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"
59 * Device linked list structure for configuration.
61 struct virtio_net_config_ll {
62 struct virtio_net dev; /* Virtio device.*/
63 struct virtio_net_config_ll *next; /* Next dev on linked list.*/
66 /* device ops to add/remove device to/from data core. */
67 struct virtio_net_device_ops const *notify_ops;
68 /* root address of the linked list of managed virtio devices */
69 static struct virtio_net_config_ll *ll_root;
71 #define VHOST_USER_F_PROTOCOL_FEATURES 30
73 /* Features supported by this lib. */
74 #define VHOST_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
75 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
76 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
77 (VHOST_SUPPORTS_MQ) | \
78 (1ULL << VHOST_F_LOG_ALL) | \
79 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))
80 static uint64_t VHOST_FEATURES = VHOST_SUPPORTED_FEATURES;
84 * Converts QEMU virtual address to Vhost virtual address. This function is
85 * used to convert the ring addresses to our address space.
88 qva_to_vva(struct virtio_net *dev, uint64_t qemu_va)
90 struct virtio_memory_regions *region;
91 uint64_t vhost_va = 0;
92 uint32_t regionidx = 0;
94 /* Find the region where the address lives. */
95 for (regionidx = 0; regionidx < dev->mem->nregions; regionidx++) {
96 region = &dev->mem->regions[regionidx];
97 if ((qemu_va >= region->userspace_address) &&
98 (qemu_va <= region->userspace_address +
99 region->memory_size)) {
100 vhost_va = qemu_va + region->guest_phys_address +
101 region->address_offset -
102 region->userspace_address;
111 * Retrieves an entry from the devices configuration linked list.
113 static struct virtio_net_config_ll *
114 get_config_ll_entry(struct vhost_device_ctx ctx)
116 struct virtio_net_config_ll *ll_dev = ll_root;
118 /* Loop through linked list until the device_fh is found. */
119 while (ll_dev != NULL) {
120 if (ll_dev->dev.device_fh == ctx.fh)
122 ll_dev = ll_dev->next;
129 * Searches the configuration core linked list and
130 * retrieves the device if it exists.
133 get_device(struct vhost_device_ctx ctx)
135 struct virtio_net_config_ll *ll_dev;
137 ll_dev = get_config_ll_entry(ctx);
142 RTE_LOG(ERR, VHOST_CONFIG,
143 "(%"PRIu64") Device not found in linked list.\n", ctx.fh);
148 * Add entry containing a device to the device configuration linked list.
151 add_config_ll_entry(struct virtio_net_config_ll *new_ll_dev)
153 struct virtio_net_config_ll *ll_dev = ll_root;
155 /* If ll_dev == NULL then this is the first device so go to else */
157 /* If the 1st device_fh != 0 then we insert our device here. */
158 if (ll_dev->dev.device_fh != 0) {
159 new_ll_dev->dev.device_fh = 0;
160 new_ll_dev->next = ll_dev;
161 ll_root = new_ll_dev;
164 * Increment through the ll until we find un unused
165 * device_fh. Insert the device at that entry.
167 while ((ll_dev->next != NULL) &&
168 (ll_dev->dev.device_fh ==
169 (ll_dev->next->dev.device_fh - 1)))
170 ll_dev = ll_dev->next;
172 new_ll_dev->dev.device_fh = ll_dev->dev.device_fh + 1;
173 new_ll_dev->next = ll_dev->next;
174 ll_dev->next = new_ll_dev;
177 ll_root = new_ll_dev;
178 ll_root->dev.device_fh = 0;
184 cleanup_vq(struct vhost_virtqueue *vq)
193 * Unmap any memory, close any file descriptors and
194 * free any memory owned by a device.
197 cleanup_device(struct virtio_net *dev)
201 /* Unmap QEMU memory file if mapped. */
203 munmap((void *)(uintptr_t)dev->mem->mapped_address,
204 (size_t)dev->mem->mapped_size);
208 for (i = 0; i < dev->virt_qp_nb; i++) {
209 cleanup_vq(dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_RXQ]);
210 cleanup_vq(dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_TXQ]);
215 * Release virtqueues and device memory.
218 free_device(struct virtio_net_config_ll *ll_dev)
222 for (i = 0; i < ll_dev->dev.virt_qp_nb; i++)
223 rte_free(ll_dev->dev.virtqueue[i * VIRTIO_QNUM]);
229 * Remove an entry from the device configuration linked list.
231 static struct virtio_net_config_ll *
232 rm_config_ll_entry(struct virtio_net_config_ll *ll_dev,
233 struct virtio_net_config_ll *ll_dev_last)
235 /* First remove the device and then clean it up. */
236 if (ll_dev == ll_root) {
237 ll_root = ll_dev->next;
238 cleanup_device(&ll_dev->dev);
242 if (likely(ll_dev_last != NULL)) {
243 ll_dev_last->next = ll_dev->next;
244 cleanup_device(&ll_dev->dev);
246 return ll_dev_last->next;
248 cleanup_device(&ll_dev->dev);
250 RTE_LOG(ERR, VHOST_CONFIG,
251 "Remove entry from config_ll failed\n");
258 init_vring_queue(struct vhost_virtqueue *vq, int qp_idx)
260 memset(vq, 0, sizeof(struct vhost_virtqueue));
265 /* Backends are set to -1 indicating an inactive device. */
268 /* always set the default vq pair to enabled */
274 init_vring_queue_pair(struct virtio_net *dev, uint32_t qp_idx)
276 uint32_t base_idx = qp_idx * VIRTIO_QNUM;
278 init_vring_queue(dev->virtqueue[base_idx + VIRTIO_RXQ], qp_idx);
279 init_vring_queue(dev->virtqueue[base_idx + VIRTIO_TXQ], qp_idx);
283 alloc_vring_queue_pair(struct virtio_net *dev, uint32_t qp_idx)
285 struct vhost_virtqueue *virtqueue = NULL;
286 uint32_t virt_rx_q_idx = qp_idx * VIRTIO_QNUM + VIRTIO_RXQ;
287 uint32_t virt_tx_q_idx = qp_idx * VIRTIO_QNUM + VIRTIO_TXQ;
289 virtqueue = rte_malloc(NULL,
290 sizeof(struct vhost_virtqueue) * VIRTIO_QNUM, 0);
291 if (virtqueue == NULL) {
292 RTE_LOG(ERR, VHOST_CONFIG,
293 "Failed to allocate memory for virt qp:%d.\n", qp_idx);
297 dev->virtqueue[virt_rx_q_idx] = virtqueue;
298 dev->virtqueue[virt_tx_q_idx] = virtqueue + VIRTIO_TXQ;
300 init_vring_queue_pair(dev, qp_idx);
302 dev->virt_qp_nb += 1;
308 * Initialise all variables in device structure.
311 init_device(struct virtio_net *dev)
317 * Virtqueues have already been malloced so
318 * we don't want to set them to NULL.
320 vq_offset = offsetof(struct virtio_net, virtqueue);
321 memset(dev, 0, vq_offset);
323 for (i = 0; i < dev->virt_qp_nb; i++)
324 init_vring_queue_pair(dev, i);
328 * Function is called from the CUSE open function. The device structure is
329 * initialised and a new entry is added to the device configuration linked
333 new_device(struct vhost_device_ctx ctx)
335 struct virtio_net_config_ll *new_ll_dev;
337 /* Setup device and virtqueues. */
338 new_ll_dev = rte_malloc(NULL, sizeof(struct virtio_net_config_ll), 0);
339 if (new_ll_dev == NULL) {
340 RTE_LOG(ERR, VHOST_CONFIG,
341 "(%"PRIu64") Failed to allocate memory for dev.\n",
346 /* Initialise device and virtqueues. */
347 init_device(&new_ll_dev->dev);
349 new_ll_dev->next = NULL;
351 /* Add entry to device configuration linked list. */
352 add_config_ll_entry(new_ll_dev);
354 return new_ll_dev->dev.device_fh;
358 * Function is called from the CUSE release function. This function will
359 * cleanup the device and remove it from device configuration linked list.
362 destroy_device(struct vhost_device_ctx ctx)
364 struct virtio_net_config_ll *ll_dev_cur_ctx, *ll_dev_last = NULL;
365 struct virtio_net_config_ll *ll_dev_cur = ll_root;
367 /* Find the linked list entry for the device to be removed. */
368 ll_dev_cur_ctx = get_config_ll_entry(ctx);
369 while (ll_dev_cur != NULL) {
371 * If the device is found or
372 * a device that doesn't exist is found then it is removed.
374 if (ll_dev_cur == ll_dev_cur_ctx) {
376 * If the device is running on a data core then call
377 * the function to remove it from the data core.
379 if ((ll_dev_cur->dev.flags & VIRTIO_DEV_RUNNING))
380 notify_ops->destroy_device(&(ll_dev_cur->dev));
381 ll_dev_cur = rm_config_ll_entry(ll_dev_cur,
384 ll_dev_last = ll_dev_cur;
385 ll_dev_cur = ll_dev_cur->next;
391 set_ifname(struct vhost_device_ctx ctx,
392 const char *if_name, unsigned int if_len)
394 struct virtio_net *dev;
397 dev = get_device(ctx);
401 len = if_len > sizeof(dev->ifname) ?
402 sizeof(dev->ifname) : if_len;
404 strncpy(dev->ifname, if_name, len);
409 * Called from CUSE IOCTL: VHOST_SET_OWNER
410 * This function just returns success at the moment unless
411 * the device hasn't been initialised.
414 set_owner(struct vhost_device_ctx ctx)
416 struct virtio_net *dev;
418 dev = get_device(ctx);
426 * Called from CUSE IOCTL: VHOST_RESET_OWNER
429 reset_owner(struct vhost_device_ctx ctx)
431 struct virtio_net *dev;
434 dev = get_device(ctx);
438 device_fh = dev->device_fh;
441 dev->device_fh = device_fh;
446 * Called from CUSE IOCTL: VHOST_GET_FEATURES
447 * The features that we support are requested.
450 get_features(struct vhost_device_ctx ctx, uint64_t *pu)
452 struct virtio_net *dev;
454 dev = get_device(ctx);
458 /* Send our supported features. */
459 *pu = VHOST_FEATURES;
464 * Called from CUSE IOCTL: VHOST_SET_FEATURES
465 * We receive the negotiated features supported by us and the virtio device.
468 set_features(struct vhost_device_ctx ctx, uint64_t *pu)
470 struct virtio_net *dev;
474 dev = get_device(ctx);
477 if (*pu & ~VHOST_FEATURES)
481 if (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) {
482 LOG_DEBUG(VHOST_CONFIG,
483 "(%"PRIu64") Mergeable RX buffers enabled\n",
485 vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
487 LOG_DEBUG(VHOST_CONFIG,
488 "(%"PRIu64") Mergeable RX buffers disabled\n",
490 vhost_hlen = sizeof(struct virtio_net_hdr);
493 for (i = 0; i < dev->virt_qp_nb; i++) {
494 uint16_t base_idx = i * VIRTIO_QNUM;
496 dev->virtqueue[base_idx + VIRTIO_RXQ]->vhost_hlen = vhost_hlen;
497 dev->virtqueue[base_idx + VIRTIO_TXQ]->vhost_hlen = vhost_hlen;
504 * Called from CUSE IOCTL: VHOST_SET_VRING_NUM
505 * The virtio device sends us the size of the descriptor ring.
508 set_vring_num(struct vhost_device_ctx ctx, struct vhost_vring_state *state)
510 struct virtio_net *dev;
512 dev = get_device(ctx);
516 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
517 dev->virtqueue[state->index]->size = state->num;
523 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
524 * same numa node as the memory of vring descriptor.
526 #ifdef RTE_LIBRTE_VHOST_NUMA
527 static struct virtio_net*
528 numa_realloc(struct virtio_net *dev, int index)
530 int oldnode, newnode;
531 struct virtio_net_config_ll *old_ll_dev, *new_ll_dev = NULL;
532 struct vhost_virtqueue *old_vq, *new_vq = NULL;
534 int realloc_dev = 0, realloc_vq = 0;
536 old_ll_dev = (struct virtio_net_config_ll *)dev;
537 old_vq = dev->virtqueue[index];
539 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
540 MPOL_F_NODE | MPOL_F_ADDR);
541 ret = ret | get_mempolicy(&oldnode, NULL, 0, old_ll_dev,
542 MPOL_F_NODE | MPOL_F_ADDR);
544 RTE_LOG(ERR, VHOST_CONFIG,
545 "Unable to get vring desc or dev numa information.\n");
548 if (oldnode != newnode)
551 ret = get_mempolicy(&oldnode, NULL, 0, old_vq,
552 MPOL_F_NODE | MPOL_F_ADDR);
554 RTE_LOG(ERR, VHOST_CONFIG,
555 "Unable to get vq numa information.\n");
558 if (oldnode != newnode)
561 if (realloc_dev == 0 && realloc_vq == 0)
565 new_ll_dev = rte_malloc_socket(NULL,
566 sizeof(struct virtio_net_config_ll), 0, newnode);
568 new_vq = rte_malloc_socket(NULL,
569 sizeof(struct vhost_virtqueue), 0, newnode);
570 if (!new_ll_dev && !new_vq)
574 memcpy(new_vq, old_vq, sizeof(*new_vq));
576 memcpy(new_ll_dev, old_ll_dev, sizeof(*new_ll_dev));
577 (new_ll_dev ? new_ll_dev : old_ll_dev)->dev.virtqueue[index] =
578 new_vq ? new_vq : old_vq;
582 if (ll_root == old_ll_dev)
583 ll_root = new_ll_dev;
585 struct virtio_net_config_ll *prev = ll_root;
586 while (prev->next != old_ll_dev)
588 prev->next = new_ll_dev;
589 new_ll_dev->next = old_ll_dev->next;
591 rte_free(old_ll_dev);
594 return realloc_dev ? &new_ll_dev->dev : dev;
597 static struct virtio_net*
598 numa_realloc(struct virtio_net *dev, int index __rte_unused)
605 * Called from CUSE IOCTL: VHOST_SET_VRING_ADDR
606 * The virtio device sends us the desc, used and avail ring addresses.
607 * This function then converts these to our address space.
610 set_vring_addr(struct vhost_device_ctx ctx, struct vhost_vring_addr *addr)
612 struct virtio_net *dev;
613 struct vhost_virtqueue *vq;
615 dev = get_device(ctx);
619 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
620 vq = dev->virtqueue[addr->index];
622 /* The addresses are converted from QEMU virtual to Vhost virtual. */
623 vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
624 addr->desc_user_addr);
626 RTE_LOG(ERR, VHOST_CONFIG,
627 "(%"PRIu64") Failed to find desc ring address.\n",
632 dev = numa_realloc(dev, addr->index);
633 vq = dev->virtqueue[addr->index];
635 vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
636 addr->avail_user_addr);
637 if (vq->avail == 0) {
638 RTE_LOG(ERR, VHOST_CONFIG,
639 "(%"PRIu64") Failed to find avail ring address.\n",
644 vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
645 addr->used_user_addr);
647 RTE_LOG(ERR, VHOST_CONFIG,
648 "(%"PRIu64") Failed to find used ring address.\n",
653 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address desc: %p\n",
654 dev->device_fh, vq->desc);
655 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address avail: %p\n",
656 dev->device_fh, vq->avail);
657 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address used: %p\n",
658 dev->device_fh, vq->used);
664 * Called from CUSE IOCTL: VHOST_SET_VRING_BASE
665 * The virtio device sends us the available ring last used index.
668 set_vring_base(struct vhost_device_ctx ctx, struct vhost_vring_state *state)
670 struct virtio_net *dev;
672 dev = get_device(ctx);
676 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
677 dev->virtqueue[state->index]->last_used_idx = state->num;
678 dev->virtqueue[state->index]->last_used_idx_res = state->num;
684 * Called from CUSE IOCTL: VHOST_GET_VRING_BASE
685 * We send the virtio device our available ring last used index.
688 get_vring_base(struct vhost_device_ctx ctx, uint32_t index,
689 struct vhost_vring_state *state)
691 struct virtio_net *dev;
693 dev = get_device(ctx);
697 state->index = index;
698 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
699 state->num = dev->virtqueue[state->index]->last_used_idx;
706 * Called from CUSE IOCTL: VHOST_SET_VRING_CALL
707 * The virtio device sends an eventfd to interrupt the guest. This fd gets
708 * copied into our process space.
711 set_vring_call(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
713 struct virtio_net *dev;
714 struct vhost_virtqueue *vq;
715 uint32_t cur_qp_idx = file->index / VIRTIO_QNUM;
717 dev = get_device(ctx);
722 * FIXME: VHOST_SET_VRING_CALL is the first per-vring message
723 * we get, so we do vring queue pair allocation here.
725 if (cur_qp_idx + 1 > dev->virt_qp_nb) {
726 if (alloc_vring_queue_pair(dev, cur_qp_idx) < 0)
730 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
731 vq = dev->virtqueue[file->index];
737 vq->callfd = file->fd;
743 * Called from CUSE IOCTL: VHOST_SET_VRING_KICK
744 * The virtio device sends an eventfd that it can use to notify us.
745 * This fd gets copied into our process space.
748 set_vring_kick(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
750 struct virtio_net *dev;
751 struct vhost_virtqueue *vq;
753 dev = get_device(ctx);
757 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
758 vq = dev->virtqueue[file->index];
763 vq->kickfd = file->fd;
769 * Called from CUSE IOCTL: VHOST_NET_SET_BACKEND
770 * To complete device initialisation when the virtio driver is loaded,
771 * we are provided with a valid fd for a tap device (not used by us).
772 * If this happens then we can add the device to a data core.
773 * When the virtio driver is removed we get fd=-1.
774 * At that point we remove the device from the data core.
775 * The device will still exist in the device configuration linked list.
778 set_backend(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
780 struct virtio_net *dev;
782 dev = get_device(ctx);
786 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
787 dev->virtqueue[file->index]->backend = file->fd;
790 * If the device isn't already running and both backend fds are set,
793 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
794 if (((int)dev->virtqueue[VIRTIO_TXQ]->backend != VIRTIO_DEV_STOPPED) &&
795 ((int)dev->virtqueue[VIRTIO_RXQ]->backend != VIRTIO_DEV_STOPPED)) {
796 return notify_ops->new_device(dev);
798 /* Otherwise we remove it. */
800 if (file->fd == VIRTIO_DEV_STOPPED)
801 notify_ops->destroy_device(dev);
806 * Function pointers are set for the device operations to allow CUSE to call
807 * functions when an IOCTL, device_add or device_release is received.
809 static const struct vhost_net_device_ops vhost_device_ops = {
810 .new_device = new_device,
811 .destroy_device = destroy_device,
813 .set_ifname = set_ifname,
815 .get_features = get_features,
816 .set_features = set_features,
818 .set_vring_num = set_vring_num,
819 .set_vring_addr = set_vring_addr,
820 .set_vring_base = set_vring_base,
821 .get_vring_base = get_vring_base,
823 .set_vring_kick = set_vring_kick,
824 .set_vring_call = set_vring_call,
826 .set_backend = set_backend,
828 .set_owner = set_owner,
829 .reset_owner = reset_owner,
833 * Called by main to setup callbacks when registering CUSE device.
835 struct vhost_net_device_ops const *
836 get_virtio_net_callbacks(void)
838 return &vhost_device_ops;
841 int rte_vhost_enable_guest_notification(struct virtio_net *dev,
842 uint16_t queue_id, int enable)
845 RTE_LOG(ERR, VHOST_CONFIG,
846 "guest notification isn't supported.\n");
850 dev->virtqueue[queue_id]->used->flags =
851 enable ? 0 : VRING_USED_F_NO_NOTIFY;
855 uint64_t rte_vhost_feature_get(void)
857 return VHOST_FEATURES;
860 int rte_vhost_feature_disable(uint64_t feature_mask)
862 VHOST_FEATURES = VHOST_FEATURES & ~feature_mask;
866 int rte_vhost_feature_enable(uint64_t feature_mask)
868 if ((feature_mask & VHOST_SUPPORTED_FEATURES) == feature_mask) {
869 VHOST_FEATURES = VHOST_FEATURES | feature_mask;
876 * Register ops so that we can add/remove device to data core.
879 rte_vhost_driver_callback_register(struct virtio_net_device_ops const * const ops)