ret; /* Restore rte_errno. */ + return NULL; +} + +/** + * Create a dedicated flow rule on e-switch table 0 (root table), to direct all + * incoming packets to table 1. + * + * Other flow rules, requested for group n, will be created in + * e-switch table n+1. + * Jump action to e-switch group n will be created to group n+1. + * + * Used when working in switchdev mode, to utilise advantages of table 1 + * and above. + * + * @param dev + * Pointer to Ethernet device. + * + * @return + * Pointer to flow on success, NULL otherwise and rte_errno is set. + */ +struct rte_flow * +mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev) +{ + const struct rte_flow_attr attr = { + .group = 0, + .priority = 0, + .ingress = 1, .egress = 0, .transfer = 1, }; @@ -2707,7 +4027,8 @@ mlx5_flow_create(struct rte_eth_dev *dev, * @param dev * Pointer to Ethernet device. * @param list - * Pointer to a TAILQ flow list. + * Pointer to a TAILQ flow list. If this parameter NULL, + * there is no flow removal from the list. * @param[in] flow * Flow to destroy. */ @@ -2715,14 +4036,21 @@ static void flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list, struct rte_flow *flow) { + struct mlx5_priv *priv = dev->data->dev_private; + /* * Update RX queue flags only if port is started, otherwise it is * already clean. */ if (dev->data->dev_started) flow_rxq_flags_trim(dev, flow); + if (flow->hairpin_flow_id) + mlx5_flow_id_release(priv->sh->flow_id_pool, + flow->hairpin_flow_id); flow_drv_destroy(dev, flow); - TAILQ_REMOVE(list, flow, next); + if (list) + TAILQ_REMOVE(list, flow, next); + flow_mreg_del_copy_action(dev, flow); rte_free(flow->fdir); rte_free(flow); } @@ -2759,8 +4087,11 @@ 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); } @@ -2782,7 +4113,15 @@ 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; @@ -2871,7 +4210,7 @@ mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev, if (!flow) { DRV_LOG(DEBUG, "Failed to create ctrl flow: rte_errno(%d)," - " type(%d), message(%s)\n", + " type(%d), message(%s)", rte_errno, error.type, error.message ? error.message : " (no stated reason)"); return -rte_errno; @@ -3591,10 +4930,10 @@ mlx5_set_query_alarm(struct mlx5_ibv_shared *sh) cont = MLX5_CNT_CONTAINER(sh, 1, 0); pools_n += rte_atomic16_read(&cont->n_valid); us = MLX5_POOL_QUERY_FREQ_US / pools_n; - DRV_LOG(DEBUG, "Set alarm for %u pools each %u us\n", pools_n, us); + DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us); if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) { sh->cmng.query_thread_on = 0; - DRV_LOG(ERR, "Cannot reinitialize query alarm\n"); + DRV_LOG(ERR, "Cannot reinitialize query alarm"); } else { sh->cmng.query_thread_on = 1; } @@ -3666,7 +5005,7 @@ next_container: (uint64_t)(uintptr_t)pool); if (ret) { DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID" - " %d\n", pool->min_dcs->id); + " %d", pool->min_dcs->id); pool->raw_hw = NULL; goto set_alarm; } @@ -3754,3 +5093,74 @@ mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external, } return 0; } + +/** + * Discover availability of metadata reg_c's. + * + * Iteratively use test flows to check availability. + * + * @param[in] dev + * Pointer to the Ethernet device structure. + * + * @return + * 0 on success, a negative errno value otherwise and rte_errno is set. + */ +int +mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev) +{ + struct mlx5_priv *priv = dev->data->dev_private; + struct mlx5_dev_config *config = &priv->config; + enum modify_reg idx; + int n = 0; + + /* reg_c[0] and reg_c[1] are reserved. */ + config->flow_mreg_c[n++] = REG_C_0; + config->flow_mreg_c[n++] = REG_C_1; + /* Discover availability of other reg_c's. */ + for (idx = REG_C_2; idx <= REG_C_7; ++idx) { + struct rte_flow_attr attr = { + .group = MLX5_FLOW_MREG_CP_TABLE_GROUP, + .priority = MLX5_FLOW_PRIO_RSVD, + .ingress = 1, + }; + struct rte_flow_item items[] = { + [0] = { + .type = RTE_FLOW_ITEM_TYPE_END, + }, + }; + struct rte_flow_action actions[] = { + [0] = { + .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG, + .conf = &(struct mlx5_flow_action_copy_mreg){ + .src = REG_C_1, + .dst = idx, + }, + }, + [1] = { + .type = RTE_FLOW_ACTION_TYPE_JUMP, + .conf = &(struct rte_flow_action_jump){ + .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP, + }, + }, + [2] = { + .type = RTE_FLOW_ACTION_TYPE_END, + }, + }; + struct rte_flow *flow; + struct rte_flow_error error; + + if (!config->dv_flow_en) + break; + /* Create internal flow, validation skips copy action. */ + flow = flow_list_create(dev, NULL, &attr, items, + actions, false, &error); + if (!flow) + continue; + if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL)) + config->flow_mreg_c[n++] = idx; + flow_list_destroy(dev, NULL, flow); + } + for (; n < MLX5_MREG_C_NUM; ++n) + config->flow_mreg_c[n] = REG_NONE; + return 0; +}