1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016 Intel Corporation
11 #include <rte_memory.h>
14 #include "virtio_user_dev.h"
15 #include "vhost_kernel_tap.h"
17 struct vhost_kernel_data {
22 struct vhost_memory_kernel {
25 struct vhost_memory_region regions[0];
28 /* vhost kernel ioctls */
29 #define VHOST_VIRTIO 0xAF
30 #define VHOST_GET_FEATURES _IOR(VHOST_VIRTIO, 0x00, __u64)
31 #define VHOST_SET_FEATURES _IOW(VHOST_VIRTIO, 0x00, __u64)
32 #define VHOST_SET_OWNER _IO(VHOST_VIRTIO, 0x01)
33 #define VHOST_RESET_OWNER _IO(VHOST_VIRTIO, 0x02)
34 #define VHOST_SET_MEM_TABLE _IOW(VHOST_VIRTIO, 0x03, struct vhost_memory_kernel)
35 #define VHOST_SET_LOG_BASE _IOW(VHOST_VIRTIO, 0x04, __u64)
36 #define VHOST_SET_LOG_FD _IOW(VHOST_VIRTIO, 0x07, int)
37 #define VHOST_SET_VRING_NUM _IOW(VHOST_VIRTIO, 0x10, struct vhost_vring_state)
38 #define VHOST_SET_VRING_ADDR _IOW(VHOST_VIRTIO, 0x11, struct vhost_vring_addr)
39 #define VHOST_SET_VRING_BASE _IOW(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
40 #define VHOST_GET_VRING_BASE _IOWR(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
41 #define VHOST_SET_VRING_KICK _IOW(VHOST_VIRTIO, 0x20, struct vhost_vring_file)
42 #define VHOST_SET_VRING_CALL _IOW(VHOST_VIRTIO, 0x21, struct vhost_vring_file)
43 #define VHOST_SET_VRING_ERR _IOW(VHOST_VIRTIO, 0x22, struct vhost_vring_file)
44 #define VHOST_NET_SET_BACKEND _IOW(VHOST_VIRTIO, 0x30, struct vhost_vring_file)
46 /* with below features, vhost kernel does not need to do the checksum and TSO,
47 * these info will be passed to virtio_user through virtio net header.
49 #define VHOST_KERNEL_GUEST_OFFLOADS_MASK \
50 ((1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
51 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
52 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
53 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
54 (1ULL << VIRTIO_NET_F_GUEST_UFO))
56 /* with below features, when flows from virtio_user to vhost kernel
57 * (1) if flows goes up through the kernel networking stack, it does not need
58 * to verify checksum, which can save CPU cycles;
59 * (2) if flows goes through a Linux bridge and outside from an interface
60 * (kernel driver), checksum and TSO will be done by GSO in kernel or even
61 * offloaded into real physical device.
63 #define VHOST_KERNEL_HOST_OFFLOADS_MASK \
64 ((1ULL << VIRTIO_NET_F_HOST_TSO4) | \
65 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
66 (1ULL << VIRTIO_NET_F_CSUM))
68 static uint64_t max_regions = 64;
71 get_vhost_kernel_max_regions(void)
74 char buf[20] = {'\0'};
76 fd = open("/sys/module/vhost/parameters/max_mem_regions", O_RDONLY);
80 if (read(fd, buf, sizeof(buf) - 1) > 0)
81 max_regions = strtoull(buf, NULL, 10);
87 vhost_kernel_ioctl(int fd, uint64_t request, void *arg)
91 ret = ioctl(fd, request, arg);
93 PMD_DRV_LOG(ERR, "Vhost-kernel ioctl %"PRIu64" failed (%s)",
94 request, strerror(errno));
102 vhost_kernel_set_owner(struct virtio_user_dev *dev)
106 struct vhost_kernel_data *data = dev->backend_data;
108 for (i = 0; i < dev->max_queue_pairs; ++i) {
109 if (data->vhostfds[i] < 0)
112 ret = vhost_kernel_ioctl(data->vhostfds[i], VHOST_SET_OWNER, NULL);
121 vhost_kernel_get_features(struct virtio_user_dev *dev, uint64_t *features)
124 unsigned int tap_features;
125 struct vhost_kernel_data *data = dev->backend_data;
127 ret = vhost_kernel_ioctl(data->vhostfds[0], VHOST_GET_FEATURES, features);
129 PMD_DRV_LOG(ERR, "Failed to get features");
133 ret = tap_support_features(&tap_features);
135 PMD_DRV_LOG(ERR, "Failed to get TAP features");
139 /* with tap as the backend, all these features are supported
140 * but not claimed by vhost-net, so we add them back when
141 * reporting to upper layer.
143 if (tap_features & IFF_VNET_HDR) {
144 *features |= VHOST_KERNEL_GUEST_OFFLOADS_MASK;
145 *features |= VHOST_KERNEL_HOST_OFFLOADS_MASK;
148 /* vhost_kernel will not declare this feature, but it does
149 * support multi-queue.
151 if (tap_features & IFF_MULTI_QUEUE)
152 *features |= (1ull << VIRTIO_NET_F_MQ);
158 vhost_kernel_set_features(struct virtio_user_dev *dev, uint64_t features)
160 struct vhost_kernel_data *data = dev->backend_data;
162 /* We don't need memory protection here */
163 features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
164 /* VHOST kernel does not know about below flags */
165 features &= ~VHOST_KERNEL_GUEST_OFFLOADS_MASK;
166 features &= ~VHOST_KERNEL_HOST_OFFLOADS_MASK;
167 features &= ~(1ULL << VIRTIO_NET_F_MQ);
169 return vhost_kernel_ioctl(data->vhostfds[0], VHOST_SET_FEATURES, &features);
173 add_memseg_list(const struct rte_memseg_list *msl, void *arg)
175 struct vhost_memory_kernel *vm = arg;
176 struct vhost_memory_region *mr;
183 if (vm->nregions >= max_regions)
186 start_addr = msl->base_va;
187 len = msl->page_sz * msl->memseg_arr.len;
189 mr = &vm->regions[vm->nregions++];
191 mr->guest_phys_addr = (uint64_t)(uintptr_t)start_addr;
192 mr->userspace_addr = (uint64_t)(uintptr_t)start_addr;
193 mr->memory_size = len;
194 mr->mmap_offset = 0; /* flags_padding */
196 PMD_DRV_LOG(DEBUG, "index=%u addr=%p len=%" PRIu64,
197 vm->nregions - 1, start_addr, len);
202 /* By default, vhost kernel module allows 64 regions, but DPDK may
203 * have much more memory regions. Below function will treat each
204 * contiguous memory space reserved by DPDK as one region.
207 vhost_kernel_set_memory_table(struct virtio_user_dev *dev)
210 struct vhost_kernel_data *data = dev->backend_data;
211 struct vhost_memory_kernel *vm;
214 vm = malloc(sizeof(struct vhost_memory_kernel) +
216 sizeof(struct vhost_memory_region));
224 * The memory lock has already been taken by memory subsystem
225 * or virtio_user_start_device().
227 ret = rte_memseg_list_walk_thread_unsafe(add_memseg_list, vm);
231 for (i = 0; i < dev->max_queue_pairs; ++i) {
232 if (data->vhostfds[i] < 0)
235 ret = vhost_kernel_ioctl(data->vhostfds[i], VHOST_SET_MEM_TABLE, vm);
246 PMD_DRV_LOG(ERR, "Failed to set memory table");
251 vhost_kernel_set_vring(struct virtio_user_dev *dev, uint64_t req, struct vhost_vring_state *state)
254 unsigned int index = state->index;
255 struct vhost_kernel_data *data = dev->backend_data;
257 /* Convert from queue index to queue-pair & offset */
258 fd = data->vhostfds[state->index / 2];
261 ret = vhost_kernel_ioctl(fd, req, state);
263 PMD_DRV_LOG(ERR, "Failed to set vring (request %" PRIu64 ")", req);
267 /* restore index back to queue index */
268 state->index = index;
274 vhost_kernel_set_vring_num(struct virtio_user_dev *dev, struct vhost_vring_state *state)
276 return vhost_kernel_set_vring(dev, VHOST_SET_VRING_NUM, state);
280 vhost_kernel_set_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
282 return vhost_kernel_set_vring(dev, VHOST_SET_VRING_BASE, state);
286 vhost_kernel_get_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
288 return vhost_kernel_set_vring(dev, VHOST_GET_VRING_BASE, state);
292 vhost_kernel_set_vring_file(struct virtio_user_dev *dev, uint64_t req,
293 struct vhost_vring_file *file)
296 unsigned int index = file->index;
297 struct vhost_kernel_data *data = dev->backend_data;
299 /* Convert from queue index to queue-pair & offset */
300 fd = data->vhostfds[file->index / 2];
303 ret = vhost_kernel_ioctl(fd, req, file);
305 PMD_DRV_LOG(ERR, "Failed to set vring file (request %" PRIu64 ")", req);
309 /* restore index back to queue index */
316 vhost_kernel_set_vring_kick(struct virtio_user_dev *dev, struct vhost_vring_file *file)
318 return vhost_kernel_set_vring_file(dev, VHOST_SET_VRING_KICK, file);
322 vhost_kernel_set_vring_call(struct virtio_user_dev *dev, struct vhost_vring_file *file)
324 return vhost_kernel_set_vring_file(dev, VHOST_SET_VRING_CALL, file);
328 vhost_kernel_set_vring_addr(struct virtio_user_dev *dev, struct vhost_vring_addr *addr)
331 unsigned int index = addr->index;
332 struct vhost_kernel_data *data = dev->backend_data;
334 /* Convert from queue index to queue-pair & offset */
335 fd = data->vhostfds[addr->index / 2];
338 ret = vhost_kernel_ioctl(fd, VHOST_SET_VRING_ADDR, addr);
340 PMD_DRV_LOG(ERR, "Failed to set vring address");
344 /* restore index back to queue index */
351 vhost_kernel_get_status(struct virtio_user_dev *dev __rte_unused, uint8_t *status __rte_unused)
357 vhost_kernel_set_status(struct virtio_user_dev *dev __rte_unused, uint8_t status __rte_unused)
363 * Set up environment to talk with a vhost kernel backend.
366 * - (-1) if fail to set up;
367 * - (>=0) if successful.
370 vhost_kernel_setup(struct virtio_user_dev *dev)
374 struct vhost_kernel_data *data;
376 data = malloc(sizeof(*data));
378 PMD_INIT_LOG(ERR, "(%s) Failed to allocate Vhost-kernel data", dev->path);
382 data->vhostfds = malloc(dev->max_queue_pairs * sizeof(int));
383 if (!data->vhostfds) {
384 PMD_INIT_LOG(ERR, "(%s) Failed to allocate Vhost FDs", dev->path);
387 data->tapfds = malloc(dev->max_queue_pairs * sizeof(int));
389 PMD_INIT_LOG(ERR, "(%s) Failed to allocate TAP FDs", dev->path);
393 for (q = 0; q < dev->max_queue_pairs; ++q) {
394 data->vhostfds[q] = -1;
395 data->tapfds[q] = -1;
398 get_vhost_kernel_max_regions();
400 for (i = 0; i < dev->max_queue_pairs; ++i) {
401 vhostfd = open(dev->path, O_RDWR);
403 PMD_DRV_LOG(ERR, "fail to open %s, %s", dev->path, strerror(errno));
407 data->vhostfds[i] = vhostfd;
410 dev->backend_data = data;
415 for (i = 0; i < dev->max_queue_pairs; i++)
416 if (data->vhostfds[i] >= 0)
417 close(data->vhostfds[i]);
421 free(data->vhostfds);
429 vhost_kernel_destroy(struct virtio_user_dev *dev)
431 struct vhost_kernel_data *data = dev->backend_data;
437 for (i = 0; i < dev->max_queue_pairs; ++i) {
438 if (data->vhostfds[i] >= 0)
439 close(data->vhostfds[i]);
440 if (data->tapfds[i] >= 0)
441 close(data->tapfds[i]);
444 free(data->vhostfds);
447 dev->backend_data = NULL;
453 vhost_kernel_set_backend(int vhostfd, int tapfd)
455 struct vhost_vring_file f;
459 if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
460 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
466 if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
467 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
476 vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
483 int req_mq = (dev->max_queue_pairs > 1);
484 struct vhost_kernel_data *data = dev->backend_data;
486 vhostfd = data->vhostfds[pair_idx];
488 if (dev->qp_enabled[pair_idx] == enable)
492 tapfd = data->tapfds[pair_idx];
493 if (vhost_kernel_set_backend(vhostfd, -1) < 0) {
494 PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
497 if (req_mq && vhost_kernel_tap_set_queue(tapfd, false) < 0) {
498 PMD_DRV_LOG(ERR, "fail to disable tap for vhost kernel");
501 dev->qp_enabled[pair_idx] = false;
505 if (data->tapfds[pair_idx] >= 0) {
506 tapfd = data->tapfds[pair_idx];
507 if (vhost_kernel_tap_set_offload(tapfd, dev->features) == -1)
509 if (req_mq && vhost_kernel_tap_set_queue(tapfd, true) < 0) {
510 PMD_DRV_LOG(ERR, "fail to enable tap for vhost kernel");
516 if ((dev->features & (1ULL << VIRTIO_NET_F_MRG_RXBUF)) ||
517 (dev->features & (1ULL << VIRTIO_F_VERSION_1)))
518 hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
520 hdr_size = sizeof(struct virtio_net_hdr);
522 tapfd = vhost_kernel_open_tap(&dev->ifname, hdr_size, req_mq,
523 (char *)dev->mac_addr, dev->features);
525 PMD_DRV_LOG(ERR, "fail to open tap for vhost kernel");
529 data->tapfds[pair_idx] = tapfd;
532 if (vhost_kernel_set_backend(vhostfd, tapfd) < 0) {
533 PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
537 dev->qp_enabled[pair_idx] = true;
542 vhost_kernel_get_backend_features(uint64_t *features)
550 vhost_kernel_update_link_state(struct virtio_user_dev *dev __rte_unused)
552 /* Nothing to update (Maybe get TAP interface link state?) */
557 vhost_kernel_get_intr_fd(struct virtio_user_dev *dev __rte_unused)
559 /* No link state interrupt with Vhost-kernel */
563 struct virtio_user_backend_ops virtio_ops_kernel = {
564 .setup = vhost_kernel_setup,
565 .destroy = vhost_kernel_destroy,
566 .get_backend_features = vhost_kernel_get_backend_features,
567 .set_owner = vhost_kernel_set_owner,
568 .get_features = vhost_kernel_get_features,
569 .set_features = vhost_kernel_set_features,
570 .set_memory_table = vhost_kernel_set_memory_table,
571 .set_vring_num = vhost_kernel_set_vring_num,
572 .set_vring_base = vhost_kernel_set_vring_base,
573 .get_vring_base = vhost_kernel_get_vring_base,
574 .set_vring_call = vhost_kernel_set_vring_call,
575 .set_vring_kick = vhost_kernel_set_vring_kick,
576 .set_vring_addr = vhost_kernel_set_vring_addr,
577 .get_status = vhost_kernel_get_status,
578 .set_status = vhost_kernel_set_status,
579 .enable_qp = vhost_kernel_enable_queue_pair,
580 .update_link_state = vhost_kernel_update_link_state,
581 .get_intr_fd = vhost_kernel_get_intr_fd,