net/mlx5: allow LRO per Rx queue

Enabling LRO offload per queue makes sense because the user will
probably want to allocate different mempool for LRO queues - the LRO
mempool mbuf size may be bigger than non LRO mempool.

Change the LRO offload to be per queue instead of per port.

If one of the queues is with LRO enabled, all the queues will be
configured via DevX.

If RSS flows direct TCP packets to queues with different LRO enabling,
these flows will not be offloaded with LRO.

Signed-off-by: Matan Azrad <matan@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>

diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index 5c40091..e812374 100644 (file)
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -195,9 +195,6 @@ TAILQ_HEAD(mlx5_flows, rte_flow);
 #define MLX5_LRO_SUPPORTED(dev) \
        (((struct mlx5_priv *)((dev)->data->dev_private))->config.lro.supported)
 
-#define MLX5_LRO_ENABLED(dev) \
-       ((dev)->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_TCP_LRO)
-
 /* LRO configurations structure. */
 struct mlx5_lro_config {
        uint32_t supported:1; /* Whether LRO is supported. */

diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index 9d11831..9629cfb 100644 (file)
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -389,7 +389,6 @@ mlx5_dev_configure(struct rte_eth_dev *dev)
        const uint8_t use_app_rss_key =
                !!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
        int ret = 0;
-       unsigned int lro_on = mlx5_lro_on(dev);
 
        if (use_app_rss_key &&
            (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len !=
@@ -454,11 +453,6 @@ mlx5_dev_configure(struct rte_eth_dev *dev)
                                j = 0;
                }
        }
-       if (lro_on && priv->config.cqe_comp) {
-               /* CQE compressing is not supported for LRO CQEs. */
-               DRV_LOG(WARNING, "Rx CQE compression isn't supported with LRO");
-               priv->config.cqe_comp = 0;
-       }
        ret = mlx5_proc_priv_init(dev);
        if (ret)
                return ret;
@@ -571,7 +565,7 @@ mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
        info->max_tx_queues = max;
        info->max_mac_addrs = MLX5_MAX_UC_MAC_ADDRESSES;
        info->rx_queue_offload_capa = mlx5_get_rx_queue_offloads(dev);
-       info->rx_offload_capa = (mlx5_get_rx_port_offloads(dev) |
+       info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
                                 info->rx_queue_offload_capa);
        info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
        info->if_index = mlx5_ifindex(dev);

diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index f7e861c..a1fdeef 100644 (file)
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -123,21 +123,6 @@ mlx5_mprq_enabled(struct rte_eth_dev *dev)
        return n == priv->rxqs_n;
 }
 
-/**
- * Check whether LRO is supported and enabled for the device.
- *
- * @param dev
- *   Pointer to Ethernet device.
- *
- * @return
- *   0 if disabled, 1 if enabled.
- */
-inline int
-mlx5_lro_on(struct rte_eth_dev *dev)
-{
-       return (MLX5_LRO_SUPPORTED(dev) && MLX5_LRO_ENABLED(dev));
-}
-
 /**
  * Allocate RX queue elements for Multi-Packet RQ.
  *
@@ -394,6 +379,8 @@ mlx5_get_rx_queue_offloads(struct rte_eth_dev *dev)
                             DEV_RX_OFFLOAD_TCP_CKSUM);
        if (config->hw_vlan_strip)
                offloads |= DEV_RX_OFFLOAD_VLAN_STRIP;
+       if (MLX5_LRO_SUPPORTED(dev))
+               offloads |= DEV_RX_OFFLOAD_TCP_LRO;
        return offloads;
 }
 
@@ -401,19 +388,14 @@ mlx5_get_rx_queue_offloads(struct rte_eth_dev *dev)
 /**
  * Returns the per-port supported offloads.
  *
- * @param dev
- *   Pointer to Ethernet device.
- *
  * @return
  *   Supported Rx offloads.
  */
 uint64_t
-mlx5_get_rx_port_offloads(struct rte_eth_dev *dev)
+mlx5_get_rx_port_offloads(void)
 {
        uint64_t offloads = DEV_RX_OFFLOAD_VLAN_FILTER;
 
-       if (MLX5_LRO_SUPPORTED(dev))
-               offloads |= DEV_RX_OFFLOAD_TCP_LRO;
        return offloads;
 }
 
@@ -889,7 +871,8 @@ mlx5_ibv_cq_new(struct rte_eth_dev *dev, struct mlx5_priv *priv,
        cq_attr.mlx5 = (struct mlx5dv_cq_init_attr){
                .comp_mask = 0,
        };
-       if (priv->config.cqe_comp && !rxq_data->hw_timestamp) {
+       if (priv->config.cqe_comp && !rxq_data->hw_timestamp &&
+           !rxq_data->lro) {
                cq_attr.mlx5.comp_mask |=
                                MLX5DV_CQ_INIT_ATTR_MASK_COMPRESSED_CQE;
 #ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
@@ -911,6 +894,10 @@ mlx5_ibv_cq_new(struct rte_eth_dev *dev, struct mlx5_priv *priv,
                        "port %u Rx CQE compression is disabled for HW"
                        " timestamp",
                        dev->data->port_id);
+       } else if (priv->config.cqe_comp && rxq_data->lro) {
+               DRV_LOG(DEBUG,
+                       "port %u Rx CQE compression is disabled for LRO",
+                       dev->data->port_id);
        }
 #ifdef HAVE_IBV_MLX5_MOD_CQE_128B_PAD
        if (priv->config.cqe_pad) {
@@ -1607,6 +1594,7 @@ mlx5_rxq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
                desc + config->rx_vec_en * MLX5_VPMD_DESCS_PER_LOOP;
        uint64_t offloads = conf->offloads |
                           dev->data->dev_conf.rxmode.offloads;
+       unsigned int lro_on_queue = !!(offloads & DEV_RX_OFFLOAD_TCP_LRO);
        const int mprq_en = mlx5_check_mprq_support(dev) > 0;
        unsigned int max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
        unsigned int non_scatter_min_mbuf_size = max_rx_pkt_len +
@@ -1646,7 +1634,7 @@ mlx5_rxq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
         * In this case scatter is, for sure, enabled and an empty mbuf may be
         * added in the start for the head-room.
         */
-       if (mlx5_lro_on(dev) && RTE_PKTMBUF_HEADROOM > 0 &&
+       if (lro_on_queue && RTE_PKTMBUF_HEADROOM > 0 &&
            non_scatter_min_mbuf_size > mb_len) {
                strd_headroom_en = 0;
                mprq_stride_size = RTE_MIN(max_rx_pkt_len,
@@ -1693,7 +1681,7 @@ mlx5_rxq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
                unsigned int size = non_scatter_min_mbuf_size;
                unsigned int sges_n;
 
-               if (mlx5_lro_on(dev) && first_mb_free_size <
+               if (lro_on_queue && first_mb_free_size <
                    MLX5_MAX_LRO_HEADER_FIX) {
                        DRV_LOG(ERR, "Not enough space in the first segment(%u)"
                                " to include the max header size(%u) for LRO",
@@ -1747,13 +1735,14 @@ mlx5_rxq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
        tmpl->rxq.vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
        /* By default, FCS (CRC) is stripped by hardware. */
        tmpl->rxq.crc_present = 0;
+       tmpl->rxq.lro = lro_on_queue;
        if (offloads & DEV_RX_OFFLOAD_KEEP_CRC) {
                if (config->hw_fcs_strip) {
                        /*
                         * RQs used for LRO-enabled TIRs should not be
                         * configured to scatter the FCS.
                         */
-                       if (mlx5_lro_on(dev))
+                       if (lro_on_queue)
                                DRV_LOG(WARNING,
                                        "port %u CRC stripping has been "
                                        "disabled but will still be performed "
@@ -2204,7 +2193,16 @@ mlx5_hrxq_new(struct rte_eth_dev *dev,
                }
        } else { /* ind_tbl->type == MLX5_IND_TBL_TYPE_DEVX */
                struct mlx5_devx_tir_attr tir_attr;
-
+               uint32_t i;
+               uint32_t lro = 1;
+
+               /* Enable TIR LRO only if all the queues were configured for. */
+               for (i = 0; i < queues_n; ++i) {
+                       if (!(*priv->rxqs)[queues[i]]->lro) {
+                               lro = 0;
+                               break;
+                       }
+               }
                memset(&tir_attr, 0, sizeof(tir_attr));
                tir_attr.disp_type = MLX5_TIRC_DISP_TYPE_INDIRECT;
                tir_attr.rx_hash_fn = MLX5_RX_HASH_FN_TOEPLITZ;
@@ -2216,7 +2214,7 @@ mlx5_hrxq_new(struct rte_eth_dev *dev,
                if (dev->data->dev_conf.lpbk_mode)
                        tir_attr.self_lb_block =
                                        MLX5_TIRC_SELF_LB_BLOCK_BLOCK_UNICAST;
-               if (mlx5_lro_on(dev)) {
+               if (lro) {
                        tir_attr.lro_timeout_period_usecs =
                                        priv->config.lro.timeout;
                        tir_attr.lro_max_msg_sz = priv->max_lro_msg_size;

diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 9b58d0a..c209d99 100644 (file)
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -115,7 +115,8 @@ struct mlx5_rxq_data {
        unsigned int strd_shift_en:1; /* Enable 2bytes shift on a stride. */
        unsigned int err_state:2; /* enum mlx5_rxq_err_state. */
        unsigned int strd_headroom_en:1; /* Enable mbuf headroom in MPRQ. */
-       unsigned int :2; /* Remaining bits. */
+       unsigned int lro:1; /* Enable LRO. */
+       unsigned int :1; /* Remaining bits. */
        volatile uint32_t *rq_db;
        volatile uint32_t *cq_db;
        uint16_t port_id;
@@ -367,9 +368,8 @@ int mlx5_hrxq_release(struct rte_eth_dev *dev, struct mlx5_hrxq *hxrq);
 int mlx5_hrxq_verify(struct rte_eth_dev *dev);
 struct mlx5_hrxq *mlx5_hrxq_drop_new(struct rte_eth_dev *dev);
 void mlx5_hrxq_drop_release(struct rte_eth_dev *dev);
-uint64_t mlx5_get_rx_port_offloads(struct rte_eth_dev *dev);
+uint64_t mlx5_get_rx_port_offloads(void);
 uint64_t mlx5_get_rx_queue_offloads(struct rte_eth_dev *dev);
-int mlx5_lro_on(struct rte_eth_dev *dev);
 
 /* mlx5_txq.c */
 

diff --git a/drivers/net/mlx5/mlx5_rxtx_vec.c b/drivers/net/mlx5/mlx5_rxtx_vec.c
index 3815ff6..3925f4d 100644 (file)
--- a/drivers/net/mlx5/mlx5_rxtx_vec.c
+++ b/drivers/net/mlx5/mlx5_rxtx_vec.c
@@ -129,6 +129,8 @@ mlx5_rxq_check_vec_support(struct mlx5_rxq_data *rxq)
                return -ENOTSUP;
        if (!ctrl->priv->config.rx_vec_en || rxq->sges_n != 0)
                return -ENOTSUP;
+       if (rxq->lro)
+               return -ENOTSUP;
        return 1;
 }
 
@@ -151,8 +153,6 @@ mlx5_check_vec_rx_support(struct rte_eth_dev *dev)
                return -ENOTSUP;
        if (mlx5_mprq_enabled(dev))
                return -ENOTSUP;
-       if (mlx5_lro_on(dev))
-               return -ENOTSUP;
        /* All the configured queues should support. */
        for (i = 0; i < priv->rxqs_n; ++i) {
                struct mlx5_rxq_data *rxq = (*priv->rxqs)[i];

diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index 8bc2174..aa323ad 100644 (file)
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -99,10 +99,14 @@ mlx5_rxq_start(struct rte_eth_dev *dev)
        struct mlx5_priv *priv = dev->data->dev_private;
        unsigned int i;
        int ret = 0;
-       unsigned int lro_on = mlx5_lro_on(dev);
-       enum mlx5_rxq_obj_type obj_type = lro_on ? MLX5_RXQ_OBJ_TYPE_DEVX_RQ :
-                                                  MLX5_RXQ_OBJ_TYPE_IBV;
+       enum mlx5_rxq_obj_type obj_type = MLX5_RXQ_OBJ_TYPE_IBV;
 
+       for (i = 0; i < priv->rxqs_n; ++i) {
+               if ((*priv->rxqs)[i]->lro) {
+                       obj_type =  MLX5_RXQ_OBJ_TYPE_DEVX_RQ;
+                       break;
+               }
+       }
        /* Allocate/reuse/resize mempool for Multi-Packet RQ. */
        if (mlx5_mprq_alloc_mp(dev)) {
                /* Should not release Rx queues but return immediately. */