--- /dev/null
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright (c) 2017 Red Hat, Inc.
+ *
+ *   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.
+ */
+
+#ifdef RTE_LIBRTE_VHOST_NUMA
+#include <numaif.h>
+#endif
+
+#include <rte_tailq.h>
+
+#include "iotlb.h"
+#include "vhost.h"
+
+struct vhost_iotlb_entry {
+       TAILQ_ENTRY(vhost_iotlb_entry) next;
+
+       uint64_t iova;
+       uint64_t uaddr;
+       uint64_t size;
+       uint8_t perm;
+};
+
+#define IOTLB_CACHE_SIZE 1024
+
+static void
+vhost_user_iotlb_cache_remove_all(struct vhost_virtqueue *vq)
+{
+       struct vhost_iotlb_entry *node, *temp_node;
+
+       rte_rwlock_write_lock(&vq->iotlb_lock);
+
+       TAILQ_FOREACH_SAFE(node, &vq->iotlb_list, next, temp_node) {
+               TAILQ_REMOVE(&vq->iotlb_list, node, next);
+               rte_mempool_put(vq->iotlb_pool, node);
+       }
+
+       vq->iotlb_cache_nr = 0;
+
+       rte_rwlock_write_unlock(&vq->iotlb_lock);
+}
+
+static void
+vhost_user_iotlb_cache_random_evict(struct vhost_virtqueue *vq)
+{
+       struct vhost_iotlb_entry *node, *temp_node;
+       int entry_idx;
+
+       rte_rwlock_write_lock(&vq->iotlb_lock);
+
+       entry_idx = rte_rand() % vq->iotlb_cache_nr;
+
+       TAILQ_FOREACH_SAFE(node, &vq->iotlb_list, next, temp_node) {
+               if (!entry_idx) {
+                       TAILQ_REMOVE(&vq->iotlb_list, node, next);
+                       rte_mempool_put(vq->iotlb_pool, node);
+                       vq->iotlb_cache_nr--;
+                       break;
+               }
+               entry_idx--;
+       }
+
+       rte_rwlock_write_unlock(&vq->iotlb_lock);
+}
+
+void
+vhost_user_iotlb_cache_insert(struct vhost_virtqueue *vq, uint64_t iova,
+                               uint64_t uaddr, uint64_t size, uint8_t perm)
+{
+       struct vhost_iotlb_entry *node, *new_node;
+       int ret;
+
+       ret = rte_mempool_get(vq->iotlb_pool, (void **)&new_node);
+       if (ret) {
+               RTE_LOG(DEBUG, VHOST_CONFIG, "IOTLB pool empty, evict one entry\n");
+               vhost_user_iotlb_cache_random_evict(vq);
+               ret = rte_mempool_get(vq->iotlb_pool, (void **)&new_node);
+               if (ret) {
+                       RTE_LOG(ERR, VHOST_CONFIG, "IOTLB pool still empty, failure\n");
+                       return;
+               }
+       }
+
+       new_node->iova = iova;
+       new_node->uaddr = uaddr;
+       new_node->size = size;
+       new_node->perm = perm;
+
+       rte_rwlock_write_lock(&vq->iotlb_lock);
+
+       TAILQ_FOREACH(node, &vq->iotlb_list, next) {
+               /*
+                * Entries must be invalidated before being updated.
+                * So if iova already in list, assume identical.
+                */
+               if (node->iova == new_node->iova) {
+                       rte_mempool_put(vq->iotlb_pool, new_node);
+                       goto unlock;
+               } else if (node->iova > new_node->iova) {
+                       TAILQ_INSERT_BEFORE(node, new_node, next);
+                       vq->iotlb_cache_nr++;
+                       goto unlock;
+               }
+       }
+
+       TAILQ_INSERT_TAIL(&vq->iotlb_list, new_node, next);
+       vq->iotlb_cache_nr++;
+
+unlock:
+       rte_rwlock_write_unlock(&vq->iotlb_lock);
+}
+
+void
+vhost_user_iotlb_cache_remove(struct vhost_virtqueue *vq,
+                                       uint64_t iova, uint64_t size)
+{
+       struct vhost_iotlb_entry *node, *temp_node;
+
+       if (unlikely(!size))
+               return;
+
+       rte_rwlock_write_lock(&vq->iotlb_lock);
+
+       TAILQ_FOREACH_SAFE(node, &vq->iotlb_list, next, temp_node) {
+               /* Sorted list */
+               if (unlikely(iova + size < node->iova))
+                       break;
+
+               if (iova < node->iova + node->size) {
+                       TAILQ_REMOVE(&vq->iotlb_list, node, next);
+                       rte_mempool_put(vq->iotlb_pool, node);
+                       vq->iotlb_cache_nr--;
+               }
+       }
+
+       rte_rwlock_write_unlock(&vq->iotlb_lock);
+}
+
+uint64_t
+vhost_user_iotlb_cache_find(struct vhost_virtqueue *vq, uint64_t iova,
+                                               uint64_t *size, uint8_t perm)
+{
+       struct vhost_iotlb_entry *node;
+       uint64_t offset, vva = 0, mapped = 0;
+
+       if (unlikely(!*size))
+               goto out;
+
+       TAILQ_FOREACH(node, &vq->iotlb_list, next) {
+               /* List sorted by iova */
+               if (unlikely(iova < node->iova))
+                       break;
+
+               if (iova >= node->iova + node->size)
+                       continue;
+
+               if (unlikely((perm & node->perm) != perm)) {
+                       vva = 0;
+                       break;
+               }
+
+               offset = iova - node->iova;
+               if (!vva)
+                       vva = node->uaddr + offset;
+
+               mapped += node->size - offset;
+               iova = node->iova + node->size;
+
+               if (mapped >= *size)
+                       break;
+       }
+
+out:
+       /* Only part of the requested chunk is mapped */
+       if (unlikely(mapped < *size))
+               *size = mapped;
+
+       return vva;
+}
+
+int
+vhost_user_iotlb_init(struct virtio_net *dev, int vq_index)
+{
+       char pool_name[RTE_MEMPOOL_NAMESIZE];
+       struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
+       int ret = -1, socket;
+
+       if (vq->iotlb_pool) {
+               /*
+                * The cache has already been initialized,
+                * just drop all entries
+                */
+               vhost_user_iotlb_cache_remove_all(vq);
+               return 0;
+       }
+
+#ifdef RTE_LIBRTE_VHOST_NUMA
+       ret = get_mempolicy(&socket, NULL, 0, vq, MPOL_F_NODE | MPOL_F_ADDR);
+#endif
+       if (ret)
+               socket = 0;
+
+       rte_rwlock_init(&vq->iotlb_lock);
+
+       TAILQ_INIT(&vq->iotlb_list);
+
+       snprintf(pool_name, sizeof(pool_name), "iotlb_cache_%d_%d",
+                       dev->vid, vq_index);
+
+       /* If already created, free it and recreate */
+       vq->iotlb_pool = rte_mempool_lookup(pool_name);
+       if (vq->iotlb_pool)
+               rte_mempool_free(vq->iotlb_pool);
+
+       vq->iotlb_pool = rte_mempool_create(pool_name,
+                       IOTLB_CACHE_SIZE, sizeof(struct vhost_iotlb_entry), 0,
+                       0, 0, NULL, NULL, NULL, socket,
+                       MEMPOOL_F_NO_CACHE_ALIGN |
+                       MEMPOOL_F_SP_PUT |
+                       MEMPOOL_F_SC_GET);
+       if (!vq->iotlb_pool) {
+               RTE_LOG(ERR, VHOST_CONFIG,
+                               "Failed to create IOTLB cache pool (%s)\n",
+                               pool_name);
+               return -1;
+       }
+
+       vq->iotlb_cache_nr = 0;
+
+       return 0;
+}
+
 
--- /dev/null
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright (c) 2017 Red Hat, Inc.
+ *
+ *   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.
+ */
+#ifndef _VHOST_IOTLB_H_
+#define _VHOST_IOTLB_H_
+
+#include "vhost.h"
+
+static __rte_always_inline void
+vhost_user_iotlb_rd_lock(struct vhost_virtqueue *vq)
+{
+       rte_rwlock_read_lock(&vq->iotlb_lock);
+}
+
+static __rte_always_inline void
+vhost_user_iotlb_rd_unlock(struct vhost_virtqueue *vq)
+{
+       rte_rwlock_read_unlock(&vq->iotlb_lock);
+}
+
+static __rte_always_inline void
+vhost_user_iotlb_wr_lock(struct vhost_virtqueue *vq)
+{
+       rte_rwlock_write_lock(&vq->iotlb_lock);
+}
+
+static __rte_always_inline void
+vhost_user_iotlb_wr_unlock(struct vhost_virtqueue *vq)
+{
+       rte_rwlock_write_unlock(&vq->iotlb_lock);
+}
+
+void vhost_user_iotlb_cache_insert(struct vhost_virtqueue *vq, uint64_t iova,
+                                       uint64_t uaddr, uint64_t size,
+                                       uint8_t perm);
+void vhost_user_iotlb_cache_remove(struct vhost_virtqueue *vq,
+                                       uint64_t iova, uint64_t size);
+uint64_t vhost_user_iotlb_cache_find(struct vhost_virtqueue *vq, uint64_t iova,
+                                       uint64_t *size, uint8_t perm);
+int vhost_user_iotlb_init(struct virtio_net *dev, int vq_index);
+
+#endif /* _VHOST_IOTLB_H_ */