1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016 Intel Corporation
10 #include <rte_memory.h>
11 #include <rte_eal_memconfig.h>
14 #include "virtio_user_dev.h"
15 #include "vhost_kernel_tap.h"
17 struct vhost_memory_kernel {
20 struct vhost_memory_region regions[0];
23 /* vhost kernel ioctls */
24 #define VHOST_VIRTIO 0xAF
25 #define VHOST_GET_FEATURES _IOR(VHOST_VIRTIO, 0x00, __u64)
26 #define VHOST_SET_FEATURES _IOW(VHOST_VIRTIO, 0x00, __u64)
27 #define VHOST_SET_OWNER _IO(VHOST_VIRTIO, 0x01)
28 #define VHOST_RESET_OWNER _IO(VHOST_VIRTIO, 0x02)
29 #define VHOST_SET_MEM_TABLE _IOW(VHOST_VIRTIO, 0x03, struct vhost_memory_kernel)
30 #define VHOST_SET_LOG_BASE _IOW(VHOST_VIRTIO, 0x04, __u64)
31 #define VHOST_SET_LOG_FD _IOW(VHOST_VIRTIO, 0x07, int)
32 #define VHOST_SET_VRING_NUM _IOW(VHOST_VIRTIO, 0x10, struct vhost_vring_state)
33 #define VHOST_SET_VRING_ADDR _IOW(VHOST_VIRTIO, 0x11, struct vhost_vring_addr)
34 #define VHOST_SET_VRING_BASE _IOW(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
35 #define VHOST_GET_VRING_BASE _IOWR(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
36 #define VHOST_SET_VRING_KICK _IOW(VHOST_VIRTIO, 0x20, struct vhost_vring_file)
37 #define VHOST_SET_VRING_CALL _IOW(VHOST_VIRTIO, 0x21, struct vhost_vring_file)
38 #define VHOST_SET_VRING_ERR _IOW(VHOST_VIRTIO, 0x22, struct vhost_vring_file)
39 #define VHOST_NET_SET_BACKEND _IOW(VHOST_VIRTIO, 0x30, struct vhost_vring_file)
41 static uint64_t max_regions = 64;
44 get_vhost_kernel_max_regions(void)
47 char buf[20] = {'\0'};
49 fd = open("/sys/module/vhost/parameters/max_mem_regions", O_RDONLY);
53 if (read(fd, buf, sizeof(buf) - 1) > 0)
54 max_regions = strtoull(buf, NULL, 10);
59 static uint64_t vhost_req_user_to_kernel[] = {
60 [VHOST_USER_SET_OWNER] = VHOST_SET_OWNER,
61 [VHOST_USER_RESET_OWNER] = VHOST_RESET_OWNER,
62 [VHOST_USER_SET_FEATURES] = VHOST_SET_FEATURES,
63 [VHOST_USER_GET_FEATURES] = VHOST_GET_FEATURES,
64 [VHOST_USER_SET_VRING_CALL] = VHOST_SET_VRING_CALL,
65 [VHOST_USER_SET_VRING_NUM] = VHOST_SET_VRING_NUM,
66 [VHOST_USER_SET_VRING_BASE] = VHOST_SET_VRING_BASE,
67 [VHOST_USER_GET_VRING_BASE] = VHOST_GET_VRING_BASE,
68 [VHOST_USER_SET_VRING_ADDR] = VHOST_SET_VRING_ADDR,
69 [VHOST_USER_SET_VRING_KICK] = VHOST_SET_VRING_KICK,
70 [VHOST_USER_SET_MEM_TABLE] = VHOST_SET_MEM_TABLE,
74 struct vhost_memory_kernel *vm;
78 add_memory_region(const struct rte_memseg_list *msl __rte_unused,
79 const struct rte_memseg *ms, size_t len, void *arg)
81 struct walk_arg *wa = arg;
82 struct vhost_memory_region *mr;
85 if (wa->region_nr >= max_regions)
88 mr = &wa->vm->regions[wa->region_nr++];
89 start_addr = ms->addr;
91 mr->guest_phys_addr = (uint64_t)(uintptr_t)start_addr;
92 mr->userspace_addr = (uint64_t)(uintptr_t)start_addr;
93 mr->memory_size = len;
99 /* By default, vhost kernel module allows 64 regions, but DPDK allows
100 * 256 segments. As a relief, below function merges those virtually
101 * adjacent memsegs into one region.
103 static struct vhost_memory_kernel *
104 prepare_vhost_memory_kernel(void)
106 struct vhost_memory_kernel *vm;
109 vm = malloc(sizeof(struct vhost_memory_kernel) +
111 sizeof(struct vhost_memory_region));
118 if (rte_memseg_contig_walk(add_memory_region, &wa) < 0) {
123 vm->nregions = wa.region_nr;
128 /* with below features, vhost kernel does not need to do the checksum and TSO,
129 * these info will be passed to virtio_user through virtio net header.
131 #define VHOST_KERNEL_GUEST_OFFLOADS_MASK \
132 ((1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
133 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
134 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
135 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
136 (1ULL << VIRTIO_NET_F_GUEST_UFO))
138 /* with below features, when flows from virtio_user to vhost kernel
139 * (1) if flows goes up through the kernel networking stack, it does not need
140 * to verify checksum, which can save CPU cycles;
141 * (2) if flows goes through a Linux bridge and outside from an interface
142 * (kernel driver), checksum and TSO will be done by GSO in kernel or even
143 * offloaded into real physical device.
145 #define VHOST_KERNEL_HOST_OFFLOADS_MASK \
146 ((1ULL << VIRTIO_NET_F_HOST_TSO4) | \
147 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
148 (1ULL << VIRTIO_NET_F_CSUM))
151 tap_supporte_mq(void)
154 unsigned int tap_features;
156 tapfd = open(PATH_NET_TUN, O_RDWR);
158 PMD_DRV_LOG(ERR, "fail to open %s: %s",
159 PATH_NET_TUN, strerror(errno));
163 if (ioctl(tapfd, TUNGETFEATURES, &tap_features) == -1) {
164 PMD_DRV_LOG(ERR, "TUNGETFEATURES failed: %s", strerror(errno));
170 return tap_features & IFF_MULTI_QUEUE;
174 vhost_kernel_ioctl(struct virtio_user_dev *dev,
175 enum vhost_user_request req,
181 struct vhost_memory_kernel *vm = NULL;
183 unsigned int queue_sel;
185 PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
187 req_kernel = vhost_req_user_to_kernel[req];
189 if (req_kernel == VHOST_SET_MEM_TABLE) {
190 vm = prepare_vhost_memory_kernel();
196 if (req_kernel == VHOST_SET_FEATURES) {
197 /* We don't need memory protection here */
198 *(uint64_t *)arg &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
200 /* VHOST kernel does not know about below flags */
201 *(uint64_t *)arg &= ~VHOST_KERNEL_GUEST_OFFLOADS_MASK;
202 *(uint64_t *)arg &= ~VHOST_KERNEL_HOST_OFFLOADS_MASK;
204 *(uint64_t *)arg &= ~(1ULL << VIRTIO_NET_F_MQ);
207 switch (req_kernel) {
208 case VHOST_SET_VRING_NUM:
209 case VHOST_SET_VRING_ADDR:
210 case VHOST_SET_VRING_BASE:
211 case VHOST_GET_VRING_BASE:
212 case VHOST_SET_VRING_KICK:
213 case VHOST_SET_VRING_CALL:
214 queue_sel = *(unsigned int *)arg;
215 vhostfd = dev->vhostfds[queue_sel / 2];
216 *(unsigned int *)arg = queue_sel % 2;
217 PMD_DRV_LOG(DEBUG, "vhostfd=%d, index=%u",
218 vhostfd, *(unsigned int *)arg);
224 for (i = 0; i < dev->max_queue_pairs; ++i) {
225 if (dev->vhostfds[i] < 0)
228 ret = ioctl(dev->vhostfds[i], req_kernel, arg);
233 ret = ioctl(vhostfd, req_kernel, arg);
236 if (!ret && req_kernel == VHOST_GET_FEATURES) {
237 /* with tap as the backend, all these features are supported
238 * but not claimed by vhost-net, so we add them back when
239 * reporting to upper layer.
241 *((uint64_t *)arg) |= VHOST_KERNEL_GUEST_OFFLOADS_MASK;
242 *((uint64_t *)arg) |= VHOST_KERNEL_HOST_OFFLOADS_MASK;
244 /* vhost_kernel will not declare this feature, but it does
245 * support multi-queue.
247 if (tap_supporte_mq())
248 *(uint64_t *)arg |= (1ull << VIRTIO_NET_F_MQ);
255 PMD_DRV_LOG(ERR, "%s failed: %s",
256 vhost_msg_strings[req], strerror(errno));
262 * Set up environment to talk with a vhost kernel backend.
265 * - (-1) if fail to set up;
266 * - (>=0) if successful.
269 vhost_kernel_setup(struct virtio_user_dev *dev)
274 get_vhost_kernel_max_regions();
276 for (i = 0; i < dev->max_queue_pairs; ++i) {
277 vhostfd = open(dev->path, O_RDWR);
279 PMD_DRV_LOG(ERR, "fail to open %s, %s",
280 dev->path, strerror(errno));
284 dev->vhostfds[i] = vhostfd;
291 vhost_kernel_set_backend(int vhostfd, int tapfd)
293 struct vhost_vring_file f;
297 if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
298 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
304 if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
305 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
314 vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
321 int req_mq = (dev->max_queue_pairs > 1);
323 vhostfd = dev->vhostfds[pair_idx];
326 if (dev->tapfds[pair_idx] >= 0) {
327 close(dev->tapfds[pair_idx]);
328 dev->tapfds[pair_idx] = -1;
330 return vhost_kernel_set_backend(vhostfd, -1);
331 } else if (dev->tapfds[pair_idx] >= 0) {
335 if ((dev->features & (1ULL << VIRTIO_NET_F_MRG_RXBUF)) ||
336 (dev->features & (1ULL << VIRTIO_F_VERSION_1)))
337 hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
339 hdr_size = sizeof(struct virtio_net_hdr);
341 tapfd = vhost_kernel_open_tap(&dev->ifname, hdr_size, req_mq,
342 (char *)dev->mac_addr);
344 PMD_DRV_LOG(ERR, "fail to open tap for vhost kernel");
348 if (vhost_kernel_set_backend(vhostfd, tapfd) < 0) {
349 PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
354 dev->tapfds[pair_idx] = tapfd;
358 struct virtio_user_backend_ops ops_kernel = {
359 .setup = vhost_kernel_setup,
360 .send_request = vhost_kernel_ioctl,
361 .enable_qp = vhost_kernel_enable_queue_pair