app/testpmd: fix offloads for newly attached port
authorViacheslav Ovsiienko <viacheslavo@nvidia.com>
Mon, 12 Jul 2021 12:40:53 +0000 (15:40 +0300)
committerAndrew Rybchenko <Andrew.Rybchenko@oktetlabs.ru>
Tue, 13 Jul 2021 09:52:05 +0000 (11:52 +0200)
For the newly attached ports (with "port attach" command) the
default offloads settings, configured from application command
line, were not applied, causing port start failure following
the attach.

For example, if scattering offload was configured in command
line and rxpkts was configured for multiple segments, the newly
attached port start was failed due to missing scattering offload
enable in the new port settings. The missing code to apply
the offloads to the new device and its queues is added.

The new local routine init_config_port_offloads() is introduced,
embracing the shared part of port offloads initialization code.

Fixes: c9cce42876f5 ("ethdev: remove deprecated attach/detach functions")
Cc: stable@dpdk.org
Signed-off-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
Acked-by: Aman Deep Singh <aman.deep.singh@intel.com>
Acked-by: Xiaoyun Li <xiaoyun.li@intel.com>
app/test-pmd/testpmd.c

index 1cdd3cd..a48f709 100644 (file)
@@ -1417,23 +1417,73 @@ check_nb_hairpinq(queueid_t hairpinq)
        return 0;
 }
 
+static void
+init_config_port_offloads(portid_t pid, uint32_t socket_id)
+{
+       struct rte_port *port = &ports[pid];
+       uint16_t data_size;
+       int ret;
+       int i;
+
+       port->dev_conf.txmode = tx_mode;
+       port->dev_conf.rxmode = rx_mode;
+
+       ret = eth_dev_info_get_print_err(pid, &port->dev_info);
+       if (ret != 0)
+               rte_exit(EXIT_FAILURE, "rte_eth_dev_info_get() failed\n");
+
+       ret = update_jumbo_frame_offload(pid);
+       if (ret != 0)
+               printf("Updating jumbo frame offload failed for port %u\n",
+                       pid);
+
+       if (!(port->dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE))
+               port->dev_conf.txmode.offloads &=
+                       ~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
+
+       /* Apply Rx offloads configuration */
+       for (i = 0; i < port->dev_info.max_rx_queues; i++)
+               port->rx_conf[i].offloads = port->dev_conf.rxmode.offloads;
+       /* Apply Tx offloads configuration */
+       for (i = 0; i < port->dev_info.max_tx_queues; i++)
+               port->tx_conf[i].offloads = port->dev_conf.txmode.offloads;
+
+       if (eth_link_speed)
+               port->dev_conf.link_speeds = eth_link_speed;
+
+       /* set flag to initialize port/queue */
+       port->need_reconfig = 1;
+       port->need_reconfig_queues = 1;
+       port->socket_id = socket_id;
+       port->tx_metadata = 0;
+
+       /*
+        * Check for maximum number of segments per MTU.
+        * Accordingly update the mbuf data size.
+        */
+       if (port->dev_info.rx_desc_lim.nb_mtu_seg_max != UINT16_MAX &&
+           port->dev_info.rx_desc_lim.nb_mtu_seg_max != 0) {
+               data_size = rx_mode.max_rx_pkt_len /
+                       port->dev_info.rx_desc_lim.nb_mtu_seg_max;
+
+               if ((data_size + RTE_PKTMBUF_HEADROOM) > mbuf_data_size[0]) {
+                       mbuf_data_size[0] = data_size + RTE_PKTMBUF_HEADROOM;
+                       TESTPMD_LOG(WARNING,
+                                   "Configured mbuf size of the first segment %hu\n",
+                                   mbuf_data_size[0]);
+               }
+       }
+}
+
 static void
 init_config(void)
 {
        portid_t pid;
-       struct rte_port *port;
        struct rte_mempool *mbp;
        unsigned int nb_mbuf_per_pool;
        lcoreid_t  lc_id;
-       uint8_t port_per_socket[RTE_MAX_NUMA_NODES];
        struct rte_gro_param gro_param;
        uint32_t gso_types;
-       uint16_t data_size;
-       bool warning = 0;
-       int k;
-       int ret;
-
-       memset(port_per_socket,0,RTE_MAX_NUMA_NODES);
 
        /* Configuration of logical cores. */
        fwd_lcores = rte_zmalloc("testpmd: fwd_lcores",
@@ -1455,30 +1505,12 @@ init_config(void)
        }
 
        RTE_ETH_FOREACH_DEV(pid) {
-               port = &ports[pid];
-               /* Apply default TxRx configuration for all ports */
-               port->dev_conf.txmode = tx_mode;
-               port->dev_conf.rxmode = rx_mode;
+               uint32_t socket_id;
 
-               ret = eth_dev_info_get_print_err(pid, &port->dev_info);
-               if (ret != 0)
-                       rte_exit(EXIT_FAILURE,
-                                "rte_eth_dev_info_get() failed\n");
-
-               ret = update_jumbo_frame_offload(pid);
-               if (ret != 0)
-                       printf("Updating jumbo frame offload failed for port %u\n",
-                               pid);
-
-               if (!(port->dev_info.tx_offload_capa &
-                     DEV_TX_OFFLOAD_MBUF_FAST_FREE))
-                       port->dev_conf.txmode.offloads &=
-                               ~DEV_TX_OFFLOAD_MBUF_FAST_FREE;
                if (numa_support) {
-                       if (port_numa[pid] != NUMA_NO_CONFIG)
-                               port_per_socket[port_numa[pid]]++;
-                       else {
-                               uint32_t socket_id = rte_eth_dev_socket_id(pid);
+                       socket_id = port_numa[pid];
+                       if (port_numa[pid] == NUMA_NO_CONFIG) {
+                               socket_id = rte_eth_dev_socket_id(pid);
 
                                /*
                                 * if socket_id is invalid,
@@ -1486,48 +1518,14 @@ init_config(void)
                                 */
                                if (check_socket_id(socket_id) < 0)
                                        socket_id = socket_ids[0];
-                               port_per_socket[socket_id]++;
-                       }
-               }
-
-               /* Apply Rx offloads configuration */
-               for (k = 0; k < port->dev_info.max_rx_queues; k++)
-                       port->rx_conf[k].offloads =
-                               port->dev_conf.rxmode.offloads;
-               /* Apply Tx offloads configuration */
-               for (k = 0; k < port->dev_info.max_tx_queues; k++)
-                       port->tx_conf[k].offloads =
-                               port->dev_conf.txmode.offloads;
-
-               if (eth_link_speed)
-                       port->dev_conf.link_speeds = eth_link_speed;
-
-               /* set flag to initialize port/queue */
-               port->need_reconfig = 1;
-               port->need_reconfig_queues = 1;
-               port->tx_metadata = 0;
-
-               /* Check for maximum number of segments per MTU. Accordingly
-                * update the mbuf data size.
-                */
-               if (port->dev_info.rx_desc_lim.nb_mtu_seg_max != UINT16_MAX &&
-                               port->dev_info.rx_desc_lim.nb_mtu_seg_max != 0) {
-                       data_size = rx_mode.max_rx_pkt_len /
-                               port->dev_info.rx_desc_lim.nb_mtu_seg_max;
-
-                       if ((data_size + RTE_PKTMBUF_HEADROOM) >
-                                                       mbuf_data_size[0]) {
-                               mbuf_data_size[0] = data_size +
-                                                RTE_PKTMBUF_HEADROOM;
-                               warning = 1;
                        }
+               } else {
+                       socket_id = (socket_num == UMA_NO_CONFIG) ?
+                                   0 : socket_num;
                }
+               /* Apply default TxRx configuration for all ports */
+               init_config_port_offloads(pid, socket_id);
        }
-
-       if (warning)
-               TESTPMD_LOG(WARNING,
-                           "Configured mbuf size of the first segment %hu\n",
-                           mbuf_data_size[0]);
        /*
         * Create pools of mbuf.
         * If NUMA support is disabled, create a single pool of mbuf in
@@ -1610,21 +1608,8 @@ init_config(void)
 void
 reconfig(portid_t new_port_id, unsigned socket_id)
 {
-       struct rte_port *port;
-       int ret;
-
        /* Reconfiguration of Ethernet ports. */
-       port = &ports[new_port_id];
-
-       ret = eth_dev_info_get_print_err(new_port_id, &port->dev_info);
-       if (ret != 0)
-               return;
-
-       /* set flag to initialize port/queue */
-       port->need_reconfig = 1;
-       port->need_reconfig_queues = 1;
-       port->socket_id = socket_id;
-
+       init_config_port_offloads(new_port_id, socket_id);
        init_port_config();
 }