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;
164 /* We don't need memory protection here */
165 features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
166 /* VHOST kernel does not know about below flags */
167 features &= ~VHOST_KERNEL_GUEST_OFFLOADS_MASK;
168 features &= ~VHOST_KERNEL_HOST_OFFLOADS_MASK;
169 features &= ~(1ULL << VIRTIO_NET_F_MQ);
171 for (i = 0; i < dev->max_queue_pairs; ++i) {
172 if (data->vhostfds[i] < 0)
175 ret = vhost_kernel_ioctl(data->vhostfds[i], VHOST_SET_FEATURES, &features);
184 add_memseg_list(const struct rte_memseg_list *msl, void *arg)
186 struct vhost_memory_kernel *vm = arg;
187 struct vhost_memory_region *mr;
194 if (vm->nregions >= max_regions)
197 start_addr = msl->base_va;
198 len = msl->page_sz * msl->memseg_arr.len;
200 mr = &vm->regions[vm->nregions++];
202 mr->guest_phys_addr = (uint64_t)(uintptr_t)start_addr;
203 mr->userspace_addr = (uint64_t)(uintptr_t)start_addr;
204 mr->memory_size = len;
205 mr->mmap_offset = 0; /* flags_padding */
207 PMD_DRV_LOG(DEBUG, "index=%u addr=%p len=%" PRIu64,
208 vm->nregions - 1, start_addr, len);
213 /* By default, vhost kernel module allows 64 regions, but DPDK may
214 * have much more memory regions. Below function will treat each
215 * contiguous memory space reserved by DPDK as one region.
218 vhost_kernel_set_memory_table(struct virtio_user_dev *dev)
221 struct vhost_kernel_data *data = dev->backend_data;
222 struct vhost_memory_kernel *vm;
225 vm = malloc(sizeof(struct vhost_memory_kernel) +
227 sizeof(struct vhost_memory_region));
235 * The memory lock has already been taken by memory subsystem
236 * or virtio_user_start_device().
238 ret = rte_memseg_list_walk_thread_unsafe(add_memseg_list, vm);
242 for (i = 0; i < dev->max_queue_pairs; ++i) {
243 if (data->vhostfds[i] < 0)
246 ret = vhost_kernel_ioctl(data->vhostfds[i], VHOST_SET_MEM_TABLE, vm);
257 PMD_DRV_LOG(ERR, "Failed to set memory table");
262 vhost_kernel_set_vring(struct virtio_user_dev *dev, uint64_t req, struct vhost_vring_state *state)
265 unsigned int index = state->index;
266 struct vhost_kernel_data *data = dev->backend_data;
268 /* Convert from queue index to queue-pair & offset */
269 fd = data->vhostfds[state->index / 2];
272 ret = vhost_kernel_ioctl(fd, req, state);
274 PMD_DRV_LOG(ERR, "Failed to set vring (request %" PRIu64 ")", req);
278 /* restore index back to queue index */
279 state->index = index;
285 vhost_kernel_set_vring_num(struct virtio_user_dev *dev, struct vhost_vring_state *state)
287 return vhost_kernel_set_vring(dev, VHOST_SET_VRING_NUM, state);
291 vhost_kernel_set_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
293 return vhost_kernel_set_vring(dev, VHOST_SET_VRING_BASE, state);
297 vhost_kernel_get_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
299 return vhost_kernel_set_vring(dev, VHOST_GET_VRING_BASE, state);
303 vhost_kernel_set_vring_file(struct virtio_user_dev *dev, uint64_t req,
304 struct vhost_vring_file *file)
307 unsigned int index = file->index;
308 struct vhost_kernel_data *data = dev->backend_data;
310 /* Convert from queue index to queue-pair & offset */
311 fd = data->vhostfds[file->index / 2];
314 ret = vhost_kernel_ioctl(fd, req, file);
316 PMD_DRV_LOG(ERR, "Failed to set vring file (request %" PRIu64 ")", req);
320 /* restore index back to queue index */
327 vhost_kernel_set_vring_kick(struct virtio_user_dev *dev, struct vhost_vring_file *file)
329 return vhost_kernel_set_vring_file(dev, VHOST_SET_VRING_KICK, file);
333 vhost_kernel_set_vring_call(struct virtio_user_dev *dev, struct vhost_vring_file *file)
335 return vhost_kernel_set_vring_file(dev, VHOST_SET_VRING_CALL, file);
339 vhost_kernel_set_vring_addr(struct virtio_user_dev *dev, struct vhost_vring_addr *addr)
342 unsigned int index = addr->index;
343 struct vhost_kernel_data *data = dev->backend_data;
345 /* Convert from queue index to queue-pair & offset */
346 fd = data->vhostfds[addr->index / 2];
349 ret = vhost_kernel_ioctl(fd, VHOST_SET_VRING_ADDR, addr);
351 PMD_DRV_LOG(ERR, "Failed to set vring address");
355 /* restore index back to queue index */
362 vhost_kernel_get_status(struct virtio_user_dev *dev __rte_unused, uint8_t *status __rte_unused)
368 vhost_kernel_set_status(struct virtio_user_dev *dev __rte_unused, uint8_t status __rte_unused)
374 * Set up environment to talk with a vhost kernel backend.
377 * - (-1) if fail to set up;
378 * - (>=0) if successful.
381 vhost_kernel_setup(struct virtio_user_dev *dev)
385 struct vhost_kernel_data *data;
387 data = malloc(sizeof(*data));
389 PMD_INIT_LOG(ERR, "(%s) Failed to allocate Vhost-kernel data", dev->path);
393 data->vhostfds = malloc(dev->max_queue_pairs * sizeof(int));
394 if (!data->vhostfds) {
395 PMD_INIT_LOG(ERR, "(%s) Failed to allocate Vhost FDs", dev->path);
398 data->tapfds = malloc(dev->max_queue_pairs * sizeof(int));
400 PMD_INIT_LOG(ERR, "(%s) Failed to allocate TAP FDs", dev->path);
404 for (q = 0; q < dev->max_queue_pairs; ++q) {
405 data->vhostfds[q] = -1;
406 data->tapfds[q] = -1;
409 get_vhost_kernel_max_regions();
411 for (i = 0; i < dev->max_queue_pairs; ++i) {
412 vhostfd = open(dev->path, O_RDWR);
414 PMD_DRV_LOG(ERR, "fail to open %s, %s", dev->path, strerror(errno));
418 data->vhostfds[i] = vhostfd;
421 dev->backend_data = data;
426 for (i = 0; i < dev->max_queue_pairs; i++)
427 if (data->vhostfds[i] >= 0)
428 close(data->vhostfds[i]);
432 free(data->vhostfds);
440 vhost_kernel_destroy(struct virtio_user_dev *dev)
442 struct vhost_kernel_data *data = dev->backend_data;
448 for (i = 0; i < dev->max_queue_pairs; ++i) {
449 if (data->vhostfds[i] >= 0)
450 close(data->vhostfds[i]);
451 if (data->tapfds[i] >= 0)
452 close(data->tapfds[i]);
455 free(data->vhostfds);
458 dev->backend_data = NULL;
464 vhost_kernel_set_backend(int vhostfd, int tapfd)
466 struct vhost_vring_file f;
470 if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
471 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
477 if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
478 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
487 vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
494 int req_mq = (dev->max_queue_pairs > 1);
495 struct vhost_kernel_data *data = dev->backend_data;
497 vhostfd = data->vhostfds[pair_idx];
499 if (dev->qp_enabled[pair_idx] == enable)
503 tapfd = data->tapfds[pair_idx];
504 if (vhost_kernel_set_backend(vhostfd, -1) < 0) {
505 PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
508 if (req_mq && vhost_kernel_tap_set_queue(tapfd, false) < 0) {
509 PMD_DRV_LOG(ERR, "fail to disable tap for vhost kernel");
512 dev->qp_enabled[pair_idx] = false;
516 if (data->tapfds[pair_idx] >= 0) {
517 tapfd = data->tapfds[pair_idx];
518 if (vhost_kernel_tap_set_offload(tapfd, dev->features) == -1)
520 if (req_mq && vhost_kernel_tap_set_queue(tapfd, true) < 0) {
521 PMD_DRV_LOG(ERR, "fail to enable tap for vhost kernel");
527 if ((dev->features & (1ULL << VIRTIO_NET_F_MRG_RXBUF)) ||
528 (dev->features & (1ULL << VIRTIO_F_VERSION_1)))
529 hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
531 hdr_size = sizeof(struct virtio_net_hdr);
533 tapfd = vhost_kernel_open_tap(&dev->ifname, hdr_size, req_mq,
534 (char *)dev->mac_addr, dev->features);
536 PMD_DRV_LOG(ERR, "fail to open tap for vhost kernel");
540 data->tapfds[pair_idx] = tapfd;
543 if (vhost_kernel_set_backend(vhostfd, tapfd) < 0) {
544 PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
548 dev->qp_enabled[pair_idx] = true;
553 vhost_kernel_get_backend_features(uint64_t *features)
561 vhost_kernel_update_link_state(struct virtio_user_dev *dev __rte_unused)
563 /* Nothing to update (Maybe get TAP interface link state?) */
568 vhost_kernel_get_intr_fd(struct virtio_user_dev *dev __rte_unused)
570 /* No link state interrupt with Vhost-kernel */
574 struct virtio_user_backend_ops virtio_ops_kernel = {
575 .setup = vhost_kernel_setup,
576 .destroy = vhost_kernel_destroy,
577 .get_backend_features = vhost_kernel_get_backend_features,
578 .set_owner = vhost_kernel_set_owner,
579 .get_features = vhost_kernel_get_features,
580 .set_features = vhost_kernel_set_features,
581 .set_memory_table = vhost_kernel_set_memory_table,
582 .set_vring_num = vhost_kernel_set_vring_num,
583 .set_vring_base = vhost_kernel_set_vring_base,
584 .get_vring_base = vhost_kernel_get_vring_base,
585 .set_vring_call = vhost_kernel_set_vring_call,
586 .set_vring_kick = vhost_kernel_set_vring_kick,
587 .set_vring_addr = vhost_kernel_set_vring_addr,
588 .get_status = vhost_kernel_get_status,
589 .set_status = vhost_kernel_set_status,
590 .enable_qp = vhost_kernel_enable_queue_pair,
591 .update_link_state = vhost_kernel_update_link_state,
592 .get_intr_fd = vhost_kernel_get_intr_fd,