net/mlx4: convert to new Tx offloads API
authorShahaf Shuler <shahafs@mellanox.com>
Wed, 10 Jan 2018 09:17:02 +0000 (11:17 +0200)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 16 Jan 2018 17:47:49 +0000 (18:47 +0100)
Ethdev Tx offloads API has changed since:

commit cba7f53b717d ("ethdev: introduce Tx queue offloads API")

This commit support the new Tx offloads API.

Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
drivers/net/mlx4/mlx4_ethdev.c
drivers/net/mlx4/mlx4_rxtx.h
drivers/net/mlx4/mlx4_txq.c

index 2f69e7d..3602f0a 100644 (file)
@@ -767,17 +767,12 @@ mlx4_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
        info->max_tx_queues = max;
        info->max_mac_addrs = RTE_DIM(priv->mac);
        info->rx_offload_capa = 0;
-       info->tx_offload_capa = 0;
+       info->tx_offload_capa = mlx4_get_tx_port_offloads(priv);
        if (priv->hw_csum) {
-               info->tx_offload_capa |= (DEV_TX_OFFLOAD_IPV4_CKSUM |
-                                         DEV_TX_OFFLOAD_UDP_CKSUM |
-                                         DEV_TX_OFFLOAD_TCP_CKSUM);
                info->rx_offload_capa |= (DEV_RX_OFFLOAD_IPV4_CKSUM |
                                          DEV_RX_OFFLOAD_UDP_CKSUM |
                                          DEV_RX_OFFLOAD_TCP_CKSUM);
        }
-       if (priv->hw_csum_l2tun)
-               info->tx_offload_capa |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
        if (mlx4_get_ifname(priv, &ifname) == 0)
                info->if_index = if_nametoindex(ifname);
        info->hash_key_size = MLX4_RSS_HASH_KEY_SIZE;
index b93e2bc..bff5ae4 100644 (file)
@@ -180,6 +180,7 @@ uint16_t mlx4_rx_burst_removed(void *dpdk_rxq, struct rte_mbuf **pkts,
 
 /* mlx4_txq.c */
 
+uint64_t mlx4_get_tx_port_offloads(struct priv *priv);
 int mlx4_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx,
                        uint16_t desc, unsigned int socket,
                        const struct rte_eth_txconf *conf);
index d651e49..7664c3e 100644 (file)
@@ -41,6 +41,7 @@
 #include <stddef.h>
 #include <stdint.h>
 #include <string.h>
+#include <inttypes.h>
 
 /* Verbs headers do not support -pedantic. */
 #ifdef PEDANTIC
@@ -181,6 +182,50 @@ mlx4_txq_fill_dv_obj_info(struct txq *txq, struct mlx4dv_obj *mlxdv)
        cq->cqe_64 = (dcq->cqe_size & 64) ? 1 : 0;
 }
 
+/**
+ * Returns the per-port supported offloads.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ *
+ * @return
+ *   Supported Tx offloads.
+ */
+uint64_t
+mlx4_get_tx_port_offloads(struct priv *priv)
+{
+       uint64_t offloads = DEV_TX_OFFLOAD_MULTI_SEGS;
+
+       if (priv->hw_csum) {
+               offloads |= (DEV_TX_OFFLOAD_IPV4_CKSUM |
+                            DEV_TX_OFFLOAD_UDP_CKSUM |
+                            DEV_TX_OFFLOAD_TCP_CKSUM);
+       }
+       if (priv->hw_csum_l2tun)
+               offloads |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
+       return offloads;
+}
+
+/**
+ * Checks if the per-queue offload configuration is valid.
+ *
+ * @param priv
+ *   Pointer to private structure.
+ * @param requested
+ *   Per-queue offloads configuration.
+ *
+ * @return
+ *   Nonzero when configuration is valid.
+ */
+static int
+mlx4_check_tx_queue_offloads(struct priv *priv, uint64_t requested)
+{
+       uint64_t mandatory = priv->dev->data->dev_conf.txmode.offloads;
+       uint64_t supported = mlx4_get_tx_port_offloads(priv);
+
+       return !((mandatory ^ requested) & supported);
+}
+
 /**
  * DPDK callback to configure a Tx queue.
  *
@@ -229,9 +274,22 @@ mlx4_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
        };
        int ret;
 
-       (void)conf; /* Thresholds configuration (ignored). */
        DEBUG("%p: configuring queue %u for %u descriptors",
              (void *)dev, idx, desc);
+       /*
+        * Don't verify port offloads for application which
+        * use the old API.
+        */
+       if ((conf->txq_flags & ETH_TXQ_FLAGS_IGNORE) &&
+           !mlx4_check_tx_queue_offloads(priv, conf->offloads)) {
+               rte_errno = ENOTSUP;
+               ERROR("%p: Tx queue offloads 0x%" PRIx64 " don't match port "
+                     "offloads 0x%" PRIx64 " or supported offloads 0x%" PRIx64,
+                     (void *)dev, conf->offloads,
+                     dev->data->dev_conf.txmode.offloads,
+                     mlx4_get_tx_port_offloads(priv));
+               return -rte_errno;
+       }
        if (idx >= dev->data->nb_tx_queues) {
                rte_errno = EOVERFLOW;
                ERROR("%p: queue index out of range (%u >= %u)",
@@ -281,8 +339,13 @@ mlx4_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
                        RTE_MIN(MLX4_PMD_TX_PER_COMP_REQ, desc / 4),
                .elts_comp_cd_init =
                        RTE_MIN(MLX4_PMD_TX_PER_COMP_REQ, desc / 4),
-               .csum = priv->hw_csum,
-               .csum_l2tun = priv->hw_csum_l2tun,
+               .csum = priv->hw_csum &&
+                       (conf->offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
+                                          DEV_TX_OFFLOAD_UDP_CKSUM |
+                                          DEV_TX_OFFLOAD_TCP_CKSUM)),
+               .csum_l2tun = priv->hw_csum_l2tun &&
+                             (conf->offloads &
+                              DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM),
                /* Enable Tx loopback for VF devices. */
                .lb = !!priv->vf,
                .bounce_buf = bounce_buf,