net/avp: add device configuration
authorAllain Legacy <allain.legacy@windriver.com>
Tue, 28 Mar 2017 11:54:01 +0000 (07:54 -0400)
committerFerruh Yigit <ferruh.yigit@intel.com>
Tue, 4 Apr 2017 17:02:18 +0000 (19:02 +0200)
Adds support for "dev_configure" operations to allow an application to
configure the device.

Signed-off-by: Allain Legacy <allain.legacy@windriver.com>
Signed-off-by: Matt Peters <matt.peters@windriver.com>
Acked-by: Vincent Jardin <vincent.jardin@6wind.com>
doc/guides/nics/features/avp.ini
drivers/net/avp/avp_ethdev.c

index 4353929..45a2185 100644 (file)
@@ -4,3 +4,7 @@
 ; Refer to default.ini for the full list of available PMD features.
 ;
 [Features]
+Link status          = Y
+VLAN offload         = Y
+Linux UIO            = Y
+x86-64               = Y
index 38ff78e..14bd885 100644 (file)
 
 
 
+static int avp_dev_configure(struct rte_eth_dev *dev);
+static void avp_dev_info_get(struct rte_eth_dev *dev,
+                            struct rte_eth_dev_info *dev_info);
+static void avp_vlan_offload_set(struct rte_eth_dev *dev, int mask);
+static int avp_dev_link_update(struct rte_eth_dev *dev,
+                              __rte_unused int wait_to_complete);
+
 #define AVP_DEV_TO_PCI(eth_dev) RTE_DEV_TO_PCI((eth_dev)->device)
 
 
@@ -97,6 +104,15 @@ static const struct rte_pci_id pci_id_avp_map[] = {
        },
 };
 
+/*
+ * dev_ops for avp, bare necessities for basic operation
+ */
+static const struct eth_dev_ops avp_eth_dev_ops = {
+       .dev_configure       = avp_dev_configure,
+       .dev_infos_get       = avp_dev_info_get,
+       .vlan_offload_set    = avp_vlan_offload_set,
+       .link_update         = avp_dev_link_update,
+};
 
 /**@{ AVP device flags */
 #define AVP_F_PROMISC (1 << 1)
@@ -182,6 +198,91 @@ struct avp_queue {
        uint64_t errors;
 };
 
+/* send a request and wait for a response
+ *
+ * @warning must be called while holding the avp->lock spinlock.
+ */
+static int
+avp_dev_process_request(struct avp_dev *avp, struct rte_avp_request *request)
+{
+       unsigned int retry = AVP_MAX_REQUEST_RETRY;
+       void *resp_addr = NULL;
+       unsigned int count;
+       int ret;
+
+       PMD_DRV_LOG(DEBUG, "Sending request %u to host\n", request->req_id);
+
+       request->result = -ENOTSUP;
+
+       /* Discard any stale responses before starting a new request */
+       while (avp_fifo_get(avp->resp_q, (void **)&resp_addr, 1))
+               PMD_DRV_LOG(DEBUG, "Discarding stale response\n");
+
+       rte_memcpy(avp->sync_addr, request, sizeof(*request));
+       count = avp_fifo_put(avp->req_q, &avp->host_sync_addr, 1);
+       if (count < 1) {
+               PMD_DRV_LOG(ERR, "Cannot send request %u to host\n",
+                           request->req_id);
+               ret = -EBUSY;
+               goto done;
+       }
+
+       while (retry--) {
+               /* wait for a response */
+               usleep(AVP_REQUEST_DELAY_USECS);
+
+               count = avp_fifo_count(avp->resp_q);
+               if (count >= 1) {
+                       /* response received */
+                       break;
+               }
+
+               if ((count < 1) && (retry == 0)) {
+                       PMD_DRV_LOG(ERR, "Timeout while waiting for a response for %u\n",
+                                   request->req_id);
+                       ret = -ETIME;
+                       goto done;
+               }
+       }
+
+       /* retrieve the response */
+       count = avp_fifo_get(avp->resp_q, (void **)&resp_addr, 1);
+       if ((count != 1) || (resp_addr != avp->host_sync_addr)) {
+               PMD_DRV_LOG(ERR, "Invalid response from host, count=%u resp=%p host_sync_addr=%p\n",
+                           count, resp_addr, avp->host_sync_addr);
+               ret = -ENODATA;
+               goto done;
+       }
+
+       /* copy to user buffer */
+       rte_memcpy(request, avp->sync_addr, sizeof(*request));
+       ret = 0;
+
+       PMD_DRV_LOG(DEBUG, "Result %d received for request %u\n",
+                   request->result, request->req_id);
+
+done:
+       return ret;
+}
+
+static int
+avp_dev_ctrl_set_config(struct rte_eth_dev *eth_dev,
+                       struct rte_avp_device_config *config)
+{
+       struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+       struct rte_avp_request request;
+       int ret;
+
+       /* setup a configure request */
+       memset(&request, 0, sizeof(request));
+       request.req_id = RTE_AVP_REQ_CFG_DEVICE;
+       memcpy(&request.config, config, sizeof(request.config));
+
+       ret = avp_dev_process_request(avp, &request);
+
+       return ret == 0 ? request.result : ret;
+}
+
 /* translate from host physical address to guest virtual address */
 static void *
 avp_dev_translate_address(struct rte_eth_dev *eth_dev,
@@ -297,6 +398,38 @@ avp_dev_check_regions(struct rte_eth_dev *eth_dev)
        return 0;
 }
 
+static void
+_avp_set_queue_counts(struct rte_eth_dev *eth_dev)
+{
+       struct rte_pci_device *pci_dev = AVP_DEV_TO_PCI(eth_dev);
+       struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+       struct rte_avp_device_info *host_info;
+       void *addr;
+
+       addr = pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR].addr;
+       host_info = (struct rte_avp_device_info *)addr;
+
+       /*
+        * the transmit direction is not negotiated beyond respecting the max
+        * number of queues because the host can handle arbitrary guest tx
+        * queues (host rx queues).
+        */
+       avp->num_tx_queues = eth_dev->data->nb_tx_queues;
+
+       /*
+        * the receive direction is more restrictive.  The host requires a
+        * minimum number of guest rx queues (host tx queues) therefore
+        * negotiate a value that is at least as large as the host minimum
+        * requirement.  If the host and guest values are not identical then a
+        * mapping will be established in the receive_queue_setup function.
+        */
+       avp->num_rx_queues = RTE_MAX(host_info->min_rx_queues,
+                                    eth_dev->data->nb_rx_queues);
+
+       PMD_DRV_LOG(DEBUG, "Requesting %u Tx and %u Rx queues from host\n",
+                   avp->num_tx_queues, avp->num_rx_queues);
+}
+
 /*
  * create a AVP device using the supplied device info by first translating it
  * to guest address space(s).
@@ -439,6 +572,7 @@ eth_avp_dev_init(struct rte_eth_dev *eth_dev)
        int ret;
 
        pci_dev = AVP_DEV_TO_PCI(eth_dev);
+       eth_dev->dev_ops = &avp_eth_dev_ops;
 
        if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
                /*
@@ -513,5 +647,112 @@ static struct eth_driver rte_avp_pmd = {
        .dev_private_size = sizeof(struct avp_adapter),
 };
 
+static int
+avp_dev_configure(struct rte_eth_dev *eth_dev)
+{
+       struct rte_pci_device *pci_dev = AVP_DEV_TO_PCI(eth_dev);
+       struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+       struct rte_avp_device_info *host_info;
+       struct rte_avp_device_config config;
+       int mask = 0;
+       void *addr;
+       int ret;
+
+       addr = pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR].addr;
+       host_info = (struct rte_avp_device_info *)addr;
+
+       /* Setup required number of queues */
+       _avp_set_queue_counts(eth_dev);
+
+       mask = (ETH_VLAN_STRIP_MASK |
+               ETH_VLAN_FILTER_MASK |
+               ETH_VLAN_EXTEND_MASK);
+       avp_vlan_offload_set(eth_dev, mask);
+
+       /* update device config */
+       memset(&config, 0, sizeof(config));
+       config.device_id = host_info->device_id;
+       config.driver_type = RTE_AVP_DRIVER_TYPE_DPDK;
+       config.driver_version = AVP_DPDK_DRIVER_VERSION;
+       config.features = avp->features;
+       config.num_tx_queues = avp->num_tx_queues;
+       config.num_rx_queues = avp->num_rx_queues;
+
+       ret = avp_dev_ctrl_set_config(eth_dev, &config);
+       if (ret < 0) {
+               PMD_DRV_LOG(ERR, "Config request failed by host, ret=%d\n",
+                           ret);
+               goto unlock;
+       }
+
+       avp->flags |= AVP_F_CONFIGURED;
+       ret = 0;
+
+unlock:
+       return ret;
+}
+
+
+static int
+avp_dev_link_update(struct rte_eth_dev *eth_dev,
+                                       __rte_unused int wait_to_complete)
+{
+       struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+       struct rte_eth_link *link = &eth_dev->data->dev_link;
+
+       link->link_speed = ETH_SPEED_NUM_10G;
+       link->link_duplex = ETH_LINK_FULL_DUPLEX;
+       link->link_status = !!(avp->flags & AVP_F_LINKUP);
+
+       return -1;
+}
+
+
+static void
+avp_dev_info_get(struct rte_eth_dev *eth_dev,
+                struct rte_eth_dev_info *dev_info)
+{
+       struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+
+       dev_info->driver_name = "rte_avp_pmd";
+       dev_info->pci_dev = RTE_DEV_TO_PCI(eth_dev->device);
+       dev_info->max_rx_queues = avp->max_rx_queues;
+       dev_info->max_tx_queues = avp->max_tx_queues;
+       dev_info->min_rx_bufsize = AVP_MIN_RX_BUFSIZE;
+       dev_info->max_rx_pktlen = avp->max_rx_pkt_len;
+       dev_info->max_mac_addrs = AVP_MAX_MAC_ADDRS;
+       if (avp->host_features & RTE_AVP_FEATURE_VLAN_OFFLOAD) {
+               dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
+               dev_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
+       }
+}
+
+static void
+avp_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
+{
+       struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+
+       if (mask & ETH_VLAN_STRIP_MASK) {
+               if (avp->host_features & RTE_AVP_FEATURE_VLAN_OFFLOAD) {
+                       if (eth_dev->data->dev_conf.rxmode.hw_vlan_strip)
+                               avp->features |= RTE_AVP_FEATURE_VLAN_OFFLOAD;
+                       else
+                               avp->features &= ~RTE_AVP_FEATURE_VLAN_OFFLOAD;
+               } else {
+                       PMD_DRV_LOG(ERR, "VLAN strip offload not supported\n");
+               }
+       }
+
+       if (mask & ETH_VLAN_FILTER_MASK) {
+               if (eth_dev->data->dev_conf.rxmode.hw_vlan_filter)
+                       PMD_DRV_LOG(ERR, "VLAN filter offload not supported\n");
+       }
+
+       if (mask & ETH_VLAN_EXTEND_MASK) {
+               if (eth_dev->data->dev_conf.rxmode.hw_vlan_extend)
+                       PMD_DRV_LOG(ERR, "VLAN extend offload not supported\n");
+       }
+}
+
 RTE_PMD_REGISTER_PCI(net_avp, rte_avp_pmd.pci_drv);
 RTE_PMD_REGISTER_PCI_TABLE(net_avp, pci_id_avp_map);