eal/windows: add missing SPDX license tag
[dpdk.git] / lib / librte_vhost / vhost_user.c
index 91482cc..a60bb94 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);
@@ -97,8 +99,15 @@ close_msg_fds(struct VhostUserMsg *msg)
 {
        int i;
 
-       for (i = 0; i < msg->fd_num; i++)
-               close(msg->fds[i]);
+       for (i = 0; i < msg->fd_num; i++) {
+               int fd = msg->fds[i];
+
+               if (fd == -1)
+                       continue;
+
+               msg->fds[i] = -1;
+               close(fd);
+       }
 }
 
 /*
@@ -132,47 +141,15 @@ get_blk_size(int fd)
        return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
 }
 
-/*
- * Reclaim all the outstanding zmbufs for a virtqueue.
- */
-static void
-drain_zmbuf_list(struct vhost_virtqueue *vq)
-{
-       struct zcopy_mbuf *zmbuf, *next;
-
-       for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
-            zmbuf != NULL; zmbuf = next) {
-               next = TAILQ_NEXT(zmbuf, next);
-
-               while (!mbuf_is_consumed(zmbuf->mbuf))
-                       usleep(1000);
-
-               TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
-               restore_mbuf(zmbuf->mbuf);
-               rte_pktmbuf_free(zmbuf->mbuf);
-               put_zmbuf(zmbuf);
-               vq->nr_zmbuf -= 1;
-       }
-}
-
 static void
 free_mem_region(struct virtio_net *dev)
 {
        uint32_t i;
        struct rte_vhost_mem_region *reg;
-       struct vhost_virtqueue *vq;
 
        if (!dev || !dev->mem)
                return;
 
-       if (dev->dequeue_zero_copy) {
-               for (i = 0; i < dev->nr_vring; i++) {
-                       vq = dev->virtqueue[i];
-                       if (vq)
-                               drain_zmbuf_list(vq);
-               }
-       }
-
        for (i = 0; i < dev->mem->nregions; i++) {
                reg = &dev->mem->regions[i];
                if (reg->host_user_addr) {
@@ -191,7 +168,7 @@ vhost_backend_cleanup(struct virtio_net *dev)
                dev->mem = NULL;
        }
 
-       free(dev->guest_pages);
+       rte_free(dev->guest_pages);
        dev->guest_pages = NULL;
 
        if (dev->log_addr) {
@@ -206,7 +183,7 @@ vhost_backend_cleanup(struct virtio_net *dev)
                        dev->inflight_info->addr = NULL;
                }
 
-               if (dev->inflight_info->fd > 0) {
+               if (dev->inflight_info->fd >= 0) {
                        close(dev->inflight_info->fd);
                        dev->inflight_info->fd = -1;
                }
@@ -228,6 +205,25 @@ 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.
@@ -315,7 +311,6 @@ vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
        uint64_t features = msg->payload.u64;
        uint64_t vhost_features = 0;
        struct rte_vdpa_device *vdpa_dev;
-       int did = -1;
 
        if (validate_msg_fds(msg, 0) != 0)
                return RTE_VHOST_MSG_RESULT_ERR;
@@ -325,6 +320,9 @@ 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;
        }
 
@@ -350,7 +348,9 @@ vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 
        dev->features = features;
        if (dev->features &
-               ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
+               ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
+                (1ULL << VIRTIO_F_VERSION_1) |
+                (1ULL << VIRTIO_F_RING_PACKED))) {
                dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
        } else {
                dev->vhost_hlen = sizeof(struct virtio_net_hdr);
@@ -384,11 +384,11 @@ vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
                }
        }
 
-       did = dev->vdpa_dev_id;
-       vdpa_dev = rte_vdpa_get_device(did);
-       if (vdpa_dev && vdpa_dev->ops->set_features)
+       vdpa_dev = dev->vdpa_dev;
+       if (vdpa_dev)
                vdpa_dev->ops->set_features(dev->vid);
 
+       dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
        return RTE_VHOST_MSG_RESULT_OK;
 }
 
@@ -431,23 +431,6 @@ vhost_user_set_vring_num(struct virtio_net **pdev,
                return RTE_VHOST_MSG_RESULT_ERR;
        }
 
-       if (dev->dequeue_zero_copy) {
-               vq->nr_zmbuf = 0;
-               vq->last_zmbuf_idx = 0;
-               vq->zmbuf_size = vq->size;
-               if (vq->zmbufs)
-                       rte_free(vq->zmbufs);
-               vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
-                                        sizeof(struct zcopy_mbuf), 0);
-               if (vq->zmbufs == NULL) {
-                       VHOST_LOG_CONFIG(WARNING,
-                               "failed to allocate mem for zero copy; "
-                               "zero copy is force disabled\n");
-                       dev->dequeue_zero_copy = 0;
-               }
-               TAILQ_INIT(&vq->zmbuf_list);
-       }
-
        if (vq_is_packed(dev)) {
                if (vq->shadow_used_packed)
                        rte_free(vq->shadow_used_packed);
@@ -464,12 +447,14 @@ 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;
                }
        }
@@ -499,7 +484,6 @@ numa_realloc(struct virtio_net *dev, int index)
        int oldnode, newnode;
        struct virtio_net *old_dev;
        struct vhost_virtqueue *old_vq, *vq;
-       struct zcopy_mbuf *new_zmbuf;
        struct vring_used_elem *new_shadow_used_split;
        struct vring_used_elem_packed *new_shadow_used_packed;
        struct batch_copy_elem *new_batch_copy_elems;
@@ -530,16 +514,6 @@ numa_realloc(struct virtio_net *dev, int index)
                        return dev;
 
                memcpy(vq, old_vq, sizeof(*vq));
-               TAILQ_INIT(&vq->zmbuf_list);
-
-               if (dev->dequeue_zero_copy) {
-                       new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
-                                       sizeof(struct zcopy_mbuf), 0, newnode);
-                       if (new_zmbuf) {
-                               rte_free(vq->zmbufs);
-                               vq->zmbufs = new_zmbuf;
-                       }
-               }
 
                if (vq_is_packed(dev)) {
                        new_shadow_used_packed = rte_malloc_socket(NULL,
@@ -656,13 +630,11 @@ ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 {
        if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
                uint64_t vva;
-               uint64_t req_size = *size;
 
-               vva = vhost_user_iotlb_cache_find(vq, ra,
+               vhost_user_iotlb_rd_lock(vq);
+               vva = vhost_iova_to_vva(dev, vq, ra,
                                        size, VHOST_ACCESS_RW);
-               if (req_size != *size)
-                       vhost_user_iotlb_miss(dev, (ra + *size),
-                                             VHOST_ACCESS_RW);
+               vhost_user_iotlb_rd_unlock(vq);
 
                return vva;
        }
@@ -670,37 +642,16 @@ ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
        return qva_to_vva(dev, ra, size);
 }
 
-/*
- * Converts vring log address to GPA
- * If IOMMU is enabled, the log address is IOVA
- * If IOMMU not enabled, the log address is already GPA
- */
 static uint64_t
-translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
-               uint64_t log_addr)
+log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
 {
-       if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
-               const uint64_t exp_size = sizeof(struct vring_used) +
-                       sizeof(struct vring_used_elem) * vq->size;
-               uint64_t hva, gpa;
-               uint64_t size = exp_size;
-
-               hva = vhost_iova_to_vva(dev, vq, log_addr,
-                                       &size, VHOST_ACCESS_RW);
-               if (size != exp_size)
-                       return 0;
+       uint64_t log_gpa;
 
-               gpa = hva_to_gpa(dev, hva, exp_size);
-               if (!gpa) {
-                       VHOST_LOG_CONFIG(ERR,
-                               "VQ: Failed to find GPA for log_addr: 0x%" PRIx64 " hva: 0x%" PRIx64 "\n",
-                               log_addr, hva);
-                       return 0;
-               }
-               return gpa;
+       vhost_user_iotlb_rd_lock(vq);
+       log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
+       vhost_user_iotlb_rd_unlock(vq);
 
-       } else
-               return log_addr;
+       return log_gpa;
 }
 
 static struct virtio_net *
@@ -712,7 +663,7 @@ translate_ring_addresses(struct virtio_net *dev, int vq_index)
 
        if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
                vq->log_guest_addr =
-                       translate_log_addr(dev, vq, addr->log_guest_addr);
+                       log_addr_to_gpa(dev, vq);
                if (vq->log_guest_addr == 0) {
                        VHOST_LOG_CONFIG(DEBUG,
                                "(%d) failed to map log_guest_addr.\n",
@@ -926,11 +877,12 @@ add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
        if (dev->nr_guest_pages == dev->max_guest_pages) {
                dev->max_guest_pages *= 2;
                old_pages = dev->guest_pages;
-               dev->guest_pages = realloc(dev->guest_pages,
-                                       dev->max_guest_pages * sizeof(*page));
-               if (!dev->guest_pages) {
+               dev->guest_pages = rte_realloc(dev->guest_pages,
+                                       dev->max_guest_pages * sizeof(*page),
+                                       RTE_CACHE_LINE_SIZE);
+               if (dev->guest_pages == NULL) {
                        VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
-                       free(old_pages);
+                       rte_free(old_pages);
                        return -1;
                }
        }
@@ -987,6 +939,12 @@ add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
                reg_size -= size;
        }
 
+       /* sort guest page array if over binary search threshold */
+       if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
+               qsort((void *)dev->guest_pages, dev->nr_guest_pages,
+                       sizeof(struct guest_page), guest_page_addrcmp);
+       }
+
        return 0;
 }
 
@@ -1040,6 +998,195 @@ vhost_memory_changed(struct VhostUserMemory *new,
        return false;
 }
 
+#ifdef RTE_LIBRTE_VHOST_POSTCOPY
+static int
+vhost_user_postcopy_region_register(struct virtio_net *dev,
+               struct rte_vhost_mem_region *reg)
+{
+       struct uffdio_register reg_struct;
+
+       /*
+        * Let's register all the mmap'ed area to ensure
+        * alignment on page boundary.
+        */
+       reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
+       reg_struct.range.len = reg->mmap_size;
+       reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
+
+       if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
+                               &reg_struct)) {
+               VHOST_LOG_CONFIG(ERR, "Failed to register ufd for region "
+                               "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
+                               (uint64_t)reg_struct.range.start,
+                               (uint64_t)reg_struct.range.start +
+                               (uint64_t)reg_struct.range.len - 1,
+                               dev->postcopy_ufd,
+                               strerror(errno));
+               return -1;
+       }
+
+       VHOST_LOG_CONFIG(INFO, "\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
+                       (uint64_t)reg_struct.range.start,
+                       (uint64_t)reg_struct.range.start +
+                       (uint64_t)reg_struct.range.len - 1);
+
+       return 0;
+}
+#else
+static int
+vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
+               struct rte_vhost_mem_region *reg __rte_unused)
+{
+       return -1;
+}
+#endif
+
+static int
+vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
+               struct VhostUserMsg *msg)
+{
+       struct VhostUserMemory *memory;
+       struct rte_vhost_mem_region *reg;
+       VhostUserMsg ack_msg;
+       uint32_t i;
+
+       if (!dev->postcopy_listening)
+               return 0;
+
+       /*
+        * We haven't a better way right now than sharing
+        * DPDK's virtual address with Qemu, so that Qemu can
+        * retrieve the region offset when handling userfaults.
+        */
+       memory = &msg->payload.memory;
+       for (i = 0; i < memory->nregions; i++) {
+               reg = &dev->mem->regions[i];
+               memory->regions[i].userspace_addr = reg->host_user_addr;
+       }
+
+       /* Send the addresses back to qemu */
+       msg->fd_num = 0;
+       send_vhost_reply(main_fd, msg);
+
+       /* Wait for qemu to acknolwedge it's got the addresses
+        * we've got to wait before we're allowed to generate faults.
+        */
+       if (read_vhost_message(main_fd, &ack_msg) <= 0) {
+               VHOST_LOG_CONFIG(ERR,
+                               "Failed to read qemu ack on postcopy set-mem-table\n");
+               return -1;
+       }
+
+       if (validate_msg_fds(&ack_msg, 0) != 0)
+               return -1;
+
+       if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
+               VHOST_LOG_CONFIG(ERR,
+                               "Bad qemu ack on postcopy set-mem-table (%d)\n",
+                               ack_msg.request.master);
+               return -1;
+       }
+
+       /* Now userfault register and we can use the memory */
+       for (i = 0; i < memory->nregions; i++) {
+               reg = &dev->mem->regions[i];
+               if (vhost_user_postcopy_region_register(dev, reg) < 0)
+                       return -1;
+       }
+
+       return 0;
+}
+
+static int
+vhost_user_mmap_region(struct virtio_net *dev,
+               struct rte_vhost_mem_region *region,
+               uint64_t mmap_offset)
+{
+       void *mmap_addr;
+       uint64_t mmap_size;
+       uint64_t alignment;
+       int populate;
+
+       /* Check for memory_size + mmap_offset overflow */
+       if (mmap_offset >= -region->size) {
+               VHOST_LOG_CONFIG(ERR,
+                               "mmap_offset (%#"PRIx64") and memory_size "
+                               "(%#"PRIx64") overflow\n",
+                               mmap_offset, region->size);
+               return -1;
+       }
+
+       mmap_size = region->size + mmap_offset;
+
+       /* mmap() without flag of MAP_ANONYMOUS, should be called with length
+        * argument aligned with hugepagesz at older longterm version Linux,
+        * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
+        *
+        * To avoid failure, make sure in caller to keep length aligned.
+        */
+       alignment = get_blk_size(region->fd);
+       if (alignment == (uint64_t)-1) {
+               VHOST_LOG_CONFIG(ERR,
+                               "couldn't get hugepage size through fstat\n");
+               return -1;
+       }
+       mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
+       if (mmap_size == 0) {
+               /*
+                * It could happen if initial mmap_size + alignment overflows
+                * the sizeof uint64, which could happen if either mmap_size or
+                * alignment value is wrong.
+                *
+                * mmap() kernel implementation would return an error, but
+                * better catch it before and provide useful info in the logs.
+                */
+               VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
+                               "or alignment (0x%" PRIx64 ") is invalid\n",
+                               region->size + mmap_offset, alignment);
+               return -1;
+       }
+
+       populate = dev->async_copy ? MAP_POPULATE : 0;
+       mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
+                       MAP_SHARED | populate, region->fd, 0);
+
+       if (mmap_addr == MAP_FAILED) {
+               VHOST_LOG_CONFIG(ERR, "mmap failed (%s).\n", strerror(errno));
+               return -1;
+       }
+
+       region->mmap_addr = mmap_addr;
+       region->mmap_size = mmap_size;
+       region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
+
+       if (dev->async_copy)
+               if (add_guest_pages(dev, region, alignment) < 0) {
+                       VHOST_LOG_CONFIG(ERR,
+                                       "adding guest pages to region failed.\n");
+                       return -1;
+               }
+
+       VHOST_LOG_CONFIG(INFO,
+                       "guest memory region size: 0x%" PRIx64 "\n"
+                       "\t guest physical addr: 0x%" PRIx64 "\n"
+                       "\t guest virtual  addr: 0x%" PRIx64 "\n"
+                       "\t host  virtual  addr: 0x%" PRIx64 "\n"
+                       "\t mmap addr : 0x%" PRIx64 "\n"
+                       "\t mmap size : 0x%" PRIx64 "\n"
+                       "\t mmap align: 0x%" PRIx64 "\n"
+                       "\t mmap off  : 0x%" PRIx64 "\n",
+                       region->size,
+                       region->guest_phys_addr,
+                       region->guest_user_addr,
+                       region->host_user_addr,
+                       (uint64_t)(uintptr_t)mmap_addr,
+                       mmap_size,
+                       alignment,
+                       mmap_offset);
+
+       return 0;
+}
+
 static int
 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
                        int main_fd)
@@ -1047,13 +1194,9 @@ vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
        struct virtio_net *dev = *pdev;
        struct VhostUserMemory *memory = &msg->payload.memory;
        struct rte_vhost_mem_region *reg;
-       void *mmap_addr;
-       uint64_t mmap_size;
+
        uint64_t mmap_offset;
-       uint64_t alignment;
        uint32_t i;
-       int populate;
-       int fd;
 
        if (validate_msg_fds(msg, memory->nregions) != 0)
                return RTE_VHOST_MSG_RESULT_ERR;
@@ -1061,7 +1204,7 @@ vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
        if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
                VHOST_LOG_CONFIG(ERR,
                        "too many memory regions (%u)\n", memory->nregions);
-               return RTE_VHOST_MSG_RESULT_ERR;
+               goto close_msg_fds;
        }
 
        if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
@@ -1074,6 +1217,13 @@ 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;
@@ -1085,16 +1235,18 @@ vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
                        vhost_user_iotlb_flush_all(dev->virtqueue[i]);
 
        dev->nr_guest_pages = 0;
-       if (!dev->guest_pages) {
+       if (dev->guest_pages == NULL) {
                dev->max_guest_pages = 8;
-               dev->guest_pages = malloc(dev->max_guest_pages *
-                                               sizeof(struct guest_page));
+               dev->guest_pages = rte_zmalloc(NULL,
+                                       dev->max_guest_pages *
+                                       sizeof(struct guest_page),
+                                       RTE_CACHE_LINE_SIZE);
                if (dev->guest_pages == NULL) {
                        VHOST_LOG_CONFIG(ERR,
                                "(%d) failed to allocate memory "
                                "for dev->guest_pages\n",
                                dev->vid);
-                       return RTE_VHOST_MSG_RESULT_ERR;
+                       goto close_msg_fds;
                }
        }
 
@@ -1104,177 +1256,42 @@ vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
                VHOST_LOG_CONFIG(ERR,
                        "(%d) failed to allocate memory for dev->mem\n",
                        dev->vid);
-               return RTE_VHOST_MSG_RESULT_ERR;
+               goto free_guest_pages;
        }
-       dev->mem->nregions = memory->nregions;
 
        for (i = 0; i < memory->nregions; i++) {
-               fd  = msg->fds[i];
                reg = &dev->mem->regions[i];
 
                reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
                reg->guest_user_addr = memory->regions[i].userspace_addr;
                reg->size            = memory->regions[i].memory_size;
-               reg->fd              = fd;
+               reg->fd              = msg->fds[i];
 
-               mmap_offset = memory->regions[i].mmap_offset;
-
-               /* Check for memory_size + mmap_offset overflow */
-               if (mmap_offset >= -reg->size) {
-                       VHOST_LOG_CONFIG(ERR,
-                               "mmap_offset (%#"PRIx64") and memory_size "
-                               "(%#"PRIx64") overflow\n",
-                               mmap_offset, reg->size);
-                       goto err_mmap;
-               }
-
-               mmap_size = reg->size + mmap_offset;
-
-               /* mmap() without flag of MAP_ANONYMOUS, should be called
-                * with length argument aligned with hugepagesz at older
-                * longterm version Linux, like 2.6.32 and 3.2.72, or
-                * mmap() will fail with EINVAL.
-                *
-                * to avoid failure, make sure in caller to keep length
-                * aligned.
+               /*
+                * Assign invalid file descriptor value to avoid double
+                * closing on error path.
                 */
-               alignment = get_blk_size(fd);
-               if (alignment == (uint64_t)-1) {
-                       VHOST_LOG_CONFIG(ERR,
-                               "couldn't get hugepage size through fstat\n");
-                       goto err_mmap;
-               }
-               mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
-               if (mmap_size == 0) {
-                       /*
-                        * It could happen if initial mmap_size + alignment
-                        * overflows the sizeof uint64, which could happen if
-                        * either mmap_size or alignment value is wrong.
-                        *
-                        * mmap() kernel implementation would return an error,
-                        * but better catch it before and provide useful info
-                        * in the logs.
-                        */
-                       VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
-                                       "or alignment (0x%" PRIx64 ") is invalid\n",
-                                       reg->size + mmap_offset, alignment);
-                       goto err_mmap;
-               }
+               msg->fds[i] = -1;
 
-               populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
-               mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
-                                MAP_SHARED | populate, fd, 0);
+               mmap_offset = memory->regions[i].mmap_offset;
 
-               if (mmap_addr == MAP_FAILED) {
-                       VHOST_LOG_CONFIG(ERR,
-                               "mmap region %u failed.\n", i);
-                       goto err_mmap;
+               if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
+                       VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
+                       goto free_mem_table;
                }
 
-               reg->mmap_addr = mmap_addr;
-               reg->mmap_size = mmap_size;
-               reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
-                                     mmap_offset;
-
-               if (dev->dequeue_zero_copy)
-                       if (add_guest_pages(dev, reg, alignment) < 0) {
-                               VHOST_LOG_CONFIG(ERR,
-                                       "adding guest pages to region %u failed.\n",
-                                       i);
-                               goto err_mmap;
-                       }
-
-               VHOST_LOG_CONFIG(INFO,
-                       "guest memory region %u, size: 0x%" PRIx64 "\n"
-                       "\t guest physical addr: 0x%" PRIx64 "\n"
-                       "\t guest virtual  addr: 0x%" PRIx64 "\n"
-                       "\t host  virtual  addr: 0x%" PRIx64 "\n"
-                       "\t mmap addr : 0x%" PRIx64 "\n"
-                       "\t mmap size : 0x%" PRIx64 "\n"
-                       "\t mmap align: 0x%" PRIx64 "\n"
-                       "\t mmap off  : 0x%" PRIx64 "\n",
-                       i, reg->size,
-                       reg->guest_phys_addr,
-                       reg->guest_user_addr,
-                       reg->host_user_addr,
-                       (uint64_t)(uintptr_t)mmap_addr,
-                       mmap_size,
-                       alignment,
-                       mmap_offset);
-
-               if (dev->postcopy_listening) {
-                       /*
-                        * We haven't a better way right now than sharing
-                        * DPDK's virtual address with Qemu, so that Qemu can
-                        * retrieve the region offset when handling userfaults.
-                        */
-                       memory->regions[i].userspace_addr =
-                               reg->host_user_addr;
-               }
+               dev->mem->nregions++;
        }
-       if (dev->postcopy_listening) {
-               /* Send the addresses back to qemu */
-               msg->fd_num = 0;
-               send_vhost_reply(main_fd, msg);
-
-               /* Wait for qemu to acknolwedge it's got the addresses
-                * we've got to wait before we're allowed to generate faults.
-                */
-               VhostUserMsg ack_msg;
-               if (read_vhost_message(main_fd, &ack_msg) <= 0) {
-                       VHOST_LOG_CONFIG(ERR,
-                               "Failed to read qemu ack on postcopy set-mem-table\n");
-                       goto err_mmap;
-               }
 
-               if (validate_msg_fds(&ack_msg, 0) != 0)
-                       goto err_mmap;
-
-               if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
-                       VHOST_LOG_CONFIG(ERR,
-                               "Bad qemu ack on postcopy set-mem-table (%d)\n",
-                               ack_msg.request.master);
-                       goto err_mmap;
-               }
-
-               /* Now userfault register and we can use the memory */
-               for (i = 0; i < memory->nregions; i++) {
-#ifdef RTE_LIBRTE_VHOST_POSTCOPY
-                       reg = &dev->mem->regions[i];
-                       struct uffdio_register reg_struct;
-
-                       /*
-                        * Let's register all the mmap'ed area to ensure
-                        * alignment on page boundary.
-                        */
-                       reg_struct.range.start =
-                               (uint64_t)(uintptr_t)reg->mmap_addr;
-                       reg_struct.range.len = reg->mmap_size;
-                       reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
-
-                       if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
-                                               &reg_struct)) {
-                               VHOST_LOG_CONFIG(ERR,
-                                       "Failed to register ufd for region %d: (ufd = %d) %s\n",
-                                       i, dev->postcopy_ufd,
-                                       strerror(errno));
-                               goto err_mmap;
-                       }
-                       VHOST_LOG_CONFIG(INFO,
-                               "\t userfaultfd registered for range : "
-                               "%" PRIx64 " - %" PRIx64 "\n",
-                               (uint64_t)reg_struct.range.start,
-                               (uint64_t)reg_struct.range.start +
-                               (uint64_t)reg_struct.range.len - 1);
-#else
-                       goto err_mmap;
-#endif
-               }
-       }
+       if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
+               goto free_mem_table;
 
        for (i = 0; i < dev->nr_vring; i++) {
                struct vhost_virtqueue *vq = dev->virtqueue[i];
 
+               if (!vq)
+                       continue;
+
                if (vq->desc || vq->avail || vq->used) {
                        /*
                         * If the memory table got updated, the ring addresses
@@ -1286,7 +1303,7 @@ vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
                        dev = translate_ring_addresses(dev, i);
                        if (!dev) {
                                dev = *pdev;
-                               goto err_mmap;
+                               goto free_mem_table;
                        }
 
                        *pdev = dev;
@@ -1297,10 +1314,15 @@ vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
 
        return RTE_VHOST_MSG_RESULT_OK;
 
-err_mmap:
+free_mem_table:
        free_mem_region(dev);
        rte_free(dev->mem);
        dev->mem = NULL;
+free_guest_pages:
+       rte_free(dev->guest_pages);
+       dev->guest_pages = NULL;
+close_msg_fds:
+       close_msg_fds(msg);
        return RTE_VHOST_MSG_RESULT_ERR;
 }
 
@@ -1320,27 +1342,48 @@ 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_BUILTIN_NUM_VQS_TO_BE_READY 2u
+
 static int
 virtio_is_ready(struct virtio_net *dev)
 {
        struct vhost_virtqueue *vq;
-       uint32_t i;
+       uint32_t i, nr_vring = dev->nr_vring;
+
+       if (dev->flags & VIRTIO_DEV_READY)
+               return 1;
 
-       if (dev->nr_vring == 0)
+       if (!dev->nr_vring)
                return 0;
 
-       for (i = 0; i < dev->nr_vring; i++) {
+       if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
+               nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
+
+               if (dev->nr_vring < nr_vring)
+                       return 0;
+       }
+
+       for (i = 0; i < nr_vring; 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;
 }
 
@@ -1431,6 +1474,7 @@ vhost_user_get_inflight_fd(struct virtio_net **pdev,
                                "failed to alloc dev inflight area\n");
                        return RTE_VHOST_MSG_RESULT_ERR;
                }
+               dev->inflight_info->fd = -1;
        }
 
        num_queues = msg->payload.inflight.num_queues;
@@ -1456,6 +1500,16 @@ vhost_user_get_inflight_fd(struct virtio_net **pdev,
        }
        memset(addr, 0, mmap_size);
 
+       if (dev->inflight_info->addr) {
+               munmap(dev->inflight_info->addr, dev->inflight_info->size);
+               dev->inflight_info->addr = NULL;
+       }
+
+       if (dev->inflight_info->fd >= 0) {
+               close(dev->inflight_info->fd);
+               dev->inflight_info->fd = -1;
+       }
+
        dev->inflight_info->addr = addr;
        dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
        dev->inflight_info->fd = msg->fds[0] = fd;
@@ -1538,10 +1592,13 @@ vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
                                "failed to alloc dev inflight area\n");
                        return RTE_VHOST_MSG_RESULT_ERR;
                }
+               dev->inflight_info->fd = -1;
        }
 
-       if (dev->inflight_info->addr)
+       if (dev->inflight_info->addr) {
                munmap(dev->inflight_info->addr, dev->inflight_info->size);
+               dev->inflight_info->addr = NULL;
+       }
 
        addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
                    fd, mmap_offset);
@@ -1550,8 +1607,10 @@ vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
                return RTE_VHOST_MSG_RESULT_ERR;
        }
 
-       if (dev->inflight_info->fd)
+       if (dev->inflight_info->fd >= 0) {
                close(dev->inflight_info->fd);
+               dev->inflight_info->fd = -1;
+       }
 
        dev->inflight_info->fd = fd;
        dev->inflight_info->addr = addr;
@@ -1559,6 +1618,9 @@ vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
 
        for (i = 0; i < num_queues; i++) {
                vq = dev->virtqueue[i];
+               if (!vq)
+                       continue;
+
                if (vq_is_packed(dev)) {
                        vq->inflight_packed = addr;
                        vq->inflight_packed->desc_num = queue_size;
@@ -1594,6 +1656,12 @@ 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);
 
@@ -1645,8 +1713,11 @@ vhost_check_queue_inflights_split(struct virtio_net *dev,
            (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
                return RTE_VHOST_MSG_RESULT_OK;
 
+       /* The frontend may still not support the inflight feature
+        * although we negotiate the protocol feature.
+        */
        if ((!vq->inflight_split))
-               return RTE_VHOST_MSG_RESULT_ERR;
+               return RTE_VHOST_MSG_RESULT_OK;
 
        if (!vq->inflight_split->version) {
                vq->inflight_split->version = INFLIGHT_VERSION;
@@ -1662,7 +1733,7 @@ vhost_check_queue_inflights_split(struct virtio_net *dev,
 
        if (inflight_split->used_idx != used->idx) {
                inflight_split->desc[last_io].inflight = 0;
-               rte_smp_mb();
+               rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
                inflight_split->used_idx = used->idx;
        }
 
@@ -1726,8 +1797,11 @@ vhost_check_queue_inflights_packed(struct virtio_net *dev,
            (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
                return RTE_VHOST_MSG_RESULT_OK;
 
+       /* The frontend may still not support the inflight feature
+        * although we negotiate the protocol feature.
+        */
        if ((!vq->inflight_packed))
-               return RTE_VHOST_MSG_RESULT_ERR;
+               return RTE_VHOST_MSG_RESULT_OK;
 
        if (!vq->inflight_packed->version) {
                vq->inflight_packed->version = INFLIGHT_VERSION;
@@ -1827,8 +1901,12 @@ vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
 
        /* Interpret ring addresses only when ring is started. */
        dev = translate_ring_addresses(dev, file.index);
-       if (!dev)
+       if (!dev) {
+               if (file.fd != VIRTIO_INVALID_EVENTFD)
+                       close(file.fd);
+
                return RTE_VHOST_MSG_RESULT_ERR;
+       }
 
        *pdev = dev;
 
@@ -1846,6 +1924,11 @@ 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;
@@ -1867,14 +1950,6 @@ vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
        return RTE_VHOST_MSG_RESULT_OK;
 }
 
-static void
-free_zmbufs(struct vhost_virtqueue *vq)
-{
-       drain_zmbuf_list(vq);
-
-       rte_free(vq->zmbufs);
-}
-
 /*
  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
  */
@@ -1929,14 +2004,19 @@ vhost_user_get_vring_base(struct virtio_net **pdev,
 
        vq->signalled_used_valid = false;
 
-       if (dev->dequeue_zero_copy)
-               free_zmbufs(vq);
        if (vq_is_packed(dev)) {
                rte_free(vq->shadow_used_packed);
                vq->shadow_used_packed = NULL;
        } else {
                rte_free(vq->shadow_used_split);
                vq->shadow_used_split = NULL;
+
+               if (vq->async_pkts_info)
+                       rte_free(vq->async_pkts_info);
+               if (vq->async_descs_split)
+                       rte_free(vq->async_descs_split);
+               vq->async_pkts_info = NULL;
+               vq->async_descs_split = NULL;
        }
 
        rte_free(vq->batch_copy_elems);
@@ -1962,8 +2042,6 @@ 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;
-       int did = -1;
 
        if (validate_msg_fds(msg, 0) != 0)
                return RTE_VHOST_MSG_RESULT_ERR;
@@ -1972,18 +2050,13 @@ vhost_user_set_vring_enable(struct virtio_net **pdev,
                "set queue enable: %d to qp idx: %d\n",
                enable, index);
 
-       did = dev->vdpa_dev_id;
-       vdpa_dev = rte_vdpa_get_device(did);
-       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);
-
-       /* On disable, rings have to be stopped being processed. */
-       if (!enable && dev->dequeue_zero_copy)
-               drain_zmbuf_list(dev->virtqueue[index]);
+       if (enable && dev->virtqueue[index]->async_registered) {
+               if (dev->virtqueue[index]->async_pkts_inflight_n) {
+                       VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
+                       "async inflight packets must be completed first\n");
+                       return RTE_VHOST_MSG_RESULT_ERR;
+               }
+       }
 
        dev->virtqueue[index]->enabled = enable;
 
@@ -2004,15 +2077,6 @@ vhost_user_get_protocol_features(struct virtio_net **pdev,
        rte_vhost_driver_get_features(dev->ifname, &features);
        rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
 
-       /*
-        * REPLY_ACK protocol feature is only mandatory for now
-        * for IOMMU feature. If IOMMU is explicitly disabled by the
-        * application, disable also REPLY_ACK feature for older buggy
-        * Qemu versions (from v2.7.0 to v2.9.0).
-        */
-       if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
-               protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
-
        msg->payload.u64 = protocol_features;
        msg->size = sizeof(msg->payload.u64);
        msg->fd_num = 0;
@@ -2070,18 +2134,18 @@ vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
                VHOST_LOG_CONFIG(ERR,
                        "invalid log base msg size: %"PRId32" != %d\n",
                        msg->size, (int)sizeof(VhostUserLog));
-               return RTE_VHOST_MSG_RESULT_ERR;
+               goto close_msg_fds;
        }
 
        size = msg->payload.log.mmap_size;
        off  = msg->payload.log.mmap_offset;
 
-       /* Don't allow mmap_offset to point outside the mmap region */
-       if (off > size) {
+       /* Check for mmap size and offset overflow. */
+       if (off >= -size) {
                VHOST_LOG_CONFIG(ERR,
-                       "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
+                       "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
                        off, size);
-               return RTE_VHOST_MSG_RESULT_ERR;
+               goto close_msg_fds;
        }
 
        VHOST_LOG_CONFIG(INFO,
@@ -2118,6 +2182,10 @@ vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
        msg->fd_num = 0;
 
        return RTE_VHOST_MSG_RESULT_REPLY;
+
+close_msg_fds:
+       close_msg_fds(msg);
+       return RTE_VHOST_MSG_RESULT_ERR;
 }
 
 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
@@ -2148,7 +2216,6 @@ vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
        struct virtio_net *dev = *pdev;
        uint8_t *mac = (uint8_t *)&msg->payload.u64;
        struct rte_vdpa_device *vdpa_dev;
-       int did = -1;
 
        if (validate_msg_fds(msg, 0) != 0)
                return RTE_VHOST_MSG_RESULT_ERR;
@@ -2162,13 +2229,11 @@ vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
         * Set the flag to inject a RARP broadcast packet at
         * rte_vhost_dequeue_burst().
         *
-        * rte_smp_wmb() is for making sure the mac is copied
-        * before the flag is set.
+        * __ATOMIC_RELEASE ordering is for making sure the mac is
+        * copied before the flag is set.
         */
-       rte_smp_wmb();
-       rte_atomic16_set(&dev->broadcast_rarp, 1);
-       did = dev->vdpa_dev_id;
-       vdpa_dev = rte_vdpa_get_device(did);
+       __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
+       vdpa_dev = dev->vdpa_dev;
        if (vdpa_dev && vdpa_dev->ops->migration_done)
                vdpa_dev->ops->migration_done(dev->vid);
 
@@ -2245,6 +2310,13 @@ is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
        if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
                return 1;
 
+       if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
+               len = sizeof(uint64_t);
+               if (ra->log_guest_addr < end &&
+                   (ra->log_guest_addr + len) > start)
+                       return 1;
+       }
+
        return 0;
 }
 
@@ -2270,6 +2342,13 @@ is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
        if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
                return 1;
 
+       if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
+               len = sizeof(uint64_t);
+               if (ra->log_guest_addr < end &&
+                   (ra->log_guest_addr + len) > start)
+                       return 1;
+       }
+
        return 0;
 }
 
@@ -2305,6 +2384,9 @@ vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
                for (i = 0; i < dev->nr_vring; i++) {
                        struct vhost_virtqueue *vq = dev->virtqueue[i];
 
+                       if (!vq)
+                               continue;
+
                        vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
                                        len, imsg->perm);
 
@@ -2316,6 +2398,9 @@ vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
                for (i = 0; i < dev->nr_vring; i++) {
                        struct vhost_virtqueue *vq = dev->virtqueue[i];
 
+                       if (!vq)
+                               continue;
+
                        vhost_user_iotlb_cache_remove(vq, imsg->iova,
                                        imsg->size);
 
@@ -2414,6 +2499,70 @@ 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-RESET: %u\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_RESET),
+                       !!(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);
@@ -2446,6 +2595,8 @@ 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. */
@@ -2529,7 +2680,7 @@ static int
 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
                        struct VhostUserMsg *msg)
 {
-       uint16_t vring_idx;
+       uint32_t vring_idx;
 
        switch (msg->request.master) {
        case VHOST_USER_SET_VRING_KICK:
@@ -2601,11 +2752,11 @@ vhost_user_msg_handler(int vid, int fd)
        struct virtio_net *dev;
        struct VhostUserMsg msg;
        struct rte_vdpa_device *vdpa_dev;
-       int did = -1;
        int ret;
        int unlock_required = 0;
        bool handled;
        int request;
+       uint32_t i;
 
        dev = get_device(vid);
        if (dev == NULL)
@@ -2678,8 +2829,10 @@ 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;
@@ -2779,30 +2932,44 @@ 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);
                }
        }
 
-       did = dev->vdpa_dev_id;
-       vdpa_dev = rte_vdpa_get_device(did);
-       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 (!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)
+               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;
 }
 
@@ -2815,11 +2982,19 @@ static int process_slave_message_reply(struct virtio_net *dev,
        if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
                return 0;
 
-       if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
+       ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
+       if (ret <= 0) {
+               if (ret < 0)
+                       VHOST_LOG_CONFIG(ERR,
+                               "vhost read slave message reply failed\n");
+               else
+                       VHOST_LOG_CONFIG(INFO,
+                               "vhost peer closed\n");
                ret = -1;
                goto out;
        }
 
+       ret = 0;
        if (msg_reply.request.slave != msg->request.slave) {
                VHOST_LOG_CONFIG(ERR,
                        "Received unexpected msg type (%u), expected %u\n",
@@ -2931,21 +3106,21 @@ 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, did, ret = 0;
+       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)
                return -ENODEV;
 
-       did = dev->vdpa_dev_id;
-       if (did < 0)
-               return -EINVAL;
+       vdpa_dev = dev->vdpa_dev;
+       if (vdpa_dev == NULL)
+               return -ENODEV;
 
        if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
            !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
@@ -2957,9 +3132,15 @@ int rte_vhost_host_notifier_ctrl(int vid, bool enable)
                        (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
                return -ENOTSUP;
 
-       vdpa_dev = rte_vdpa_get_device(did);
-       if (!vdpa_dev)
-               return -ENODEV;
+       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);
@@ -2969,7 +3150,7 @@ 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;
@@ -2984,7 +3165,7 @@ 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);
                }