vhost: do not inline unlikely fragmented buffers code
[dpdk.git] / lib / librte_vhost / vhost.c
index 4b9d781..981837b 100644 (file)
@@ -1,34 +1,5 @@
-/*-
- *   BSD LICENSE
- *
- *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
- *   All rights reserved.
- *
- *   Redistribution and use in source and binary forms, with or without
- *   modification, are permitted provided that the following conditions
- *   are met:
- *
- *     * Redistributions of source code must retain the above copyright
- *       notice, this list of conditions and the following disclaimer.
- *     * Redistributions in binary form must reproduce the above copyright
- *       notice, this list of conditions and the following disclaimer in
- *       the documentation and/or other materials provided with the
- *       distribution.
- *     * Neither the name of Intel Corporation nor the names of its
- *       contributors may be used to endorse or promote products derived
- *       from this software without specific prior written permission.
- *
- *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2010-2017 Intel Corporation
  */
 
 #include <linux/vhost.h>
 #include <stdint.h>
 #include <stdlib.h>
 #ifdef RTE_LIBRTE_VHOST_NUMA
+#include <numa.h>
 #include <numaif.h>
 #endif
 
+#include <rte_errno.h>
 #include <rte_ethdev.h>
 #include <rte_log.h>
 #include <rte_string_fns.h>
 #include <rte_memory.h>
 #include <rte_malloc.h>
-#include <rte_virtio_net.h>
+#include <rte_vhost.h>
+#include <rte_rwlock.h>
 
+#include "iotlb.h"
 #include "vhost.h"
+#include "vhost_user.h"
 
 struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
 
-struct virtio_net *
-get_device(int vid)
+/* Called with iotlb_lock read-locked */
+uint64_t
+__vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
+                   uint64_t iova, uint64_t *size, uint8_t perm)
 {
-       struct virtio_net *dev = vhost_devices[vid];
+       uint64_t vva, tmp_size;
 
-       if (unlikely(!dev)) {
-               RTE_LOG(ERR, VHOST_CONFIG,
-                       "(%d) device not found.\n", vid);
+       if (unlikely(!*size))
+               return 0;
+
+       tmp_size = *size;
+
+       vva = vhost_user_iotlb_cache_find(vq, iova, &tmp_size, perm);
+       if (tmp_size == *size)
+               return vva;
+
+       iova += tmp_size;
+
+       if (!vhost_user_iotlb_pending_miss(vq, iova, perm)) {
+               /*
+                * iotlb_lock is read-locked for a full burst,
+                * but it only protects the iotlb cache.
+                * In case of IOTLB miss, we might block on the socket,
+                * which could cause a deadlock with QEMU if an IOTLB update
+                * is being handled. We can safely unlock here to avoid it.
+                */
+               vhost_user_iotlb_rd_unlock(vq);
+
+               vhost_user_iotlb_pending_insert(vq, iova, perm);
+               if (vhost_user_iotlb_miss(dev, iova, perm)) {
+                       RTE_LOG(ERR, VHOST_CONFIG,
+                               "IOTLB miss req failed for IOVA 0x%" PRIx64 "\n",
+                               iova);
+                       vhost_user_iotlb_pending_remove(vq, iova, 1, perm);
+               }
+
+               vhost_user_iotlb_rd_lock(vq);
        }
 
-       return dev;
+       return 0;
 }
 
-static void
+#define VHOST_LOG_PAGE 4096
+
+/*
+ * Atomically set a bit in memory.
+ */
+static __rte_always_inline void
+vhost_set_bit(unsigned int nr, volatile uint8_t *addr)
+{
+#if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
+       /*
+        * __sync_ built-ins are deprecated, but __atomic_ ones
+        * are sub-optimized in older GCC versions.
+        */
+       __sync_fetch_and_or_1(addr, (1U << nr));
+#else
+       __atomic_fetch_or(addr, (1U << nr), __ATOMIC_RELAXED);
+#endif
+}
+
+static __rte_always_inline void
+vhost_log_page(uint8_t *log_base, uint64_t page)
+{
+       vhost_set_bit(page % 8, &log_base[page / 8]);
+}
+
+void
+__vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
+{
+       uint64_t page;
+
+       if (unlikely(!dev->log_base || !len))
+               return;
+
+       if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
+               return;
+
+       /* To make sure guest memory updates are committed before logging */
+       rte_smp_wmb();
+
+       page = addr / VHOST_LOG_PAGE;
+       while (page * VHOST_LOG_PAGE < addr + len) {
+               vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
+               page += 1;
+       }
+}
+
+void
+__vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
+       unsigned long *log_base;
+       int i;
+
+       if (unlikely(!dev->log_base))
+               return;
+
+       rte_smp_wmb();
+
+       log_base = (unsigned long *)(uintptr_t)dev->log_base;
+
+       for (i = 0; i < vq->log_cache_nb_elem; i++) {
+               struct log_cache_entry *elem = vq->log_cache + i;
+
+#if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
+               /*
+                * '__sync' builtins are deprecated, but '__atomic' ones
+                * are sub-optimized in older GCC versions.
+                */
+               __sync_fetch_and_or(log_base + elem->offset, elem->val);
+#else
+               __atomic_fetch_or(log_base + elem->offset, elem->val,
+                               __ATOMIC_RELAXED);
+#endif
+       }
+
+       rte_smp_wmb();
+
+       vq->log_cache_nb_elem = 0;
+}
+
+static __rte_always_inline void
+vhost_log_cache_page(struct virtio_net *dev, struct vhost_virtqueue *vq,
+                       uint64_t page)
+{
+       uint32_t bit_nr = page % (sizeof(unsigned long) << 3);
+       uint32_t offset = page / (sizeof(unsigned long) << 3);
+       int i;
+
+       for (i = 0; i < vq->log_cache_nb_elem; i++) {
+               struct log_cache_entry *elem = vq->log_cache + i;
+
+               if (elem->offset == offset) {
+                       elem->val |= (1UL << bit_nr);
+                       return;
+               }
+       }
+
+       if (unlikely(i >= VHOST_LOG_CACHE_NR)) {
+               /*
+                * No more room for a new log cache entry,
+                * so write the dirty log map directly.
+                */
+               rte_smp_wmb();
+               vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
+
+               return;
+       }
+
+       vq->log_cache[i].offset = offset;
+       vq->log_cache[i].val = (1UL << bit_nr);
+       vq->log_cache_nb_elem++;
+}
+
+void
+__vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
+                       uint64_t addr, uint64_t len)
+{
+       uint64_t page;
+
+       if (unlikely(!dev->log_base || !len))
+               return;
+
+       if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
+               return;
+
+       page = addr / VHOST_LOG_PAGE;
+       while (page * VHOST_LOG_PAGE < addr + len) {
+               vhost_log_cache_page(dev, vq, page);
+               page += 1;
+       }
+}
+
+void *
+vhost_alloc_copy_ind_table(struct virtio_net *dev, struct vhost_virtqueue *vq,
+               uint64_t desc_addr, uint64_t desc_len)
+{
+       void *idesc;
+       uint64_t src, dst;
+       uint64_t len, remain = desc_len;
+
+       idesc = rte_malloc(__func__, desc_len, 0);
+       if (unlikely(!idesc))
+               return NULL;
+
+       dst = (uint64_t)(uintptr_t)idesc;
+
+       while (remain) {
+               len = remain;
+               src = vhost_iova_to_vva(dev, vq, desc_addr, &len,
+                               VHOST_ACCESS_RO);
+               if (unlikely(!src || !len)) {
+                       rte_free(idesc);
+                       return NULL;
+               }
+
+               rte_memcpy((void *)(uintptr_t)dst, (void *)(uintptr_t)src, len);
+
+               remain -= len;
+               dst += len;
+               desc_addr += len;
+       }
+
+       return idesc;
+}
+
+void
 cleanup_vq(struct vhost_virtqueue *vq, int destroy)
 {
        if ((vq->callfd >= 0) && (destroy != 0))
@@ -84,10 +253,20 @@ cleanup_device(struct virtio_net *dev, int destroy)
 
        vhost_backend_cleanup(dev);
 
-       for (i = 0; i < dev->virt_qp_nb; i++) {
-               cleanup_vq(dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_RXQ], destroy);
-               cleanup_vq(dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_TXQ], destroy);
-       }
+       for (i = 0; i < dev->nr_vring; i++)
+               cleanup_vq(dev->virtqueue[i], destroy);
+}
+
+void
+free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
+       if (vq_is_packed(dev))
+               rte_free(vq->shadow_used_packed);
+       else
+               rte_free(vq->shadow_used_split);
+       rte_free(vq->batch_copy_elems);
+       rte_mempool_free(vq->iotlb_pool);
+       rte_free(vq);
 }
 
 /*
@@ -97,96 +276,190 @@ static void
 free_device(struct virtio_net *dev)
 {
        uint32_t i;
-       struct vhost_virtqueue *rxq, *txq;
 
-       for (i = 0; i < dev->virt_qp_nb; i++) {
-               rxq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_RXQ];
-               txq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_TXQ];
+       for (i = 0; i < dev->nr_vring; i++)
+               free_vq(dev, dev->virtqueue[i]);
+
+       rte_free(dev);
+}
+
+static int
+vring_translate_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
+       uint64_t req_size, size;
+
+       req_size = sizeof(struct vring_desc) * vq->size;
+       size = req_size;
+       vq->desc = (struct vring_desc *)(uintptr_t)vhost_iova_to_vva(dev, vq,
+                                               vq->ring_addrs.desc_user_addr,
+                                               &size, VHOST_ACCESS_RW);
+       if (!vq->desc || size != req_size)
+               return -1;
+
+       req_size = sizeof(struct vring_avail);
+       req_size += sizeof(uint16_t) * vq->size;
+       if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
+               req_size += sizeof(uint16_t);
+       size = req_size;
+       vq->avail = (struct vring_avail *)(uintptr_t)vhost_iova_to_vva(dev, vq,
+                                               vq->ring_addrs.avail_user_addr,
+                                               &size, VHOST_ACCESS_RW);
+       if (!vq->avail || size != req_size)
+               return -1;
+
+       req_size = sizeof(struct vring_used);
+       req_size += sizeof(struct vring_used_elem) * vq->size;
+       if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
+               req_size += sizeof(uint16_t);
+       size = req_size;
+       vq->used = (struct vring_used *)(uintptr_t)vhost_iova_to_vva(dev, vq,
+                                               vq->ring_addrs.used_user_addr,
+                                               &size, VHOST_ACCESS_RW);
+       if (!vq->used || size != req_size)
+               return -1;
+
+       return 0;
+}
+
+static int
+vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
+       uint64_t req_size, size;
+
+       req_size = sizeof(struct vring_packed_desc) * vq->size;
+       size = req_size;
+       vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
+               vhost_iova_to_vva(dev, vq, vq->ring_addrs.desc_user_addr,
+                               &size, VHOST_ACCESS_RW);
+       if (!vq->desc_packed || size != req_size)
+               return -1;
+
+       req_size = sizeof(struct vring_packed_desc_event);
+       size = req_size;
+       vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
+               vhost_iova_to_vva(dev, vq, vq->ring_addrs.avail_user_addr,
+                               &size, VHOST_ACCESS_RW);
+       if (!vq->driver_event || size != req_size)
+               return -1;
+
+       req_size = sizeof(struct vring_packed_desc_event);
+       size = req_size;
+       vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
+               vhost_iova_to_vva(dev, vq, vq->ring_addrs.used_user_addr,
+                               &size, VHOST_ACCESS_RW);
+       if (!vq->device_event || size != req_size)
+               return -1;
+
+       return 0;
+}
+
+int
+vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
 
-               rte_free(rxq->shadow_used_ring);
-               rte_free(txq->shadow_used_ring);
+       if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
+               goto out;
 
-               /* rxq and txq are allocated together as queue-pair */
-               rte_free(rxq);
+       if (vq_is_packed(dev)) {
+               if (vring_translate_packed(dev, vq) < 0)
+                       return -1;
+       } else {
+               if (vring_translate_split(dev, vq) < 0)
+                       return -1;
        }
+out:
+       vq->access_ok = 1;
 
-       rte_free(dev);
+       return 0;
+}
+
+void
+vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq)
+{
+       if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
+               vhost_user_iotlb_wr_lock(vq);
+
+       vq->access_ok = 0;
+       vq->desc = NULL;
+       vq->avail = NULL;
+       vq->used = NULL;
+
+       if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
+               vhost_user_iotlb_wr_unlock(vq);
 }
 
 static void
-init_vring_queue(struct vhost_virtqueue *vq, int qp_idx)
+init_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 {
+       struct vhost_virtqueue *vq;
+
+       if (vring_idx >= VHOST_MAX_VRING) {
+               RTE_LOG(ERR, VHOST_CONFIG,
+                               "Failed not init vring, out of bound (%d)\n",
+                               vring_idx);
+               return;
+       }
+
+       vq = dev->virtqueue[vring_idx];
+
        memset(vq, 0, sizeof(struct vhost_virtqueue));
 
        vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
        vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
 
+       vhost_user_iotlb_init(dev, vring_idx);
        /* Backends are set to -1 indicating an inactive device. */
        vq->backend = -1;
 
-       /* always set the default vq pair to enabled */
-       if (qp_idx == 0)
-               vq->enabled = 1;
-
        TAILQ_INIT(&vq->zmbuf_list);
 }
 
 static void
-init_vring_queue_pair(struct virtio_net *dev, uint32_t qp_idx)
-{
-       uint32_t base_idx = qp_idx * VIRTIO_QNUM;
-
-       init_vring_queue(dev->virtqueue[base_idx + VIRTIO_RXQ], qp_idx);
-       init_vring_queue(dev->virtqueue[base_idx + VIRTIO_TXQ], qp_idx);
-}
-
-static void
-reset_vring_queue(struct vhost_virtqueue *vq, int qp_idx)
+reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 {
+       struct vhost_virtqueue *vq;
        int callfd;
 
+       if (vring_idx >= VHOST_MAX_VRING) {
+               RTE_LOG(ERR, VHOST_CONFIG,
+                               "Failed not init vring, out of bound (%d)\n",
+                               vring_idx);
+               return;
+       }
+
+       vq = dev->virtqueue[vring_idx];
        callfd = vq->callfd;
-       init_vring_queue(vq, qp_idx);
+       init_vring_queue(dev, vring_idx);
        vq->callfd = callfd;
 }
 
-static void
-reset_vring_queue_pair(struct virtio_net *dev, uint32_t qp_idx)
-{
-       uint32_t base_idx = qp_idx * VIRTIO_QNUM;
-
-       reset_vring_queue(dev->virtqueue[base_idx + VIRTIO_RXQ], qp_idx);
-       reset_vring_queue(dev->virtqueue[base_idx + VIRTIO_TXQ], qp_idx);
-}
-
 int
-alloc_vring_queue_pair(struct virtio_net *dev, uint32_t qp_idx)
+alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
 {
-       struct vhost_virtqueue *virtqueue = NULL;
-       uint32_t virt_rx_q_idx = qp_idx * VIRTIO_QNUM + VIRTIO_RXQ;
-       uint32_t virt_tx_q_idx = qp_idx * VIRTIO_QNUM + VIRTIO_TXQ;
+       struct vhost_virtqueue *vq;
 
-       virtqueue = rte_malloc(NULL,
-                              sizeof(struct vhost_virtqueue) * VIRTIO_QNUM, 0);
-       if (virtqueue == NULL) {
+       vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
+       if (vq == NULL) {
                RTE_LOG(ERR, VHOST_CONFIG,
-                       "Failed to allocate memory for virt qp:%d.\n", qp_idx);
+                       "Failed to allocate memory for vring:%u.\n", vring_idx);
                return -1;
        }
 
-       dev->virtqueue[virt_rx_q_idx] = virtqueue;
-       dev->virtqueue[virt_tx_q_idx] = virtqueue + VIRTIO_TXQ;
-
-       init_vring_queue_pair(dev, qp_idx);
+       dev->virtqueue[vring_idx] = vq;
+       init_vring_queue(dev, vring_idx);
+       rte_spinlock_init(&vq->access_lock);
+       vq->avail_wrap_counter = 1;
+       vq->used_wrap_counter = 1;
+       vq->signalled_used_valid = false;
 
-       dev->virt_qp_nb += 1;
+       dev->nr_vring += 1;
 
        return 0;
 }
 
 /*
  * Reset some variables in device structure, while keeping few
- * others untouched, such as vid, ifname, virt_qp_nb: they
+ * others untouched, such as vid, ifname, nr_vring: they
  * should be same unless the device is removed.
  */
 void
@@ -196,10 +469,10 @@ reset_device(struct virtio_net *dev)
 
        dev->features = 0;
        dev->protocol_features = 0;
-       dev->flags = 0;
+       dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
 
-       for (i = 0; i < dev->virt_qp_nb; i++)
-               reset_vring_queue_pair(dev, i);
+       for (i = 0; i < dev->nr_vring; i++)
+               reset_vring_queue(dev, i);
 }
 
 /*
@@ -212,30 +485,51 @@ vhost_new_device(void)
        struct virtio_net *dev;
        int i;
 
-       dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
-       if (dev == NULL) {
-               RTE_LOG(ERR, VHOST_CONFIG,
-                       "Failed to allocate memory for new dev.\n");
-               return -1;
-       }
-
        for (i = 0; i < MAX_VHOST_DEVICE; i++) {
                if (vhost_devices[i] == NULL)
                        break;
        }
+
        if (i == MAX_VHOST_DEVICE) {
                RTE_LOG(ERR, VHOST_CONFIG,
                        "Failed to find a free slot for new device.\n");
-               rte_free(dev);
+               return -1;
+       }
+
+       dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
+       if (dev == NULL) {
+               RTE_LOG(ERR, VHOST_CONFIG,
+                       "Failed to allocate memory for new dev.\n");
                return -1;
        }
 
        vhost_devices[i] = dev;
        dev->vid = i;
+       dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
+       dev->slave_req_fd = -1;
+       dev->vdpa_dev_id = -1;
+       dev->postcopy_ufd = -1;
+       rte_spinlock_init(&dev->slave_req_lock);
 
        return i;
 }
 
+void
+vhost_destroy_device_notify(struct virtio_net *dev)
+{
+       struct rte_vdpa_device *vdpa_dev;
+       int did;
+
+       if (dev->flags & VIRTIO_DEV_RUNNING) {
+               did = dev->vdpa_dev_id;
+               vdpa_dev = rte_vdpa_get_device(did);
+               if (vdpa_dev && vdpa_dev->ops->dev_close)
+                       vdpa_dev->ops->dev_close(dev->vid);
+               dev->flags &= ~VIRTIO_DEV_RUNNING;
+               dev->notify_ops->destroy_device(dev->vid);
+       }
+}
+
 /*
  * Invoked when there is the vhost-user connection is broken (when
  * the virtio device is being detached).
@@ -248,10 +542,7 @@ vhost_destroy_device(int vid)
        if (dev == NULL)
                return;
 
-       if (dev->flags & VIRTIO_DEV_RUNNING) {
-               dev->flags &= ~VIRTIO_DEV_RUNNING;
-               dev->notify_ops->destroy_device(vid);
-       }
+       vhost_destroy_device_notify(dev);
 
        cleanup_device(dev, 1);
        free_device(dev);
@@ -259,6 +550,20 @@ vhost_destroy_device(int vid)
        vhost_devices[vid] = NULL;
 }
 
+void
+vhost_attach_vdpa_device(int vid, int did)
+{
+       struct virtio_net *dev = get_device(vid);
+
+       if (dev == NULL)
+               return;
+
+       if (rte_vdpa_get_device(did) == NULL)
+               return;
+
+       dev->vdpa_dev_id = did;
+}
+
 void
 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
 {
@@ -287,18 +592,32 @@ vhost_enable_dequeue_zero_copy(int vid)
        dev->dequeue_zero_copy = 1;
 }
 
+void
+vhost_set_builtin_virtio_net(int vid, bool enable)
+{
+       struct virtio_net *dev = get_device(vid);
+
+       if (dev == NULL)
+               return;
+
+       if (enable)
+               dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
+       else
+               dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
+}
+
 int
 rte_vhost_get_mtu(int vid, uint16_t *mtu)
 {
        struct virtio_net *dev = get_device(vid);
 
-       if (!dev)
+       if (dev == NULL || mtu == NULL)
                return -ENODEV;
 
        if (!(dev->flags & VIRTIO_DEV_READY))
                return -EAGAIN;
 
-       if (!(dev->features & VIRTIO_NET_F_MTU))
+       if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
                return -ENOTSUP;
 
        *mtu = dev->mtu;
@@ -314,14 +633,15 @@ rte_vhost_get_numa_node(int vid)
        int numa_node;
        int ret;
 
-       if (dev == NULL)
+       if (dev == NULL || numa_available() != 0)
                return -1;
 
        ret = get_mempolicy(&numa_node, NULL, 0, dev,
                            MPOL_F_NODE | MPOL_F_ADDR);
        if (ret < 0) {
                RTE_LOG(ERR, VHOST_CONFIG,
-                       "(%d) failed to query numa node: %d\n", vid, ret);
+                       "(%d) failed to query numa node: %s\n",
+                       vid, rte_strerror(errno));
                return -1;
        }
 
@@ -340,7 +660,18 @@ rte_vhost_get_queue_num(int vid)
        if (dev == NULL)
                return 0;
 
-       return dev->virt_qp_nb;
+       return dev->nr_vring / 2;
+}
+
+uint16_t
+rte_vhost_get_vring_num(int vid)
+{
+       struct virtio_net *dev = get_device(vid);
+
+       if (dev == NULL)
+               return 0;
+
+       return dev->nr_vring;
 }
 
 int
@@ -348,7 +679,7 @@ rte_vhost_get_ifname(int vid, char *buf, size_t len)
 {
        struct virtio_net *dev = get_device(vid);
 
-       if (dev == NULL)
+       if (dev == NULL || buf == NULL)
                return -1;
 
        len = RTE_MIN(len, sizeof(dev->ifname));
@@ -365,7 +696,7 @@ rte_vhost_get_negotiated_features(int vid, uint64_t *features)
        struct virtio_net *dev;
 
        dev = get_device(vid);
-       if (!dev)
+       if (dev == NULL || features == NULL)
                return -1;
 
        *features = dev->features;
@@ -380,11 +711,11 @@ rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
        size_t size;
 
        dev = get_device(vid);
-       if (!dev)
+       if (dev == NULL || mem == NULL)
                return -1;
 
        size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
-       m = malloc(size);
+       m = malloc(sizeof(struct rte_vhost_memory) + size);
        if (!m)
                return -1;
 
@@ -403,7 +734,7 @@ rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
        struct vhost_virtqueue *vq;
 
        dev = get_device(vid);
-       if (!dev)
+       if (dev == NULL || vring == NULL)
                return -1;
 
        if (vring_idx >= VHOST_MAX_VRING)
@@ -425,6 +756,31 @@ rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
        return 0;
 }
 
+int
+rte_vhost_vring_call(int vid, uint16_t vring_idx)
+{
+       struct virtio_net *dev;
+       struct vhost_virtqueue *vq;
+
+       dev = get_device(vid);
+       if (!dev)
+               return -1;
+
+       if (vring_idx >= VHOST_MAX_VRING)
+               return -1;
+
+       vq = dev->virtqueue[vring_idx];
+       if (!vq)
+               return -1;
+
+       if (vq_is_packed(dev))
+               vhost_vring_call_packed(dev, vq);
+       else
+               vhost_vring_call_split(dev, vq);
+
+       return 0;
+}
+
 uint16_t
 rte_vhost_avail_entries(int vid, uint16_t queue_id)
 {
@@ -442,20 +798,181 @@ rte_vhost_avail_entries(int vid, uint16_t queue_id)
        return *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
 }
 
+static inline void
+vhost_enable_notify_split(struct virtio_net *dev,
+               struct vhost_virtqueue *vq, int enable)
+{
+       if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
+               if (enable)
+                       vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
+               else
+                       vq->used->flags |= VRING_USED_F_NO_NOTIFY;
+       } else {
+               if (enable)
+                       vhost_avail_event(vq) = vq->last_avail_idx;
+       }
+}
+
+static inline void
+vhost_enable_notify_packed(struct virtio_net *dev,
+               struct vhost_virtqueue *vq, int enable)
+{
+       uint16_t flags;
+
+       if (!enable) {
+               vq->device_event->flags = VRING_EVENT_F_DISABLE;
+               return;
+       }
+
+       flags = VRING_EVENT_F_ENABLE;
+       if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
+               flags = VRING_EVENT_F_DESC;
+               vq->device_event->off_wrap = vq->last_avail_idx |
+                       vq->avail_wrap_counter << 15;
+       }
+
+       rte_smp_wmb();
+
+       vq->device_event->flags = flags;
+}
+
 int
 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
+{
+       struct virtio_net *dev = get_device(vid);
+       struct vhost_virtqueue *vq;
+
+       if (!dev)
+               return -1;
+
+       vq = dev->virtqueue[queue_id];
+
+       if (vq_is_packed(dev))
+               vhost_enable_notify_packed(dev, vq, enable);
+       else
+               vhost_enable_notify_split(dev, vq, enable);
+
+       return 0;
+}
+
+void
+rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
+{
+       struct virtio_net *dev = get_device(vid);
+
+       if (dev == NULL)
+               return;
+
+       vhost_log_write(dev, addr, len);
+}
+
+void
+rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
+                        uint64_t offset, uint64_t len)
+{
+       struct virtio_net *dev;
+       struct vhost_virtqueue *vq;
+
+       dev = get_device(vid);
+       if (dev == NULL)
+               return;
+
+       if (vring_idx >= VHOST_MAX_VRING)
+               return;
+       vq = dev->virtqueue[vring_idx];
+       if (!vq)
+               return;
+
+       vhost_log_used_vring(dev, vq, offset, len);
+}
+
+uint32_t
+rte_vhost_rx_queue_count(int vid, uint16_t qid)
+{
+       struct virtio_net *dev;
+       struct vhost_virtqueue *vq;
+
+       dev = get_device(vid);
+       if (dev == NULL)
+               return 0;
+
+       if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
+               RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
+                       dev->vid, __func__, qid);
+               return 0;
+       }
+
+       vq = dev->virtqueue[qid];
+       if (vq == NULL)
+               return 0;
+
+       if (unlikely(vq->enabled == 0 || vq->avail == NULL))
+               return 0;
+
+       return *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
+}
+
+int rte_vhost_get_vdpa_device_id(int vid)
 {
        struct virtio_net *dev = get_device(vid);
 
        if (dev == NULL)
                return -1;
 
-       if (enable) {
-               RTE_LOG(ERR, VHOST_CONFIG,
-                       "guest notification isn't supported.\n");
+       return dev->vdpa_dev_id;
+}
+
+int rte_vhost_get_log_base(int vid, uint64_t *log_base,
+               uint64_t *log_size)
+{
+       struct virtio_net *dev = get_device(vid);
+
+       if (dev == NULL || log_base == NULL || log_size == NULL)
+               return -1;
+
+       *log_base = dev->log_base;
+       *log_size = dev->log_size;
+
+       return 0;
+}
+
+int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
+               uint16_t *last_avail_idx, uint16_t *last_used_idx)
+{
+       struct virtio_net *dev = get_device(vid);
+
+       if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
+               return -1;
+
+       *last_avail_idx = dev->virtqueue[queue_id]->last_avail_idx;
+       *last_used_idx = dev->virtqueue[queue_id]->last_used_idx;
+
+       return 0;
+}
+
+int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
+               uint16_t last_avail_idx, uint16_t last_used_idx)
+{
+       struct virtio_net *dev = get_device(vid);
+
+       if (!dev)
+               return -1;
+
+       dev->virtqueue[queue_id]->last_avail_idx = last_avail_idx;
+       dev->virtqueue[queue_id]->last_used_idx = last_used_idx;
+
+       return 0;
+}
+
+int rte_vhost_extern_callback_register(int vid,
+               struct rte_vhost_user_extern_ops const * const ops, void *ctx)
+{
+       struct virtio_net *dev = get_device(vid);
+
+       if (dev == NULL || ops == NULL)
                return -1;
-       }
 
-       dev->virtqueue[queue_id]->used->flags = VRING_USED_F_NO_NOTIFY;
+       dev->extern_ops = *ops;
+       dev->extern_data = ctx;
        return 0;
 }