net/virtio-user: support reply-ack
[dpdk.git] / drivers / net / virtio / virtio_user / virtio_user_dev.c
index e21e3ec..0a6991b 100644 (file)
@@ -13,6 +13,7 @@
 #include <sys/types.h>
 #include <sys/stat.h>
 
+#include <rte_string_fns.h>
 #include <rte_eal_memconfig.h>
 
 #include "vhost.h"
@@ -43,15 +44,26 @@ virtio_user_kick_queue(struct virtio_user_dev *dev, uint32_t queue_sel)
        struct vhost_vring_file file;
        struct vhost_vring_state state;
        struct vring *vring = &dev->vrings[queue_sel];
+       struct vring_packed *pq_vring = &dev->packed_vrings[queue_sel];
        struct vhost_vring_addr addr = {
                .index = queue_sel,
-               .desc_user_addr = (uint64_t)(uintptr_t)vring->desc,
-               .avail_user_addr = (uint64_t)(uintptr_t)vring->avail,
-               .used_user_addr = (uint64_t)(uintptr_t)vring->used,
                .log_guest_addr = 0,
                .flags = 0, /* disable log */
        };
 
+       if (dev->features & (1ULL << VIRTIO_F_RING_PACKED)) {
+               addr.desc_user_addr =
+                       (uint64_t)(uintptr_t)pq_vring->desc;
+               addr.avail_user_addr =
+                       (uint64_t)(uintptr_t)pq_vring->driver;
+               addr.used_user_addr =
+                       (uint64_t)(uintptr_t)pq_vring->device;
+       } else {
+               addr.desc_user_addr = (uint64_t)(uintptr_t)vring->desc;
+               addr.avail_user_addr = (uint64_t)(uintptr_t)vring->avail;
+               addr.used_user_addr = (uint64_t)(uintptr_t)vring->used;
+       }
+
        state.index = queue_sel;
        state.num = vring->num;
        dev->ops->send_request(dev, VHOST_USER_SET_VRING_NUM, &state);
@@ -113,7 +125,6 @@ is_vhost_user_by_type(const char *path)
 int
 virtio_user_start_device(struct virtio_user_dev *dev)
 {
-       struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
        uint64_t features;
        int ret;
 
@@ -130,7 +141,7 @@ virtio_user_start_device(struct virtio_user_dev *dev)
         * replaced when we get proper supports from the
         * memory subsystem in the future.
         */
-       rte_rwlock_read_lock(&mcfg->memory_hotplug_lock);
+       rte_mcfg_mem_read_lock();
        pthread_mutex_lock(&dev->mutex);
 
        if (is_vhost_user_by_type(dev->path) && dev->vhostfd < 0)
@@ -140,8 +151,10 @@ virtio_user_start_device(struct virtio_user_dev *dev)
        if (virtio_user_queue_setup(dev, virtio_user_create_queue) < 0)
                goto error;
 
-       /* Step 1: set features */
+       /* Step 1: negotiate protocol features & set features */
        features = dev->features;
+
+
        /* Strip VIRTIO_NET_F_MAC, as MAC address is handled in vdev init */
        features &= ~(1ull << VIRTIO_NET_F_MAC);
        /* Strip VIRTIO_NET_F_CTRL_VQ, as devices do not really need to know */
@@ -168,12 +181,12 @@ virtio_user_start_device(struct virtio_user_dev *dev)
 
        dev->started = true;
        pthread_mutex_unlock(&dev->mutex);
-       rte_rwlock_read_unlock(&mcfg->memory_hotplug_lock);
+       rte_mcfg_mem_read_unlock();
 
        return 0;
 error:
        pthread_mutex_unlock(&dev->mutex);
-       rte_rwlock_read_unlock(&mcfg->memory_hotplug_lock);
+       rte_mcfg_mem_read_unlock();
        /* TODO: free resource here or caller to check */
        return -1;
 }
@@ -213,17 +226,13 @@ out:
 static inline void
 parse_mac(struct virtio_user_dev *dev, const char *mac)
 {
-       int i, r;
-       uint32_t tmp[ETHER_ADDR_LEN];
+       struct rte_ether_addr tmp;
 
        if (!mac)
                return;
 
-       r = sscanf(mac, "%x:%x:%x:%x:%x:%x", &tmp[0],
-                       &tmp[1], &tmp[2], &tmp[3], &tmp[4], &tmp[5]);
-       if (r == ETHER_ADDR_LEN) {
-               for (i = 0; i < ETHER_ADDR_LEN; ++i)
-                       dev->mac_addr[i] = (uint8_t)tmp[i];
+       if (rte_ether_unformat_addr(mac, &tmp) == 0) {
+               memcpy(dev->mac_addr, &tmp, RTE_ETHER_ADDR_LEN);
                dev->mac_specified = 1;
        } else {
                /* ignore the wrong mac, use random mac */
@@ -308,9 +317,9 @@ virtio_user_fill_intr_handle(struct virtio_user_dev *dev)
 
 static void
 virtio_user_mem_event_cb(enum rte_mem_event type __rte_unused,
-                                                const void *addr __rte_unused,
-                                                size_t len __rte_unused,
-                                                void *arg)
+                        const void *addr,
+                        size_t len __rte_unused,
+                        void *arg)
 {
        struct virtio_user_dev *dev = arg;
        struct rte_memseg_list *msl;
@@ -411,22 +420,30 @@ virtio_user_dev_setup(struct virtio_user_dev *dev)
         1ULL << VIRTIO_F_IN_ORDER              |       \
         1ULL << VIRTIO_F_VERSION_1             |       \
         1ULL << VIRTIO_F_RING_PACKED           |       \
-        1ULL << VIRTIO_RING_F_EVENT_IDX)
+        1ULL << VHOST_USER_F_PROTOCOL_FEATURES)
+
+#define VIRTIO_USER_SUPPORTED_PROTOCOL_FEATURES                \
+       (1ULL << VHOST_USER_PROTOCOL_F_MQ |             \
+        1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK)
 
 int
 virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
                     int cq, int queue_size, const char *mac, char **ifname,
-                    int mrg_rxbuf, int in_order, int packed_vq)
+                    int server, int mrg_rxbuf, int in_order, int packed_vq)
 {
+       uint64_t protocol_features = 0;
+
        pthread_mutex_init(&dev->mutex, NULL);
-       snprintf(dev->path, PATH_MAX, "%s", path);
+       strlcpy(dev->path, path, PATH_MAX);
        dev->started = 0;
        dev->max_queue_pairs = queues;
        dev->queue_pairs = 1; /* mq disabled by default */
        dev->queue_size = queue_size;
+       dev->is_server = server;
        dev->mac_specified = 0;
        dev->frontend_features = 0;
        dev->unsupported_features = ~VIRTIO_USER_SUPPORTED_FEATURES;
+       dev->protocol_features = VIRTIO_USER_SUPPORTED_PROTOCOL_FEATURES;
        parse_mac(dev, mac);
 
        if (*ifname) {
@@ -439,6 +456,10 @@ virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
                return -1;
        }
 
+       if (!is_vhost_user_by_type(dev->path))
+               dev->unsupported_features |=
+                       (1ULL << VHOST_USER_F_PROTOCOL_FEATURES);
+
        if (!dev->is_server) {
                if (dev->ops->send_request(dev, VHOST_USER_SET_OWNER,
                                           NULL) < 0) {
@@ -453,6 +474,27 @@ virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
                                     strerror(errno));
                        return -1;
                }
+
+
+               if (dev->device_features &
+                               (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) {
+                       if (dev->ops->send_request(dev,
+                                       VHOST_USER_GET_PROTOCOL_FEATURES,
+                                       &protocol_features))
+                               return -1;
+
+                       dev->protocol_features &= protocol_features;
+
+                       if (dev->ops->send_request(dev,
+                                       VHOST_USER_SET_PROTOCOL_FEATURES,
+                                       &dev->protocol_features))
+                               return -1;
+
+                       if (!(dev->protocol_features &
+                                       (1ULL << VHOST_USER_PROTOCOL_F_MQ)))
+                               dev->unsupported_features |=
+                                       (1ull << VIRTIO_NET_F_MQ);
+               }
        } else {
                /* We just pretend vhost-user can support all these features.
                 * Note that this could be problematic that if some feature is
@@ -462,21 +504,16 @@ virtio_user_dev_init(struct virtio_user_dev *dev, char *path, int queues,
                dev->device_features = VIRTIO_USER_SUPPORTED_FEATURES;
        }
 
+
+
        if (!mrg_rxbuf)
                dev->unsupported_features |= (1ull << VIRTIO_NET_F_MRG_RXBUF);
 
        if (!in_order)
                dev->unsupported_features |= (1ull << VIRTIO_F_IN_ORDER);
 
-       if (packed_vq) {
-               if (cq) {
-                       PMD_INIT_LOG(ERR, "control vq not supported yet with "
-                                         "packed virtqueues\n");
-                       return -1;
-               }
-       } else {
+       if (!packed_vq)
                dev->unsupported_features |= (1ull << VIRTIO_F_RING_PACKED);
-       }
 
        if (dev->mac_specified)
                dev->frontend_features |= (1ull << VIRTIO_NET_F_MAC);
@@ -537,7 +574,8 @@ virtio_user_dev_uninit(struct virtio_user_dev *dev)
                close(dev->kickfds[i]);
        }
 
-       close(dev->vhostfd);
+       if (dev->vhostfd >= 0)
+               close(dev->vhostfd);
 
        if (dev->is_server && dev->listenfd >= 0) {
                close(dev->listenfd);
@@ -545,8 +583,11 @@ virtio_user_dev_uninit(struct virtio_user_dev *dev)
        }
 
        if (dev->vhostfds) {
-               for (i = 0; i < dev->max_queue_pairs; ++i)
+               for (i = 0; i < dev->max_queue_pairs; ++i) {
                        close(dev->vhostfds[i]);
+                       if (dev->tapfds[i] >= 0)
+                               close(dev->tapfds[i]);
+               }
                free(dev->vhostfds);
                free(dev->tapfds);
        }
@@ -613,6 +654,10 @@ virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring,
 
                queues = *(uint16_t *)(uintptr_t)vring->desc[idx_data].addr;
                status = virtio_user_handle_mq(dev, queues);
+       } else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
+                  hdr->class == VIRTIO_NET_CTRL_MAC ||
+                  hdr->class == VIRTIO_NET_CTRL_VLAN) {
+               status = 0;
        }
 
        /* Update status */
@@ -621,6 +666,98 @@ virtio_user_handle_ctrl_msg(struct virtio_user_dev *dev, struct vring *vring,
        return n_descs;
 }
 
+static inline int
+desc_is_avail(struct vring_packed_desc *desc, bool wrap_counter)
+{
+       uint16_t flags = __atomic_load_n(&desc->flags, __ATOMIC_ACQUIRE);
+
+       return wrap_counter == !!(flags & VRING_PACKED_DESC_F_AVAIL) &&
+               wrap_counter != !!(flags & VRING_PACKED_DESC_F_USED);
+}
+
+static uint32_t
+virtio_user_handle_ctrl_msg_packed(struct virtio_user_dev *dev,
+                                  struct vring_packed *vring,
+                                  uint16_t idx_hdr)
+{
+       struct virtio_net_ctrl_hdr *hdr;
+       virtio_net_ctrl_ack status = ~0;
+       uint16_t idx_data, idx_status;
+       /* initialize to one, header is first */
+       uint32_t n_descs = 1;
+
+       /* locate desc for header, data, and status */
+       idx_data = idx_hdr + 1;
+       if (idx_data >= dev->queue_size)
+               idx_data -= dev->queue_size;
+
+       n_descs++;
+
+       idx_status = idx_data;
+       while (vring->desc[idx_status].flags & VRING_DESC_F_NEXT) {
+               idx_status++;
+               if (idx_status >= dev->queue_size)
+                       idx_status -= dev->queue_size;
+               n_descs++;
+       }
+
+       hdr = (void *)(uintptr_t)vring->desc[idx_hdr].addr;
+       if (hdr->class == VIRTIO_NET_CTRL_MQ &&
+           hdr->cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
+               uint16_t queues;
+
+               queues = *(uint16_t *)(uintptr_t)
+                               vring->desc[idx_data].addr;
+               status = virtio_user_handle_mq(dev, queues);
+       } else if (hdr->class == VIRTIO_NET_CTRL_RX  ||
+                  hdr->class == VIRTIO_NET_CTRL_MAC ||
+                  hdr->class == VIRTIO_NET_CTRL_VLAN) {
+               status = 0;
+       }
+
+       /* Update status */
+       *(virtio_net_ctrl_ack *)(uintptr_t)
+               vring->desc[idx_status].addr = status;
+
+       /* Update used descriptor */
+       vring->desc[idx_hdr].id = vring->desc[idx_status].id;
+       vring->desc[idx_hdr].len = sizeof(status);
+
+       return n_descs;
+}
+
+void
+virtio_user_handle_cq_packed(struct virtio_user_dev *dev, uint16_t queue_idx)
+{
+       struct virtio_user_queue *vq = &dev->packed_queues[queue_idx];
+       struct vring_packed *vring = &dev->packed_vrings[queue_idx];
+       uint16_t n_descs, flags;
+
+       /* Perform a load-acquire barrier in desc_is_avail to
+        * enforce the ordering between desc flags and desc
+        * content.
+        */
+       while (desc_is_avail(&vring->desc[vq->used_idx],
+                            vq->used_wrap_counter)) {
+
+               n_descs = virtio_user_handle_ctrl_msg_packed(dev, vring,
+                               vq->used_idx);
+
+               flags = VRING_DESC_F_WRITE;
+               if (vq->used_wrap_counter)
+                       flags |= VRING_PACKED_DESC_F_AVAIL_USED;
+
+               __atomic_store_n(&vring->desc[vq->used_idx].flags, flags,
+                                __ATOMIC_RELEASE);
+
+               vq->used_idx += n_descs;
+               if (vq->used_idx >= dev->queue_size) {
+                       vq->used_idx -= dev->queue_size;
+                       vq->used_wrap_counter ^= 1;
+               }
+       }
+}
+
 void
 virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
 {
@@ -630,17 +767,19 @@ virtio_user_handle_cq(struct virtio_user_dev *dev, uint16_t queue_idx)
        struct vring *vring = &dev->vrings[queue_idx];
 
        /* Consume avail ring, using used ring idx as first one */
-       while (vring->used->idx != vring->avail->idx) {
-               avail_idx = (vring->used->idx) & (vring->num - 1);
+       while (__atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
+              != vring->avail->idx) {
+               avail_idx = __atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
+                           & (vring->num - 1);
                desc_idx = vring->avail->ring[avail_idx];
 
                n_descs = virtio_user_handle_ctrl_msg(dev, vring, desc_idx);
 
                /* Update used ring */
                uep = &vring->used->ring[avail_idx];
-               uep->id = avail_idx;
+               uep->id = desc_idx;
                uep->len = n_descs;
 
-               vring->used->idx++;
+               __atomic_add_fetch(&vring->used->idx, 1, __ATOMIC_RELAXED);
        }
 }