1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2020 Red Hat Inc.
11 #include <rte_memory.h>
14 #include "virtio_user_dev.h"
16 /* vhost kernel & vdpa ioctls */
17 #define VHOST_VIRTIO 0xAF
18 #define VHOST_GET_FEATURES _IOR(VHOST_VIRTIO, 0x00, __u64)
19 #define VHOST_SET_FEATURES _IOW(VHOST_VIRTIO, 0x00, __u64)
20 #define VHOST_SET_OWNER _IO(VHOST_VIRTIO, 0x01)
21 #define VHOST_RESET_OWNER _IO(VHOST_VIRTIO, 0x02)
22 #define VHOST_SET_LOG_BASE _IOW(VHOST_VIRTIO, 0x04, __u64)
23 #define VHOST_SET_LOG_FD _IOW(VHOST_VIRTIO, 0x07, int)
24 #define VHOST_SET_VRING_NUM _IOW(VHOST_VIRTIO, 0x10, struct vhost_vring_state)
25 #define VHOST_SET_VRING_ADDR _IOW(VHOST_VIRTIO, 0x11, struct vhost_vring_addr)
26 #define VHOST_SET_VRING_BASE _IOW(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
27 #define VHOST_GET_VRING_BASE _IOWR(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
28 #define VHOST_SET_VRING_KICK _IOW(VHOST_VIRTIO, 0x20, struct vhost_vring_file)
29 #define VHOST_SET_VRING_CALL _IOW(VHOST_VIRTIO, 0x21, struct vhost_vring_file)
30 #define VHOST_SET_VRING_ERR _IOW(VHOST_VIRTIO, 0x22, struct vhost_vring_file)
31 #define VHOST_NET_SET_BACKEND _IOW(VHOST_VIRTIO, 0x30, struct vhost_vring_file)
32 #define VHOST_VDPA_GET_DEVICE_ID _IOR(VHOST_VIRTIO, 0x70, __u32)
33 #define VHOST_VDPA_GET_STATUS _IOR(VHOST_VIRTIO, 0x71, __u8)
34 #define VHOST_VDPA_SET_STATUS _IOW(VHOST_VIRTIO, 0x72, __u8)
35 #define VHOST_VDPA_SET_VRING_ENABLE _IOW(VHOST_VIRTIO, 0x75, \
36 struct vhost_vring_state)
37 #define VHOST_SET_BACKEND_FEATURES _IOW(VHOST_VIRTIO, 0x25, __u64)
38 #define VHOST_GET_BACKEND_FEATURES _IOR(VHOST_VIRTIO, 0x26, __u64)
40 static uint64_t vhost_req_user_to_vdpa[] = {
41 [VHOST_USER_RESET_OWNER] = VHOST_RESET_OWNER,
42 [VHOST_USER_SET_VRING_CALL] = VHOST_SET_VRING_CALL,
43 [VHOST_USER_SET_VRING_NUM] = VHOST_SET_VRING_NUM,
44 [VHOST_USER_SET_VRING_BASE] = VHOST_SET_VRING_BASE,
45 [VHOST_USER_GET_VRING_BASE] = VHOST_GET_VRING_BASE,
46 [VHOST_USER_SET_VRING_ADDR] = VHOST_SET_VRING_ADDR,
47 [VHOST_USER_SET_VRING_KICK] = VHOST_SET_VRING_KICK,
48 [VHOST_USER_SET_STATUS] = VHOST_VDPA_SET_STATUS,
49 [VHOST_USER_GET_STATUS] = VHOST_VDPA_GET_STATUS,
50 [VHOST_USER_SET_VRING_ENABLE] = VHOST_VDPA_SET_VRING_ENABLE,
51 [VHOST_USER_GET_PROTOCOL_FEATURES] = VHOST_GET_BACKEND_FEATURES,
52 [VHOST_USER_SET_PROTOCOL_FEATURES] = VHOST_SET_BACKEND_FEATURES,
55 /* no alignment requirement */
56 struct vhost_iotlb_msg {
60 #define VHOST_ACCESS_RO 0x1
61 #define VHOST_ACCESS_WO 0x2
62 #define VHOST_ACCESS_RW 0x3
64 #define VHOST_IOTLB_MISS 1
65 #define VHOST_IOTLB_UPDATE 2
66 #define VHOST_IOTLB_INVALIDATE 3
67 #define VHOST_IOTLB_ACCESS_FAIL 4
68 #define VHOST_IOTLB_BATCH_BEGIN 5
69 #define VHOST_IOTLB_BATCH_END 6
73 #define VHOST_IOTLB_MSG_V2 0x2
79 struct vhost_iotlb_msg iotlb;
86 vhost_vdpa_ioctl(int fd, uint64_t request, void *arg)
90 ret = ioctl(fd, request, arg);
92 PMD_DRV_LOG(ERR, "Vhost-vDPA ioctl %"PRIu64" failed (%s)",
93 request, strerror(errno));
101 vhost_vdpa_set_owner(struct virtio_user_dev *dev)
103 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_OWNER, NULL);
107 vhost_vdpa_get_backend_features(struct virtio_user_dev *dev, uint64_t *features)
109 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_GET_BACKEND_FEATURES, features);
113 vhost_vdpa_set_backend_features(struct virtio_user_dev *dev, uint64_t features)
115 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_BACKEND_FEATURES, &features);
119 vhost_vdpa_get_features(struct virtio_user_dev *dev, uint64_t *features)
123 ret = vhost_vdpa_ioctl(dev->vhostfd, VHOST_GET_FEATURES, features);
125 PMD_DRV_LOG(ERR, "Failed to get features");
129 /* Multiqueue not supported for now */
130 *features &= ~(1ULL << VIRTIO_NET_F_MQ);
136 vhost_vdpa_set_features(struct virtio_user_dev *dev, uint64_t features)
139 features |= 1ULL << VIRTIO_F_IOMMU_PLATFORM;
141 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_FEATURES, &features);
145 vhost_vdpa_iotlb_batch_begin(struct virtio_user_dev *dev)
147 struct vhost_msg msg = {};
149 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_BATCH)))
152 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
153 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
157 msg.type = VHOST_IOTLB_MSG_V2;
158 msg.iotlb.type = VHOST_IOTLB_BATCH_BEGIN;
160 if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
161 PMD_DRV_LOG(ERR, "Failed to send IOTLB batch begin (%s)",
170 vhost_vdpa_iotlb_batch_end(struct virtio_user_dev *dev)
172 struct vhost_msg msg = {};
174 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_BATCH)))
177 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
178 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
182 msg.type = VHOST_IOTLB_MSG_V2;
183 msg.iotlb.type = VHOST_IOTLB_BATCH_END;
185 if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
186 PMD_DRV_LOG(ERR, "Failed to send IOTLB batch end (%s)",
195 vhost_vdpa_dma_map(struct virtio_user_dev *dev, void *addr,
196 uint64_t iova, size_t len)
198 struct vhost_msg msg = {};
200 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
201 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
205 msg.type = VHOST_IOTLB_MSG_V2;
206 msg.iotlb.type = VHOST_IOTLB_UPDATE;
207 msg.iotlb.iova = iova;
208 msg.iotlb.uaddr = (uint64_t)(uintptr_t)addr;
209 msg.iotlb.size = len;
210 msg.iotlb.perm = VHOST_ACCESS_RW;
212 PMD_DRV_LOG(DEBUG, "%s: iova: 0x%" PRIx64 ", addr: %p, len: 0x%zx",
213 __func__, iova, addr, len);
215 if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
216 PMD_DRV_LOG(ERR, "Failed to send IOTLB update (%s)",
225 vhost_vdpa_dma_unmap(struct virtio_user_dev *dev, __rte_unused void *addr,
226 uint64_t iova, size_t len)
228 struct vhost_msg msg = {};
230 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
231 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
235 msg.type = VHOST_IOTLB_MSG_V2;
236 msg.iotlb.type = VHOST_IOTLB_INVALIDATE;
237 msg.iotlb.iova = iova;
238 msg.iotlb.size = len;
240 PMD_DRV_LOG(DEBUG, "%s: iova: 0x%" PRIx64 ", len: 0x%zx",
241 __func__, iova, len);
243 if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
244 PMD_DRV_LOG(ERR, "Failed to send IOTLB invalidate (%s)",
253 vhost_vdpa_dma_map_batch(struct virtio_user_dev *dev, void *addr,
254 uint64_t iova, size_t len)
258 if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
261 ret = vhost_vdpa_dma_map(dev, addr, iova, len);
263 if (vhost_vdpa_iotlb_batch_end(dev) < 0)
270 vhost_vdpa_dma_unmap_batch(struct virtio_user_dev *dev, void *addr,
271 uint64_t iova, size_t len)
275 if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
278 ret = vhost_vdpa_dma_unmap(dev, addr, iova, len);
280 if (vhost_vdpa_iotlb_batch_end(dev) < 0)
287 vhost_vdpa_map_contig(const struct rte_memseg_list *msl,
288 const struct rte_memseg *ms, size_t len, void *arg)
290 struct virtio_user_dev *dev = arg;
295 return vhost_vdpa_dma_map(dev, ms->addr, ms->iova, len);
299 vhost_vdpa_map(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
302 struct virtio_user_dev *dev = arg;
304 /* skip external memory that isn't a heap */
305 if (msl->external && !msl->heap)
308 /* skip any segments with invalid IOVA addresses */
309 if (ms->iova == RTE_BAD_IOVA)
312 /* if IOVA mode is VA, we've already mapped the internal segments */
313 if (!msl->external && rte_eal_iova_mode() == RTE_IOVA_VA)
316 return vhost_vdpa_dma_map(dev, ms->addr, ms->iova, ms->len);
320 vhost_vdpa_set_memory_table(struct virtio_user_dev *dev)
324 if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
327 vhost_vdpa_dma_unmap(dev, NULL, 0, SIZE_MAX);
329 if (rte_eal_iova_mode() == RTE_IOVA_VA) {
330 /* with IOVA as VA mode, we can get away with mapping contiguous
331 * chunks rather than going page-by-page.
333 ret = rte_memseg_contig_walk_thread_unsafe(
334 vhost_vdpa_map_contig, dev);
337 /* we have to continue the walk because we've skipped the
338 * external segments during the config walk.
341 ret = rte_memseg_walk_thread_unsafe(vhost_vdpa_map, dev);
344 if (vhost_vdpa_iotlb_batch_end(dev) < 0)
350 /* with below features, vhost vdpa does not need to do the checksum and TSO,
351 * these info will be passed to virtio_user through virtio net header.
353 #define VHOST_VDPA_GUEST_OFFLOADS_MASK \
354 ((1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
355 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
356 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
357 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
358 (1ULL << VIRTIO_NET_F_GUEST_UFO))
360 #define VHOST_VDPA_HOST_OFFLOADS_MASK \
361 ((1ULL << VIRTIO_NET_F_HOST_TSO4) | \
362 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
363 (1ULL << VIRTIO_NET_F_CSUM))
366 vhost_vdpa_send_request(struct virtio_user_dev *dev,
367 enum vhost_user_request req,
373 PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
375 req_vdpa = vhost_req_user_to_vdpa[req];
378 case VHOST_SET_VRING_NUM:
379 case VHOST_SET_VRING_ADDR:
380 case VHOST_SET_VRING_BASE:
381 case VHOST_GET_VRING_BASE:
382 case VHOST_SET_VRING_KICK:
383 case VHOST_SET_VRING_CALL:
384 PMD_DRV_LOG(DEBUG, "vhostfd=%d, index=%u",
385 dev->vhostfd, *(unsigned int *)arg);
391 ret = ioctl(dev->vhostfd, req_vdpa, arg);
393 PMD_DRV_LOG(ERR, "%s failed: %s",
394 vhost_msg_strings[req], strerror(errno));
400 * Set up environment to talk with a vhost vdpa backend.
403 * - (-1) if fail to set up;
404 * - (>=0) if successful.
407 vhost_vdpa_setup(struct virtio_user_dev *dev)
409 uint32_t did = (uint32_t)-1;
411 dev->vhostfd = open(dev->path, O_RDWR);
412 if (dev->vhostfd < 0) {
413 PMD_DRV_LOG(ERR, "Failed to open %s: %s\n",
414 dev->path, strerror(errno));
418 if (ioctl(dev->vhostfd, VHOST_VDPA_GET_DEVICE_ID, &did) < 0 ||
419 did != VIRTIO_ID_NETWORK) {
420 PMD_DRV_LOG(ERR, "Invalid vdpa device ID: %u\n", did);
428 vhost_vdpa_enable_queue_pair(struct virtio_user_dev *dev,
434 if (dev->qp_enabled[pair_idx] == enable)
437 for (i = 0; i < 2; ++i) {
438 struct vhost_vring_state state = {
439 .index = pair_idx * 2 + i,
443 if (vhost_vdpa_send_request(dev, VHOST_USER_SET_VRING_ENABLE, &state))
447 dev->qp_enabled[pair_idx] = enable;
452 struct virtio_user_backend_ops virtio_ops_vdpa = {
453 .setup = vhost_vdpa_setup,
454 .set_owner = vhost_vdpa_set_owner,
455 .get_features = vhost_vdpa_get_features,
456 .set_features = vhost_vdpa_set_features,
457 .get_protocol_features = vhost_vdpa_get_backend_features,
458 .set_protocol_features = vhost_vdpa_set_backend_features,
459 .set_memory_table = vhost_vdpa_set_memory_table,
460 .send_request = vhost_vdpa_send_request,
461 .enable_qp = vhost_vdpa_enable_queue_pair,
462 .dma_map = vhost_vdpa_dma_map_batch,
463 .dma_unmap = vhost_vdpa_dma_unmap_batch,