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 (1ULL << VHOST_F_LOG_ALL) | \
78 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))
79 static uint64_t VHOST_FEATURES = VHOST_SUPPORTED_FEATURES;
83 * Converts QEMU virtual address to Vhost virtual address. This function is
84 * used to convert the ring addresses to our address space.
87 qva_to_vva(struct virtio_net *dev, uint64_t qemu_va)
89 struct virtio_memory_regions *region;
90 uint64_t vhost_va = 0;
91 uint32_t regionidx = 0;
93 /* Find the region where the address lives. */
94 for (regionidx = 0; regionidx < dev->mem->nregions; regionidx++) {
95 region = &dev->mem->regions[regionidx];
96 if ((qemu_va >= region->userspace_address) &&
97 (qemu_va <= region->userspace_address +
98 region->memory_size)) {
99 vhost_va = qemu_va + region->guest_phys_address +
100 region->address_offset -
101 region->userspace_address;
110 * Retrieves an entry from the devices configuration linked list.
112 static struct virtio_net_config_ll *
113 get_config_ll_entry(struct vhost_device_ctx ctx)
115 struct virtio_net_config_ll *ll_dev = ll_root;
117 /* Loop through linked list until the device_fh is found. */
118 while (ll_dev != NULL) {
119 if (ll_dev->dev.device_fh == ctx.fh)
121 ll_dev = ll_dev->next;
128 * Searches the configuration core linked list and
129 * retrieves the device if it exists.
132 get_device(struct vhost_device_ctx ctx)
134 struct virtio_net_config_ll *ll_dev;
136 ll_dev = get_config_ll_entry(ctx);
141 RTE_LOG(ERR, VHOST_CONFIG,
142 "(%"PRIu64") Device not found in linked list.\n", ctx.fh);
147 * Add entry containing a device to the device configuration linked list.
150 add_config_ll_entry(struct virtio_net_config_ll *new_ll_dev)
152 struct virtio_net_config_ll *ll_dev = ll_root;
154 /* If ll_dev == NULL then this is the first device so go to else */
156 /* If the 1st device_fh != 0 then we insert our device here. */
157 if (ll_dev->dev.device_fh != 0) {
158 new_ll_dev->dev.device_fh = 0;
159 new_ll_dev->next = ll_dev;
160 ll_root = new_ll_dev;
163 * Increment through the ll until we find un unused
164 * device_fh. Insert the device at that entry.
166 while ((ll_dev->next != NULL) &&
167 (ll_dev->dev.device_fh ==
168 (ll_dev->next->dev.device_fh - 1)))
169 ll_dev = ll_dev->next;
171 new_ll_dev->dev.device_fh = ll_dev->dev.device_fh + 1;
172 new_ll_dev->next = ll_dev->next;
173 ll_dev->next = new_ll_dev;
176 ll_root = new_ll_dev;
177 ll_root->dev.device_fh = 0;
183 cleanup_vq(struct vhost_virtqueue *vq)
192 * Unmap any memory, close any file descriptors and
193 * free any memory owned by a device.
196 cleanup_device(struct virtio_net *dev)
200 /* Unmap QEMU memory file if mapped. */
202 munmap((void *)(uintptr_t)dev->mem->mapped_address,
203 (size_t)dev->mem->mapped_size);
207 for (i = 0; i < dev->virt_qp_nb; i++) {
208 cleanup_vq(dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_RXQ]);
209 cleanup_vq(dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_TXQ]);
214 * Release virtqueues and device memory.
217 free_device(struct virtio_net_config_ll *ll_dev)
221 for (i = 0; i < ll_dev->dev.virt_qp_nb; i++)
222 rte_free(ll_dev->dev.virtqueue[i * VIRTIO_QNUM]);
228 * Remove an entry from the device configuration linked list.
230 static struct virtio_net_config_ll *
231 rm_config_ll_entry(struct virtio_net_config_ll *ll_dev,
232 struct virtio_net_config_ll *ll_dev_last)
234 /* First remove the device and then clean it up. */
235 if (ll_dev == ll_root) {
236 ll_root = ll_dev->next;
237 cleanup_device(&ll_dev->dev);
241 if (likely(ll_dev_last != NULL)) {
242 ll_dev_last->next = ll_dev->next;
243 cleanup_device(&ll_dev->dev);
245 return ll_dev_last->next;
247 cleanup_device(&ll_dev->dev);
249 RTE_LOG(ERR, VHOST_CONFIG,
250 "Remove entry from config_ll failed\n");
257 init_vring_queue(struct vhost_virtqueue *vq)
259 memset(vq, 0, sizeof(struct vhost_virtqueue));
264 /* Backends are set to -1 indicating an inactive device. */
269 init_vring_queue_pair(struct virtio_net *dev, uint32_t qp_idx)
271 init_vring_queue(dev->virtqueue[qp_idx * VIRTIO_QNUM + VIRTIO_RXQ]);
272 init_vring_queue(dev->virtqueue[qp_idx * VIRTIO_QNUM + VIRTIO_TXQ]);
276 alloc_vring_queue_pair(struct virtio_net *dev, uint32_t qp_idx)
278 struct vhost_virtqueue *virtqueue = NULL;
279 uint32_t virt_rx_q_idx = qp_idx * VIRTIO_QNUM + VIRTIO_RXQ;
280 uint32_t virt_tx_q_idx = qp_idx * VIRTIO_QNUM + VIRTIO_TXQ;
282 virtqueue = rte_malloc(NULL,
283 sizeof(struct vhost_virtqueue) * VIRTIO_QNUM, 0);
284 if (virtqueue == NULL) {
285 RTE_LOG(ERR, VHOST_CONFIG,
286 "Failed to allocate memory for virt qp:%d.\n", qp_idx);
290 dev->virtqueue[virt_rx_q_idx] = virtqueue;
291 dev->virtqueue[virt_tx_q_idx] = virtqueue + VIRTIO_TXQ;
293 init_vring_queue_pair(dev, qp_idx);
295 dev->virt_qp_nb += 1;
301 * Initialise all variables in device structure.
304 init_device(struct virtio_net *dev)
310 * Virtqueues have already been malloced so
311 * we don't want to set them to NULL.
313 vq_offset = offsetof(struct virtio_net, virtqueue);
314 memset(dev, 0, vq_offset);
316 for (i = 0; i < dev->virt_qp_nb; i++)
317 init_vring_queue_pair(dev, i);
321 * Function is called from the CUSE open function. The device structure is
322 * initialised and a new entry is added to the device configuration linked
326 new_device(struct vhost_device_ctx ctx)
328 struct virtio_net_config_ll *new_ll_dev;
330 /* Setup device and virtqueues. */
331 new_ll_dev = rte_malloc(NULL, sizeof(struct virtio_net_config_ll), 0);
332 if (new_ll_dev == NULL) {
333 RTE_LOG(ERR, VHOST_CONFIG,
334 "(%"PRIu64") Failed to allocate memory for dev.\n",
339 /* Initialise device and virtqueues. */
340 init_device(&new_ll_dev->dev);
342 new_ll_dev->next = NULL;
344 /* Add entry to device configuration linked list. */
345 add_config_ll_entry(new_ll_dev);
347 return new_ll_dev->dev.device_fh;
351 * Function is called from the CUSE release function. This function will
352 * cleanup the device and remove it from device configuration linked list.
355 destroy_device(struct vhost_device_ctx ctx)
357 struct virtio_net_config_ll *ll_dev_cur_ctx, *ll_dev_last = NULL;
358 struct virtio_net_config_ll *ll_dev_cur = ll_root;
360 /* Find the linked list entry for the device to be removed. */
361 ll_dev_cur_ctx = get_config_ll_entry(ctx);
362 while (ll_dev_cur != NULL) {
364 * If the device is found or
365 * a device that doesn't exist is found then it is removed.
367 if (ll_dev_cur == ll_dev_cur_ctx) {
369 * If the device is running on a data core then call
370 * the function to remove it from the data core.
372 if ((ll_dev_cur->dev.flags & VIRTIO_DEV_RUNNING))
373 notify_ops->destroy_device(&(ll_dev_cur->dev));
374 ll_dev_cur = rm_config_ll_entry(ll_dev_cur,
377 ll_dev_last = ll_dev_cur;
378 ll_dev_cur = ll_dev_cur->next;
384 set_ifname(struct vhost_device_ctx ctx,
385 const char *if_name, unsigned int if_len)
387 struct virtio_net *dev;
390 dev = get_device(ctx);
394 len = if_len > sizeof(dev->ifname) ?
395 sizeof(dev->ifname) : if_len;
397 strncpy(dev->ifname, if_name, len);
402 * Called from CUSE IOCTL: VHOST_SET_OWNER
403 * This function just returns success at the moment unless
404 * the device hasn't been initialised.
407 set_owner(struct vhost_device_ctx ctx)
409 struct virtio_net *dev;
411 dev = get_device(ctx);
419 * Called from CUSE IOCTL: VHOST_RESET_OWNER
422 reset_owner(struct vhost_device_ctx ctx)
424 struct virtio_net *dev;
427 dev = get_device(ctx);
431 device_fh = dev->device_fh;
434 dev->device_fh = device_fh;
439 * Called from CUSE IOCTL: VHOST_GET_FEATURES
440 * The features that we support are requested.
443 get_features(struct vhost_device_ctx ctx, uint64_t *pu)
445 struct virtio_net *dev;
447 dev = get_device(ctx);
451 /* Send our supported features. */
452 *pu = VHOST_FEATURES;
457 * Called from CUSE IOCTL: VHOST_SET_FEATURES
458 * We receive the negotiated features supported by us and the virtio device.
461 set_features(struct vhost_device_ctx ctx, uint64_t *pu)
463 struct virtio_net *dev;
467 dev = get_device(ctx);
470 if (*pu & ~VHOST_FEATURES)
474 if (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) {
475 LOG_DEBUG(VHOST_CONFIG,
476 "(%"PRIu64") Mergeable RX buffers enabled\n",
478 vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
480 LOG_DEBUG(VHOST_CONFIG,
481 "(%"PRIu64") Mergeable RX buffers disabled\n",
483 vhost_hlen = sizeof(struct virtio_net_hdr);
486 for (i = 0; i < dev->virt_qp_nb; i++) {
487 uint16_t base_idx = i * VIRTIO_QNUM;
489 dev->virtqueue[base_idx + VIRTIO_RXQ]->vhost_hlen = vhost_hlen;
490 dev->virtqueue[base_idx + VIRTIO_TXQ]->vhost_hlen = vhost_hlen;
497 * Called from CUSE IOCTL: VHOST_SET_VRING_NUM
498 * The virtio device sends us the size of the descriptor ring.
501 set_vring_num(struct vhost_device_ctx ctx, struct vhost_vring_state *state)
503 struct virtio_net *dev;
505 dev = get_device(ctx);
509 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
510 dev->virtqueue[state->index]->size = state->num;
516 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
517 * same numa node as the memory of vring descriptor.
519 #ifdef RTE_LIBRTE_VHOST_NUMA
520 static struct virtio_net*
521 numa_realloc(struct virtio_net *dev, int index)
523 int oldnode, newnode;
524 struct virtio_net_config_ll *old_ll_dev, *new_ll_dev = NULL;
525 struct vhost_virtqueue *old_vq, *new_vq = NULL;
527 int realloc_dev = 0, realloc_vq = 0;
529 old_ll_dev = (struct virtio_net_config_ll *)dev;
530 old_vq = dev->virtqueue[index];
532 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
533 MPOL_F_NODE | MPOL_F_ADDR);
534 ret = ret | get_mempolicy(&oldnode, NULL, 0, old_ll_dev,
535 MPOL_F_NODE | MPOL_F_ADDR);
537 RTE_LOG(ERR, VHOST_CONFIG,
538 "Unable to get vring desc or dev numa information.\n");
541 if (oldnode != newnode)
544 ret = get_mempolicy(&oldnode, NULL, 0, old_vq,
545 MPOL_F_NODE | MPOL_F_ADDR);
547 RTE_LOG(ERR, VHOST_CONFIG,
548 "Unable to get vq numa information.\n");
551 if (oldnode != newnode)
554 if (realloc_dev == 0 && realloc_vq == 0)
558 new_ll_dev = rte_malloc_socket(NULL,
559 sizeof(struct virtio_net_config_ll), 0, newnode);
561 new_vq = rte_malloc_socket(NULL,
562 sizeof(struct vhost_virtqueue), 0, newnode);
563 if (!new_ll_dev && !new_vq)
567 memcpy(new_vq, old_vq, sizeof(*new_vq));
569 memcpy(new_ll_dev, old_ll_dev, sizeof(*new_ll_dev));
570 (new_ll_dev ? new_ll_dev : old_ll_dev)->dev.virtqueue[index] =
571 new_vq ? new_vq : old_vq;
575 if (ll_root == old_ll_dev)
576 ll_root = new_ll_dev;
578 struct virtio_net_config_ll *prev = ll_root;
579 while (prev->next != old_ll_dev)
581 prev->next = new_ll_dev;
582 new_ll_dev->next = old_ll_dev->next;
584 rte_free(old_ll_dev);
587 return realloc_dev ? &new_ll_dev->dev : dev;
590 static struct virtio_net*
591 numa_realloc(struct virtio_net *dev, int index __rte_unused)
598 * Called from CUSE IOCTL: VHOST_SET_VRING_ADDR
599 * The virtio device sends us the desc, used and avail ring addresses.
600 * This function then converts these to our address space.
603 set_vring_addr(struct vhost_device_ctx ctx, struct vhost_vring_addr *addr)
605 struct virtio_net *dev;
606 struct vhost_virtqueue *vq;
608 dev = get_device(ctx);
612 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
613 vq = dev->virtqueue[addr->index];
615 /* The addresses are converted from QEMU virtual to Vhost virtual. */
616 vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
617 addr->desc_user_addr);
619 RTE_LOG(ERR, VHOST_CONFIG,
620 "(%"PRIu64") Failed to find desc ring address.\n",
625 dev = numa_realloc(dev, addr->index);
626 vq = dev->virtqueue[addr->index];
628 vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
629 addr->avail_user_addr);
630 if (vq->avail == 0) {
631 RTE_LOG(ERR, VHOST_CONFIG,
632 "(%"PRIu64") Failed to find avail ring address.\n",
637 vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
638 addr->used_user_addr);
640 RTE_LOG(ERR, VHOST_CONFIG,
641 "(%"PRIu64") Failed to find used ring address.\n",
646 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address desc: %p\n",
647 dev->device_fh, vq->desc);
648 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address avail: %p\n",
649 dev->device_fh, vq->avail);
650 LOG_DEBUG(VHOST_CONFIG, "(%"PRIu64") mapped address used: %p\n",
651 dev->device_fh, vq->used);
657 * Called from CUSE IOCTL: VHOST_SET_VRING_BASE
658 * The virtio device sends us the available ring last used index.
661 set_vring_base(struct vhost_device_ctx ctx, struct vhost_vring_state *state)
663 struct virtio_net *dev;
665 dev = get_device(ctx);
669 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
670 dev->virtqueue[state->index]->last_used_idx = state->num;
671 dev->virtqueue[state->index]->last_used_idx_res = state->num;
677 * Called from CUSE IOCTL: VHOST_GET_VRING_BASE
678 * We send the virtio device our available ring last used index.
681 get_vring_base(struct vhost_device_ctx ctx, uint32_t index,
682 struct vhost_vring_state *state)
684 struct virtio_net *dev;
686 dev = get_device(ctx);
690 state->index = index;
691 /* State->index refers to the queue index. The txq is 1, rxq is 0. */
692 state->num = dev->virtqueue[state->index]->last_used_idx;
699 * Called from CUSE IOCTL: VHOST_SET_VRING_CALL
700 * The virtio device sends an eventfd to interrupt the guest. This fd gets
701 * copied into our process space.
704 set_vring_call(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
706 struct virtio_net *dev;
707 struct vhost_virtqueue *vq;
708 uint32_t cur_qp_idx = file->index / VIRTIO_QNUM;
710 dev = get_device(ctx);
715 * FIXME: VHOST_SET_VRING_CALL is the first per-vring message
716 * we get, so we do vring queue pair allocation here.
718 if (cur_qp_idx + 1 > dev->virt_qp_nb) {
719 if (alloc_vring_queue_pair(dev, cur_qp_idx) < 0)
723 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
724 vq = dev->virtqueue[file->index];
730 vq->callfd = file->fd;
736 * Called from CUSE IOCTL: VHOST_SET_VRING_KICK
737 * The virtio device sends an eventfd that it can use to notify us.
738 * This fd gets copied into our process space.
741 set_vring_kick(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
743 struct virtio_net *dev;
744 struct vhost_virtqueue *vq;
746 dev = get_device(ctx);
750 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
751 vq = dev->virtqueue[file->index];
756 vq->kickfd = file->fd;
762 * Called from CUSE IOCTL: VHOST_NET_SET_BACKEND
763 * To complete device initialisation when the virtio driver is loaded,
764 * we are provided with a valid fd for a tap device (not used by us).
765 * If this happens then we can add the device to a data core.
766 * When the virtio driver is removed we get fd=-1.
767 * At that point we remove the device from the data core.
768 * The device will still exist in the device configuration linked list.
771 set_backend(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
773 struct virtio_net *dev;
775 dev = get_device(ctx);
779 /* file->index refers to the queue index. The txq is 1, rxq is 0. */
780 dev->virtqueue[file->index]->backend = file->fd;
783 * If the device isn't already running and both backend fds are set,
786 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
787 if (((int)dev->virtqueue[VIRTIO_TXQ]->backend != VIRTIO_DEV_STOPPED) &&
788 ((int)dev->virtqueue[VIRTIO_RXQ]->backend != VIRTIO_DEV_STOPPED)) {
789 return notify_ops->new_device(dev);
791 /* Otherwise we remove it. */
793 if (file->fd == VIRTIO_DEV_STOPPED)
794 notify_ops->destroy_device(dev);
799 * Function pointers are set for the device operations to allow CUSE to call
800 * functions when an IOCTL, device_add or device_release is received.
802 static const struct vhost_net_device_ops vhost_device_ops = {
803 .new_device = new_device,
804 .destroy_device = destroy_device,
806 .set_ifname = set_ifname,
808 .get_features = get_features,
809 .set_features = set_features,
811 .set_vring_num = set_vring_num,
812 .set_vring_addr = set_vring_addr,
813 .set_vring_base = set_vring_base,
814 .get_vring_base = get_vring_base,
816 .set_vring_kick = set_vring_kick,
817 .set_vring_call = set_vring_call,
819 .set_backend = set_backend,
821 .set_owner = set_owner,
822 .reset_owner = reset_owner,
826 * Called by main to setup callbacks when registering CUSE device.
828 struct vhost_net_device_ops const *
829 get_virtio_net_callbacks(void)
831 return &vhost_device_ops;
834 int rte_vhost_enable_guest_notification(struct virtio_net *dev,
835 uint16_t queue_id, int enable)
838 RTE_LOG(ERR, VHOST_CONFIG,
839 "guest notification isn't supported.\n");
843 dev->virtqueue[queue_id]->used->flags =
844 enable ? 0 : VRING_USED_F_NO_NOTIFY;
848 uint64_t rte_vhost_feature_get(void)
850 return VHOST_FEATURES;
853 int rte_vhost_feature_disable(uint64_t feature_mask)
855 VHOST_FEATURES = VHOST_FEATURES & ~feature_mask;
859 int rte_vhost_feature_enable(uint64_t feature_mask)
861 if ((feature_mask & VHOST_SUPPORTED_FEATURES) == feature_mask) {
862 VHOST_FEATURES = VHOST_FEATURES | feature_mask;
869 * Register ops so that we can add/remove device to data core.
872 rte_vhost_driver_callback_register(struct virtio_net_device_ops const * const ops)