ceph/src/dpdk/lib/librte_eal/linuxapp/eal/eal_vfio.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <string.h>
  35 #include <fcntl.h>
  36 #include <unistd.h>
  37 #include <sys/ioctl.h>
  38
  39 #include <rte_log.h>
  40 #include <rte_memory.h>
  41 #include <rte_eal_memconfig.h>
  42
  43 #include "eal_filesystem.h"
  44 #include "eal_vfio.h"
  45 #include "eal_private.h"
  46
  47 #ifdef VFIO_PRESENT
  48
  49 /* per-process VFIO config */
  50 static struct vfio_config vfio_cfg;
  51
  52 static int vfio_type1_dma_map(int);
  53 static int vfio_noiommu_dma_map(int);
  54
  55 /* IOMMU types we support */
  56 static const struct vfio_iommu_type iommu_types[] = {
  57         /* x86 IOMMU, otherwise known as type 1 */
  58         { RTE_VFIO_TYPE1, "Type 1", &vfio_type1_dma_map},
  59         /* IOMMU-less mode */
  60         { RTE_VFIO_NOIOMMU, "No-IOMMU", &vfio_noiommu_dma_map},
  61 };
  62
  63 int
  64 vfio_get_group_fd(int iommu_group_no)
  65 {
  66         int i;
  67         int vfio_group_fd;
  68         char filename[PATH_MAX];
  69
  70         /* check if we already have the group descriptor open */
  71         for (i = 0; i < vfio_cfg.vfio_group_idx; i++)
  72                 if (vfio_cfg.vfio_groups[i].group_no == iommu_group_no)
  73                         return vfio_cfg.vfio_groups[i].fd;
  74
  75         /* if primary, try to open the group */
  76         if (internal_config.process_type == RTE_PROC_PRIMARY) {
  77                 /* try regular group format */
  78                 snprintf(filename, sizeof(filename),
  79                                  VFIO_GROUP_FMT, iommu_group_no);
  80                 vfio_group_fd = open(filename, O_RDWR);
  81                 if (vfio_group_fd < 0) {
  82                         /* if file not found, it's not an error */
  83                         if (errno != ENOENT) {
  84                                 RTE_LOG(ERR, EAL, "Cannot open %s: %s\n", filename,
  85                                                 strerror(errno));
  86                                 return -1;
  87                         }
  88
  89                         /* special case: try no-IOMMU path as well */
  90                         snprintf(filename, sizeof(filename),
  91                                         VFIO_NOIOMMU_GROUP_FMT, iommu_group_no);
  92                         vfio_group_fd = open(filename, O_RDWR);
  93                         if (vfio_group_fd < 0) {
  94                                 if (errno != ENOENT) {
  95                                         RTE_LOG(ERR, EAL, "Cannot open %s: %s\n", filename,
  96                                                         strerror(errno));
  97                                         return -1;
  98                                 }
  99                                 return 0;
 100                         }
 101                         /* noiommu group found */
 102                 }
 103
 104                 /* if the fd is valid, create a new group for it */
 105                 if (vfio_cfg.vfio_group_idx == VFIO_MAX_GROUPS) {
 106                         RTE_LOG(ERR, EAL, "Maximum number of VFIO groups reached!\n");
 107                         close(vfio_group_fd);
 108                         return -1;
 109                 }
 110                 vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].group_no = iommu_group_no;
 111                 vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].fd = vfio_group_fd;
 112                 return vfio_group_fd;
 113         }
 114         /* if we're in a secondary process, request group fd from the primary
 115          * process via our socket
 116          */
 117         else {
 118                 int socket_fd, ret;
 119
 120                 socket_fd = vfio_mp_sync_connect_to_primary();
 121
 122                 if (socket_fd < 0) {
 123                         RTE_LOG(ERR, EAL, "  cannot connect to primary process!\n");
 124                         return -1;
 125                 }
 126                 if (vfio_mp_sync_send_request(socket_fd, SOCKET_REQ_GROUP) < 0) {
 127                         RTE_LOG(ERR, EAL, "  cannot request container fd!\n");
 128                         close(socket_fd);
 129                         return -1;
 130                 }
 131                 if (vfio_mp_sync_send_request(socket_fd, iommu_group_no) < 0) {
 132                         RTE_LOG(ERR, EAL, "  cannot send group number!\n");
 133                         close(socket_fd);
 134                         return -1;
 135                 }
 136                 ret = vfio_mp_sync_receive_request(socket_fd);
 137                 switch (ret) {
 138                 case SOCKET_NO_FD:
 139                         close(socket_fd);
 140                         return 0;
 141                 case SOCKET_OK:
 142                         vfio_group_fd = vfio_mp_sync_receive_fd(socket_fd);
 143                         /* if we got the fd, return it */
 144                         if (vfio_group_fd > 0) {
 145                                 close(socket_fd);
 146                                 return vfio_group_fd;
 147                         }
 148                         /* fall-through on error */
 149                 default:
 150                         RTE_LOG(ERR, EAL, "  cannot get container fd!\n");
 151                         close(socket_fd);
 152                         return -1;
 153                 }
 154         }
 155         return -1;
 156 }
 157
 158 static void
 159 clear_current_group(void)
 160 {
 161         vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].group_no = 0;
 162         vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].fd = -1;
 163 }
 164
 165 int vfio_setup_device(const char *sysfs_base, const char *dev_addr,
 166                 int *vfio_dev_fd, struct vfio_device_info *device_info)
 167 {
 168         struct vfio_group_status group_status = {
 169                         .argsz = sizeof(group_status)
 170         };
 171         int vfio_group_fd;
 172         int iommu_group_no;
 173         int ret;
 174
 175         /* get group number */
 176         ret = vfio_get_group_no(sysfs_base, dev_addr, &iommu_group_no);
 177         if (ret == 0) {
 178                 RTE_LOG(WARNING, EAL, "  %s not managed by VFIO driver, skipping\n",
 179                         dev_addr);
 180                 return 1;
 181         }
 182
 183         /* if negative, something failed */
 184         if (ret < 0)
 185                 return -1;
 186
 187         /* get the actual group fd */
 188         vfio_group_fd = vfio_get_group_fd(iommu_group_no);
 189         if (vfio_group_fd < 0)
 190                 return -1;
 191
 192         /* store group fd */
 193         vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].group_no = iommu_group_no;
 194         vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].fd = vfio_group_fd;
 195
 196         /* if group_fd == 0, that means the device isn't managed by VFIO */
 197         if (vfio_group_fd == 0) {
 198                 RTE_LOG(WARNING, EAL, "  %s not managed by VFIO driver, skipping\n",
 199                                 dev_addr);
 200                 /* we store 0 as group fd to distinguish between existing but
 201                  * unbound VFIO groups, and groups that don't exist at all.
 202                  */
 203                 vfio_cfg.vfio_group_idx++;
 204                 return 1;
 205         }
 206
 207         /*
 208          * at this point, we know that this group is viable (meaning, all devices
 209          * are either bound to VFIO or not bound to anything)
 210          */
 211
 212         /* check if the group is viable */
 213         ret = ioctl(vfio_group_fd, VFIO_GROUP_GET_STATUS, &group_status);
 214         if (ret) {
 215                 RTE_LOG(ERR, EAL, "  %s cannot get group status, "
 216                                 "error %i (%s)\n", dev_addr, errno, strerror(errno));
 217                 close(vfio_group_fd);
 218                 clear_current_group();
 219                 return -1;
 220         } else if (!(group_status.flags & VFIO_GROUP_FLAGS_VIABLE)) {
 221                 RTE_LOG(ERR, EAL, "  %s VFIO group is not viable!\n", dev_addr);
 222                 close(vfio_group_fd);
 223                 clear_current_group();
 224                 return -1;
 225         }
 226
 227         /* check if group does not have a container yet */
 228         if (!(group_status.flags & VFIO_GROUP_FLAGS_CONTAINER_SET)) {
 229
 230                 /* add group to a container */
 231                 ret = ioctl(vfio_group_fd, VFIO_GROUP_SET_CONTAINER,
 232                                 &vfio_cfg.vfio_container_fd);
 233                 if (ret) {
 234                         RTE_LOG(ERR, EAL, "  %s cannot add VFIO group to container, "
 235                                         "error %i (%s)\n", dev_addr, errno, strerror(errno));
 236                         close(vfio_group_fd);
 237                         clear_current_group();
 238                         return -1;
 239                 }
 240                 /*
 241                  * at this point we know that this group has been successfully
 242                  * initialized, so we increment vfio_group_idx to indicate that we can
 243                  * add new groups.
 244                  */
 245                 vfio_cfg.vfio_group_idx++;
 246         }
 247
 248         /*
 249          * pick an IOMMU type and set up DMA mappings for container
 250          *
 251          * needs to be done only once, only when at least one group is assigned to
 252          * a container and only in primary process
 253          */
 254         if (internal_config.process_type == RTE_PROC_PRIMARY &&
 255                         vfio_cfg.vfio_container_has_dma == 0) {
 256                 /* select an IOMMU type which we will be using */
 257                 const struct vfio_iommu_type *t =
 258                                 vfio_set_iommu_type(vfio_cfg.vfio_container_fd);
 259                 if (!t) {
 260                         RTE_LOG(ERR, EAL, "  %s failed to select IOMMU type\n", dev_addr);
 261                         return -1;
 262                 }
 263                 ret = t->dma_map_func(vfio_cfg.vfio_container_fd);
 264                 if (ret) {
 265                         RTE_LOG(ERR, EAL, "  %s DMA remapping failed, "
 266                                         "error %i (%s)\n", dev_addr, errno, strerror(errno));
 267                         return -1;
 268                 }
 269                 vfio_cfg.vfio_container_has_dma = 1;
 270         }
 271
 272         /* get a file descriptor for the device */
 273         *vfio_dev_fd = ioctl(vfio_group_fd, VFIO_GROUP_GET_DEVICE_FD, dev_addr);
 274         if (*vfio_dev_fd < 0) {
 275                 /* if we cannot get a device fd, this simply means that this
 276                 * particular port is not bound to VFIO
 277                 */
 278                 RTE_LOG(WARNING, EAL, "  %s not managed by VFIO driver, skipping\n",
 279                                 dev_addr);
 280                 return 1;
 281         }
 282
 283         /* test and setup the device */
 284         ret = ioctl(*vfio_dev_fd, VFIO_DEVICE_GET_INFO, device_info);
 285         if (ret) {
 286                 RTE_LOG(ERR, EAL, "  %s cannot get device info, "
 287                                 "error %i (%s)\n", dev_addr, errno, strerror(errno));
 288                 close(*vfio_dev_fd);
 289                 return -1;
 290         }
 291
 292         return 0;
 293 }
 294
 295 int
 296 vfio_enable(const char *modname)
 297 {
 298         /* initialize group list */
 299         int i;
 300         int vfio_available;
 301
 302         for (i = 0; i < VFIO_MAX_GROUPS; i++) {
 303                 vfio_cfg.vfio_groups[i].fd = -1;
 304                 vfio_cfg.vfio_groups[i].group_no = -1;
 305         }
 306
 307         /* inform the user that we are probing for VFIO */
 308         RTE_LOG(INFO, EAL, "Probing VFIO support...\n");
 309
 310         /* check if vfio-pci module is loaded */
 311         vfio_available = rte_eal_check_module(modname);
 312
 313         /* return error directly */
 314         if (vfio_available == -1) {
 315                 RTE_LOG(INFO, EAL, "Could not get loaded module details!\n");
 316                 return -1;
 317         }
 318
 319         /* return 0 if VFIO modules not loaded */
 320         if (vfio_available == 0) {
 321                 RTE_LOG(DEBUG, EAL, "VFIO modules not loaded, "
 322                         "skipping VFIO support...\n");
 323                 return 0;
 324         }
 325
 326         vfio_cfg.vfio_container_fd = vfio_get_container_fd();
 327
 328         /* check if we have VFIO driver enabled */
 329         if (vfio_cfg.vfio_container_fd != -1) {
 330                 RTE_LOG(NOTICE, EAL, "VFIO support initialized\n");
 331                 vfio_cfg.vfio_enabled = 1;
 332         } else {
 333                 RTE_LOG(NOTICE, EAL, "VFIO support could not be initialized\n");
 334         }
 335
 336         return 0;
 337 }
 338
 339 int
 340 vfio_is_enabled(const char *modname)
 341 {
 342         const int mod_available = rte_eal_check_module(modname);
 343         return vfio_cfg.vfio_enabled && mod_available;
 344 }
 345
 346 const struct vfio_iommu_type *
 347 vfio_set_iommu_type(int vfio_container_fd)
 348 {
 349         unsigned idx;
 350         for (idx = 0; idx < RTE_DIM(iommu_types); idx++) {
 351                 const struct vfio_iommu_type *t = &iommu_types[idx];
 352
 353                 int ret = ioctl(vfio_container_fd, VFIO_SET_IOMMU,
 354                                 t->type_id);
 355                 if (!ret) {
 356                         RTE_LOG(NOTICE, EAL, "  using IOMMU type %d (%s)\n",
 357                                         t->type_id, t->name);
 358                         return t;
 359                 }
 360                 /* not an error, there may be more supported IOMMU types */
 361                 RTE_LOG(DEBUG, EAL, "  set IOMMU type %d (%s) failed, "
 362                                 "error %i (%s)\n", t->type_id, t->name, errno,
 363                                 strerror(errno));
 364         }
 365         /* if we didn't find a suitable IOMMU type, fail */
 366         return NULL;
 367 }
 368
 369 int
 370 vfio_has_supported_extensions(int vfio_container_fd)
 371 {
 372         int ret;
 373         unsigned idx, n_extensions = 0;
 374         for (idx = 0; idx < RTE_DIM(iommu_types); idx++) {
 375                 const struct vfio_iommu_type *t = &iommu_types[idx];
 376
 377                 ret = ioctl(vfio_container_fd, VFIO_CHECK_EXTENSION,
 378                                 t->type_id);
 379                 if (ret < 0) {
 380                         RTE_LOG(ERR, EAL, "  could not get IOMMU type, "
 381                                 "error %i (%s)\n", errno,
 382                                 strerror(errno));
 383                         close(vfio_container_fd);
 384                         return -1;
 385                 } else if (ret == 1) {
 386                         /* we found a supported extension */
 387                         n_extensions++;
 388                 }
 389                 RTE_LOG(DEBUG, EAL, "  IOMMU type %d (%s) is %s\n",
 390                                 t->type_id, t->name,
 391                                 ret ? "supported" : "not supported");
 392         }
 393
 394         /* if we didn't find any supported IOMMU types, fail */
 395         if (!n_extensions) {
 396                 close(vfio_container_fd);
 397                 return -1;
 398         }
 399
 400         return 0;
 401 }
 402
 403 int
 404 vfio_get_container_fd(void)
 405 {
 406         int ret, vfio_container_fd;
 407
 408         /* if we're in a primary process, try to open the container */
 409         if (internal_config.process_type == RTE_PROC_PRIMARY) {
 410                 vfio_container_fd = open(VFIO_CONTAINER_PATH, O_RDWR);
 411                 if (vfio_container_fd < 0) {
 412                         RTE_LOG(ERR, EAL, "  cannot open VFIO container, "
 413                                         "error %i (%s)\n", errno, strerror(errno));
 414                         return -1;
 415                 }
 416
 417                 /* check VFIO API version */
 418                 ret = ioctl(vfio_container_fd, VFIO_GET_API_VERSION);
 419                 if (ret != VFIO_API_VERSION) {
 420                         if (ret < 0)
 421                                 RTE_LOG(ERR, EAL, "  could not get VFIO API version, "
 422                                                 "error %i (%s)\n", errno, strerror(errno));
 423                         else
 424                                 RTE_LOG(ERR, EAL, "  unsupported VFIO API version!\n");
 425                         close(vfio_container_fd);
 426                         return -1;
 427                 }
 428
 429                 ret = vfio_has_supported_extensions(vfio_container_fd);
 430                 if (ret) {
 431                         RTE_LOG(ERR, EAL, "  no supported IOMMU "
 432                                         "extensions found!\n");
 433                         return -1;
 434                 }
 435
 436                 return vfio_container_fd;
 437         } else {
 438                 /*
 439                  * if we're in a secondary process, request container fd from the
 440                  * primary process via our socket
 441                  */
 442                 int socket_fd;
 443
 444                 socket_fd = vfio_mp_sync_connect_to_primary();
 445                 if (socket_fd < 0) {
 446                         RTE_LOG(ERR, EAL, "  cannot connect to primary process!\n");
 447                         return -1;
 448                 }
 449                 if (vfio_mp_sync_send_request(socket_fd, SOCKET_REQ_CONTAINER) < 0) {
 450                         RTE_LOG(ERR, EAL, "  cannot request container fd!\n");
 451                         close(socket_fd);
 452                         return -1;
 453                 }
 454                 vfio_container_fd = vfio_mp_sync_receive_fd(socket_fd);
 455                 if (vfio_container_fd < 0) {
 456                         RTE_LOG(ERR, EAL, "  cannot get container fd!\n");
 457                         close(socket_fd);
 458                         return -1;
 459                 }
 460                 close(socket_fd);
 461                 return vfio_container_fd;
 462         }
 463
 464         return -1;
 465 }
 466
 467 int
 468 vfio_get_group_no(const char *sysfs_base,
 469                 const char *dev_addr, int *iommu_group_no)
 470 {
 471         char linkname[PATH_MAX];
 472         char filename[PATH_MAX];
 473         char *tok[16], *group_tok, *end;
 474         int ret;
 475
 476         memset(linkname, 0, sizeof(linkname));
 477         memset(filename, 0, sizeof(filename));
 478
 479         /* try to find out IOMMU group for this device */
 480         snprintf(linkname, sizeof(linkname),
 481                          "%s/%s/iommu_group", sysfs_base, dev_addr);
 482
 483         ret = readlink(linkname, filename, sizeof(filename));
 484
 485         /* if the link doesn't exist, no VFIO for us */
 486         if (ret < 0)
 487                 return 0;
 488
 489         ret = rte_strsplit(filename, sizeof(filename),
 490                         tok, RTE_DIM(tok), '/');
 491
 492         if (ret <= 0) {
 493                 RTE_LOG(ERR, EAL, "  %s cannot get IOMMU group\n", dev_addr);
 494                 return -1;
 495         }
 496
 497         /* IOMMU group is always the last token */
 498         errno = 0;
 499         group_tok = tok[ret - 1];
 500         end = group_tok;
 501         *iommu_group_no = strtol(group_tok, &end, 10);
 502         if ((end != group_tok && *end != '\0') || errno != 0) {
 503                 RTE_LOG(ERR, EAL, "  %s error parsing IOMMU number!\n", dev_addr);
 504                 return -1;
 505         }
 506
 507         return 1;
 508 }
 509
 510 static int
 511 vfio_type1_dma_map(int vfio_container_fd)
 512 {
 513         const struct rte_memseg *ms = rte_eal_get_physmem_layout();
 514         int i, ret;
 515
 516         /* map all DPDK segments for DMA. use 1:1 PA to IOVA mapping */
 517         for (i = 0; i < RTE_MAX_MEMSEG; i++) {
 518                 struct vfio_iommu_type1_dma_map dma_map;
 519
 520                 if (ms[i].addr == NULL)
 521                         break;
 522
 523                 memset(&dma_map, 0, sizeof(dma_map));
 524                 dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map);
 525                 dma_map.vaddr = ms[i].addr_64;
 526                 dma_map.size = ms[i].len;
 527                 dma_map.iova = ms[i].phys_addr;
 528                 dma_map.flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE;
 529
 530                 ret = ioctl(vfio_container_fd, VFIO_IOMMU_MAP_DMA, &dma_map);
 531
 532                 if (ret) {
 533                         RTE_LOG(ERR, EAL, "  cannot set up DMA remapping, "
 534                                         "error %i (%s)\n", errno, strerror(errno));
 535                         return -1;
 536                 }
 537         }
 538
 539         return 0;
 540 }
 541
 542 static int
 543 vfio_noiommu_dma_map(int __rte_unused vfio_container_fd)
 544 {
 545         /* No-IOMMU mode does not need DMA mapping */
 546         return 0;
 547 }
 548
 549 #endif