4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <linux/vhost.h>
35 #include <linux/virtio_net.h>
42 #ifdef RTE_LIBRTE_VHOST_NUMA
46 #include <sys/socket.h>
48 #include <rte_ethdev.h>
50 #include <rte_string_fns.h>
51 #include <rte_memory.h>
52 #include <rte_malloc.h>
53 #include <rte_virtio_net.h>
55 #include "vhost-net.h"
56 #include "virtio-net.h"
58 #define MAX_VHOST_DEVICE 1024
59 static struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
61 /* device ops to add/remove device to/from data core. */
62 struct virtio_net_device_ops const *notify_ops;
64 #define VHOST_USER_F_PROTOCOL_FEATURES 30
66 /* Features supported by this lib. */
67 #define VHOST_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
68 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
69 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
70 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
71 (VHOST_SUPPORTS_MQ) | \
72 (1ULL << VIRTIO_F_VERSION_1) | \
73 (1ULL << VHOST_F_LOG_ALL) | \
74 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES) | \
75 (1ULL << VIRTIO_NET_F_HOST_TSO4) | \
76 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
77 (1ULL << VIRTIO_NET_F_CSUM) | \
78 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
79 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
80 (1ULL << VIRTIO_NET_F_GUEST_TSO6))
82 static uint64_t VHOST_FEATURES = VHOST_SUPPORTED_FEATURES;
86 * Converts QEMU virtual address to Vhost virtual address. This function is
87 * used to convert the ring addresses to our address space.
90 qva_to_vva(struct virtio_net *dev, uint64_t qemu_va)
92 struct virtio_memory_regions *region;
93 uint64_t vhost_va = 0;
94 uint32_t regionidx = 0;
96 /* Find the region where the address lives. */
97 for (regionidx = 0; regionidx < dev->mem->nregions; regionidx++) {
98 region = &dev->mem->regions[regionidx];
99 if ((qemu_va >= region->userspace_address) &&
100 (qemu_va <= region->userspace_address +
101 region->memory_size)) {
102 vhost_va = qemu_va + region->guest_phys_address +
103 region->address_offset -
104 region->userspace_address;
113 get_device(struct vhost_device_ctx ctx)
115 struct virtio_net *dev = vhost_devices[ctx.fh];
117 if (unlikely(!dev)) {
118 RTE_LOG(ERR, VHOST_CONFIG,
119 "(%"PRIu64") device not found.\n", ctx.fh);
126 cleanup_vq(struct vhost_virtqueue *vq, int destroy)
128 if ((vq->callfd >= 0) && (destroy != 0))
135 * Unmap any memory, close any file descriptors and
136 * free any memory owned by a device.
139 cleanup_device(struct virtio_net *dev, int destroy)
143 vhost_backend_cleanup(dev);
145 for (i = 0; i < dev->virt_qp_nb; i++) {
146 cleanup_vq(dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_RXQ], destroy);
147 cleanup_vq(dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_TXQ], destroy);
152 * Release virtqueues and device memory.
155 free_device(struct virtio_net *dev)
159 for (i = 0; i < dev->virt_qp_nb; i++)
160 rte_free(dev->virtqueue[i * VIRTIO_QNUM]);
166 init_vring_queue(struct vhost_virtqueue *vq, int qp_idx)
168 memset(vq, 0, sizeof(struct vhost_virtqueue));
173 /* Backends are set to -1 indicating an inactive device. */
176 /* always set the default vq pair to enabled */
182 init_vring_queue_pair(struct virtio_net *dev, uint32_t qp_idx)
184 uint32_t base_idx = qp_idx * VIRTIO_QNUM;
186 init_vring_queue(dev->virtqueue[base_idx + VIRTIO_RXQ], qp_idx);
187 init_vring_queue(dev->virtqueue[base_idx + VIRTIO_TXQ], qp_idx);
191 reset_vring_queue(struct vhost_virtqueue *vq, int qp_idx)
196 init_vring_queue(vq, qp_idx);
201 reset_vring_queue_pair(struct virtio_net *dev, uint32_t qp_idx)
203 uint32_t base_idx = qp_idx * VIRTIO_QNUM;
205 reset_vring_queue(dev->virtqueue[base_idx + VIRTIO_RXQ], qp_idx);
206 reset_vring_queue(dev->virtqueue[base_idx + VIRTIO_TXQ], qp_idx);
210 alloc_vring_queue_pair(struct virtio_net *dev, uint32_t qp_idx)
212 struct vhost_virtqueue *virtqueue = NULL;
213 uint32_t virt_rx_q_idx = qp_idx * VIRTIO_QNUM + VIRTIO_RXQ;
214 uint32_t virt_tx_q_idx = qp_idx * VIRTIO_QNUM + VIRTIO_TXQ;
216 virtqueue = rte_malloc(NULL,
217 sizeof(struct vhost_virtqueue) * VIRTIO_QNUM, 0);
218 if (virtqueue == NULL) {
219 RTE_LOG(ERR, VHOST_CONFIG,
220 "Failed to allocate memory for virt qp:%d.\n", qp_idx);
224 dev->virtqueue[virt_rx_q_idx] = virtqueue;
225 dev->virtqueue[virt_tx_q_idx] = virtqueue + VIRTIO_TXQ;
227 init_vring_queue_pair(dev, qp_idx);
229 dev->virt_qp_nb += 1;
235 * Reset some variables in device structure, while keeping few
236 * others untouched, such as device_fh, ifname, virt_qp_nb: they
237 * should be same unless the device is removed.
240 reset_device(struct virtio_net *dev)
245 dev->protocol_features = 0;
248 for (i = 0; i < dev->virt_qp_nb; i++)
249 reset_vring_queue_pair(dev, i);
253 * Function is called from the CUSE open function. The device structure is
254 * initialised and a new entry is added to the device configuration linked
258 vhost_new_device(struct vhost_device_ctx ctx)
260 struct virtio_net *dev;
263 dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
265 RTE_LOG(ERR, VHOST_CONFIG,
266 "(%"PRIu64") Failed to allocate memory for dev.\n",
271 for (i = 0; i < MAX_VHOST_DEVICE; i++) {
272 if (vhost_devices[i] == NULL)
275 if (i == MAX_VHOST_DEVICE) {
276 RTE_LOG(ERR, VHOST_CONFIG,
277 "Failed to find a free slot for new device.\n");
281 vhost_devices[i] = dev;
288 * Function is called from the CUSE release function. This function will
289 * cleanup the device and remove it from device configuration linked list.
292 vhost_destroy_device(struct vhost_device_ctx ctx)
294 struct virtio_net *dev = get_device(ctx);
296 if (dev->flags & VIRTIO_DEV_RUNNING)
297 notify_ops->destroy_device(dev);
299 cleanup_device(dev, 1);
302 vhost_devices[ctx.fh] = NULL;
306 vhost_set_ifname(struct vhost_device_ctx ctx,
307 const char *if_name, unsigned int if_len)
309 struct virtio_net *dev;
312 dev = get_device(ctx);
316 len = if_len > sizeof(dev->ifname) ?
317 sizeof(dev->ifname) : if_len;
319 strncpy(dev->ifname, if_name, len);
324 * Called from CUSE IOCTL: VHOST_SET_OWNER
325 * This function just returns success at the moment unless
326 * the device hasn't been initialised.
329 vhost_set_owner(struct vhost_device_ctx ctx)
331 struct virtio_net *dev;
333 dev = get_device(ctx);
341 * Called from CUSE IOCTL: VHOST_RESET_OWNER
344 vhost_reset_owner(struct vhost_device_ctx ctx)
346 struct virtio_net *dev;
348 dev = get_device(ctx);
352 if (dev->flags & VIRTIO_DEV_RUNNING)
353 notify_ops->destroy_device(dev);
355 cleanup_device(dev, 0);
361 * Called from CUSE IOCTL: VHOST_GET_FEATURES
362 * The features that we support are requested.
365 vhost_get_features(struct vhost_device_ctx ctx, uint64_t *pu)
367 struct virtio_net *dev;
369 dev = get_device(ctx);
373 /* Send our supported features. */
374 *pu = VHOST_FEATURES;
379 * Called from CUSE IOCTL: VHOST_SET_FEATURES
380 * We receive the negotiated features supported by us and the virtio device.
383 vhost_set_features(struct vhost_device_ctx ctx, uint64_t *pu)
385 struct virtio_net *dev;
389 dev = get_device(ctx);
392 if (*pu & ~VHOST_FEATURES)
397 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
398 vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
400 vhost_hlen = sizeof(struct virtio_net_hdr);
402 LOG_DEBUG(VHOST_CONFIG,
403 "(%"PRIu64") Mergeable RX buffers %s, virtio 1 %s\n",
405 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
406 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
408 for (i = 0; i < dev->virt_qp_nb; i++) {
409 uint16_t base_idx = i * VIRTIO_QNUM;
411 dev->virtqueue[base_idx + VIRTIO_RXQ]->vhost_hlen = vhost_hlen;
412 dev->virtqueue[base_idx + VIRTIO_TXQ]->vhost_hlen = vhost_hlen;
419 * Called from CUSE IOCTL: VHOST_SET_VRING_NUM
420 * The virtio device sends us the size of the descriptor ring.
423 vhost_set_vring_num(struct vhost_device_ctx ctx,
424 struct vhost_vring_state *state)
426 struct virtio_net *dev;
428 dev = get_device(ctx);
432 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
433 dev->virtqueue[state->index]->size = state->num;
439 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
440 * same numa node as the memory of vring descriptor.
442 #ifdef RTE_LIBRTE_VHOST_NUMA
443 static struct virtio_net*
444 numa_realloc(struct virtio_net *dev, int index)
446 int oldnode, newnode;
447 struct virtio_net *old_dev, *new_dev = NULL;
448 struct vhost_virtqueue *old_vq, *new_vq = NULL;
450 int realloc_dev = 0, realloc_vq = 0;
453 old_vq = dev->virtqueue[index];
455 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
456 MPOL_F_NODE | MPOL_F_ADDR);
457 ret = ret | get_mempolicy(&oldnode, NULL, 0, old_dev,
458 MPOL_F_NODE | MPOL_F_ADDR);
460 RTE_LOG(ERR, VHOST_CONFIG,
461 "Unable to get vring desc or dev numa information.\n");
464 if (oldnode != newnode)
467 ret = get_mempolicy(&oldnode, NULL, 0, old_vq,
468 MPOL_F_NODE | MPOL_F_ADDR);
470 RTE_LOG(ERR, VHOST_CONFIG,
471 "Unable to get vq numa information.\n");
474 if (oldnode != newnode)
477 if (realloc_dev == 0 && realloc_vq == 0)
481 new_dev = rte_malloc_socket(NULL,
482 sizeof(struct virtio_net), 0, newnode);
484 new_vq = rte_malloc_socket(NULL,
485 sizeof(struct vhost_virtqueue), 0, newnode);
486 if (!new_dev && !new_vq)
490 memcpy(new_vq, old_vq, sizeof(*new_vq));
492 memcpy(new_dev, old_dev, sizeof(*new_dev));
494 (new_dev ? new_dev : old_dev)->virtqueue[index] =
495 new_vq ? new_vq : old_vq;
501 vhost_devices[new_dev->device_fh] = new_dev;
504 return realloc_dev ? new_dev : dev;
507 static struct virtio_net*
508 numa_realloc(struct virtio_net *dev, int index __rte_unused)
515 * Called from CUSE IOCTL: VHOST_SET_VRING_ADDR
516 * The virtio device sends us the desc, used and avail ring addresses.
517 * This function then converts these to our address space.
520 vhost_set_vring_addr(struct vhost_device_ctx ctx, struct vhost_vring_addr *addr)
522 struct virtio_net *dev;
523 struct vhost_virtqueue *vq;
525 dev = get_device(ctx);
526 if ((dev == NULL) || (dev->mem == NULL))
529 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
530 vq = dev->virtqueue[addr->index];
532 /* The addresses are converted from QEMU virtual to Vhost virtual. */
533 vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
534 addr->desc_user_addr);
536 RTE_LOG(ERR, VHOST_CONFIG,
537 "(%"PRIu64") Failed to find desc ring address.\n",
542 dev = numa_realloc(dev, addr->index);
543 vq = dev->virtqueue[addr->index];
545 vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
546 addr->avail_user_addr);
547 if (vq->avail == 0) {
548 RTE_LOG(ERR, VHOST_CONFIG,
549 "(%"PRIu64") Failed to find avail ring address.\n",
554 vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
555 addr->used_user_addr);
557 RTE_LOG(ERR, VHOST_CONFIG,
558 "(%"PRIu64") Failed to find used ring address.\n",
563 vq->log_guest_addr = addr->log_guest_addr;
565 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address desc: %p\n",
566 dev->device_fh, vq->desc);
567 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address avail: %p\n",
568 dev->device_fh, vq->avail);
569 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address used: %p\n",
570 dev->device_fh, vq->used);
571 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") log_guest_addr: %"PRIx64"\n",
572 dev->device_fh, vq->log_guest_addr);
578 * Called from CUSE IOCTL: VHOST_SET_VRING_BASE
579 * The virtio device sends us the available ring last used index.
582 vhost_set_vring_base(struct vhost_device_ctx ctx,
583 struct vhost_vring_state *state)
585 struct virtio_net *dev;
587 dev = get_device(ctx);
591 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
592 dev->virtqueue[state->index]->last_used_idx = state->num;
593 dev->virtqueue[state->index]->last_used_idx_res = state->num;
599 * Called from CUSE IOCTL: VHOST_GET_VRING_BASE
600 * We send the virtio device our available ring last used index.
603 vhost_get_vring_base(struct vhost_device_ctx ctx, uint32_t index,
604 struct vhost_vring_state *state)
606 struct virtio_net *dev;
608 dev = get_device(ctx);
612 state->index = index;
613 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
614 state->num = dev->virtqueue[state->index]->last_used_idx;
621 * Called from CUSE IOCTL: VHOST_SET_VRING_CALL
622 * The virtio device sends an eventfd to interrupt the guest. This fd gets
623 * copied into our process space.
626 vhost_set_vring_call(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
628 struct virtio_net *dev;
629 struct vhost_virtqueue *vq;
630 uint32_t cur_qp_idx = file->index / VIRTIO_QNUM;
632 dev = get_device(ctx);
637 * FIXME: VHOST_SET_VRING_CALL is the first per-vring message
638 * we get, so we do vring queue pair allocation here.
640 if (cur_qp_idx + 1 > dev->virt_qp_nb) {
641 if (alloc_vring_queue_pair(dev, cur_qp_idx) < 0)
645 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
646 vq = dev->virtqueue[file->index];
652 vq->callfd = file->fd;
658 * Called from CUSE IOCTL: VHOST_SET_VRING_KICK
659 * The virtio device sends an eventfd that it can use to notify us.
660 * This fd gets copied into our process space.
663 vhost_set_vring_kick(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
665 struct virtio_net *dev;
666 struct vhost_virtqueue *vq;
668 dev = get_device(ctx);
672 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
673 vq = dev->virtqueue[file->index];
678 vq->kickfd = file->fd;
684 * Called from CUSE IOCTL: VHOST_NET_SET_BACKEND
685 * To complete device initialisation when the virtio driver is loaded,
686 * we are provided with a valid fd for a tap device (not used by us).
687 * If this happens then we can add the device to a data core.
688 * When the virtio driver is removed we get fd=-1.
689 * At that point we remove the device from the data core.
690 * The device will still exist in the device configuration linked list.
693 vhost_set_backend(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
695 struct virtio_net *dev;
697 dev = get_device(ctx);
701 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
702 dev->virtqueue[file->index]->backend = file->fd;
705 * If the device isn't already running and both backend fds are set,
708 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
709 if (((int)dev->virtqueue[VIRTIO_TXQ]->backend != VIRTIO_DEV_STOPPED) &&
710 ((int)dev->virtqueue[VIRTIO_RXQ]->backend != VIRTIO_DEV_STOPPED)) {
711 return notify_ops->new_device(dev);
713 /* Otherwise we remove it. */
715 if (file->fd == VIRTIO_DEV_STOPPED)
716 notify_ops->destroy_device(dev);
720 int rte_vhost_enable_guest_notification(struct virtio_net *dev,
721 uint16_t queue_id, int enable)
724 RTE_LOG(ERR, VHOST_CONFIG,
725 "guest notification isn't supported.\n");
729 dev->virtqueue[queue_id]->used->flags = VRING_USED_F_NO_NOTIFY;
733 uint64_t rte_vhost_feature_get(void)
735 return VHOST_FEATURES;
738 int rte_vhost_feature_disable(uint64_t feature_mask)
740 VHOST_FEATURES = VHOST_FEATURES & ~feature_mask;
744 int rte_vhost_feature_enable(uint64_t feature_mask)
746 if ((feature_mask & VHOST_SUPPORTED_FEATURES) == feature_mask) {
747 VHOST_FEATURES = VHOST_FEATURES | feature_mask;
754 * Register ops so that we can add/remove device to data core.
757 rte_vhost_driver_callback_register(struct virtio_net_device_ops const * const ops)