net/qede: fix VF port creation sequence
authorHarish Patil <harish.patil@cavium.com>
Sat, 3 Feb 2018 06:03:17 +0000 (22:03 -0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Mon, 5 Feb 2018 18:56:04 +0000 (19:56 +0100)
Few adjustments are required to effectively handle VF vport create/delete
sequence. The problem is exposed by recent ethdev TX offload changes
which requires port to be in down state before applying TX offloads.

 - Move vport creation from dev_init() to dev_configure()
 - Force to stop vport if it was already started due to previous run
   (restart case)
 - Move link state enable/disable to dev_init() and dev_close()
   respectively.
 - For MTU change, recreate vport with new MTU value and restore old
   config. This is necessary since VF MTU value can be changed only upon
   vport creation.

Fixes: ec94dbc57362 ("qede: add base driver")
Cc: stable@dpdk.org
Signed-off-by: Harish Patil <harish.patil@cavium.com>
drivers/net/qede/qede_ethdev.c
drivers/net/qede/qede_ethdev.h
drivers/net/qede/qede_rxtx.c

index 20a1c31..46e0e36 100644 (file)
@@ -442,55 +442,59 @@ static void qede_reset_queue_stats(struct qede_dev *qdev, bool xstats)
 }
 
 static int
-qede_start_vport(struct qede_dev *qdev, uint16_t mtu)
+qede_stop_vport(struct ecore_dev *edev)
 {
-       struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
-       struct ecore_sp_vport_start_params params;
        struct ecore_hwfn *p_hwfn;
+       uint8_t vport_id;
        int rc;
        int i;
 
-       memset(&params, 0, sizeof(params));
-       params.vport_id = 0;
-       params.mtu = mtu;
-       /* @DPDK - Disable FW placement */
-       params.zero_placement_offset = 1;
+       vport_id = 0;
        for_each_hwfn(edev, i) {
                p_hwfn = &edev->hwfns[i];
-               params.concrete_fid = p_hwfn->hw_info.concrete_fid;
-               params.opaque_fid = p_hwfn->hw_info.opaque_fid;
-               rc = ecore_sp_vport_start(p_hwfn, &params);
+               rc = ecore_sp_vport_stop(p_hwfn, p_hwfn->hw_info.opaque_fid,
+                                        vport_id);
                if (rc != ECORE_SUCCESS) {
-                       DP_ERR(edev, "Start V-PORT failed %d\n", rc);
+                       DP_ERR(edev, "Stop V-PORT failed rc = %d\n", rc);
                        return rc;
                }
        }
-       ecore_reset_vport_stats(edev);
-       if (IS_PF(edev))
-               qede_reset_queue_stats(qdev, true);
-       DP_INFO(edev, "VPORT started with MTU = %u\n", mtu);
+
+       DP_INFO(edev, "vport stopped\n");
 
        return 0;
 }
 
 static int
-qede_stop_vport(struct ecore_dev *edev)
+qede_start_vport(struct qede_dev *qdev, uint16_t mtu)
 {
+       struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
+       struct ecore_sp_vport_start_params params;
        struct ecore_hwfn *p_hwfn;
-       uint8_t vport_id;
        int rc;
        int i;
 
-       vport_id = 0;
+       if (qdev->vport_started)
+               qede_stop_vport(edev);
+
+       memset(&params, 0, sizeof(params));
+       params.vport_id = 0;
+       params.mtu = mtu;
+       /* @DPDK - Disable FW placement */
+       params.zero_placement_offset = 1;
        for_each_hwfn(edev, i) {
                p_hwfn = &edev->hwfns[i];
-               rc = ecore_sp_vport_stop(p_hwfn, p_hwfn->hw_info.opaque_fid,
-                                        vport_id);
+               params.concrete_fid = p_hwfn->hw_info.concrete_fid;
+               params.opaque_fid = p_hwfn->hw_info.opaque_fid;
+               rc = ecore_sp_vport_start(p_hwfn, &params);
                if (rc != ECORE_SUCCESS) {
-                       DP_ERR(edev, "Stop V-PORT failed rc = %d\n", rc);
+                       DP_ERR(edev, "Start V-PORT failed %d\n", rc);
                        return rc;
                }
        }
+       ecore_reset_vport_stats(edev);
+       qdev->vport_started = true;
+       DP_INFO(edev, "VPORT started with MTU = %u\n", mtu);
 
        return 0;
 }
@@ -1194,6 +1198,8 @@ static int qede_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
                DP_INFO(edev, "No offloads are supported with VLAN Q-in-Q"
                        " and classification is based on outer tag only\n");
 
+       qdev->vlan_offload_mask = mask;
+
        DP_INFO(edev, "vlan offload mask %d vlan-strip %d vlan-filter %d\n",
                mask, rxmode->hw_vlan_strip, rxmode->hw_vlan_filter);
 
@@ -1267,13 +1273,6 @@ static int qede_dev_start(struct rte_eth_dev *eth_dev)
 
        PMD_INIT_FUNC_TRACE(edev);
 
-       /* Update MTU only if it has changed */
-       if (qdev->mtu != qdev->new_mtu) {
-               if (qede_update_mtu(eth_dev, qdev->new_mtu))
-                       goto err;
-               qdev->mtu = qdev->new_mtu;
-       }
-
        /* Configure TPA parameters */
        if (rxmode->enable_lro) {
                if (qede_enable_tpa(eth_dev, true))
@@ -1287,6 +1286,9 @@ static int qede_dev_start(struct rte_eth_dev *eth_dev)
        if (qede_start_queues(eth_dev))
                goto err;
 
+       if (IS_PF(edev))
+               qede_reset_queue_stats(qdev, true);
+
        /* Newer SR-IOV PF driver expects RX/TX queues to be started before
         * enabling RSS. Hence RSS configuration is deferred upto this point.
         * Also, we would like to retain similar behavior in PF case, so we
@@ -1300,9 +1302,6 @@ static int qede_dev_start(struct rte_eth_dev *eth_dev)
        if (qede_activate_vport(eth_dev, true))
                goto err;
 
-       /* Bring-up the link */
-       qede_dev_set_link_state(eth_dev, true);
-
        /* Update link status */
        qede_link_update(eth_dev, 0);
 
@@ -1337,9 +1336,6 @@ static void qede_dev_stop(struct rte_eth_dev *eth_dev)
        /* Disable traffic */
        ecore_hw_stop_fastpath(edev); /* TBD - loop */
 
-       /* Bring the link down */
-       qede_dev_set_link_state(eth_dev, false);
-
        DP_INFO(edev, "Device is stopped\n");
 }
 
@@ -1464,20 +1460,11 @@ static int qede_dev_configure(struct rte_eth_dev *eth_dev)
        if (qede_check_fdir_support(eth_dev))
                return -ENOTSUP;
 
-       /* Deallocate resources if held previously. It is needed only if the
-        * queue count has been changed from previous configuration. If its
-        * going to change then it means RX/TX queue setup will be called
-        * again and the fastpath pointers will be reinitialized there.
-        */
-       if (qdev->num_tx_queues != eth_dev->data->nb_tx_queues ||
-           qdev->num_rx_queues != eth_dev->data->nb_rx_queues) {
-               qede_dealloc_fp_resc(eth_dev);
-               /* Proceed with updated queue count */
-               qdev->num_tx_queues = eth_dev->data->nb_tx_queues;
-               qdev->num_rx_queues = eth_dev->data->nb_rx_queues;
-               if (qede_alloc_fp_resc(qdev))
-                       return -ENOMEM;
-       }
+       qede_dealloc_fp_resc(eth_dev);
+       qdev->num_tx_queues = eth_dev->data->nb_tx_queues;
+       qdev->num_rx_queues = eth_dev->data->nb_rx_queues;
+       if (qede_alloc_fp_resc(qdev))
+               return -ENOMEM;
 
        /* If jumbo enabled adjust MTU */
        if (eth_dev->data->dev_conf.rxmode.jumbo_frame)
@@ -1485,19 +1472,9 @@ static int qede_dev_configure(struct rte_eth_dev *eth_dev)
                                eth_dev->data->dev_conf.rxmode.max_rx_pkt_len -
                                ETHER_HDR_LEN - ETHER_CRC_LEN;
 
-       /* VF's MTU has to be set using vport-start where as
-        * PF's MTU can be updated via vport-update.
-        */
-       if (IS_VF(edev)) {
-               if (qede_start_vport(qdev, eth_dev->data->mtu))
-                       return -1;
-       } else {
-               if (qede_update_mtu(eth_dev, eth_dev->data->mtu))
-                       return -1;
-       }
-
+       if (qede_start_vport(qdev, eth_dev->data->mtu))
+               return -1;
        qdev->mtu = eth_dev->data->mtu;
-       qdev->new_mtu = qdev->mtu;
 
        /* Enable VLAN offloads by default */
        ret = qede_vlan_offload_set(eth_dev, ETH_VLAN_STRIP_MASK  |
@@ -1711,12 +1688,15 @@ static void qede_dev_close(struct rte_eth_dev *eth_dev)
                qede_dev_stop(eth_dev);
 
        qede_stop_vport(edev);
+       qdev->vport_started = false;
        qede_fdir_dealloc_resc(eth_dev);
        qede_dealloc_fp_resc(eth_dev);
 
        eth_dev->data->nb_rx_queues = 0;
        eth_dev->data->nb_tx_queues = 0;
 
+       /* Bring the link down */
+       qede_dev_set_link_state(eth_dev, false);
        qdev->ops->common->slowpath_stop(edev);
        qdev->ops->common->remove(edev);
        rte_intr_disable(&pci_dev->intr_handle);
@@ -2387,8 +2367,6 @@ static int qede_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
        int i;
 
        PMD_INIT_FUNC_TRACE(edev);
-       if (IS_VF(edev))
-               return -ENOTSUP;
        qede_dev_info_get(dev, &dev_info);
        max_rx_pkt_len = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
        frame_size = max_rx_pkt_len + QEDE_ETH_OVERHEAD;
@@ -2415,7 +2393,9 @@ static int qede_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
                restart = true;
        }
        rte_delay_ms(1000);
-       qdev->new_mtu = mtu;
+       qede_start_vport(qdev, mtu); /* Recreate vport */
+       qdev->mtu = mtu;
+
        /* Fix up RX buf size for all queues of the port */
        for_each_rss(i) {
                fp = &qdev->fp_array[i];
@@ -2429,17 +2409,33 @@ static int qede_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
                                rx_buf_size = frame_size;
                        rx_buf_size = QEDE_CEIL_TO_CACHE_LINE_SIZE(rx_buf_size);
                        fp->rxq->rx_buf_size = rx_buf_size;
-                       DP_INFO(edev, "buf_size adjusted to %u\n", rx_buf_size);
+                       DP_INFO(edev, "RX buffer size %u\n", rx_buf_size);
                }
        }
        if (max_rx_pkt_len > ETHER_MAX_LEN)
                dev->data->dev_conf.rxmode.jumbo_frame = 1;
        else
                dev->data->dev_conf.rxmode.jumbo_frame = 0;
+
+       /* Restore config lost due to vport stop */
+       qede_mac_addr_set(dev, &qdev->primary_mac);
+       if (dev->data->promiscuous)
+               qede_promiscuous_enable(dev);
+       else
+               qede_promiscuous_disable(dev);
+
+       if (dev->data->all_multicast)
+               qede_allmulticast_enable(dev);
+       else
+               qede_allmulticast_disable(dev);
+
+       qede_vlan_offload_set(dev, qdev->vlan_offload_mask);
+
        if (!dev->data->dev_started && restart) {
                qede_dev_start(dev);
                dev->data->dev_started = 1;
        }
+
        /* update max frame size */
        dev->data->dev_conf.rxmode.max_rx_pkt_len = max_rx_pkt_len;
        /* Reassign back */
@@ -3115,29 +3111,28 @@ static int qede_common_dev_init(struct rte_eth_dev *eth_dev, bool is_vf)
                do_once = false;
        }
 
+       /* Bring-up the link */
+       qede_dev_set_link_state(eth_dev, true);
+
        adapter->num_tx_queues = 0;
        adapter->num_rx_queues = 0;
        SLIST_INIT(&adapter->fdir_info.fdir_list_head);
        SLIST_INIT(&adapter->vlan_list_head);
        SLIST_INIT(&adapter->uc_list_head);
        adapter->mtu = ETHER_MTU;
-       adapter->new_mtu = ETHER_MTU;
-       if (!is_vf) {
-               if (qede_start_vport(adapter, adapter->mtu))
-                       return -1;
-       } else {
-               /* VF tunnel offloads is enabled by default in PF driver */
-               adapter->vxlan.enable = true;
-               adapter->vxlan.num_filters = 0;
-               adapter->vxlan.filter_type = ETH_TUNNEL_FILTER_IMAC |
-                                            ETH_TUNNEL_FILTER_IVLAN;
-               adapter->vxlan.udp_port = QEDE_VXLAN_DEF_PORT;
-               adapter->geneve.enable = true;
-               adapter->vxlan.num_filters = 0;
-               adapter->vxlan.filter_type = ETH_TUNNEL_FILTER_IMAC |
-                                            ETH_TUNNEL_FILTER_IVLAN;
-               adapter->vxlan.udp_port = QEDE_GENEVE_DEF_PORT;
-       }
+       adapter->vport_started = false;
+
+       /* VF tunnel offloads is enabled by default in PF driver */
+       adapter->vxlan.enable = true;
+       adapter->vxlan.num_filters = 0;
+       adapter->vxlan.filter_type = ETH_TUNNEL_FILTER_IMAC |
+                                    ETH_TUNNEL_FILTER_IVLAN;
+       adapter->vxlan.udp_port = QEDE_VXLAN_DEF_PORT;
+       adapter->geneve.enable = true;
+       adapter->vxlan.num_filters = 0;
+       adapter->vxlan.filter_type = ETH_TUNNEL_FILTER_IMAC |
+                                    ETH_TUNNEL_FILTER_IVLAN;
+       adapter->vxlan.udp_port = QEDE_GENEVE_DEF_PORT;
 
        DP_INFO(edev, "MAC address : %02x:%02x:%02x:%02x:%02x:%02x\n",
                adapter->primary_mac.addr_bytes[0],
index c7bb415..23f7e0e 100644 (file)
@@ -187,7 +187,6 @@ struct qede_dev {
        struct ecore_sb_info *sb_array;
        struct qede_fastpath *fp_array;
        uint16_t mtu;
-       uint16_t new_mtu;
        bool enable_tx_switching;
        bool rss_enable;
        struct rte_eth_rss_conf rss_conf;
@@ -211,6 +210,8 @@ struct qede_dev {
        struct qede_fdir_info fdir_info;
        bool vlan_strip_flg;
        char drv_ver[QEDE_PMD_DRV_VER_STR_SIZE];
+       bool vport_started;
+       int vlan_offload_mask;
        void *ethdev;
 };
 
index 810f0f3..169ede8 100644 (file)
@@ -417,6 +417,8 @@ int qede_alloc_fp_resc(struct qede_dev *qdev)
 
        for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
                fp = &qdev->fp_array[sb_idx];
+               if (!fp)
+                       continue;
                fp->sb_info = rte_calloc("sb", 1, sizeof(struct ecore_sb_info),
                                RTE_CACHE_LINE_SIZE);
                if (!fp->sb_info) {
@@ -448,6 +450,8 @@ void qede_dealloc_fp_resc(struct rte_eth_dev *eth_dev)
 
        for (sb_idx = 0; sb_idx < QEDE_RXTX_MAX(qdev); sb_idx++) {
                fp = &qdev->fp_array[sb_idx];
+               if (!fp)
+                       continue;
                DP_INFO(edev, "Free sb_info index 0x%x\n",
                                fp->sb_info->igu_sb_id);
                if (fp->sb_info) {