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, struct vhost_vring_state)
36 #define VHOST_SET_BACKEND_FEATURES _IOW(VHOST_VIRTIO, 0x25, __u64)
37 #define VHOST_GET_BACKEND_FEATURES _IOR(VHOST_VIRTIO, 0x26, __u64)
39 static uint64_t vhost_req_user_to_vdpa[] = {
40 [VHOST_USER_RESET_OWNER] = VHOST_RESET_OWNER,
41 [VHOST_USER_SET_VRING_CALL] = VHOST_SET_VRING_CALL,
42 [VHOST_USER_SET_VRING_ADDR] = VHOST_SET_VRING_ADDR,
43 [VHOST_USER_SET_VRING_KICK] = VHOST_SET_VRING_KICK,
44 [VHOST_USER_SET_STATUS] = VHOST_VDPA_SET_STATUS,
45 [VHOST_USER_GET_STATUS] = VHOST_VDPA_GET_STATUS,
48 /* no alignment requirement */
49 struct vhost_iotlb_msg {
53 #define VHOST_ACCESS_RO 0x1
54 #define VHOST_ACCESS_WO 0x2
55 #define VHOST_ACCESS_RW 0x3
57 #define VHOST_IOTLB_MISS 1
58 #define VHOST_IOTLB_UPDATE 2
59 #define VHOST_IOTLB_INVALIDATE 3
60 #define VHOST_IOTLB_ACCESS_FAIL 4
61 #define VHOST_IOTLB_BATCH_BEGIN 5
62 #define VHOST_IOTLB_BATCH_END 6
66 #define VHOST_IOTLB_MSG_V2 0x2
72 struct vhost_iotlb_msg iotlb;
79 vhost_vdpa_ioctl(int fd, uint64_t request, void *arg)
83 ret = ioctl(fd, request, arg);
85 PMD_DRV_LOG(ERR, "Vhost-vDPA ioctl %"PRIu64" failed (%s)",
86 request, strerror(errno));
94 vhost_vdpa_set_owner(struct virtio_user_dev *dev)
96 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_OWNER, NULL);
100 vhost_vdpa_get_backend_features(struct virtio_user_dev *dev, uint64_t *features)
102 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_GET_BACKEND_FEATURES, features);
106 vhost_vdpa_set_backend_features(struct virtio_user_dev *dev, uint64_t features)
108 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_BACKEND_FEATURES, &features);
112 vhost_vdpa_get_features(struct virtio_user_dev *dev, uint64_t *features)
116 ret = vhost_vdpa_ioctl(dev->vhostfd, VHOST_GET_FEATURES, features);
118 PMD_DRV_LOG(ERR, "Failed to get features");
122 /* Multiqueue not supported for now */
123 *features &= ~(1ULL << VIRTIO_NET_F_MQ);
129 vhost_vdpa_set_features(struct virtio_user_dev *dev, uint64_t features)
132 features |= 1ULL << VIRTIO_F_IOMMU_PLATFORM;
134 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_FEATURES, &features);
138 vhost_vdpa_iotlb_batch_begin(struct virtio_user_dev *dev)
140 struct vhost_msg msg = {};
142 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_BATCH)))
145 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
146 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
150 msg.type = VHOST_IOTLB_MSG_V2;
151 msg.iotlb.type = VHOST_IOTLB_BATCH_BEGIN;
153 if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
154 PMD_DRV_LOG(ERR, "Failed to send IOTLB batch begin (%s)",
163 vhost_vdpa_iotlb_batch_end(struct virtio_user_dev *dev)
165 struct vhost_msg msg = {};
167 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_BATCH)))
170 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
171 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
175 msg.type = VHOST_IOTLB_MSG_V2;
176 msg.iotlb.type = VHOST_IOTLB_BATCH_END;
178 if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
179 PMD_DRV_LOG(ERR, "Failed to send IOTLB batch end (%s)",
188 vhost_vdpa_dma_map(struct virtio_user_dev *dev, void *addr,
189 uint64_t iova, size_t len)
191 struct vhost_msg msg = {};
193 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
194 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
198 msg.type = VHOST_IOTLB_MSG_V2;
199 msg.iotlb.type = VHOST_IOTLB_UPDATE;
200 msg.iotlb.iova = iova;
201 msg.iotlb.uaddr = (uint64_t)(uintptr_t)addr;
202 msg.iotlb.size = len;
203 msg.iotlb.perm = VHOST_ACCESS_RW;
205 PMD_DRV_LOG(DEBUG, "%s: iova: 0x%" PRIx64 ", addr: %p, len: 0x%zx",
206 __func__, iova, addr, len);
208 if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
209 PMD_DRV_LOG(ERR, "Failed to send IOTLB update (%s)",
218 vhost_vdpa_dma_unmap(struct virtio_user_dev *dev, __rte_unused void *addr,
219 uint64_t iova, size_t len)
221 struct vhost_msg msg = {};
223 if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
224 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
228 msg.type = VHOST_IOTLB_MSG_V2;
229 msg.iotlb.type = VHOST_IOTLB_INVALIDATE;
230 msg.iotlb.iova = iova;
231 msg.iotlb.size = len;
233 PMD_DRV_LOG(DEBUG, "%s: iova: 0x%" PRIx64 ", len: 0x%zx",
234 __func__, iova, len);
236 if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
237 PMD_DRV_LOG(ERR, "Failed to send IOTLB invalidate (%s)",
246 vhost_vdpa_dma_map_batch(struct virtio_user_dev *dev, void *addr,
247 uint64_t iova, size_t len)
251 if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
254 ret = vhost_vdpa_dma_map(dev, addr, iova, len);
256 if (vhost_vdpa_iotlb_batch_end(dev) < 0)
263 vhost_vdpa_dma_unmap_batch(struct virtio_user_dev *dev, void *addr,
264 uint64_t iova, size_t len)
268 if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
271 ret = vhost_vdpa_dma_unmap(dev, addr, iova, len);
273 if (vhost_vdpa_iotlb_batch_end(dev) < 0)
280 vhost_vdpa_map_contig(const struct rte_memseg_list *msl,
281 const struct rte_memseg *ms, size_t len, void *arg)
283 struct virtio_user_dev *dev = arg;
288 return vhost_vdpa_dma_map(dev, ms->addr, ms->iova, len);
292 vhost_vdpa_map(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
295 struct virtio_user_dev *dev = arg;
297 /* skip external memory that isn't a heap */
298 if (msl->external && !msl->heap)
301 /* skip any segments with invalid IOVA addresses */
302 if (ms->iova == RTE_BAD_IOVA)
305 /* if IOVA mode is VA, we've already mapped the internal segments */
306 if (!msl->external && rte_eal_iova_mode() == RTE_IOVA_VA)
309 return vhost_vdpa_dma_map(dev, ms->addr, ms->iova, ms->len);
313 vhost_vdpa_set_memory_table(struct virtio_user_dev *dev)
317 if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
320 vhost_vdpa_dma_unmap(dev, NULL, 0, SIZE_MAX);
322 if (rte_eal_iova_mode() == RTE_IOVA_VA) {
323 /* with IOVA as VA mode, we can get away with mapping contiguous
324 * chunks rather than going page-by-page.
326 ret = rte_memseg_contig_walk_thread_unsafe(
327 vhost_vdpa_map_contig, dev);
330 /* we have to continue the walk because we've skipped the
331 * external segments during the config walk.
334 ret = rte_memseg_walk_thread_unsafe(vhost_vdpa_map, dev);
337 if (vhost_vdpa_iotlb_batch_end(dev) < 0)
344 vhost_vdpa_set_vring_enable(struct virtio_user_dev *dev, struct vhost_vring_state *state)
346 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_VDPA_SET_VRING_ENABLE, state);
350 vhost_vdpa_set_vring_num(struct virtio_user_dev *dev, struct vhost_vring_state *state)
352 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_VRING_NUM, state);
356 vhost_vdpa_set_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
358 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_VRING_BASE, state);
362 vhost_vdpa_get_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
364 return vhost_vdpa_ioctl(dev->vhostfd, VHOST_GET_VRING_BASE, state);
367 /* with below features, vhost vdpa does not need to do the checksum and TSO,
368 * these info will be passed to virtio_user through virtio net header.
370 #define VHOST_VDPA_GUEST_OFFLOADS_MASK \
371 ((1ULL << VIRTIO_NET_F_GUEST_CSUM) | \
372 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
373 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
374 (1ULL << VIRTIO_NET_F_GUEST_ECN) | \
375 (1ULL << VIRTIO_NET_F_GUEST_UFO))
377 #define VHOST_VDPA_HOST_OFFLOADS_MASK \
378 ((1ULL << VIRTIO_NET_F_HOST_TSO4) | \
379 (1ULL << VIRTIO_NET_F_HOST_TSO6) | \
380 (1ULL << VIRTIO_NET_F_CSUM))
383 vhost_vdpa_send_request(struct virtio_user_dev *dev,
384 enum vhost_user_request req,
390 PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
392 req_vdpa = vhost_req_user_to_vdpa[req];
395 case VHOST_SET_VRING_ADDR:
396 case VHOST_SET_VRING_KICK:
397 case VHOST_SET_VRING_CALL:
398 PMD_DRV_LOG(DEBUG, "vhostfd=%d, index=%u",
399 dev->vhostfd, *(unsigned int *)arg);
405 ret = ioctl(dev->vhostfd, req_vdpa, arg);
407 PMD_DRV_LOG(ERR, "%s failed: %s",
408 vhost_msg_strings[req], strerror(errno));
414 * Set up environment to talk with a vhost vdpa backend.
417 * - (-1) if fail to set up;
418 * - (>=0) if successful.
421 vhost_vdpa_setup(struct virtio_user_dev *dev)
423 uint32_t did = (uint32_t)-1;
425 dev->vhostfd = open(dev->path, O_RDWR);
426 if (dev->vhostfd < 0) {
427 PMD_DRV_LOG(ERR, "Failed to open %s: %s\n",
428 dev->path, strerror(errno));
432 if (ioctl(dev->vhostfd, VHOST_VDPA_GET_DEVICE_ID, &did) < 0 ||
433 did != VIRTIO_ID_NETWORK) {
434 PMD_DRV_LOG(ERR, "Invalid vdpa device ID: %u\n", did);
442 vhost_vdpa_enable_queue_pair(struct virtio_user_dev *dev,
448 if (dev->qp_enabled[pair_idx] == enable)
451 for (i = 0; i < 2; ++i) {
452 struct vhost_vring_state state = {
453 .index = pair_idx * 2 + i,
457 if (vhost_vdpa_set_vring_enable(dev, &state))
461 dev->qp_enabled[pair_idx] = enable;
466 struct virtio_user_backend_ops virtio_ops_vdpa = {
467 .setup = vhost_vdpa_setup,
468 .set_owner = vhost_vdpa_set_owner,
469 .get_features = vhost_vdpa_get_features,
470 .set_features = vhost_vdpa_set_features,
471 .get_protocol_features = vhost_vdpa_get_backend_features,
472 .set_protocol_features = vhost_vdpa_set_backend_features,
473 .set_memory_table = vhost_vdpa_set_memory_table,
474 .set_vring_num = vhost_vdpa_set_vring_num,
475 .set_vring_base = vhost_vdpa_set_vring_base,
476 .get_vring_base = vhost_vdpa_get_vring_base,
477 .send_request = vhost_vdpa_send_request,
478 .enable_qp = vhost_vdpa_enable_queue_pair,
479 .dma_map = vhost_vdpa_dma_map_batch,
480 .dma_unmap = vhost_vdpa_dma_unmap_batch,