drivers/net/virtio/virtio_user/vhost_vdpa.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2020 Red Hat Inc.
   3  */
   4
   5 #include <sys/ioctl.h>
   6 #include <sys/types.h>
   7 #include <sys/stat.h>
   8 #include <fcntl.h>
   9 #include <unistd.h>
  10
  11 #include <rte_memory.h>
  12
  13 #include "vhost.h"
  14 #include "virtio_user_dev.h"
  15
  16 #define VHOST_VDPA_SUPPORTED_BACKEND_FEATURES           \
  17         (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2   |       \
  18         1ULL << VHOST_BACKEND_F_IOTLB_BATCH)
  19
  20 /* vhost kernel & vdpa ioctls */
  21 #define VHOST_VIRTIO 0xAF
  22 #define VHOST_GET_FEATURES _IOR(VHOST_VIRTIO, 0x00, __u64)
  23 #define VHOST_SET_FEATURES _IOW(VHOST_VIRTIO, 0x00, __u64)
  24 #define VHOST_SET_OWNER _IO(VHOST_VIRTIO, 0x01)
  25 #define VHOST_RESET_OWNER _IO(VHOST_VIRTIO, 0x02)
  26 #define VHOST_SET_LOG_BASE _IOW(VHOST_VIRTIO, 0x04, __u64)
  27 #define VHOST_SET_LOG_FD _IOW(VHOST_VIRTIO, 0x07, int)
  28 #define VHOST_SET_VRING_NUM _IOW(VHOST_VIRTIO, 0x10, struct vhost_vring_state)
  29 #define VHOST_SET_VRING_ADDR _IOW(VHOST_VIRTIO, 0x11, struct vhost_vring_addr)
  30 #define VHOST_SET_VRING_BASE _IOW(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
  31 #define VHOST_GET_VRING_BASE _IOWR(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
  32 #define VHOST_SET_VRING_KICK _IOW(VHOST_VIRTIO, 0x20, struct vhost_vring_file)
  33 #define VHOST_SET_VRING_CALL _IOW(VHOST_VIRTIO, 0x21, struct vhost_vring_file)
  34 #define VHOST_SET_VRING_ERR _IOW(VHOST_VIRTIO, 0x22, struct vhost_vring_file)
  35 #define VHOST_NET_SET_BACKEND _IOW(VHOST_VIRTIO, 0x30, struct vhost_vring_file)
  36 #define VHOST_VDPA_GET_DEVICE_ID _IOR(VHOST_VIRTIO, 0x70, __u32)
  37 #define VHOST_VDPA_GET_STATUS _IOR(VHOST_VIRTIO, 0x71, __u8)
  38 #define VHOST_VDPA_SET_STATUS _IOW(VHOST_VIRTIO, 0x72, __u8)
  39 #define VHOST_VDPA_SET_VRING_ENABLE _IOW(VHOST_VIRTIO, 0x75, struct vhost_vring_state)
  40 #define VHOST_SET_BACKEND_FEATURES _IOW(VHOST_VIRTIO, 0x25, __u64)
  41 #define VHOST_GET_BACKEND_FEATURES _IOR(VHOST_VIRTIO, 0x26, __u64)
  42
  43 /* no alignment requirement */
  44 struct vhost_iotlb_msg {
  45         uint64_t iova;
  46         uint64_t size;
  47         uint64_t uaddr;
  48 #define VHOST_ACCESS_RO      0x1
  49 #define VHOST_ACCESS_WO      0x2
  50 #define VHOST_ACCESS_RW      0x3
  51         uint8_t perm;
  52 #define VHOST_IOTLB_MISS           1
  53 #define VHOST_IOTLB_UPDATE         2
  54 #define VHOST_IOTLB_INVALIDATE     3
  55 #define VHOST_IOTLB_ACCESS_FAIL    4
  56 #define VHOST_IOTLB_BATCH_BEGIN    5
  57 #define VHOST_IOTLB_BATCH_END      6
  58         uint8_t type;
  59 };
  60
  61 #define VHOST_IOTLB_MSG_V2 0x2
  62
  63 struct vhost_msg {
  64         uint32_t type;
  65         uint32_t reserved;
  66         union {
  67                 struct vhost_iotlb_msg iotlb;
  68                 uint8_t padding[64];
  69         };
  70 };
  71
  72
  73 static int
  74 vhost_vdpa_ioctl(int fd, uint64_t request, void *arg)
  75 {
  76         int ret;
  77
  78         ret = ioctl(fd, request, arg);
  79         if (ret) {
  80                 PMD_DRV_LOG(ERR, "Vhost-vDPA ioctl %"PRIu64" failed (%s)",
  81                                 request, strerror(errno));
  82                 return -1;
  83         }
  84
  85         return 0;
  86 }
  87
  88 static int
  89 vhost_vdpa_set_owner(struct virtio_user_dev *dev)
  90 {
  91         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_OWNER, NULL);
  92 }
  93
  94 static int
  95 vhost_vdpa_get_protocol_features(struct virtio_user_dev *dev, uint64_t *features)
  96 {
  97         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_GET_BACKEND_FEATURES, features);
  98 }
  99
 100 static int
 101 vhost_vdpa_set_protocol_features(struct virtio_user_dev *dev, uint64_t features)
 102 {
 103         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_BACKEND_FEATURES, &features);
 104 }
 105
 106 static int
 107 vhost_vdpa_get_features(struct virtio_user_dev *dev, uint64_t *features)
 108 {
 109         int ret;
 110
 111         ret = vhost_vdpa_ioctl(dev->vhostfd, VHOST_GET_FEATURES, features);
 112         if (ret) {
 113                 PMD_DRV_LOG(ERR, "Failed to get features");
 114                 return -1;
 115         }
 116
 117         /* Multiqueue not supported for now */
 118         *features &= ~(1ULL << VIRTIO_NET_F_MQ);
 119
 120         /* Negotiated vDPA backend features */
 121         ret = vhost_vdpa_get_protocol_features(dev, &dev->protocol_features);
 122         if (ret < 0) {
 123                 PMD_DRV_LOG(ERR, "Failed to get backend features");
 124                 return -1;
 125         }
 126
 127         dev->protocol_features &= VHOST_VDPA_SUPPORTED_BACKEND_FEATURES;
 128
 129         ret = vhost_vdpa_set_protocol_features(dev, dev->protocol_features);
 130         if (ret < 0) {
 131                 PMD_DRV_LOG(ERR, "Failed to set backend features");
 132                 return -1;
 133         }
 134
 135         return 0;
 136 }
 137
 138 static int
 139 vhost_vdpa_set_features(struct virtio_user_dev *dev, uint64_t features)
 140 {
 141         /* WORKAROUND */
 142         features |= 1ULL << VIRTIO_F_IOMMU_PLATFORM;
 143
 144         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_FEATURES, &features);
 145 }
 146
 147 static int
 148 vhost_vdpa_iotlb_batch_begin(struct virtio_user_dev *dev)
 149 {
 150         struct vhost_msg msg = {};
 151
 152         if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_BATCH)))
 153                 return 0;
 154
 155         if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
 156                 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
 157                 return -1;
 158         }
 159
 160         msg.type = VHOST_IOTLB_MSG_V2;
 161         msg.iotlb.type = VHOST_IOTLB_BATCH_BEGIN;
 162
 163         if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
 164                 PMD_DRV_LOG(ERR, "Failed to send IOTLB batch begin (%s)",
 165                                 strerror(errno));
 166                 return -1;
 167         }
 168
 169         return 0;
 170 }
 171
 172 static int
 173 vhost_vdpa_iotlb_batch_end(struct virtio_user_dev *dev)
 174 {
 175         struct vhost_msg msg = {};
 176
 177         if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_BATCH)))
 178                 return 0;
 179
 180         if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
 181                 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
 182                 return -1;
 183         }
 184
 185         msg.type = VHOST_IOTLB_MSG_V2;
 186         msg.iotlb.type = VHOST_IOTLB_BATCH_END;
 187
 188         if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
 189                 PMD_DRV_LOG(ERR, "Failed to send IOTLB batch end (%s)",
 190                                 strerror(errno));
 191                 return -1;
 192         }
 193
 194         return 0;
 195 }
 196
 197 static int
 198 vhost_vdpa_dma_map(struct virtio_user_dev *dev, void *addr,
 199                                   uint64_t iova, size_t len)
 200 {
 201         struct vhost_msg msg = {};
 202
 203         if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
 204                 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
 205                 return -1;
 206         }
 207
 208         msg.type = VHOST_IOTLB_MSG_V2;
 209         msg.iotlb.type = VHOST_IOTLB_UPDATE;
 210         msg.iotlb.iova = iova;
 211         msg.iotlb.uaddr = (uint64_t)(uintptr_t)addr;
 212         msg.iotlb.size = len;
 213         msg.iotlb.perm = VHOST_ACCESS_RW;
 214
 215         PMD_DRV_LOG(DEBUG, "%s: iova: 0x%" PRIx64 ", addr: %p, len: 0x%zx",
 216                         __func__, iova, addr, len);
 217
 218         if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
 219                 PMD_DRV_LOG(ERR, "Failed to send IOTLB update (%s)",
 220                                 strerror(errno));
 221                 return -1;
 222         }
 223
 224         return 0;
 225 }
 226
 227 static int
 228 vhost_vdpa_dma_unmap(struct virtio_user_dev *dev, __rte_unused void *addr,
 229                                   uint64_t iova, size_t len)
 230 {
 231         struct vhost_msg msg = {};
 232
 233         if (!(dev->protocol_features & (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2))) {
 234                 PMD_DRV_LOG(ERR, "IOTLB_MSG_V2 not supported by the backend.");
 235                 return -1;
 236         }
 237
 238         msg.type = VHOST_IOTLB_MSG_V2;
 239         msg.iotlb.type = VHOST_IOTLB_INVALIDATE;
 240         msg.iotlb.iova = iova;
 241         msg.iotlb.size = len;
 242
 243         PMD_DRV_LOG(DEBUG, "%s: iova: 0x%" PRIx64 ", len: 0x%zx",
 244                         __func__, iova, len);
 245
 246         if (write(dev->vhostfd, &msg, sizeof(msg)) != sizeof(msg)) {
 247                 PMD_DRV_LOG(ERR, "Failed to send IOTLB invalidate (%s)",
 248                                 strerror(errno));
 249                 return -1;
 250         }
 251
 252         return 0;
 253 }
 254
 255 static int
 256 vhost_vdpa_dma_map_batch(struct virtio_user_dev *dev, void *addr,
 257                                   uint64_t iova, size_t len)
 258 {
 259         int ret;
 260
 261         if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
 262                 return -1;
 263
 264         ret = vhost_vdpa_dma_map(dev, addr, iova, len);
 265
 266         if (vhost_vdpa_iotlb_batch_end(dev) < 0)
 267                 return -1;
 268
 269         return ret;
 270 }
 271
 272 static int
 273 vhost_vdpa_dma_unmap_batch(struct virtio_user_dev *dev, void *addr,
 274                                   uint64_t iova, size_t len)
 275 {
 276         int ret;
 277
 278         if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
 279                 return -1;
 280
 281         ret = vhost_vdpa_dma_unmap(dev, addr, iova, len);
 282
 283         if (vhost_vdpa_iotlb_batch_end(dev) < 0)
 284                 return -1;
 285
 286         return ret;
 287 }
 288
 289 static int
 290 vhost_vdpa_map_contig(const struct rte_memseg_list *msl,
 291                 const struct rte_memseg *ms, size_t len, void *arg)
 292 {
 293         struct virtio_user_dev *dev = arg;
 294
 295         if (msl->external)
 296                 return 0;
 297
 298         return vhost_vdpa_dma_map(dev, ms->addr, ms->iova, len);
 299 }
 300
 301 static int
 302 vhost_vdpa_map(const struct rte_memseg_list *msl, const struct rte_memseg *ms,
 303                 void *arg)
 304 {
 305         struct virtio_user_dev *dev = arg;
 306
 307         /* skip external memory that isn't a heap */
 308         if (msl->external && !msl->heap)
 309                 return 0;
 310
 311         /* skip any segments with invalid IOVA addresses */
 312         if (ms->iova == RTE_BAD_IOVA)
 313                 return 0;
 314
 315         /* if IOVA mode is VA, we've already mapped the internal segments */
 316         if (!msl->external && rte_eal_iova_mode() == RTE_IOVA_VA)
 317                 return 0;
 318
 319         return vhost_vdpa_dma_map(dev, ms->addr, ms->iova, ms->len);
 320 }
 321
 322 static int
 323 vhost_vdpa_set_memory_table(struct virtio_user_dev *dev)
 324 {
 325         int ret;
 326
 327         if (vhost_vdpa_iotlb_batch_begin(dev) < 0)
 328                 return -1;
 329
 330         vhost_vdpa_dma_unmap(dev, NULL, 0, SIZE_MAX);
 331
 332         if (rte_eal_iova_mode() == RTE_IOVA_VA) {
 333                 /* with IOVA as VA mode, we can get away with mapping contiguous
 334                  * chunks rather than going page-by-page.
 335                  */
 336                 ret = rte_memseg_contig_walk_thread_unsafe(
 337                                 vhost_vdpa_map_contig, dev);
 338                 if (ret)
 339                         goto batch_end;
 340                 /* we have to continue the walk because we've skipped the
 341                  * external segments during the config walk.
 342                  */
 343         }
 344         ret = rte_memseg_walk_thread_unsafe(vhost_vdpa_map, dev);
 345
 346 batch_end:
 347         if (vhost_vdpa_iotlb_batch_end(dev) < 0)
 348                 return -1;
 349
 350         return ret;
 351 }
 352
 353 static int
 354 vhost_vdpa_set_vring_enable(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 355 {
 356         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_VDPA_SET_VRING_ENABLE, state);
 357 }
 358
 359 static int
 360 vhost_vdpa_set_vring_num(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 361 {
 362         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_VRING_NUM, state);
 363 }
 364
 365 static int
 366 vhost_vdpa_set_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 367 {
 368         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_VRING_BASE, state);
 369 }
 370
 371 static int
 372 vhost_vdpa_get_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 373 {
 374         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_GET_VRING_BASE, state);
 375 }
 376
 377 static int
 378 vhost_vdpa_set_vring_call(struct virtio_user_dev *dev, struct vhost_vring_file *file)
 379 {
 380         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_VRING_CALL, file);
 381 }
 382
 383 static int
 384 vhost_vdpa_set_vring_kick(struct virtio_user_dev *dev, struct vhost_vring_file *file)
 385 {
 386         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_VRING_KICK, file);
 387 }
 388
 389 static int
 390 vhost_vdpa_set_vring_addr(struct virtio_user_dev *dev, struct vhost_vring_addr *addr)
 391 {
 392         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_SET_VRING_ADDR, addr);
 393 }
 394
 395 static int
 396 vhost_vdpa_get_status(struct virtio_user_dev *dev, uint8_t *status)
 397 {
 398         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_VDPA_GET_STATUS, status);
 399 }
 400
 401 static int
 402 vhost_vdpa_set_status(struct virtio_user_dev *dev, uint8_t status)
 403 {
 404         return vhost_vdpa_ioctl(dev->vhostfd, VHOST_VDPA_SET_STATUS, &status);
 405 }
 406
 407 /**
 408  * Set up environment to talk with a vhost vdpa backend.
 409  *
 410  * @return
 411  *   - (-1) if fail to set up;
 412  *   - (>=0) if successful.
 413  */
 414 static int
 415 vhost_vdpa_setup(struct virtio_user_dev *dev)
 416 {
 417         uint32_t did = (uint32_t)-1;
 418
 419         dev->vhostfd = open(dev->path, O_RDWR);
 420         if (dev->vhostfd < 0) {
 421                 PMD_DRV_LOG(ERR, "Failed to open %s: %s\n",
 422                                 dev->path, strerror(errno));
 423                 return -1;
 424         }
 425
 426         if (ioctl(dev->vhostfd, VHOST_VDPA_GET_DEVICE_ID, &did) < 0 ||
 427                         did != VIRTIO_ID_NETWORK) {
 428                 PMD_DRV_LOG(ERR, "Invalid vdpa device ID: %u\n", did);
 429                 return -1;
 430         }
 431
 432         return 0;
 433 }
 434
 435 static int
 436 vhost_vdpa_destroy(struct virtio_user_dev *dev __rte_unused)
 437 {
 438         return 0;
 439 }
 440
 441 static int
 442 vhost_vdpa_enable_queue_pair(struct virtio_user_dev *dev,
 443                                uint16_t pair_idx,
 444                                int enable)
 445 {
 446         int i;
 447
 448         if (dev->qp_enabled[pair_idx] == enable)
 449                 return 0;
 450
 451         for (i = 0; i < 2; ++i) {
 452                 struct vhost_vring_state state = {
 453                         .index = pair_idx * 2 + i,
 454                         .num   = enable,
 455                 };
 456
 457                 if (vhost_vdpa_set_vring_enable(dev, &state))
 458                         return -1;
 459         }
 460
 461         dev->qp_enabled[pair_idx] = enable;
 462
 463         return 0;
 464 }
 465
 466 static int
 467 vhost_vdpa_get_backend_features(uint64_t *features)
 468 {
 469         *features = 0;
 470
 471         return 0;
 472 }
 473
 474 struct virtio_user_backend_ops virtio_ops_vdpa = {
 475         .setup = vhost_vdpa_setup,
 476         .destroy = vhost_vdpa_destroy,
 477         .get_backend_features = vhost_vdpa_get_backend_features,
 478         .set_owner = vhost_vdpa_set_owner,
 479         .get_features = vhost_vdpa_get_features,
 480         .set_features = vhost_vdpa_set_features,
 481         .set_memory_table = vhost_vdpa_set_memory_table,
 482         .set_vring_num = vhost_vdpa_set_vring_num,
 483         .set_vring_base = vhost_vdpa_set_vring_base,
 484         .get_vring_base = vhost_vdpa_get_vring_base,
 485         .set_vring_call = vhost_vdpa_set_vring_call,
 486         .set_vring_kick = vhost_vdpa_set_vring_kick,
 487         .set_vring_addr = vhost_vdpa_set_vring_addr,
 488         .get_status = vhost_vdpa_get_status,
 489         .set_status = vhost_vdpa_set_status,
 490         .enable_qp = vhost_vdpa_enable_queue_pair,
 491         .dma_map = vhost_vdpa_dma_map_batch,
 492         .dma_unmap = vhost_vdpa_dma_unmap_batch,
 493 };