src/gallium/drivers/vc4/vc4_bufmgr.c

   1 /*
   2  * Copyright © 2014-2015 Broadcom
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include <errno.h>
  25 #include <err.h>
  26 #include <sys/mman.h>
  27 #include <fcntl.h>
  28 #include <xf86drm.h>
  29 #include <xf86drmMode.h>
  30
  31 #include "util/u_memory.h"
  32 #include "util/ralloc.h"
  33
  34 #include "vc4_context.h"
  35 #include "vc4_screen.h"
  36
  37 static bool dump_stats = false;
  38
  39 static void
  40 vc4_bo_cache_free_all(struct vc4_bo_cache *cache);
  41
  42 static void
  43 vc4_bo_dump_stats(struct vc4_screen *screen)
  44 {
  45         struct vc4_bo_cache *cache = &screen->bo_cache;
  46
  47         fprintf(stderr, "  BOs allocated:   %d\n", screen->bo_count);
  48         fprintf(stderr, "  BOs size:        %dkb\n", screen->bo_size / 1024);
  49         fprintf(stderr, "  BOs cached:      %d\n", cache->bo_count);
  50         fprintf(stderr, "  BOs cached size: %dkb\n", cache->bo_size / 1024);
  51
  52         if (!list_empty(&cache->time_list)) {
  53                 struct vc4_bo *first = LIST_ENTRY(struct vc4_bo,
  54                                                   cache->time_list.next,
  55                                                   time_list);
  56                 struct vc4_bo *last = LIST_ENTRY(struct vc4_bo,
  57                                                   cache->time_list.prev,
  58                                                   time_list);
  59
  60                 fprintf(stderr, "  oldest cache time: %ld\n",
  61                         (long)first->free_time);
  62                 fprintf(stderr, "  newest cache time: %ld\n",
  63                         (long)last->free_time);
  64
  65                 struct timespec time;
  66                 clock_gettime(CLOCK_MONOTONIC, &time);
  67                 fprintf(stderr, "  now:               %ld\n",
  68                         time.tv_sec);
  69         }
  70 }
  71
  72 static void
  73 vc4_bo_remove_from_cache(struct vc4_bo_cache *cache, struct vc4_bo *bo)
  74 {
  75         list_del(&bo->time_list);
  76         list_del(&bo->size_list);
  77         cache->bo_count--;
  78         cache->bo_size -= bo->size;
  79 }
  80
  81 static struct vc4_bo *
  82 vc4_bo_from_cache(struct vc4_screen *screen, uint32_t size, const char *name)
  83 {
  84         struct vc4_bo_cache *cache = &screen->bo_cache;
  85         uint32_t page_index = size / 4096 - 1;
  86
  87         if (cache->size_list_size <= page_index)
  88                 return NULL;
  89
  90         struct vc4_bo *bo = NULL;
  91         pipe_mutex_lock(cache->lock);
  92         if (!list_empty(&cache->size_list[page_index])) {
  93                 bo = LIST_ENTRY(struct vc4_bo, cache->size_list[page_index].next,
  94                                 size_list);
  95
  96                 /* Check that the BO has gone idle.  If not, then we want to
  97                  * allocate something new instead, since we assume that the
  98                  * user will proceed to CPU map it and fill it with stuff.
  99                  */
 100                 if (!vc4_bo_wait(bo, 0, NULL)) {
 101                         pipe_mutex_unlock(cache->lock);
 102                         return NULL;
 103                 }
 104
 105                 pipe_reference_init(&bo->reference, 1);
 106                 vc4_bo_remove_from_cache(cache, bo);
 107
 108                 bo->name = name;
 109         }
 110         pipe_mutex_unlock(cache->lock);
 111         return bo;
 112 }
 113
 114 struct vc4_bo *
 115 vc4_bo_alloc(struct vc4_screen *screen, uint32_t size, const char *name)
 116 {
 117         struct vc4_bo *bo;
 118         int ret;
 119
 120         size = align(size, 4096);
 121
 122         bo = vc4_bo_from_cache(screen, size, name);
 123         if (bo) {
 124                 if (dump_stats) {
 125                         fprintf(stderr, "Allocated %s %dkb from cache:\n",
 126                                 name, size / 1024);
 127                         vc4_bo_dump_stats(screen);
 128                 }
 129                 return bo;
 130         }
 131
 132         bo = CALLOC_STRUCT(vc4_bo);
 133         if (!bo)
 134                 return NULL;
 135
 136         pipe_reference_init(&bo->reference, 1);
 137         bo->screen = screen;
 138         bo->size = size;
 139         bo->name = name;
 140         bo->private = true;
 141
 142         bool cleared_and_retried = false;
 143 retry:
 144         if (!using_vc4_simulator) {
 145                 struct drm_vc4_create_bo create;
 146                 memset(&create, 0, sizeof(create));
 147
 148                 create.size = size;
 149
 150                 ret = drmIoctl(screen->fd, DRM_IOCTL_VC4_CREATE_BO, &create);
 151                 bo->handle = create.handle;
 152         } else {
 153                 struct drm_mode_create_dumb create;
 154                 memset(&create, 0, sizeof(create));
 155
 156                 create.width = 128;
 157                 create.bpp = 8;
 158                 create.height = (size + 127) / 128;
 159
 160                 ret = drmIoctl(screen->fd, DRM_IOCTL_MODE_CREATE_DUMB, &create);
 161                 bo->handle = create.handle;
 162                 assert(create.size >= size);
 163         }
 164         if (ret != 0) {
 165                 if (!list_empty(&screen->bo_cache.time_list) &&
 166                     !cleared_and_retried) {
 167                         cleared_and_retried = true;
 168                         vc4_bo_cache_free_all(&screen->bo_cache);
 169                         goto retry;
 170                 }
 171                 fprintf(stderr, "create ioctl failure\n");
 172                 abort();
 173         }
 174
 175         screen->bo_count++;
 176         screen->bo_size += bo->size;
 177         if (dump_stats) {
 178                 fprintf(stderr, "Allocated %s %dkb:\n", name, size / 1024);
 179                 vc4_bo_dump_stats(screen);
 180         }
 181
 182         return bo;
 183 }
 184
 185 void
 186 vc4_bo_last_unreference(struct vc4_bo *bo)
 187 {
 188         struct vc4_screen *screen = bo->screen;
 189
 190         struct timespec time;
 191         clock_gettime(CLOCK_MONOTONIC, &time);
 192         pipe_mutex_lock(screen->bo_cache.lock);
 193         vc4_bo_last_unreference_locked_timed(bo, time.tv_sec);
 194         pipe_mutex_unlock(screen->bo_cache.lock);
 195 }
 196
 197 static void
 198 vc4_bo_free(struct vc4_bo *bo)
 199 {
 200         struct vc4_screen *screen = bo->screen;
 201
 202         if (bo->map) {
 203 #ifdef USE_VC4_SIMULATOR
 204                 if (bo->simulator_winsys_map) {
 205                         free(bo->map);
 206                         bo->map = bo->simulator_winsys_map;
 207                 }
 208 #endif
 209                 munmap(bo->map, bo->size);
 210         }
 211
 212         struct drm_gem_close c;
 213         memset(&c, 0, sizeof(c));
 214         c.handle = bo->handle;
 215         int ret = drmIoctl(screen->fd, DRM_IOCTL_GEM_CLOSE, &c);
 216         if (ret != 0)
 217                 fprintf(stderr, "close object %d: %s\n", bo->handle, strerror(errno));
 218
 219         screen->bo_count--;
 220         screen->bo_size -= bo->size;
 221
 222         if (dump_stats) {
 223                 fprintf(stderr, "Freed %s%s%dkb:\n",
 224                         bo->name ? bo->name : "",
 225                         bo->name ? " " : "",
 226                         bo->size / 1024);
 227                 vc4_bo_dump_stats(screen);
 228         }
 229
 230         free(bo);
 231 }
 232
 233 static void
 234 free_stale_bos(struct vc4_screen *screen, time_t time)
 235 {
 236         struct vc4_bo_cache *cache = &screen->bo_cache;
 237         bool freed_any = false;
 238
 239         list_for_each_entry_safe(struct vc4_bo, bo, &cache->time_list,
 240                                  time_list) {
 241                 if (dump_stats && !freed_any) {
 242                         fprintf(stderr, "Freeing stale BOs:\n");
 243                         vc4_bo_dump_stats(screen);
 244                         freed_any = true;
 245                 }
 246
 247                 /* If it's more than a second old, free it. */
 248                 if (time - bo->free_time > 2) {
 249                         vc4_bo_remove_from_cache(cache, bo);
 250                         vc4_bo_free(bo);
 251                 } else {
 252                         break;
 253                 }
 254         }
 255
 256         if (dump_stats && freed_any) {
 257                 fprintf(stderr, "Freed stale BOs:\n");
 258                 vc4_bo_dump_stats(screen);
 259         }
 260 }
 261
 262 static void
 263 vc4_bo_cache_free_all(struct vc4_bo_cache *cache)
 264 {
 265         pipe_mutex_lock(cache->lock);
 266         list_for_each_entry_safe(struct vc4_bo, bo, &cache->time_list,
 267                                  time_list) {
 268                 vc4_bo_remove_from_cache(cache, bo);
 269                 vc4_bo_free(bo);
 270         }
 271         pipe_mutex_unlock(cache->lock);
 272 }
 273
 274 void
 275 vc4_bo_last_unreference_locked_timed(struct vc4_bo *bo, time_t time)
 276 {
 277         struct vc4_screen *screen = bo->screen;
 278         struct vc4_bo_cache *cache = &screen->bo_cache;
 279         uint32_t page_index = bo->size / 4096 - 1;
 280
 281         if (!bo->private) {
 282                 vc4_bo_free(bo);
 283                 return;
 284         }
 285
 286         if (cache->size_list_size <= page_index) {
 287                 struct list_head *new_list =
 288                         ralloc_array(screen, struct list_head, page_index + 1);
 289
 290                 /* Move old list contents over (since the array has moved, and
 291                  * therefore the pointers to the list heads have to change).
 292                  */
 293                 for (int i = 0; i < cache->size_list_size; i++) {
 294                         struct list_head *old_head = &cache->size_list[i];
 295                         if (list_empty(old_head))
 296                                 list_inithead(&new_list[i]);
 297                         else {
 298                                 new_list[i].next = old_head->next;
 299                                 new_list[i].prev = old_head->prev;
 300                                 new_list[i].next->prev = &new_list[i];
 301                                 new_list[i].prev->next = &new_list[i];
 302                         }
 303                 }
 304                 for (int i = cache->size_list_size; i < page_index + 1; i++)
 305                         list_inithead(&new_list[i]);
 306
 307                 cache->size_list = new_list;
 308                 cache->size_list_size = page_index + 1;
 309         }
 310
 311         bo->free_time = time;
 312         list_addtail(&bo->size_list, &cache->size_list[page_index]);
 313         list_addtail(&bo->time_list, &cache->time_list);
 314         cache->bo_count++;
 315         cache->bo_size += bo->size;
 316         if (dump_stats) {
 317                 fprintf(stderr, "Freed %s %dkb to cache:\n",
 318                         bo->name, bo->size / 1024);
 319                 vc4_bo_dump_stats(screen);
 320         }
 321         bo->name = NULL;
 322
 323         free_stale_bos(screen, time);
 324 }
 325
 326 static struct vc4_bo *
 327 vc4_bo_open_handle(struct vc4_screen *screen,
 328                    uint32_t winsys_stride,
 329                    uint32_t handle, uint32_t size)
 330 {
 331         struct vc4_bo *bo = CALLOC_STRUCT(vc4_bo);
 332
 333         assert(size);
 334
 335         pipe_reference_init(&bo->reference, 1);
 336         bo->screen = screen;
 337         bo->handle = handle;
 338         bo->size = size;
 339         bo->name = "winsys";
 340         bo->private = false;
 341
 342 #ifdef USE_VC4_SIMULATOR
 343         vc4_bo_map(bo);
 344         bo->simulator_winsys_map = bo->map;
 345         bo->simulator_winsys_stride = winsys_stride;
 346         bo->map = malloc(bo->size);
 347 #endif
 348
 349         return bo;
 350 }
 351
 352 struct vc4_bo *
 353 vc4_bo_open_name(struct vc4_screen *screen, uint32_t name,
 354                  uint32_t winsys_stride)
 355 {
 356         struct drm_gem_open o = {
 357                 .name = name
 358         };
 359         int ret = drmIoctl(screen->fd, DRM_IOCTL_GEM_OPEN, &o);
 360         if (ret) {
 361                 fprintf(stderr, "Failed to open bo %d: %s\n",
 362                         name, strerror(errno));
 363                 return NULL;
 364         }
 365
 366         return vc4_bo_open_handle(screen, winsys_stride, o.handle, o.size);
 367 }
 368
 369 struct vc4_bo *
 370 vc4_bo_open_dmabuf(struct vc4_screen *screen, int fd, uint32_t winsys_stride)
 371 {
 372         uint32_t handle;
 373         int ret = drmPrimeFDToHandle(screen->fd, fd, &handle);
 374         int size;
 375         if (ret) {
 376                 fprintf(stderr, "Failed to get vc4 handle for dmabuf %d\n", fd);
 377                 return NULL;
 378         }
 379
 380         /* Determine the size of the bo we were handed. */
 381         size = lseek(fd, 0, SEEK_END);
 382         if (size == -1) {
 383                 fprintf(stderr, "Couldn't get size of dmabuf fd %d.\n", fd);
 384                 return NULL;
 385         }
 386
 387         return vc4_bo_open_handle(screen, winsys_stride, handle, size);
 388 }
 389
 390 int
 391 vc4_bo_get_dmabuf(struct vc4_bo *bo)
 392 {
 393         int fd;
 394         int ret = drmPrimeHandleToFD(bo->screen->fd, bo->handle,
 395                                      O_CLOEXEC, &fd);
 396         if (ret != 0) {
 397                 fprintf(stderr, "Failed to export gem bo %d to dmabuf\n",
 398                         bo->handle);
 399                 return -1;
 400         }
 401         bo->private = false;
 402
 403         return fd;
 404 }
 405
 406 struct vc4_bo *
 407 vc4_bo_alloc_shader(struct vc4_screen *screen, const void *data, uint32_t size)
 408 {
 409         struct vc4_bo *bo;
 410         int ret;
 411
 412         bo = CALLOC_STRUCT(vc4_bo);
 413         if (!bo)
 414                 return NULL;
 415
 416         pipe_reference_init(&bo->reference, 1);
 417         bo->screen = screen;
 418         bo->size = align(size, 4096);
 419         bo->name = "code";
 420         bo->private = false; /* Make sure it doesn't go back to the cache. */
 421
 422         if (!using_vc4_simulator) {
 423                 struct drm_vc4_create_shader_bo create = {
 424                         .size = size,
 425                         .data = (uintptr_t)data,
 426                 };
 427
 428                 ret = drmIoctl(screen->fd, DRM_IOCTL_VC4_CREATE_SHADER_BO,
 429                                &create);
 430                 bo->handle = create.handle;
 431         } else {
 432                 struct drm_mode_create_dumb create;
 433                 memset(&create, 0, sizeof(create));
 434
 435                 create.width = 128;
 436                 create.bpp = 8;
 437                 create.height = (size + 127) / 128;
 438
 439                 ret = drmIoctl(screen->fd, DRM_IOCTL_MODE_CREATE_DUMB, &create);
 440                 bo->handle = create.handle;
 441                 assert(create.size >= size);
 442
 443                 vc4_bo_map(bo);
 444                 memcpy(bo->map, data, size);
 445         }
 446         if (ret != 0) {
 447                 fprintf(stderr, "create shader ioctl failure\n");
 448                 abort();
 449         }
 450
 451         screen->bo_count++;
 452         screen->bo_size += bo->size;
 453         if (dump_stats) {
 454                 fprintf(stderr, "Allocated shader %dkb:\n", bo->size / 1024);
 455                 vc4_bo_dump_stats(screen);
 456         }
 457
 458         return bo;
 459 }
 460
 461 bool
 462 vc4_bo_flink(struct vc4_bo *bo, uint32_t *name)
 463 {
 464         struct drm_gem_flink flink = {
 465                 .handle = bo->handle,
 466         };
 467         int ret = drmIoctl(bo->screen->fd, DRM_IOCTL_GEM_FLINK, &flink);
 468         if (ret) {
 469                 fprintf(stderr, "Failed to flink bo %d: %s\n",
 470                         bo->handle, strerror(errno));
 471                 free(bo);
 472                 return false;
 473         }
 474
 475         bo->private = false;
 476         *name = flink.name;
 477
 478         return true;
 479 }
 480
 481 static int vc4_wait_seqno_ioctl(int fd, uint64_t seqno, uint64_t timeout_ns)
 482 {
 483         if (using_vc4_simulator)
 484                 return 0;
 485
 486         struct drm_vc4_wait_seqno wait = {
 487                 .seqno = seqno,
 488                 .timeout_ns = timeout_ns,
 489         };
 490         int ret = drmIoctl(fd, DRM_IOCTL_VC4_WAIT_SEQNO, &wait);
 491         if (ret == -1)
 492                 return -errno;
 493         else
 494                 return 0;
 495
 496 }
 497
 498 bool
 499 vc4_wait_seqno(struct vc4_screen *screen, uint64_t seqno, uint64_t timeout_ns,
 500                const char *reason)
 501 {
 502         if (screen->finished_seqno >= seqno)
 503                 return true;
 504
 505         if (unlikely(vc4_debug & VC4_DEBUG_PERF) && timeout_ns && reason) {
 506                 if (vc4_wait_seqno_ioctl(screen->fd, seqno, 0) == -ETIME) {
 507                         fprintf(stderr, "Blocking on seqno %lld for %s\n",
 508                                 (long long)seqno, reason);
 509                 }
 510         }
 511
 512         int ret = vc4_wait_seqno_ioctl(screen->fd, seqno, timeout_ns);
 513         if (ret) {
 514                 if (ret != -ETIME) {
 515                         fprintf(stderr, "wait failed: %d\n", ret);
 516                         abort();
 517                 }
 518
 519                 return false;
 520         }
 521
 522         screen->finished_seqno = seqno;
 523         return true;
 524 }
 525
 526 static int vc4_wait_bo_ioctl(int fd, uint32_t handle, uint64_t timeout_ns)
 527 {
 528         if (using_vc4_simulator)
 529                 return 0;
 530
 531         struct drm_vc4_wait_bo wait = {
 532                 .handle = handle,
 533                 .timeout_ns = timeout_ns,
 534         };
 535         int ret = drmIoctl(fd, DRM_IOCTL_VC4_WAIT_BO, &wait);
 536         if (ret == -1)
 537                 return -errno;
 538         else
 539                 return 0;
 540
 541 }
 542
 543 bool
 544 vc4_bo_wait(struct vc4_bo *bo, uint64_t timeout_ns, const char *reason)
 545 {
 546         struct vc4_screen *screen = bo->screen;
 547
 548         if (unlikely(vc4_debug & VC4_DEBUG_PERF) && timeout_ns && reason) {
 549                 if (vc4_wait_bo_ioctl(screen->fd, bo->handle, 0) == -ETIME) {
 550                         fprintf(stderr, "Blocking on %s BO for %s\n",
 551                                 bo->name, reason);
 552                 }
 553         }
 554
 555         int ret = vc4_wait_bo_ioctl(screen->fd, bo->handle, timeout_ns);
 556         if (ret) {
 557                 if (ret != -ETIME) {
 558                         fprintf(stderr, "wait failed: %d\n", ret);
 559                         abort();
 560                 }
 561
 562                 return false;
 563         }
 564
 565         return true;
 566 }
 567
 568 void *
 569 vc4_bo_map_unsynchronized(struct vc4_bo *bo)
 570 {
 571         uint64_t offset;
 572         int ret;
 573
 574         if (bo->map)
 575                 return bo->map;
 576
 577         if (!using_vc4_simulator) {
 578                 struct drm_vc4_mmap_bo map;
 579                 memset(&map, 0, sizeof(map));
 580                 map.handle = bo->handle;
 581                 ret = drmIoctl(bo->screen->fd, DRM_IOCTL_VC4_MMAP_BO, &map);
 582                 offset = map.offset;
 583         } else {
 584                 struct drm_mode_map_dumb map;
 585                 memset(&map, 0, sizeof(map));
 586                 map.handle = bo->handle;
 587                 ret = drmIoctl(bo->screen->fd, DRM_IOCTL_MODE_MAP_DUMB, &map);
 588                 offset = map.offset;
 589         }
 590         if (ret != 0) {
 591                 fprintf(stderr, "map ioctl failure\n");
 592                 abort();
 593         }
 594
 595         bo->map = mmap(NULL, bo->size, PROT_READ | PROT_WRITE, MAP_SHARED,
 596                        bo->screen->fd, offset);
 597         if (bo->map == MAP_FAILED) {
 598                 fprintf(stderr, "mmap of bo %d (offset 0x%016llx, size %d) failed\n",
 599                         bo->handle, (long long)offset, bo->size);
 600                 abort();
 601         }
 602
 603         return bo->map;
 604 }
 605
 606 void *
 607 vc4_bo_map(struct vc4_bo *bo)
 608 {
 609         void *map = vc4_bo_map_unsynchronized(bo);
 610
 611         bool ok = vc4_bo_wait(bo, PIPE_TIMEOUT_INFINITE, "bo map");
 612         if (!ok) {
 613                 fprintf(stderr, "BO wait for map failed\n");
 614                 abort();
 615         }
 616
 617         return map;
 618 }
 619
 620 void
 621 vc4_bufmgr_destroy(struct pipe_screen *pscreen)
 622 {
 623         struct vc4_screen *screen = vc4_screen(pscreen);
 624         struct vc4_bo_cache *cache = &screen->bo_cache;
 625
 626         vc4_bo_cache_free_all(cache);
 627
 628         if (dump_stats) {
 629                 fprintf(stderr, "BO stats after screen destroy:\n");
 630                 vc4_bo_dump_stats(screen);
 631         }
 632 }