1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
5 #include <linux/vhost.h>
6 #include <linux/virtio_net.h>
10 #ifdef RTE_LIBRTE_VHOST_NUMA
14 #include <rte_errno.h>
15 #include <rte_ethdev.h>
17 #include <rte_string_fns.h>
18 #include <rte_memory.h>
19 #include <rte_malloc.h>
20 #include <rte_vhost.h>
21 #include <rte_rwlock.h>
25 #include "vhost_user.h"
27 struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
29 /* Called with iotlb_lock read-locked */
31 __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
32 uint64_t iova, uint64_t size, uint8_t perm)
34 uint64_t vva, tmp_size;
41 vva = vhost_user_iotlb_cache_find(vq, iova, &tmp_size, perm);
45 if (!vhost_user_iotlb_pending_miss(vq, iova + tmp_size, perm)) {
47 * iotlb_lock is read-locked for a full burst,
48 * but it only protects the iotlb cache.
49 * In case of IOTLB miss, we might block on the socket,
50 * which could cause a deadlock with QEMU if an IOTLB update
51 * is being handled. We can safely unlock here to avoid it.
53 vhost_user_iotlb_rd_unlock(vq);
55 vhost_user_iotlb_pending_insert(vq, iova + tmp_size, perm);
56 vhost_user_iotlb_miss(dev, iova + tmp_size, perm);
58 vhost_user_iotlb_rd_lock(vq);
67 struct virtio_net *dev = vhost_devices[vid];
70 RTE_LOG(ERR, VHOST_CONFIG,
71 "(%d) device not found.\n", vid);
78 cleanup_vq(struct vhost_virtqueue *vq, int destroy)
80 if ((vq->callfd >= 0) && (destroy != 0))
87 * Unmap any memory, close any file descriptors and
88 * free any memory owned by a device.
91 cleanup_device(struct virtio_net *dev, int destroy)
95 vhost_backend_cleanup(dev);
97 for (i = 0; i < dev->nr_vring; i++)
98 cleanup_vq(dev->virtqueue[i], destroy);
102 free_vq(struct vhost_virtqueue *vq)
104 rte_free(vq->shadow_used_ring);
105 rte_free(vq->batch_copy_elems);
106 rte_mempool_free(vq->iotlb_pool);
111 * Release virtqueues and device memory.
114 free_device(struct virtio_net *dev)
118 for (i = 0; i < dev->nr_vring; i++)
119 free_vq(dev->virtqueue[i]);
125 vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
129 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
132 size = sizeof(struct vring_desc) * vq->size;
133 vq->desc = (struct vring_desc *)(uintptr_t)vhost_iova_to_vva(dev, vq,
134 vq->ring_addrs.desc_user_addr,
135 size, VHOST_ACCESS_RW);
139 size = sizeof(struct vring_avail);
140 size += sizeof(uint16_t) * vq->size;
141 vq->avail = (struct vring_avail *)(uintptr_t)vhost_iova_to_vva(dev, vq,
142 vq->ring_addrs.avail_user_addr,
143 size, VHOST_ACCESS_RW);
147 size = sizeof(struct vring_used);
148 size += sizeof(struct vring_used_elem) * vq->size;
149 vq->used = (struct vring_used *)(uintptr_t)vhost_iova_to_vva(dev, vq,
150 vq->ring_addrs.used_user_addr,
151 size, VHOST_ACCESS_RW);
162 vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq)
164 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
165 vhost_user_iotlb_wr_lock(vq);
172 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
173 vhost_user_iotlb_wr_unlock(vq);
177 init_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
179 struct vhost_virtqueue *vq;
181 if (vring_idx >= VHOST_MAX_VRING) {
182 RTE_LOG(ERR, VHOST_CONFIG,
183 "Failed not init vring, out of bound (%d)\n",
188 vq = dev->virtqueue[vring_idx];
190 memset(vq, 0, sizeof(struct vhost_virtqueue));
192 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
193 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
195 vhost_user_iotlb_init(dev, vring_idx);
196 /* Backends are set to -1 indicating an inactive device. */
199 TAILQ_INIT(&vq->zmbuf_list);
203 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
205 struct vhost_virtqueue *vq;
208 if (vring_idx >= VHOST_MAX_VRING) {
209 RTE_LOG(ERR, VHOST_CONFIG,
210 "Failed not init vring, out of bound (%d)\n",
215 vq = dev->virtqueue[vring_idx];
217 init_vring_queue(dev, vring_idx);
222 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
224 struct vhost_virtqueue *vq;
226 vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
228 RTE_LOG(ERR, VHOST_CONFIG,
229 "Failed to allocate memory for vring:%u.\n", vring_idx);
233 dev->virtqueue[vring_idx] = vq;
234 init_vring_queue(dev, vring_idx);
242 * Reset some variables in device structure, while keeping few
243 * others untouched, such as vid, ifname, nr_vring: they
244 * should be same unless the device is removed.
247 reset_device(struct virtio_net *dev)
252 dev->protocol_features = 0;
255 for (i = 0; i < dev->nr_vring; i++)
256 reset_vring_queue(dev, i);
260 * Invoked when there is a new vhost-user connection established (when
261 * there is a new virtio device being attached).
264 vhost_new_device(void)
266 struct virtio_net *dev;
269 dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
271 RTE_LOG(ERR, VHOST_CONFIG,
272 "Failed to allocate memory for new dev.\n");
276 for (i = 0; i < MAX_VHOST_DEVICE; i++) {
277 if (vhost_devices[i] == NULL)
280 if (i == MAX_VHOST_DEVICE) {
281 RTE_LOG(ERR, VHOST_CONFIG,
282 "Failed to find a free slot for new device.\n");
287 vhost_devices[i] = dev;
289 dev->slave_req_fd = -1;
295 * Invoked when there is the vhost-user connection is broken (when
296 * the virtio device is being detached).
299 vhost_destroy_device(int vid)
301 struct virtio_net *dev = get_device(vid);
306 if (dev->flags & VIRTIO_DEV_RUNNING) {
307 dev->flags &= ~VIRTIO_DEV_RUNNING;
308 dev->notify_ops->destroy_device(vid);
311 cleanup_device(dev, 1);
314 vhost_devices[vid] = NULL;
318 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
320 struct virtio_net *dev;
323 dev = get_device(vid);
327 len = if_len > sizeof(dev->ifname) ?
328 sizeof(dev->ifname) : if_len;
330 strncpy(dev->ifname, if_name, len);
331 dev->ifname[sizeof(dev->ifname) - 1] = '\0';
335 vhost_enable_dequeue_zero_copy(int vid)
337 struct virtio_net *dev = get_device(vid);
342 dev->dequeue_zero_copy = 1;
346 rte_vhost_get_mtu(int vid, uint16_t *mtu)
348 struct virtio_net *dev = get_device(vid);
353 if (!(dev->flags & VIRTIO_DEV_READY))
356 if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
365 rte_vhost_get_numa_node(int vid)
367 #ifdef RTE_LIBRTE_VHOST_NUMA
368 struct virtio_net *dev = get_device(vid);
375 ret = get_mempolicy(&numa_node, NULL, 0, dev,
376 MPOL_F_NODE | MPOL_F_ADDR);
378 RTE_LOG(ERR, VHOST_CONFIG,
379 "(%d) failed to query numa node: %s\n",
380 vid, rte_strerror(errno));
392 rte_vhost_get_queue_num(int vid)
394 struct virtio_net *dev = get_device(vid);
399 return dev->nr_vring / 2;
403 rte_vhost_get_vring_num(int vid)
405 struct virtio_net *dev = get_device(vid);
410 return dev->nr_vring;
414 rte_vhost_get_ifname(int vid, char *buf, size_t len)
416 struct virtio_net *dev = get_device(vid);
421 len = RTE_MIN(len, sizeof(dev->ifname));
423 strncpy(buf, dev->ifname, len);
430 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
432 struct virtio_net *dev;
434 dev = get_device(vid);
438 *features = dev->features;
443 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
445 struct virtio_net *dev;
446 struct rte_vhost_memory *m;
449 dev = get_device(vid);
453 size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
454 m = malloc(sizeof(struct rte_vhost_memory) + size);
458 m->nregions = dev->mem->nregions;
459 memcpy(m->regions, dev->mem->regions, size);
466 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
467 struct rte_vhost_vring *vring)
469 struct virtio_net *dev;
470 struct vhost_virtqueue *vq;
472 dev = get_device(vid);
476 if (vring_idx >= VHOST_MAX_VRING)
479 vq = dev->virtqueue[vring_idx];
483 vring->desc = vq->desc;
484 vring->avail = vq->avail;
485 vring->used = vq->used;
486 vring->log_guest_addr = vq->log_guest_addr;
488 vring->callfd = vq->callfd;
489 vring->kickfd = vq->kickfd;
490 vring->size = vq->size;
496 rte_vhost_vring_call(int vid, uint16_t vring_idx)
498 struct virtio_net *dev;
499 struct vhost_virtqueue *vq;
501 dev = get_device(vid);
505 if (vring_idx >= VHOST_MAX_VRING)
508 vq = dev->virtqueue[vring_idx];
512 vhost_vring_call(vq);
517 rte_vhost_avail_entries(int vid, uint16_t queue_id)
519 struct virtio_net *dev;
520 struct vhost_virtqueue *vq;
522 dev = get_device(vid);
526 vq = dev->virtqueue[queue_id];
530 return *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
534 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
536 struct virtio_net *dev = get_device(vid);
542 RTE_LOG(ERR, VHOST_CONFIG,
543 "guest notification isn't supported.\n");
547 dev->virtqueue[queue_id]->used->flags = VRING_USED_F_NO_NOTIFY;
552 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
554 struct virtio_net *dev = get_device(vid);
559 vhost_log_write(dev, addr, len);
563 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
564 uint64_t offset, uint64_t len)
566 struct virtio_net *dev;
567 struct vhost_virtqueue *vq;
569 dev = get_device(vid);
573 if (vring_idx >= VHOST_MAX_VRING)
575 vq = dev->virtqueue[vring_idx];
579 vhost_log_used_vring(dev, vq, offset, len);
583 rte_vhost_rx_queue_count(int vid, uint16_t qid)
585 struct virtio_net *dev;
586 struct vhost_virtqueue *vq;
588 dev = get_device(vid);
592 if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
593 RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
594 dev->vid, __func__, qid);
598 vq = dev->virtqueue[qid];
602 if (unlikely(vq->enabled == 0 || vq->avail == NULL))
605 return *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;