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>
41 #ifdef RTE_LIBRTE_VHOST_NUMA
45 #include <sys/socket.h>
47 #include <rte_ethdev.h>
49 #include <rte_string_fns.h>
50 #include <rte_memory.h>
51 #include <rte_malloc.h>
52 #include <rte_virtio_net.h>
54 #include "vhost-net.h"
55 #include "virtio-net.h"
58 * Device linked list structure for configuration.
60 struct virtio_net_config_ll {
61 struct virtio_net dev; /* Virtio device.*/
62 struct virtio_net_config_ll *next; /* Next dev on linked list.*/
65 /* device ops to add/remove device to/from data core. */
66 struct virtio_net_device_ops const *notify_ops;
67 /* root address of the linked list of managed virtio devices */
68 static struct virtio_net_config_ll *ll_root;
70 /* Features supported by this lib. */
71 #define VHOST_SUPPORTED_FEATURES ((1ULL << VIRTIO_NET_F_MRG_RXBUF) | \
72 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
73 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
74 (1ULL << VHOST_F_LOG_ALL))
75 static uint64_t VHOST_FEATURES = VHOST_SUPPORTED_FEATURES;
79 * Converts QEMU virtual address to Vhost virtual address. This function is
80 * used to convert the ring addresses to our address space.
83 qva_to_vva(struct virtio_net *dev, uint64_t qemu_va)
85 struct virtio_memory_regions *region;
86 uint64_t vhost_va = 0;
87 uint32_t regionidx = 0;
89 /* Find the region where the address lives. */
90 for (regionidx = 0; regionidx < dev->mem->nregions; regionidx++) {
91 region = &dev->mem->regions[regionidx];
92 if ((qemu_va >= region->userspace_address) &&
93 (qemu_va <= region->userspace_address +
94 region->memory_size)) {
95 vhost_va = qemu_va + region->guest_phys_address +
96 region->address_offset -
97 region->userspace_address;
106 * Retrieves an entry from the devices configuration linked list.
108 static struct virtio_net_config_ll *
109 get_config_ll_entry(struct vhost_device_ctx ctx)
111 struct virtio_net_config_ll *ll_dev = ll_root;
113 /* Loop through linked list until the device_fh is found. */
114 while (ll_dev != NULL) {
115 if (ll_dev->dev.device_fh == ctx.fh)
117 ll_dev = ll_dev->next;
124 * Searches the configuration core linked list and
125 * retrieves the device if it exists.
128 get_device(struct vhost_device_ctx ctx)
130 struct virtio_net_config_ll *ll_dev;
132 ll_dev = get_config_ll_entry(ctx);
137 RTE_LOG(ERR, VHOST_CONFIG,
138 "(%"PRIu64") Device not found in linked list.\n", ctx.fh);
143 * Add entry containing a device to the device configuration linked list.
146 add_config_ll_entry(struct virtio_net_config_ll *new_ll_dev)
148 struct virtio_net_config_ll *ll_dev = ll_root;
150 /* If ll_dev == NULL then this is the first device so go to else */
152 /* If the 1st device_fh != 0 then we insert our device here. */
153 if (ll_dev->dev.device_fh != 0) {
154 new_ll_dev->dev.device_fh = 0;
155 new_ll_dev->next = ll_dev;
156 ll_root = new_ll_dev;
159 * Increment through the ll until we find un unused
160 * device_fh. Insert the device at that entry.
162 while ((ll_dev->next != NULL) &&
163 (ll_dev->dev.device_fh ==
164 (ll_dev->next->dev.device_fh - 1)))
165 ll_dev = ll_dev->next;
167 new_ll_dev->dev.device_fh = ll_dev->dev.device_fh + 1;
168 new_ll_dev->next = ll_dev->next;
169 ll_dev->next = new_ll_dev;
172 ll_root = new_ll_dev;
173 ll_root->dev.device_fh = 0;
179 * Unmap any memory, close any file descriptors and
180 * free any memory owned by a device.
183 cleanup_device(struct virtio_net *dev)
185 /* Unmap QEMU memory file if mapped. */
187 munmap((void *)(uintptr_t)dev->mem->mapped_address,
188 (size_t)dev->mem->mapped_size);
192 /* Close any event notifiers opened by device. */
193 if (dev->virtqueue[VIRTIO_RXQ]->callfd >= 0)
194 close(dev->virtqueue[VIRTIO_RXQ]->callfd);
195 if (dev->virtqueue[VIRTIO_RXQ]->kickfd >= 0)
196 close(dev->virtqueue[VIRTIO_RXQ]->kickfd);
197 if (dev->virtqueue[VIRTIO_TXQ]->callfd >= 0)
198 close(dev->virtqueue[VIRTIO_TXQ]->callfd);
199 if (dev->virtqueue[VIRTIO_TXQ]->kickfd >= 0)
200 close(dev->virtqueue[VIRTIO_TXQ]->kickfd);
204 * Release virtqueues and device memory.
207 free_device(struct virtio_net_config_ll *ll_dev)
209 /* Free any malloc'd memory */
210 rte_free(ll_dev->dev.virtqueue[VIRTIO_RXQ]);
211 rte_free(ll_dev->dev.virtqueue[VIRTIO_TXQ]);
216 * Remove an entry from the device configuration linked list.
218 static struct virtio_net_config_ll *
219 rm_config_ll_entry(struct virtio_net_config_ll *ll_dev,
220 struct virtio_net_config_ll *ll_dev_last)
222 /* First remove the device and then clean it up. */
223 if (ll_dev == ll_root) {
224 ll_root = ll_dev->next;
225 cleanup_device(&ll_dev->dev);
229 if (likely(ll_dev_last != NULL)) {
230 ll_dev_last->next = ll_dev->next;
231 cleanup_device(&ll_dev->dev);
233 return ll_dev_last->next;
235 cleanup_device(&ll_dev->dev);
237 RTE_LOG(ERR, VHOST_CONFIG,
238 "Remove entry from config_ll failed\n");
245 * Initialise all variables in device structure.
248 init_device(struct virtio_net *dev)
253 * Virtqueues have already been malloced so
254 * we don't want to set them to NULL.
256 vq_offset = offsetof(struct virtio_net, mem);
258 /* Set everything to 0. */
259 memset((void *)(uintptr_t)((uint64_t)(uintptr_t)dev + vq_offset), 0,
260 (sizeof(struct virtio_net) - (size_t)vq_offset));
261 memset(dev->virtqueue[VIRTIO_RXQ], 0, sizeof(struct vhost_virtqueue));
262 memset(dev->virtqueue[VIRTIO_TXQ], 0, sizeof(struct vhost_virtqueue));
264 dev->virtqueue[VIRTIO_RXQ]->kickfd = -1;
265 dev->virtqueue[VIRTIO_RXQ]->callfd = -1;
266 dev->virtqueue[VIRTIO_TXQ]->kickfd = -1;
267 dev->virtqueue[VIRTIO_TXQ]->callfd = -1;
269 /* Backends are set to -1 indicating an inactive device. */
270 dev->virtqueue[VIRTIO_RXQ]->backend = VIRTIO_DEV_STOPPED;
271 dev->virtqueue[VIRTIO_TXQ]->backend = VIRTIO_DEV_STOPPED;
275 * Function is called from the CUSE open function. The device structure is
276 * initialised and a new entry is added to the device configuration linked
280 new_device(struct vhost_device_ctx ctx)
282 struct virtio_net_config_ll *new_ll_dev;
283 struct vhost_virtqueue *virtqueue_rx, *virtqueue_tx;
285 /* Setup device and virtqueues. */
286 new_ll_dev = rte_malloc(NULL, sizeof(struct virtio_net_config_ll), 0);
287 if (new_ll_dev == NULL) {
288 RTE_LOG(ERR, VHOST_CONFIG,
289 "(%"PRIu64") Failed to allocate memory for dev.\n",
294 virtqueue_rx = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
295 if (virtqueue_rx == NULL) {
296 rte_free(new_ll_dev);
297 RTE_LOG(ERR, VHOST_CONFIG,
298 "(%"PRIu64") Failed to allocate memory for rxq.\n",
303 virtqueue_tx = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
304 if (virtqueue_tx == NULL) {
305 rte_free(virtqueue_rx);
306 rte_free(new_ll_dev);
307 RTE_LOG(ERR, VHOST_CONFIG,
308 "(%"PRIu64") Failed to allocate memory for txq.\n",
313 new_ll_dev->dev.virtqueue[VIRTIO_RXQ] = virtqueue_rx;
314 new_ll_dev->dev.virtqueue[VIRTIO_TXQ] = virtqueue_tx;
316 /* Initialise device and virtqueues. */
317 init_device(&new_ll_dev->dev);
319 new_ll_dev->next = NULL;
321 /* Add entry to device configuration linked list. */
322 add_config_ll_entry(new_ll_dev);
324 return new_ll_dev->dev.device_fh;
328 * Function is called from the CUSE release function. This function will
329 * cleanup the device and remove it from device configuration linked list.
332 destroy_device(struct vhost_device_ctx ctx)
334 struct virtio_net_config_ll *ll_dev_cur_ctx, *ll_dev_last = NULL;
335 struct virtio_net_config_ll *ll_dev_cur = ll_root;
337 /* Find the linked list entry for the device to be removed. */
338 ll_dev_cur_ctx = get_config_ll_entry(ctx);
339 while (ll_dev_cur != NULL) {
341 * If the device is found or
342 * a device that doesn't exist is found then it is removed.
344 if (ll_dev_cur == ll_dev_cur_ctx) {
346 * If the device is running on a data core then call
347 * the function to remove it from the data core.
349 if ((ll_dev_cur->dev.flags & VIRTIO_DEV_RUNNING))
350 notify_ops->destroy_device(&(ll_dev_cur->dev));
351 ll_dev_cur = rm_config_ll_entry(ll_dev_cur,
354 ll_dev_last = ll_dev_cur;
355 ll_dev_cur = ll_dev_cur->next;
361 set_ifname(struct vhost_device_ctx ctx,
362 const char *if_name, unsigned int if_len)
364 struct virtio_net *dev;
367 dev = get_device(ctx);
371 len = if_len > sizeof(dev->ifname) ?
372 sizeof(dev->ifname) : if_len;
374 strncpy(dev->ifname, if_name, len);
379 * Called from CUSE IOCTL: VHOST_SET_OWNER
380 * This function just returns success at the moment unless
381 * the device hasn't been initialised.
384 set_owner(struct vhost_device_ctx ctx)
386 struct virtio_net *dev;
388 dev = get_device(ctx);
396 * Called from CUSE IOCTL: VHOST_RESET_OWNER
399 reset_owner(struct vhost_device_ctx ctx)
401 struct virtio_net *dev;
404 dev = get_device(ctx);
408 device_fh = dev->device_fh;
411 dev->device_fh = device_fh;
416 * Called from CUSE IOCTL: VHOST_GET_FEATURES
417 * The features that we support are requested.
420 get_features(struct vhost_device_ctx ctx, uint64_t *pu)
422 struct virtio_net *dev;
424 dev = get_device(ctx);
428 /* Send our supported features. */
429 *pu = VHOST_FEATURES;
434 * Called from CUSE IOCTL: VHOST_SET_FEATURES
435 * We receive the negotiated features supported by us and the virtio device.
438 set_features(struct vhost_device_ctx ctx, uint64_t *pu)
440 struct virtio_net *dev;
442 dev = get_device(ctx);
445 if (*pu & ~VHOST_FEATURES)
448 /* Store the negotiated feature list for the device. */
451 /* Set the vhost_hlen depending on if VIRTIO_NET_F_MRG_RXBUF is set. */
452 if (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) {
453 LOG_DEBUG(VHOST_CONFIG,
454 "(%"PRIu64") Mergeable RX buffers enabled\n",
456 dev->virtqueue[VIRTIO_RXQ]->vhost_hlen =
457 sizeof(struct virtio_net_hdr_mrg_rxbuf);
458 dev->virtqueue[VIRTIO_TXQ]->vhost_hlen =
459 sizeof(struct virtio_net_hdr_mrg_rxbuf);
461 LOG_DEBUG(VHOST_CONFIG,
462 "(%"PRIu64") Mergeable RX buffers disabled\n",
464 dev->virtqueue[VIRTIO_RXQ]->vhost_hlen =
465 sizeof(struct virtio_net_hdr);
466 dev->virtqueue[VIRTIO_TXQ]->vhost_hlen =
467 sizeof(struct virtio_net_hdr);
473 * Called from CUSE IOCTL: VHOST_SET_VRING_NUM
474 * The virtio device sends us the size of the descriptor ring.
477 set_vring_num(struct vhost_device_ctx ctx, struct vhost_vring_state *state)
479 struct virtio_net *dev;
481 dev = get_device(ctx);
485 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
486 dev->virtqueue[state->index]->size = state->num;
492 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
493 * same numa node as the memory of vring descriptor.
495 #ifdef RTE_LIBRTE_VHOST_NUMA
496 static struct virtio_net*
497 numa_realloc(struct virtio_net *dev, int index)
499 int oldnode, newnode;
500 struct virtio_net_config_ll *old_ll_dev, *new_ll_dev = NULL;
501 struct vhost_virtqueue *old_vq, *new_vq = NULL;
503 int realloc_dev = 0, realloc_vq = 0;
505 old_ll_dev = (struct virtio_net_config_ll *)dev;
506 old_vq = dev->virtqueue[index];
508 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
509 MPOL_F_NODE | MPOL_F_ADDR);
510 ret = ret | get_mempolicy(&oldnode, NULL, 0, old_ll_dev,
511 MPOL_F_NODE | MPOL_F_ADDR);
513 RTE_LOG(ERR, VHOST_CONFIG,
514 "Unable to get vring desc or dev numa information.\n");
517 if (oldnode != newnode)
520 ret = get_mempolicy(&oldnode, NULL, 0, old_vq,
521 MPOL_F_NODE | MPOL_F_ADDR);
523 RTE_LOG(ERR, VHOST_CONFIG,
524 "Unable to get vq numa information.\n");
527 if (oldnode != newnode)
530 if (realloc_dev == 0 && realloc_vq == 0)
534 new_ll_dev = rte_malloc_socket(NULL,
535 sizeof(struct virtio_net_config_ll), 0, newnode);
537 new_vq = rte_malloc_socket(NULL,
538 sizeof(struct vhost_virtqueue), 0, newnode);
539 if (!new_ll_dev && !new_vq)
543 memcpy(new_vq, old_vq, sizeof(*new_vq));
545 memcpy(new_ll_dev, old_ll_dev, sizeof(*new_ll_dev));
546 (new_ll_dev ? new_ll_dev : old_ll_dev)->dev.virtqueue[index] =
547 new_vq ? new_vq : old_vq;
551 if (ll_root == old_ll_dev)
552 ll_root = new_ll_dev;
554 struct virtio_net_config_ll *prev = ll_root;
555 while (prev->next != old_ll_dev)
557 prev->next = new_ll_dev;
558 new_ll_dev->next = old_ll_dev->next;
560 rte_free(old_ll_dev);
563 return realloc_dev ? &new_ll_dev->dev : dev;
566 static struct virtio_net*
567 numa_realloc(struct virtio_net *dev, int index __rte_unused)
574 * Called from CUSE IOCTL: VHOST_SET_VRING_ADDR
575 * The virtio device sends us the desc, used and avail ring addresses.
576 * This function then converts these to our address space.
579 set_vring_addr(struct vhost_device_ctx ctx, struct vhost_vring_addr *addr)
581 struct virtio_net *dev;
582 struct vhost_virtqueue *vq;
584 dev = get_device(ctx);
588 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
589 vq = dev->virtqueue[addr->index];
591 /* The addresses are converted from QEMU virtual to Vhost virtual. */
592 vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
593 addr->desc_user_addr);
595 RTE_LOG(ERR, VHOST_CONFIG,
596 "(%"PRIu64") Failed to find desc ring address.\n",
601 dev = numa_realloc(dev, addr->index);
602 vq = dev->virtqueue[addr->index];
604 vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
605 addr->avail_user_addr);
606 if (vq->avail == 0) {
607 RTE_LOG(ERR, VHOST_CONFIG,
608 "(%"PRIu64") Failed to find avail ring address.\n",
613 vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
614 addr->used_user_addr);
616 RTE_LOG(ERR, VHOST_CONFIG,
617 "(%"PRIu64") Failed to find used ring address.\n",
622 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address desc: %p\n",
623 dev->device_fh, vq->desc);
624 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address avail: %p\n",
625 dev->device_fh, vq->avail);
626 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address used: %p\n",
627 dev->device_fh, vq->used);
633 * Called from CUSE IOCTL: VHOST_SET_VRING_BASE
634 * The virtio device sends us the available ring last used index.
637 set_vring_base(struct vhost_device_ctx ctx, struct vhost_vring_state *state)
639 struct virtio_net *dev;
641 dev = get_device(ctx);
645 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
646 dev->virtqueue[state->index]->last_used_idx = state->num;
647 dev->virtqueue[state->index]->last_used_idx_res = state->num;
653 * Called from CUSE IOCTL: VHOST_GET_VRING_BASE
654 * We send the virtio device our available ring last used index.
657 get_vring_base(struct vhost_device_ctx ctx, uint32_t index,
658 struct vhost_vring_state *state)
660 struct virtio_net *dev;
662 dev = get_device(ctx);
666 state->index = index;
667 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
668 state->num = dev->virtqueue[state->index]->last_used_idx;
675 * Called from CUSE IOCTL: VHOST_SET_VRING_CALL
676 * The virtio device sends an eventfd to interrupt the guest. This fd gets
677 * copied into our process space.
680 set_vring_call(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
682 struct virtio_net *dev;
683 struct vhost_virtqueue *vq;
685 dev = get_device(ctx);
689 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
690 vq = dev->virtqueue[file->index];
695 vq->callfd = file->fd;
701 * Called from CUSE IOCTL: VHOST_SET_VRING_KICK
702 * The virtio device sends an eventfd that it can use to notify us.
703 * This fd gets copied into our process space.
706 set_vring_kick(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
708 struct virtio_net *dev;
709 struct vhost_virtqueue *vq;
711 dev = get_device(ctx);
715 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
716 vq = dev->virtqueue[file->index];
721 vq->kickfd = file->fd;
727 * Called from CUSE IOCTL: VHOST_NET_SET_BACKEND
728 * To complete device initialisation when the virtio driver is loaded,
729 * we are provided with a valid fd for a tap device (not used by us).
730 * If this happens then we can add the device to a data core.
731 * When the virtio driver is removed we get fd=-1.
732 * At that point we remove the device from the data core.
733 * The device will still exist in the device configuration linked list.
736 set_backend(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
738 struct virtio_net *dev;
740 dev = get_device(ctx);
744 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
745 dev->virtqueue[file->index]->backend = file->fd;
748 * If the device isn't already running and both backend fds are set,
751 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
752 if (((int)dev->virtqueue[VIRTIO_TXQ]->backend != VIRTIO_DEV_STOPPED) &&
753 ((int)dev->virtqueue[VIRTIO_RXQ]->backend != VIRTIO_DEV_STOPPED)) {
754 return notify_ops->new_device(dev);
756 /* Otherwise we remove it. */
758 if (file->fd == VIRTIO_DEV_STOPPED)
759 notify_ops->destroy_device(dev);
764 * Function pointers are set for the device operations to allow CUSE to call
765 * functions when an IOCTL, device_add or device_release is received.
767 static const struct vhost_net_device_ops vhost_device_ops = {
768 .new_device = new_device,
769 .destroy_device = destroy_device,
771 .set_ifname = set_ifname,
773 .get_features = get_features,
774 .set_features = set_features,
776 .set_vring_num = set_vring_num,
777 .set_vring_addr = set_vring_addr,
778 .set_vring_base = set_vring_base,
779 .get_vring_base = get_vring_base,
781 .set_vring_kick = set_vring_kick,
782 .set_vring_call = set_vring_call,
784 .set_backend = set_backend,
786 .set_owner = set_owner,
787 .reset_owner = reset_owner,
791 * Called by main to setup callbacks when registering CUSE device.
793 struct vhost_net_device_ops const *
794 get_virtio_net_callbacks(void)
796 return &vhost_device_ops;
799 int rte_vhost_enable_guest_notification(struct virtio_net *dev,
800 uint16_t queue_id, int enable)
803 RTE_LOG(ERR, VHOST_CONFIG,
804 "guest notification isn't supported.\n");
808 dev->virtqueue[queue_id]->used->flags =
809 enable ? 0 : VRING_USED_F_NO_NOTIFY;
813 uint64_t rte_vhost_feature_get(void)
815 return VHOST_FEATURES;
818 int rte_vhost_feature_disable(uint64_t feature_mask)
820 VHOST_FEATURES = VHOST_FEATURES & ~feature_mask;
824 int rte_vhost_feature_enable(uint64_t feature_mask)
826 if ((feature_mask & VHOST_SUPPORTED_FEATURES) == feature_mask) {
827 VHOST_FEATURES = VHOST_FEATURES | feature_mask;
834 * Register ops so that we can add/remove device to data core.
837 rte_vhost_driver_callback_register(struct virtio_net_device_ops const * const ops)