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 add_memseg_list(const struct rte_memseg_list *msl, void *arg)
76 struct vhost_memory_kernel *vm = arg;
77 struct vhost_memory_region *mr;
81 if (vm->nregions >= max_regions)
84 start_addr = msl->base_va;
85 len = msl->page_sz * msl->memseg_arr.len;
87 mr = &vm->regions[vm->nregions++];
89 mr->guest_phys_addr = (uint64_t)(uintptr_t)start_addr;
90 mr->userspace_addr = (uint64_t)(uintptr_t)start_addr;
91 mr->memory_size = len;
92 mr->mmap_offset = 0; /* flags_padding */
94 PMD_DRV_LOG(DEBUG, "index=%u addr=%p len=%" PRIu64,
95 vm->nregions - 1, start_addr, len);
100 /* By default, vhost kernel module allows 64 regions, but DPDK may
101 * have much more memory regions. Below function will treat each
102 * contiguous memory space reserved by DPDK as one region.
104 static struct vhost_memory_kernel *
105 prepare_vhost_memory_kernel(void)
107 struct vhost_memory_kernel *vm;
109 vm = malloc(sizeof(struct vhost_memory_kernel) +
111 sizeof(struct vhost_memory_region));
119 * The memory lock has already been taken by memory subsystem
120 * or virtio_user_start_device().
122 if (rte_memseg_list_walk_thread_unsafe(add_memseg_list, vm) < 0) {
130 /* with below features, vhost kernel does not need to do the checksum and TSO,
131 * these info will be passed to virtio_user through virtio net header.
133 #define VHOST_KERNEL_GUEST_OFFLOADS_MASK \
134 ((1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
135 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
136 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
137 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
138 (1ULL << VIRTIO_NET_F_GUEST_UFO))
140 /* with below features, when flows from virtio_user to vhost kernel
141 * (1) if flows goes up through the kernel networking stack, it does not need
142 * to verify checksum, which can save CPU cycles;
143 * (2) if flows goes through a Linux bridge and outside from an interface
144 * (kernel driver), checksum and TSO will be done by GSO in kernel or even
145 * offloaded into real physical device.
147 #define VHOST_KERNEL_HOST_OFFLOADS_MASK \
148 ((1ULL << VIRTIO_NET_F_HOST_TSO4) | \
149 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
150 (1ULL << VIRTIO_NET_F_CSUM))
153 tap_support_features(void)
156 unsigned int tap_features;
158 tapfd = open(PATH_NET_TUN, O_RDWR);
160 PMD_DRV_LOG(ERR, "fail to open %s: %s",
161 PATH_NET_TUN, strerror(errno));
165 if (ioctl(tapfd, TUNGETFEATURES, &tap_features) == -1) {
166 PMD_DRV_LOG(ERR, "TUNGETFEATURES failed: %s", strerror(errno));
176 vhost_kernel_ioctl(struct virtio_user_dev *dev,
177 enum vhost_user_request req,
183 struct vhost_memory_kernel *vm = NULL;
185 unsigned int queue_sel;
186 unsigned int features;
188 PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
190 req_kernel = vhost_req_user_to_kernel[req];
192 if (req_kernel == VHOST_SET_MEM_TABLE) {
193 vm = prepare_vhost_memory_kernel();
199 if (req_kernel == VHOST_SET_FEATURES) {
200 /* We don't need memory protection here */
201 *(uint64_t *)arg &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
203 /* VHOST kernel does not know about below flags */
204 *(uint64_t *)arg &= ~VHOST_KERNEL_GUEST_OFFLOADS_MASK;
205 *(uint64_t *)arg &= ~VHOST_KERNEL_HOST_OFFLOADS_MASK;
207 *(uint64_t *)arg &= ~(1ULL << VIRTIO_NET_F_MQ);
210 switch (req_kernel) {
211 case VHOST_SET_VRING_NUM:
212 case VHOST_SET_VRING_ADDR:
213 case VHOST_SET_VRING_BASE:
214 case VHOST_GET_VRING_BASE:
215 case VHOST_SET_VRING_KICK:
216 case VHOST_SET_VRING_CALL:
217 queue_sel = *(unsigned int *)arg;
218 vhostfd = dev->vhostfds[queue_sel / 2];
219 *(unsigned int *)arg = queue_sel % 2;
220 PMD_DRV_LOG(DEBUG, "vhostfd=%d, index=%u",
221 vhostfd, *(unsigned int *)arg);
227 for (i = 0; i < dev->max_queue_pairs; ++i) {
228 if (dev->vhostfds[i] < 0)
231 ret = ioctl(dev->vhostfds[i], req_kernel, arg);
236 ret = ioctl(vhostfd, req_kernel, arg);
239 if (!ret && req_kernel == VHOST_GET_FEATURES) {
240 features = tap_support_features();
241 /* with tap as the backend, all these features are supported
242 * but not claimed by vhost-net, so we add them back when
243 * reporting to upper layer.
245 if (features & IFF_VNET_HDR) {
246 *((uint64_t *)arg) |= VHOST_KERNEL_GUEST_OFFLOADS_MASK;
247 *((uint64_t *)arg) |= VHOST_KERNEL_HOST_OFFLOADS_MASK;
250 /* vhost_kernel will not declare this feature, but it does
251 * support multi-queue.
253 if (features & IFF_MULTI_QUEUE)
254 *(uint64_t *)arg |= (1ull << VIRTIO_NET_F_MQ);
261 PMD_DRV_LOG(ERR, "%s failed: %s",
262 vhost_msg_strings[req], strerror(errno));
268 * Set up environment to talk with a vhost kernel backend.
271 * - (-1) if fail to set up;
272 * - (>=0) if successful.
275 vhost_kernel_setup(struct virtio_user_dev *dev)
280 get_vhost_kernel_max_regions();
282 for (i = 0; i < dev->max_queue_pairs; ++i) {
283 vhostfd = open(dev->path, O_RDWR);
285 PMD_DRV_LOG(ERR, "fail to open %s, %s",
286 dev->path, strerror(errno));
290 dev->vhostfds[i] = vhostfd;
297 vhost_kernel_set_backend(int vhostfd, int tapfd)
299 struct vhost_vring_file f;
303 if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
304 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
310 if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
311 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
320 vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
327 int req_mq = (dev->max_queue_pairs > 1);
329 vhostfd = dev->vhostfds[pair_idx];
332 if (dev->tapfds[pair_idx] >= 0) {
333 close(dev->tapfds[pair_idx]);
334 dev->tapfds[pair_idx] = -1;
336 return vhost_kernel_set_backend(vhostfd, -1);
337 } else if (dev->tapfds[pair_idx] >= 0) {
341 if ((dev->features & (1ULL << VIRTIO_NET_F_MRG_RXBUF)) ||
342 (dev->features & (1ULL << VIRTIO_F_VERSION_1)))
343 hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
345 hdr_size = sizeof(struct virtio_net_hdr);
347 tapfd = vhost_kernel_open_tap(&dev->ifname, hdr_size, req_mq,
348 (char *)dev->mac_addr, dev->features);
350 PMD_DRV_LOG(ERR, "fail to open tap for vhost kernel");
354 if (vhost_kernel_set_backend(vhostfd, tapfd) < 0) {
355 PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
360 dev->tapfds[pair_idx] = tapfd;
364 struct virtio_user_backend_ops ops_kernel = {
365 .setup = vhost_kernel_setup,
366 .send_request = vhost_kernel_ioctl,
367 .enable_qp = vhost_kernel_enable_queue_pair