vhost: fix guest notification setting
[dpdk.git] / lib / librte_vhost / vhost_user.c
index 3405cd8..c3c924f 100644 (file)
@@ -87,6 +87,8 @@ static const char *vhost_message_str[VHOST_USER_MAX] = {
        [VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
        [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
        [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
+       [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
+       [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
 };
 
 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
@@ -228,6 +230,25 @@ vhost_backend_cleanup(struct virtio_net *dev)
        dev->postcopy_listening = 0;
 }
 
+static void
+vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
+                             int enable)
+{
+       struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
+       struct vhost_virtqueue *vq = dev->virtqueue[index];
+
+       /* Configure guest notifications on enable */
+       if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
+               vhost_enable_guest_notification(dev, vq, vq->notif_enable);
+
+       if (vdpa_dev && vdpa_dev->ops->set_vring_state)
+               vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
+
+       if (dev->notify_ops->vring_state_changed)
+               dev->notify_ops->vring_state_changed(dev->vid,
+                               index, enable);
+}
+
 /*
  * This function just returns success at the moment unless
  * the device hasn't been initialised.
@@ -324,6 +345,9 @@ vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
                VHOST_LOG_CONFIG(ERR,
                        "(%d) received invalid negotiated features.\n",
                        dev->vid);
+               dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
+               dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
+
                return RTE_VHOST_MSG_RESULT_ERR;
        }
 
@@ -384,9 +408,10 @@ vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
        }
 
        vdpa_dev = dev->vdpa_dev;
-       if (vdpa_dev && vdpa_dev->ops->set_features)
+       if (vdpa_dev)
                vdpa_dev->ops->set_features(dev->vid);
 
+       dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
        return RTE_VHOST_MSG_RESULT_OK;
 }
 
@@ -462,12 +487,14 @@ vhost_user_set_vring_num(struct virtio_net **pdev,
        } else {
                if (vq->shadow_used_split)
                        rte_free(vq->shadow_used_split);
+
                vq->shadow_used_split = rte_malloc(NULL,
                                vq->size * sizeof(struct vring_used_elem),
                                RTE_CACHE_LINE_SIZE);
+
                if (!vq->shadow_used_split) {
                        VHOST_LOG_CONFIG(ERR,
-                                       "failed to allocate memory for shadow used ring.\n");
+                                       "failed to allocate memory for vq internal data.\n");
                        return RTE_VHOST_MSG_RESULT_ERR;
                }
        }
@@ -1056,6 +1083,13 @@ vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
        }
 
        if (dev->mem) {
+               if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
+                       struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
+
+                       if (vdpa_dev && vdpa_dev->ops->dev_close)
+                               vdpa_dev->ops->dev_close(dev->vid);
+                       dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
+               }
                free_mem_region(dev);
                rte_free(dev->mem);
                dev->mem = NULL;
@@ -1145,7 +1179,8 @@ vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
                        goto err_mmap;
                }
 
-               populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
+               populate = (dev->dequeue_zero_copy || dev->async_copy) ?
+                       MAP_POPULATE : 0;
                mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
                                 MAP_SHARED | populate, fd, 0);
 
@@ -1160,7 +1195,7 @@ vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
                reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
                                      mmap_offset;
 
-               if (dev->dequeue_zero_copy)
+               if (dev->dequeue_zero_copy || dev->async_copy)
                        if (add_guest_pages(dev, reg, alignment) < 0) {
                                VHOST_LOG_CONFIG(ERR,
                                        "adding guest pages to region %u failed.\n",
@@ -1304,27 +1339,41 @@ vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
 
        return rings_ok &&
               vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
-              vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
+              vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
+              vq->enabled;
 }
 
+#define VIRTIO_DEV_NUM_VQS_TO_BE_READY 2u
+
 static int
 virtio_is_ready(struct virtio_net *dev)
 {
        struct vhost_virtqueue *vq;
        uint32_t i;
 
-       if (dev->nr_vring == 0)
+       if (dev->flags & VIRTIO_DEV_READY)
+               return 1;
+
+       if (dev->nr_vring < VIRTIO_DEV_NUM_VQS_TO_BE_READY)
                return 0;
 
-       for (i = 0; i < dev->nr_vring; i++) {
+       for (i = 0; i < VIRTIO_DEV_NUM_VQS_TO_BE_READY; i++) {
                vq = dev->virtqueue[i];
 
                if (!vq_is_ready(dev, vq))
                        return 0;
        }
 
-       VHOST_LOG_CONFIG(INFO,
-               "virtio is now ready for processing.\n");
+       /* If supported, ensure the frontend is really done with config */
+       if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
+               if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
+                       return 0;
+
+       dev->flags |= VIRTIO_DEV_READY;
+
+       if (!(dev->flags & VIRTIO_DEV_RUNNING))
+               VHOST_LOG_CONFIG(INFO,
+                       "virtio is now ready for processing.\n");
        return 1;
 }
 
@@ -1594,6 +1643,12 @@ vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
                "vring call idx:%d file:%d\n", file.index, file.fd);
 
        vq = dev->virtqueue[file.index];
+
+       if (vq->ready) {
+               vq->ready = 0;
+               vhost_user_notify_queue_state(dev, file.index, 0);
+       }
+
        if (vq->callfd >= 0)
                close(vq->callfd);
 
@@ -1852,6 +1907,11 @@ vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
                                dev->vid, file.index, 1);
        }
 
+       if (vq->ready) {
+               vq->ready = 0;
+               vhost_user_notify_queue_state(dev, file.index, 0);
+       }
+
        if (vq->kickfd >= 0)
                close(vq->kickfd);
        vq->kickfd = file.fd;
@@ -1879,6 +1939,7 @@ free_zmbufs(struct vhost_virtqueue *vq)
        drain_zmbuf_list(vq);
 
        rte_free(vq->zmbufs);
+       vq->zmbufs = NULL;
 }
 
 /*
@@ -1943,6 +2004,12 @@ vhost_user_get_vring_base(struct virtio_net **pdev,
        } else {
                rte_free(vq->shadow_used_split);
                vq->shadow_used_split = NULL;
+               if (vq->async_pkts_pending)
+                       rte_free(vq->async_pkts_pending);
+               if (vq->async_pending_info)
+                       rte_free(vq->async_pending_info);
+               vq->async_pkts_pending = NULL;
+               vq->async_pending_info = NULL;
        }
 
        rte_free(vq->batch_copy_elems);
@@ -1968,7 +2035,6 @@ vhost_user_set_vring_enable(struct virtio_net **pdev,
        struct virtio_net *dev = *pdev;
        int enable = (int)msg->payload.state.num;
        int index = (int)msg->payload.state.index;
-       struct rte_vdpa_device *vdpa_dev;
 
        if (validate_msg_fds(msg, 0) != 0)
                return RTE_VHOST_MSG_RESULT_ERR;
@@ -1977,13 +2043,13 @@ vhost_user_set_vring_enable(struct virtio_net **pdev,
                "set queue enable: %d to qp idx: %d\n",
                enable, index);
 
-       vdpa_dev = dev->vdpa_dev;
-       if (vdpa_dev && vdpa_dev->ops->set_vring_state)
-               vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
-
-       if (dev->notify_ops->vring_state_changed)
-               dev->notify_ops->vring_state_changed(dev->vid,
-                               index, enable);
+       if (!enable && dev->virtqueue[index]->async_registered) {
+               if (dev->virtqueue[index]->async_pkts_inflight_n) {
+                       VHOST_LOG_CONFIG(ERR, "failed to disable vring. "
+                       "async inflight packets must be completed first\n");
+                       return RTE_VHOST_MSG_RESULT_ERR;
+               }
+       }
 
        /* On disable, rings have to be stopped being processed. */
        if (!enable && dev->dequeue_zero_copy)
@@ -2420,6 +2486,68 @@ vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
        return RTE_VHOST_MSG_RESULT_REPLY;
 }
 
+static int
+vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
+                     int main_fd __rte_unused)
+{
+       struct virtio_net *dev = *pdev;
+
+       if (validate_msg_fds(msg, 0) != 0)
+               return RTE_VHOST_MSG_RESULT_ERR;
+
+       msg->payload.u64 = dev->status;
+       msg->size = sizeof(msg->payload.u64);
+       msg->fd_num = 0;
+
+       return RTE_VHOST_MSG_RESULT_REPLY;
+}
+
+static int
+vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
+                       int main_fd __rte_unused)
+{
+       struct virtio_net *dev = *pdev;
+
+       if (validate_msg_fds(msg, 0) != 0)
+               return RTE_VHOST_MSG_RESULT_ERR;
+
+       /* As per Virtio specification, the device status is 8bits long */
+       if (msg->payload.u64 > UINT8_MAX) {
+               VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
+                               msg->payload.u64);
+               return RTE_VHOST_MSG_RESULT_ERR;
+       }
+
+       dev->status = msg->payload.u64;
+
+       if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
+           (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
+               VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
+               /*
+                * Clear the bit to let the driver know about the feature
+                * negotiation failure
+                */
+               dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
+       }
+
+       VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
+                       "\t-ACKNOWLEDGE: %u\n"
+                       "\t-DRIVER: %u\n"
+                       "\t-FEATURES_OK: %u\n"
+                       "\t-DRIVER_OK: %u\n"
+                       "\t-DEVICE_NEED_RESET: %u\n"
+                       "\t-FAILED: %u\n",
+                       dev->status,
+                       !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
+                       !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
+                       !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
+                       !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
+                       !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
+                       !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
+
+       return RTE_VHOST_MSG_RESULT_OK;
+}
+
 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
                                        struct VhostUserMsg *msg,
                                        int main_fd);
@@ -2452,6 +2580,8 @@ static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
        [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
        [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
        [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
+       [VHOST_USER_SET_STATUS] = vhost_user_set_status,
+       [VHOST_USER_GET_STATUS] = vhost_user_get_status,
 };
 
 /* return bytes# of read on success or negative val on failure. */
@@ -2611,6 +2741,7 @@ vhost_user_msg_handler(int vid, int fd)
        int unlock_required = 0;
        bool handled;
        int request;
+       uint32_t i;
 
        dev = get_device(vid);
        if (dev == NULL)
@@ -2683,8 +2814,10 @@ vhost_user_msg_handler(int vid, int fd)
        case VHOST_USER_SEND_RARP:
        case VHOST_USER_NET_SET_MTU:
        case VHOST_USER_SET_SLAVE_REQ_FD:
-               vhost_user_lock_all_queue_pairs(dev);
-               unlock_required = 1;
+               if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
+                       vhost_user_lock_all_queue_pairs(dev);
+                       unlock_required = 1;
+               }
                break;
        default:
                break;
@@ -2784,29 +2917,44 @@ skip_to_post_handle:
                return -1;
        }
 
-       if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
-               dev->flags |= VIRTIO_DEV_READY;
-
-               if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
-                       if (dev->dequeue_zero_copy) {
-                               VHOST_LOG_CONFIG(INFO,
-                                               "dequeue zero copy is enabled\n");
-                       }
+       for (i = 0; i < dev->nr_vring; i++) {
+               struct vhost_virtqueue *vq = dev->virtqueue[i];
+               bool cur_ready = vq_is_ready(dev, vq);
 
-                       if (dev->notify_ops->new_device(dev->vid) == 0)
-                               dev->flags |= VIRTIO_DEV_RUNNING;
+               if (cur_ready != (vq && vq->ready)) {
+                       vq->ready = cur_ready;
+                       vhost_user_notify_queue_state(dev, i, cur_ready);
                }
        }
 
+
+       if (!virtio_is_ready(dev))
+               goto out;
+
+       /*
+        * Virtio is now ready. If not done already, it is time
+        * to notify the application it can process the rings and
+        * configure the vDPA device if present.
+        */
+
+       if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
+               if (dev->notify_ops->new_device(dev->vid) == 0)
+                       dev->flags |= VIRTIO_DEV_RUNNING;
+       }
+
        vdpa_dev = dev->vdpa_dev;
-       if (vdpa_dev && virtio_is_ready(dev) &&
-                       !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
-                       msg.request.master == VHOST_USER_SET_VRING_CALL) {
-               if (vdpa_dev->ops->dev_conf)
-                       vdpa_dev->ops->dev_conf(dev->vid);
-               dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
+       if (!vdpa_dev)
+               goto out;
+
+       if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
+               if (vdpa_dev->ops->dev_conf(dev->vid))
+                       VHOST_LOG_CONFIG(ERR,
+                                        "Failed to configure vDPA device\n");
+               else
+                       dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
        }
 
+out:
        return 0;
 }
 
@@ -2943,13 +3091,13 @@ static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
        return process_slave_message_reply(dev, &msg);
 }
 
-int rte_vhost_host_notifier_ctrl(int vid, bool enable)
+int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
 {
        struct virtio_net *dev;
        struct rte_vdpa_device *vdpa_dev;
        int vfio_device_fd, ret = 0;
        uint64_t offset, size;
-       unsigned int i;
+       unsigned int i, q_start, q_last;
 
        dev = get_device(vid);
        if (!dev)
@@ -2969,6 +3117,16 @@ int rte_vhost_host_notifier_ctrl(int vid, bool enable)
                        (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
                return -ENOTSUP;
 
+       if (qid == RTE_VHOST_QUEUE_ALL) {
+               q_start = 0;
+               q_last = dev->nr_vring - 1;
+       } else {
+               if (qid >= dev->nr_vring)
+                       return -EINVAL;
+               q_start = qid;
+               q_last = qid;
+       }
+
        RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
        RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
 
@@ -2977,7 +3135,7 @@ int rte_vhost_host_notifier_ctrl(int vid, bool enable)
                return -ENOTSUP;
 
        if (enable) {
-               for (i = 0; i < dev->nr_vring; i++) {
+               for (i = q_start; i <= q_last; i++) {
                        if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
                                        &size) < 0) {
                                ret = -ENOTSUP;
@@ -2992,7 +3150,7 @@ int rte_vhost_host_notifier_ctrl(int vid, bool enable)
                }
        } else {
 disable:
-               for (i = 0; i < dev->nr_vring; i++) {
+               for (i = q_start; i <= q_last; i++) {
                        vhost_user_slave_set_vring_host_notifier(dev, i, -1,
                                        0, 0);
                }