drivers: remove direct access to interrupt handle
[dpdk.git] / drivers / net / netvsc / hn_ethdev.c
index b330bf3..8a95040 100644 (file)
@@ -9,6 +9,11 @@
 #include <stdio.h>
 #include <errno.h>
 #include <unistd.h>
+#include <dirent.h>
+#include <net/if.h>
+#include <net/if_arp.h>
+#include <netinet/in.h>
+#include <sys/ioctl.h>
 
 #include <rte_ethdev.h>
 #include <rte_memcpy.h>
 #include <rte_atomic.h>
 #include <rte_branch_prediction.h>
 #include <rte_ether.h>
-#include <rte_ethdev_driver.h>
+#include <ethdev_driver.h>
 #include <rte_cycles.h>
 #include <rte_errno.h>
 #include <rte_memory.h>
 #include <rte_eal.h>
 #include <rte_dev.h>
 #include <rte_bus_vmbus.h>
+#include <rte_alarm.h>
 
 #include "hn_logs.h"
 #include "hn_var.h"
 #include "hn_nvs.h"
 #include "ndis.h"
 
-#define HN_TX_OFFLOAD_CAPS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
-                           DEV_TX_OFFLOAD_TCP_CKSUM  | \
-                           DEV_TX_OFFLOAD_UDP_CKSUM  | \
-                           DEV_TX_OFFLOAD_TCP_TSO    | \
-                           DEV_TX_OFFLOAD_MULTI_SEGS | \
-                           DEV_TX_OFFLOAD_VLAN_INSERT)
+#define HN_TX_OFFLOAD_CAPS (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM | \
+                           RTE_ETH_TX_OFFLOAD_TCP_CKSUM  | \
+                           RTE_ETH_TX_OFFLOAD_UDP_CKSUM  | \
+                           RTE_ETH_TX_OFFLOAD_TCP_TSO    | \
+                           RTE_ETH_TX_OFFLOAD_MULTI_SEGS | \
+                           RTE_ETH_TX_OFFLOAD_VLAN_INSERT)
 
-#define HN_RX_OFFLOAD_CAPS (DEV_RX_OFFLOAD_CHECKSUM | \
-                           DEV_RX_OFFLOAD_VLAN_STRIP)
+#define HN_RX_OFFLOAD_CAPS (RTE_ETH_RX_OFFLOAD_CHECKSUM | \
+                           RTE_ETH_RX_OFFLOAD_VLAN_STRIP | \
+                           RTE_ETH_RX_OFFLOAD_RSS_HASH)
 
-int hn_logtype_init;
-int hn_logtype_driver;
+#define NETVSC_ARG_LATENCY "latency"
+#define NETVSC_ARG_RXBREAK "rx_copybreak"
+#define NETVSC_ARG_TXBREAK "tx_copybreak"
+#define NETVSC_ARG_RX_EXTMBUF_ENABLE "rx_extmbuf_enable"
+
+/* The max number of retry when hot adding a VF device */
+#define NETVSC_MAX_HOTADD_RETRY 10
 
 struct hn_xstats_name_off {
        char name[RTE_ETH_XSTATS_NAME_SIZE];
@@ -57,6 +69,7 @@ static const struct hn_xstats_name_off hn_stat_strings[] = {
        { "good_bytes",             offsetof(struct hn_stats, bytes) },
        { "errors",                 offsetof(struct hn_stats, errors) },
        { "ring full",              offsetof(struct hn_stats, ring_full) },
+       { "channel full",           offsetof(struct hn_stats, channel_full) },
        { "multicast_packets",      offsetof(struct hn_stats, multicast) },
        { "broadcast_packets",      offsetof(struct hn_stats, broadcast) },
        { "undersize_packets",      offsetof(struct hn_stats, size_bins[0]) },
@@ -69,6 +82,18 @@ static const struct hn_xstats_name_off hn_stat_strings[] = {
        { "size_1519_max_packets",  offsetof(struct hn_stats, size_bins[7]) },
 };
 
+/* The default RSS key.
+ * This value is the same as MLX5 so that flows will be
+ * received on same path for both VF and synthetic NIC.
+ */
+static const uint8_t rss_default_key[NDIS_HASH_KEYSIZE_TOEPLITZ] = {
+       0x2c, 0xc6, 0x81, 0xd1, 0x5b, 0xdb, 0xf4, 0xf7,
+       0xfc, 0xa2, 0x83, 0x19, 0xdb, 0x1a, 0x3e, 0x94,
+       0x6b, 0x9e, 0x38, 0xd9, 0x2c, 0x9c, 0x03, 0xd1,
+       0xad, 0x99, 0x44, 0xa7, 0xd9, 0x56, 0x3d, 0x59,
+       0x06, 0x3c, 0x25, 0xf3, 0xfc, 0x1f, 0xdc, 0x2a,
+};
+
 static struct rte_eth_dev *
 eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
 {
@@ -108,9 +133,9 @@ eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
        eth_dev->device = &dev->device;
 
        /* interrupt is simulated */
-       dev->intr_handle.type = RTE_INTR_HANDLE_EXT;
+       rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_EXT);
        eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
-       eth_dev->intr_handle = &dev->intr_handle;
+       eth_dev->intr_handle = dev->intr_handle;
 
        return eth_dev;
 }
@@ -118,8 +143,6 @@ eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
 static void
 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
 {
-       /* mac_addrs must not be freed alone because part of dev_private */
-       eth_dev->data->mac_addrs = NULL;
        /* free ether device */
        rte_eth_dev_release_port(eth_dev);
 
@@ -127,24 +150,36 @@ eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
        eth_dev->intr_handle = NULL;
 }
 
-/* handle "latency=X" from devargs */
-static int hn_set_latency(const char *key, const char *value, void *opaque)
+static int hn_set_parameter(const char *key, const char *value, void *opaque)
 {
        struct hn_data *hv = opaque;
        char *endp = NULL;
-       unsigned long lat;
-
-       errno = 0;
-       lat = strtoul(value, &endp, 0);
+       unsigned long v;
 
+       v = strtoul(value, &endp, 0);
        if (*value == '\0' || *endp != '\0') {
                PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
                return -EINVAL;
        }
 
-       PMD_DRV_LOG(DEBUG, "set latency %lu usec", lat);
+       if (!strcmp(key, NETVSC_ARG_LATENCY)) {
+               /* usec to nsec */
+               hv->latency = v * 1000;
+               PMD_DRV_LOG(DEBUG, "set latency %u usec", hv->latency);
+       } else if (!strcmp(key, NETVSC_ARG_RXBREAK)) {
+               hv->rx_copybreak = v;
+               PMD_DRV_LOG(DEBUG, "rx copy break set to %u",
+                           hv->rx_copybreak);
+       } else if (!strcmp(key, NETVSC_ARG_TXBREAK)) {
+               hv->tx_copybreak = v;
+               PMD_DRV_LOG(DEBUG, "tx copy break set to %u",
+                           hv->tx_copybreak);
+       } else if (!strcmp(key, NETVSC_ARG_RX_EXTMBUF_ENABLE)) {
+               hv->rx_extmbuf_enable = v;
+               PMD_DRV_LOG(DEBUG, "rx extmbuf enable set to %u",
+                           hv->rx_extmbuf_enable);
+       }
 
-       hv->latency = lat * 1000;       /* usec to nsec */
        return 0;
 }
 
@@ -154,7 +189,10 @@ static int hn_parse_args(const struct rte_eth_dev *dev)
        struct hn_data *hv = dev->data->dev_private;
        struct rte_devargs *devargs = dev->device->devargs;
        static const char * const valid_keys[] = {
-               "latency",
+               NETVSC_ARG_LATENCY,
+               NETVSC_ARG_RXBREAK,
+               NETVSC_ARG_TXBREAK,
+               NETVSC_ARG_RX_EXTMBUF_ENABLE,
                NULL
        };
        struct rte_kvargs *kvlist;
@@ -168,15 +206,13 @@ static int hn_parse_args(const struct rte_eth_dev *dev)
 
        kvlist = rte_kvargs_parse(devargs->args, valid_keys);
        if (!kvlist) {
-               PMD_DRV_LOG(NOTICE, "invalid parameters");
+               PMD_DRV_LOG(ERR, "invalid parameters");
                return -EINVAL;
        }
 
-       ret = rte_kvargs_process(kvlist, "latency", hn_set_latency, hv);
-       if (ret)
-               PMD_DRV_LOG(ERR, "Unable to process latency arg\n");
-
+       ret = rte_kvargs_process(kvlist, NULL, hn_set_parameter, hv);
        rte_kvargs_free(kvlist);
+
        return ret;
 }
 
@@ -187,7 +223,7 @@ static int hn_parse_args(const struct rte_eth_dev *dev)
  */
 int
 hn_dev_link_update(struct rte_eth_dev *dev,
-                  int wait_to_complete)
+                  int wait_to_complete __rte_unused)
 {
        struct hn_data *hv = dev->data->dev_private;
        struct rte_eth_link link, old;
@@ -201,59 +237,226 @@ hn_dev_link_update(struct rte_eth_dev *dev,
 
        hn_rndis_get_linkspeed(hv);
 
-       hn_vf_link_update(dev, wait_to_complete);
-
        link = (struct rte_eth_link) {
-               .link_duplex = ETH_LINK_FULL_DUPLEX,
-               .link_autoneg = ETH_LINK_SPEED_FIXED,
+               .link_duplex = RTE_ETH_LINK_FULL_DUPLEX,
+               .link_autoneg = RTE_ETH_LINK_SPEED_FIXED,
                .link_speed = hv->link_speed / 10000,
        };
 
        if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
-               link.link_status = ETH_LINK_UP;
+               link.link_status = RTE_ETH_LINK_UP;
        else
-               link.link_status = ETH_LINK_DOWN;
+               link.link_status = RTE_ETH_LINK_DOWN;
 
        if (old.link_status == link.link_status)
                return 0;
 
        PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
-                    (link.link_status == ETH_LINK_UP) ? "up" : "down");
+                    (link.link_status == RTE_ETH_LINK_UP) ? "up" : "down");
 
        return rte_eth_linkstatus_set(dev, &link);
 }
 
-static void hn_dev_info_get(struct rte_eth_dev *dev,
-                           struct rte_eth_dev_info *dev_info)
+static int hn_dev_info_get(struct rte_eth_dev *dev,
+                          struct rte_eth_dev_info *dev_info)
 {
        struct hn_data *hv = dev->data->dev_private;
+       int rc;
 
-       dev_info->speed_capa = ETH_LINK_SPEED_10G;
+       dev_info->speed_capa = RTE_ETH_LINK_SPEED_10G;
        dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
        dev_info->max_rx_pktlen  = HN_MAX_XFER_LEN;
        dev_info->max_mac_addrs  = 1;
 
        dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
-       dev_info->flow_type_rss_offloads =
-               ETH_RSS_IPV4 | ETH_RSS_IPV6 | ETH_RSS_TCP | ETH_RSS_UDP;
+       dev_info->flow_type_rss_offloads = hv->rss_offloads;
+       dev_info->reta_size = RTE_ETH_RSS_RETA_SIZE_128;
 
        dev_info->max_rx_queues = hv->max_queues;
        dev_info->max_tx_queues = hv->max_queues;
 
-       hn_rndis_get_offload(hv, dev_info);
-       hn_vf_info_get(hv, dev_info);
+       dev_info->tx_desc_lim.nb_min = 1;
+       dev_info->tx_desc_lim.nb_max = 4096;
+
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return 0;
+
+       /* fills in rx and tx offload capability */
+       rc = hn_rndis_get_offload(hv, dev_info);
+       if (rc != 0)
+               return rc;
+
+       /* merges the offload and queues of vf */
+       return hn_vf_info_get(hv, dev_info);
 }
 
-static void
+static int hn_rss_reta_update(struct rte_eth_dev *dev,
+                             struct rte_eth_rss_reta_entry64 *reta_conf,
+                             uint16_t reta_size)
+{
+       struct hn_data *hv = dev->data->dev_private;
+       unsigned int i;
+       int err;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (reta_size != NDIS_HASH_INDCNT) {
+               PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
+               return -EINVAL;
+       }
+
+       for (i = 0; i < NDIS_HASH_INDCNT; i++) {
+               uint16_t idx = i / RTE_ETH_RETA_GROUP_SIZE;
+               uint16_t shift = i % RTE_ETH_RETA_GROUP_SIZE;
+               uint64_t mask = (uint64_t)1 << shift;
+
+               if (reta_conf[idx].mask & mask)
+                       hv->rss_ind[i] = reta_conf[idx].reta[shift];
+       }
+
+       err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
+       if (err) {
+               PMD_DRV_LOG(NOTICE,
+                       "rss disable failed");
+               return err;
+       }
+
+       err = hn_rndis_conf_rss(hv, 0);
+       if (err) {
+               PMD_DRV_LOG(NOTICE,
+                           "reta reconfig failed");
+               return err;
+       }
+
+       return hn_vf_reta_hash_update(dev, reta_conf, reta_size);
+}
+
+static int hn_rss_reta_query(struct rte_eth_dev *dev,
+                            struct rte_eth_rss_reta_entry64 *reta_conf,
+                            uint16_t reta_size)
+{
+       struct hn_data *hv = dev->data->dev_private;
+       unsigned int i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (reta_size != NDIS_HASH_INDCNT) {
+               PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
+               return -EINVAL;
+       }
+
+       for (i = 0; i < NDIS_HASH_INDCNT; i++) {
+               uint16_t idx = i / RTE_ETH_RETA_GROUP_SIZE;
+               uint16_t shift = i % RTE_ETH_RETA_GROUP_SIZE;
+               uint64_t mask = (uint64_t)1 << shift;
+
+               if (reta_conf[idx].mask & mask)
+                       reta_conf[idx].reta[shift] = hv->rss_ind[i];
+       }
+       return 0;
+}
+
+static void hn_rss_hash_init(struct hn_data *hv,
+                            const struct rte_eth_rss_conf *rss_conf)
+{
+       /* Convert from DPDK RSS hash flags to NDIS hash flags */
+       hv->rss_hash = NDIS_HASH_FUNCTION_TOEPLITZ;
+
+       if (rss_conf->rss_hf & RTE_ETH_RSS_IPV4)
+               hv->rss_hash |= NDIS_HASH_IPV4;
+       if (rss_conf->rss_hf & RTE_ETH_RSS_NONFRAG_IPV4_TCP)
+               hv->rss_hash |= NDIS_HASH_TCP_IPV4;
+       if (rss_conf->rss_hf & RTE_ETH_RSS_IPV6)
+               hv->rss_hash |=  NDIS_HASH_IPV6;
+       if (rss_conf->rss_hf & RTE_ETH_RSS_IPV6_EX)
+               hv->rss_hash |=  NDIS_HASH_IPV6_EX;
+       if (rss_conf->rss_hf & RTE_ETH_RSS_NONFRAG_IPV6_TCP)
+               hv->rss_hash |= NDIS_HASH_TCP_IPV6;
+       if (rss_conf->rss_hf & RTE_ETH_RSS_IPV6_TCP_EX)
+               hv->rss_hash |= NDIS_HASH_TCP_IPV6_EX;
+
+       memcpy(hv->rss_key, rss_conf->rss_key ? : rss_default_key,
+              NDIS_HASH_KEYSIZE_TOEPLITZ);
+}
+
+static int hn_rss_hash_update(struct rte_eth_dev *dev,
+                             struct rte_eth_rss_conf *rss_conf)
+{
+       struct hn_data *hv = dev->data->dev_private;
+       int err;
+
+       PMD_INIT_FUNC_TRACE();
+
+       err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
+       if (err) {
+               PMD_DRV_LOG(NOTICE,
+                           "rss disable failed");
+               return err;
+       }
+
+       hn_rss_hash_init(hv, rss_conf);
+
+       if (rss_conf->rss_hf != 0) {
+               err = hn_rndis_conf_rss(hv, 0);
+               if (err) {
+                       PMD_DRV_LOG(NOTICE,
+                                   "rss reconfig failed (RSS disabled)");
+                       return err;
+               }
+       }
+
+       return hn_vf_rss_hash_update(dev, rss_conf);
+}
+
+static int hn_rss_hash_conf_get(struct rte_eth_dev *dev,
+                               struct rte_eth_rss_conf *rss_conf)
+{
+       struct hn_data *hv = dev->data->dev_private;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (hv->ndis_ver < NDIS_VERSION_6_20) {
+               PMD_DRV_LOG(DEBUG, "RSS not supported on this host");
+               return -EOPNOTSUPP;
+       }
+
+       rss_conf->rss_key_len = NDIS_HASH_KEYSIZE_TOEPLITZ;
+       if (rss_conf->rss_key)
+               memcpy(rss_conf->rss_key, hv->rss_key,
+                      NDIS_HASH_KEYSIZE_TOEPLITZ);
+
+       rss_conf->rss_hf = 0;
+       if (hv->rss_hash & NDIS_HASH_IPV4)
+               rss_conf->rss_hf |= RTE_ETH_RSS_IPV4;
+
+       if (hv->rss_hash & NDIS_HASH_TCP_IPV4)
+               rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_TCP;
+
+       if (hv->rss_hash & NDIS_HASH_IPV6)
+               rss_conf->rss_hf |= RTE_ETH_RSS_IPV6;
+
+       if (hv->rss_hash & NDIS_HASH_IPV6_EX)
+               rss_conf->rss_hf |= RTE_ETH_RSS_IPV6_EX;
+
+       if (hv->rss_hash & NDIS_HASH_TCP_IPV6)
+               rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV6_TCP;
+
+       if (hv->rss_hash & NDIS_HASH_TCP_IPV6_EX)
+               rss_conf->rss_hf |= RTE_ETH_RSS_IPV6_TCP_EX;
+
+       return 0;
+}
+
+static int
 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
 {
        struct hn_data *hv = dev->data->dev_private;
 
        hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
-       hn_vf_promiscuous_enable(dev);
+       return hn_vf_promiscuous_enable(dev);
 }
 
-static void
+static int
 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
 {
        struct hn_data *hv = dev->data->dev_private;
@@ -263,10 +466,10 @@ hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
        if (dev->data->all_multicast)
                filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
        hn_rndis_set_rxfilter(hv, filter);
-       hn_vf_promiscuous_disable(dev);
+       return hn_vf_promiscuous_disable(dev);
 }
 
-static void
+static int
 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
 {
        struct hn_data *hv = dev->data->dev_private;
@@ -274,22 +477,22 @@ hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
        hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
                              NDIS_PACKET_TYPE_ALL_MULTICAST |
                        NDIS_PACKET_TYPE_BROADCAST);
-       hn_vf_allmulticast_enable(dev);
+       return hn_vf_allmulticast_enable(dev);
 }
 
-static void
+static int
 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
 {
        struct hn_data *hv = dev->data->dev_private;
 
        hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
                             NDIS_PACKET_TYPE_BROADCAST);
-       hn_vf_allmulticast_disable(dev);
+       return hn_vf_allmulticast_disable(dev);
 }
 
 static int
 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
-                    struct ether_addr *mc_addr_set,
+                    struct rte_ether_addr *mc_addr_set,
                     uint32_t nb_mc_addr)
 {
        /* No filtering on the synthetic path, but can do it on VF */
@@ -348,20 +551,144 @@ static int hn_subchan_configure(struct hn_data *hv,
        return err;
 }
 
+static void netvsc_hotplug_retry(void *args)
+{
+       int ret;
+       struct hn_data *hv = args;
+       struct rte_eth_dev *dev = &rte_eth_devices[hv->port_id];
+       struct rte_devargs *d = &hv->devargs;
+       char buf[256];
+
+       DIR *di;
+       struct dirent *dir;
+       struct ifreq req;
+       struct rte_ether_addr eth_addr;
+       int s;
+
+       PMD_DRV_LOG(DEBUG, "%s: retry count %d",
+                   __func__, hv->eal_hot_plug_retry);
+
+       if (hv->eal_hot_plug_retry++ > NETVSC_MAX_HOTADD_RETRY)
+               return;
+
+       snprintf(buf, sizeof(buf), "/sys/bus/pci/devices/%s/net", d->name);
+       di = opendir(buf);
+       if (!di) {
+               PMD_DRV_LOG(DEBUG, "%s: can't open directory %s, "
+                           "retrying in 1 second", __func__, buf);
+               goto retry;
+       }
+
+       while ((dir = readdir(di))) {
+               /* Skip . and .. directories */
+               if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
+                       continue;
+
+               /* trying to get mac address if this is a network device*/
+               s = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
+               if (s == -1) {
+                       PMD_DRV_LOG(ERR, "Failed to create socket errno %d",
+                                   errno);
+                       break;
+               }
+               strlcpy(req.ifr_name, dir->d_name, sizeof(req.ifr_name));
+               ret = ioctl(s, SIOCGIFHWADDR, &req);
+               close(s);
+               if (ret == -1) {
+                       PMD_DRV_LOG(ERR,
+                                   "Failed to send SIOCGIFHWADDR for device %s",
+                                   dir->d_name);
+                       break;
+               }
+               if (req.ifr_hwaddr.sa_family != ARPHRD_ETHER) {
+                       closedir(di);
+                       return;
+               }
+               memcpy(eth_addr.addr_bytes, req.ifr_hwaddr.sa_data,
+                      RTE_DIM(eth_addr.addr_bytes));
+
+               if (rte_is_same_ether_addr(&eth_addr, dev->data->mac_addrs)) {
+                       PMD_DRV_LOG(NOTICE,
+                                   "Found matching MAC address, adding device %s network name %s",
+                                   d->name, dir->d_name);
+                       ret = rte_eal_hotplug_add(d->bus->name, d->name,
+                                                 d->args);
+                       if (ret) {
+                               PMD_DRV_LOG(ERR,
+                                           "Failed to add PCI device %s",
+                                           d->name);
+                               break;
+                       }
+               }
+               /* When the code reaches here, we either have already added
+                * the device, or its MAC address did not match.
+                */
+               closedir(di);
+               return;
+       }
+       closedir(di);
+retry:
+       /* The device is still being initialized, retry after 1 second */
+       rte_eal_alarm_set(1000000, netvsc_hotplug_retry, hv);
+}
+
+static void
+netvsc_hotadd_callback(const char *device_name, enum rte_dev_event_type type,
+                      void *arg)
+{
+       struct hn_data *hv = arg;
+       struct rte_devargs *d = &hv->devargs;
+       int ret;
+
+       PMD_DRV_LOG(INFO, "Device notification type=%d device_name=%s",
+                   type, device_name);
+
+       switch (type) {
+       case RTE_DEV_EVENT_ADD:
+               /* if we already has a VF, don't check on hot add */
+               if (hv->vf_ctx.vf_state > vf_removed)
+                       break;
+
+               ret = rte_devargs_parse(d, device_name);
+               if (ret) {
+                       PMD_DRV_LOG(ERR,
+                                   "devargs parsing failed ret=%d", ret);
+                       return;
+               }
+
+               if (!strcmp(d->bus->name, "pci")) {
+                       /* Start the process of figuring out if this
+                        * PCI device is a VF device
+                        */
+                       hv->eal_hot_plug_retry = 0;
+                       rte_eal_alarm_set(1000000, netvsc_hotplug_retry, hv);
+               }
+
+               /* We will switch to VF on RDNIS configure message
+                * sent from VSP
+                */
+
+               break;
+       default:
+               break;
+       }
+}
+
 static int hn_dev_configure(struct rte_eth_dev *dev)
 {
-       const struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
+       struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
+       struct rte_eth_rss_conf *rss_conf = &dev_conf->rx_adv_conf.rss_conf;
        const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
        const struct rte_eth_txmode *txmode = &dev_conf->txmode;
-
-       const struct rte_eth_rss_conf *rss_conf =
-               &dev_conf->rx_adv_conf.rss_conf;
        struct hn_data *hv = dev->data->dev_private;
        uint64_t unsupported;
-       int err, subchan;
+       int i, err, subchan;
 
        PMD_INIT_FUNC_TRACE();
 
+       if (dev_conf->rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG)
+               dev_conf->rxmode.offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
+
        unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
        if (unsupported) {
                PMD_DRV_LOG(NOTICE,
@@ -378,6 +705,8 @@ static int hn_dev_configure(struct rte_eth_dev *dev)
                return -EINVAL;
        }
 
+       hv->vlan_strip = !!(rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP);
+
        err = hn_rndis_conf_offload(hv, txmode->offloads,
                                    rxmode->offloads);
        if (err) {
@@ -388,6 +717,12 @@ static int hn_dev_configure(struct rte_eth_dev *dev)
 
        hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
                                 dev->data->nb_tx_queues);
+
+       for (i = 0; i < NDIS_HASH_INDCNT; i++)
+               hv->rss_ind[i] = i % dev->data->nb_rx_queues;
+
+       hn_rss_hash_init(hv, rss_conf);
+
        subchan = hv->num_queues - 1;
        if (subchan > 0) {
                err = hn_subchan_configure(hv, subchan);
@@ -397,15 +732,24 @@ static int hn_dev_configure(struct rte_eth_dev *dev)
                        return err;
                }
 
-               err = hn_rndis_conf_rss(hv, rss_conf);
+               err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
                if (err) {
                        PMD_DRV_LOG(NOTICE,
-                                   "rss configuration failed");
+                               "rss disable failed");
                        return err;
                }
+
+               if (rss_conf->rss_hf != 0) {
+                       err = hn_rndis_conf_rss(hv, 0);
+                       if (err) {
+                               PMD_DRV_LOG(NOTICE,
+                                           "initial RSS config failed");
+                               return err;
+                       }
+               }
        }
 
-       return hn_vf_configure(dev, dev_conf);
+       return hn_vf_configure_locked(dev, dev_conf);
 }
 
 static int hn_dev_stats_get(struct rte_eth_dev *dev,
@@ -452,7 +796,7 @@ static int hn_dev_stats_get(struct rte_eth_dev *dev,
        return 0;
 }
 
-static void
+static int
 hn_dev_stats_reset(struct rte_eth_dev *dev)
 {
        unsigned int i;
@@ -475,13 +819,20 @@ hn_dev_stats_reset(struct rte_eth_dev *dev)
 
                memset(&rxq->stats, 0, sizeof(struct hn_stats));
        }
+
+       return 0;
 }
 
-static void
+static int
 hn_dev_xstats_reset(struct rte_eth_dev *dev)
 {
-       hn_dev_stats_reset(dev);
-       hn_vf_xstats_reset(dev);
+       int ret;
+
+       ret = hn_dev_stats_reset(dev);
+       if (ret != 0)
+               return 0;
+
+       return hn_vf_xstats_reset(dev);
 }
 
 static int
@@ -572,9 +923,11 @@ hn_dev_xstats_get(struct rte_eth_dev *dev,
                        continue;
 
                stats = (const char *)&txq->stats;
-               for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
-                       xstats[count++].value = *(const uint64_t *)
+               for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
+                       xstats[count].id = count;
+                       xstats[count].value = *(const uint64_t *)
                                (stats + hn_stat_strings[t].offset);
+               }
        }
 
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
@@ -584,12 +937,14 @@ hn_dev_xstats_get(struct rte_eth_dev *dev,
                        continue;
 
                stats = (const char *)&rxq->stats;
-               for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
-                       xstats[count++].value = *(const uint64_t *)
+               for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
+                       xstats[count].id = count;
+                       xstats[count].value = *(const uint64_t *)
                                (stats + hn_stat_strings[t].offset);
+               }
        }
 
-       ret = hn_vf_xstats_get(dev, xstats + count, n - count);
+       ret = hn_vf_xstats_get(dev, xstats, count, n);
        if (ret < 0)
                return ret;
 
@@ -604,6 +959,14 @@ hn_dev_start(struct rte_eth_dev *dev)
 
        PMD_INIT_FUNC_TRACE();
 
+       /* Register to monitor hot plug events */
+       error = rte_dev_event_callback_register(NULL, netvsc_hotadd_callback,
+                                               hv);
+       if (error) {
+               PMD_DRV_LOG(ERR, "failed to register device event callback");
+               return error;
+       }
+
        error = hn_rndis_set_rxfilter(hv,
                                      NDIS_PACKET_TYPE_BROADCAST |
                                      NDIS_PACKET_TYPE_ALL_MULTICAST |
@@ -615,26 +978,42 @@ hn_dev_start(struct rte_eth_dev *dev)
        if (error)
                hn_rndis_set_rxfilter(hv, 0);
 
+       /* Initialize Link state */
+       if (error == 0)
+               hn_dev_link_update(dev, 0);
+
        return error;
 }
 
-static void
+static int
 hn_dev_stop(struct rte_eth_dev *dev)
 {
        struct hn_data *hv = dev->data->dev_private;
 
        PMD_INIT_FUNC_TRACE();
+       dev->data->dev_started = 0;
 
+       rte_dev_event_callback_unregister(NULL, netvsc_hotadd_callback, hv);
        hn_rndis_set_rxfilter(hv, 0);
-       hn_vf_stop(dev);
+       return hn_vf_stop(dev);
 }
 
-static void
-hn_dev_close(struct rte_eth_dev *dev __rte_unused)
+static int
+hn_dev_close(struct rte_eth_dev *dev)
 {
-       PMD_INIT_LOG(DEBUG, "close");
+       int ret;
+       struct hn_data *hv = dev->data->dev_private;
 
-       hn_vf_close(dev);
+       PMD_INIT_FUNC_TRACE();
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return 0;
+
+       rte_eal_alarm_cancel(netvsc_hotplug_retry, &hv->devargs);
+
+       ret = hn_vf_close(dev);
+       hn_dev_free_queues(dev);
+
+       return ret;
 }
 
 static const struct eth_dev_ops hn_eth_dev_ops = {
@@ -643,12 +1022,18 @@ static const struct eth_dev_ops hn_eth_dev_ops = {
        .dev_stop               = hn_dev_stop,
        .dev_close              = hn_dev_close,
        .dev_infos_get          = hn_dev_info_get,
+       .txq_info_get           = hn_dev_tx_queue_info,
+       .rxq_info_get           = hn_dev_rx_queue_info,
        .dev_supported_ptypes_get = hn_vf_supported_ptypes,
        .promiscuous_enable     = hn_dev_promiscuous_enable,
        .promiscuous_disable    = hn_dev_promiscuous_disable,
        .allmulticast_enable    = hn_dev_allmulticast_enable,
        .allmulticast_disable   = hn_dev_allmulticast_disable,
        .set_mc_addr_list       = hn_dev_mc_addr_list,
+       .reta_update            = hn_rss_reta_update,
+       .reta_query             = hn_rss_reta_query,
+       .rss_hash_update        = hn_rss_hash_update,
+       .rss_hash_conf_get      = hn_rss_hash_conf_get,
        .tx_queue_setup         = hn_dev_tx_queue_setup,
        .tx_queue_release       = hn_dev_tx_queue_release,
        .tx_done_cleanup        = hn_dev_tx_done_cleanup,
@@ -714,6 +1099,9 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)
 
        vmbus = container_of(device, struct rte_vmbus_device, device);
        eth_dev->dev_ops = &hn_eth_dev_ops;
+       eth_dev->rx_queue_count = hn_dev_rx_queue_count;
+       eth_dev->rx_descriptor_status = hn_dev_rx_queue_status;
+       eth_dev->tx_descriptor_status = hn_dev_tx_descriptor_status;
        eth_dev->tx_pkt_burst = &hn_xmit_pkts;
        eth_dev->rx_pkt_burst = &hn_recv_pkts;
 
@@ -724,14 +1112,32 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
                return 0;
 
-       /* Since Hyper-V only supports one MAC address, just use local data */
-       eth_dev->data->mac_addrs = &hv->mac_addr;
+       eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
+
+       /* Since Hyper-V only supports one MAC address */
+       eth_dev->data->mac_addrs = rte_calloc("hv_mac", HN_MAX_MAC_ADDRS,
+                                             sizeof(struct rte_ether_addr), 0);
+       if (eth_dev->data->mac_addrs == NULL) {
+               PMD_INIT_LOG(ERR,
+                            "Failed to allocate memory store MAC addresses");
+               return -ENOMEM;
+       }
 
        hv->vmbus = vmbus;
        hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
        hv->chim_res  = &vmbus->resource[HV_SEND_BUF_MAP];
        hv->port_id = eth_dev->data->port_id;
        hv->latency = HN_CHAN_LATENCY_NS;
+       hv->rx_copybreak = HN_RXCOPY_THRESHOLD;
+       hv->tx_copybreak = HN_TXCOPY_THRESHOLD;
+       hv->rx_extmbuf_enable = HN_RX_EXTMBUF_ENABLE;
+       hv->max_queues = 1;
+
+       rte_rwlock_init(&hv->vf_lock);
+       hv->vf_ctx.vf_vsc_switched = false;
+       hv->vf_ctx.vf_vsp_reported = false;
+       hv->vf_ctx.vf_attached = false;
+       hv->vf_ctx.vf_state = vf_unknown;
 
        err = hn_parse_args(eth_dev);
        if (err)
@@ -758,18 +1164,22 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)
        if (!hv->primary)
                return -ENOMEM;
 
-       err = hn_attach(hv, ETHER_MTU);
+       err = hn_attach(hv, RTE_ETHER_MTU);
        if  (err)
                goto failed;
 
-       err = hn_tx_pool_init(eth_dev);
+       err = hn_chim_init(eth_dev);
        if (err)
                goto failed;
 
-       err = hn_rndis_get_eaddr(hv, hv->mac_addr.addr_bytes);
+       err = hn_rndis_get_eaddr(hv, eth_dev->data->mac_addrs->addr_bytes);
        if (err)
                goto failed;
 
+       /* Multi queue requires later versions of windows server */
+       if (hv->nvs_ver < NVS_VERSION_5)
+               return 0;
+
        max_chan = rte_vmbus_max_channels(vmbus);
        PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
        if (max_chan <= 0)
@@ -781,12 +1191,10 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)
        hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
 
        /* If VF was reported but not added, do it now */
-       if (hv->vf_present && !hv->vf_dev) {
+       if (hv->vf_ctx.vf_vsp_reported && !hv->vf_ctx.vf_vsc_switched) {
                PMD_INIT_LOG(DEBUG, "Adding VF device");
 
                err = hn_vf_add(eth_dev, hv);
-               if (err)
-                       goto failed;
        }
 
        return 0;
@@ -794,6 +1202,7 @@ eth_hn_dev_init(struct rte_eth_dev *eth_dev)
 failed:
        PMD_INIT_LOG(NOTICE, "device init failed");
 
+       hn_chim_uninit(eth_dev);
        hn_detach(hv);
        return err;
 }
@@ -802,25 +1211,25 @@ static int
 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
 {
        struct hn_data *hv = eth_dev->data->dev_private;
+       int ret, ret_stop;
 
        PMD_INIT_FUNC_TRACE();
 
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
                return 0;
 
-       hn_dev_stop(eth_dev);
+       ret_stop = hn_dev_stop(eth_dev);
        hn_dev_close(eth_dev);
 
-       eth_dev->dev_ops = NULL;
-       eth_dev->tx_pkt_burst = NULL;
-       eth_dev->rx_pkt_burst = NULL;
-
        hn_detach(hv);
+       hn_chim_uninit(eth_dev);
        rte_vmbus_chan_close(hv->primary->chan);
        rte_free(hv->primary);
-       rte_eth_dev_owner_delete(hv->owner.id);
+       ret = rte_eth_dev_owner_delete(hv->owner.id);
+       if (ret != 0)
+               return ret;
 
-       return 0;
+       return ret_stop;
 }
 
 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
@@ -831,15 +1240,23 @@ static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
 
        PMD_INIT_FUNC_TRACE();
 
+       ret = rte_dev_event_monitor_start();
+       if (ret) {
+               PMD_DRV_LOG(ERR, "Failed to start device event monitoring");
+               return ret;
+       }
+
        eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
        if (!eth_dev)
                return -ENOMEM;
 
        ret = eth_hn_dev_init(eth_dev);
-       if (ret)
+       if (ret) {
                eth_dev_vmbus_release(eth_dev);
-       else
+               rte_dev_event_monitor_stop();
+       } else {
                rte_eth_dev_probing_finish(eth_dev);
+       }
 
        return ret;
 }
@@ -853,13 +1270,14 @@ static int eth_hn_remove(struct rte_vmbus_device *dev)
 
        eth_dev = rte_eth_dev_allocated(dev->device.name);
        if (!eth_dev)
-               return -ENODEV;
+               return 0; /* port already released */
 
        ret = eth_hn_dev_uninit(eth_dev);
        if (ret)
                return ret;
 
        eth_dev_vmbus_release(eth_dev);
+       rte_dev_event_monitor_stop();
        return 0;
 }
 
@@ -878,13 +1296,10 @@ static struct rte_vmbus_driver rte_netvsc_pmd = {
 
 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
-
-RTE_INIT(hn_init_log)
-{
-       hn_logtype_init = rte_log_register("pmd.net.netvsc.init");
-       if (hn_logtype_init >= 0)
-               rte_log_set_level(hn_logtype_init, RTE_LOG_NOTICE);
-       hn_logtype_driver = rte_log_register("pmd.net.netvsc.driver");
-       if (hn_logtype_driver >= 0)
-               rte_log_set_level(hn_logtype_driver, RTE_LOG_NOTICE);
-}
+RTE_LOG_REGISTER_SUFFIX(hn_logtype_init, init, NOTICE);
+RTE_LOG_REGISTER_SUFFIX(hn_logtype_driver, driver, NOTICE);
+RTE_PMD_REGISTER_PARAM_STRING(net_netvsc,
+                             NETVSC_ARG_LATENCY "=<uint32> "
+                             NETVSC_ARG_RXBREAK "=<uint32> "
+                             NETVSC_ARG_TXBREAK "=<uint32> "
+                             NETVSC_ARG_RX_EXTMBUF_ENABLE "=<0|1>");