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
 172                 free(bo);
 173                 return NULL;
 174         }
 175
 176         screen->bo_count++;
 177         screen->bo_size += bo->size;
 178         if (dump_stats) {
 179                 fprintf(stderr, "Allocated %s %dkb:\n", name, size / 1024);
 180                 vc4_bo_dump_stats(screen);
 181         }
 182
 183         return bo;
 184 }
 185
 186 void
 187 vc4_bo_last_unreference(struct vc4_bo *bo)
 188 {
 189         struct vc4_screen *screen = bo->screen;
 190
 191         struct timespec time;
 192         clock_gettime(CLOCK_MONOTONIC, &time);
 193         pipe_mutex_lock(screen->bo_cache.lock);
 194         vc4_bo_last_unreference_locked_timed(bo, time.tv_sec);
 195         pipe_mutex_unlock(screen->bo_cache.lock);
 196 }
 197
 198 static void
 199 vc4_bo_free(struct vc4_bo *bo)
 200 {
 201         struct vc4_screen *screen = bo->screen;
 202
 203         if (bo->map) {
 204 #ifdef USE_VC4_SIMULATOR
 205                 if (bo->simulator_winsys_map) {
 206                         free(bo->map);
 207                         bo->map = bo->simulator_winsys_map;
 208                 }
 209 #endif
 210                 munmap(bo->map, bo->size);
 211         }
 212
 213         struct drm_gem_close c;
 214         memset(&c, 0, sizeof(c));
 215         c.handle = bo->handle;
 216         int ret = drmIoctl(screen->fd, DRM_IOCTL_GEM_CLOSE, &c);
 217         if (ret != 0)
 218                 fprintf(stderr, "close object %d: %s\n", bo->handle, strerror(errno));
 219
 220         screen->bo_count--;
 221         screen->bo_size -= bo->size;
 222
 223         if (dump_stats) {
 224                 fprintf(stderr, "Freed %s%s%dkb:\n",
 225                         bo->name ? bo->name : "",
 226                         bo->name ? " " : "",
 227                         bo->size / 1024);
 228                 vc4_bo_dump_stats(screen);
 229         }
 230
 231         free(bo);
 232 }
 233
 234 static void
 235 free_stale_bos(struct vc4_screen *screen, time_t time)
 236 {
 237         struct vc4_bo_cache *cache = &screen->bo_cache;
 238         bool freed_any = false;
 239
 240         list_for_each_entry_safe(struct vc4_bo, bo, &cache->time_list,
 241                                  time_list) {
 242                 if (dump_stats && !freed_any) {
 243                         fprintf(stderr, "Freeing stale BOs:\n");
 244                         vc4_bo_dump_stats(screen);
 245                         freed_any = true;
 246                 }
 247
 248                 /* If it's more than a second old, free it. */
 249                 if (time - bo->free_time > 2) {
 250                         vc4_bo_remove_from_cache(cache, bo);
 251                         vc4_bo_free(bo);
 252                 } else {
 253                         break;
 254                 }
 255         }
 256
 257         if (dump_stats && freed_any) {
 258                 fprintf(stderr, "Freed stale BOs:\n");
 259                 vc4_bo_dump_stats(screen);
 260         }
 261 }
 262
 263 static void
 264 vc4_bo_cache_free_all(struct vc4_bo_cache *cache)
 265 {
 266         pipe_mutex_lock(cache->lock);
 267         list_for_each_entry_safe(struct vc4_bo, bo, &cache->time_list,
 268                                  time_list) {
 269                 vc4_bo_remove_from_cache(cache, bo);
 270                 vc4_bo_free(bo);
 271         }
 272         pipe_mutex_unlock(cache->lock);
 273 }
 274
 275 void
 276 vc4_bo_last_unreference_locked_timed(struct vc4_bo *bo, time_t time)
 277 {
 278         struct vc4_screen *screen = bo->screen;
 279         struct vc4_bo_cache *cache = &screen->bo_cache;
 280         uint32_t page_index = bo->size / 4096 - 1;
 281
 282         if (!bo->private) {
 283                 vc4_bo_free(bo);
 284                 return;
 285         }
 286
 287         if (cache->size_list_size <= page_index) {
 288                 struct list_head *new_list =
 289                         ralloc_array(screen, struct list_head, page_index + 1);
 290
 291                 /* Move old list contents over (since the array has moved, and
 292                  * therefore the pointers to the list heads have to change).
 293                  */
 294                 for (int i = 0; i < cache->size_list_size; i++) {
 295                         struct list_head *old_head = &cache->size_list[i];
 296                         if (list_empty(old_head))
 297                                 list_inithead(&new_list[i]);
 298                         else {
 299                                 new_list[i].next = old_head->next;
 300                                 new_list[i].prev = old_head->prev;
 301                                 new_list[i].next->prev = &new_list[i];
 302                                 new_list[i].prev->next = &new_list[i];
 303                         }
 304                 }
 305                 for (int i = cache->size_list_size; i < page_index + 1; i++)
 306                         list_inithead(&new_list[i]);
 307
 308                 cache->size_list = new_list;
 309                 cache->size_list_size = page_index + 1;
 310         }
 311
 312         bo->free_time = time;
 313         list_addtail(&bo->size_list, &cache->size_list[page_index]);
 314         list_addtail(&bo->time_list, &cache->time_list);
 315         cache->bo_count++;
 316         cache->bo_size += bo->size;
 317         if (dump_stats) {
 318                 fprintf(stderr, "Freed %s %dkb to cache:\n",
 319                         bo->name, bo->size / 1024);
 320                 vc4_bo_dump_stats(screen);
 321         }
 322         bo->name = NULL;
 323
 324         free_stale_bos(screen, time);
 325 }
 326
 327 static struct vc4_bo *
 328 vc4_bo_open_handle(struct vc4_screen *screen,
 329                    uint32_t winsys_stride,
 330                    uint32_t handle, uint32_t size)
 331 {
 332         struct vc4_bo *bo = CALLOC_STRUCT(vc4_bo);
 333
 334         assert(size);
 335
 336         pipe_reference_init(&bo->reference, 1);
 337         bo->screen = screen;
 338         bo->handle = handle;
 339         bo->size = size;
 340         bo->name = "winsys";
 341         bo->private = false;
 342
 343 #ifdef USE_VC4_SIMULATOR
 344         vc4_bo_map(bo);
 345         bo->simulator_winsys_map = bo->map;
 346         bo->simulator_winsys_stride = winsys_stride;
 347         bo->map = malloc(bo->size);
 348 #endif
 349
 350         return bo;
 351 }
 352
 353 struct vc4_bo *
 354 vc4_bo_open_name(struct vc4_screen *screen, uint32_t name,
 355                  uint32_t winsys_stride)
 356 {
 357         struct drm_gem_open o = {
 358                 .name = name
 359         };
 360         int ret = drmIoctl(screen->fd, DRM_IOCTL_GEM_OPEN, &o);
 361         if (ret) {
 362                 fprintf(stderr, "Failed to open bo %d: %s\n",
 363                         name, strerror(errno));
 364                 return NULL;
 365         }
 366
 367         return vc4_bo_open_handle(screen, winsys_stride, o.handle, o.size);
 368 }
 369
 370 struct vc4_bo *
 371 vc4_bo_open_dmabuf(struct vc4_screen *screen, int fd, uint32_t winsys_stride)
 372 {
 373         uint32_t handle;
 374         int ret = drmPrimeFDToHandle(screen->fd, fd, &handle);
 375         int size;
 376         if (ret) {
 377                 fprintf(stderr, "Failed to get vc4 handle for dmabuf %d\n", fd);
 378                 return NULL;
 379         }
 380
 381         /* Determine the size of the bo we were handed. */
 382         size = lseek(fd, 0, SEEK_END);
 383         if (size == -1) {
 384                 fprintf(stderr, "Couldn't get size of dmabuf fd %d.\n", fd);
 385                 return NULL;
 386         }
 387
 388         return vc4_bo_open_handle(screen, winsys_stride, handle, size);
 389 }
 390
 391 int
 392 vc4_bo_get_dmabuf(struct vc4_bo *bo)
 393 {
 394         int fd;
 395         int ret = drmPrimeHandleToFD(bo->screen->fd, bo->handle,
 396                                      O_CLOEXEC, &fd);
 397         if (ret != 0) {
 398                 fprintf(stderr, "Failed to export gem bo %d to dmabuf\n",
 399                         bo->handle);
 400                 return -1;
 401         }
 402         bo->private = false;
 403
 404         return fd;
 405 }
 406
 407 struct vc4_bo *
 408 vc4_bo_alloc_shader(struct vc4_screen *screen, const void *data, uint32_t size)
 409 {
 410         struct vc4_bo *bo;
 411         int ret;
 412
 413         bo = CALLOC_STRUCT(vc4_bo);
 414         if (!bo)
 415                 return NULL;
 416
 417         pipe_reference_init(&bo->reference, 1);
 418         bo->screen = screen;
 419         bo->size = align(size, 4096);
 420         bo->name = "code";
 421         bo->private = false; /* Make sure it doesn't go back to the cache. */
 422
 423         if (!using_vc4_simulator) {
 424                 struct drm_vc4_create_shader_bo create = {
 425                         .size = size,
 426                         .data = (uintptr_t)data,
 427                 };
 428
 429                 ret = drmIoctl(screen->fd, DRM_IOCTL_VC4_CREATE_SHADER_BO,
 430                                &create);
 431                 bo->handle = create.handle;
 432         } else {
 433                 struct drm_mode_create_dumb create;
 434                 memset(&create, 0, sizeof(create));
 435
 436                 create.width = 128;
 437                 create.bpp = 8;
 438                 create.height = (size + 127) / 128;
 439
 440                 ret = drmIoctl(screen->fd, DRM_IOCTL_MODE_CREATE_DUMB, &create);
 441                 bo->handle = create.handle;
 442                 assert(create.size >= size);
 443
 444                 vc4_bo_map(bo);
 445                 memcpy(bo->map, data, size);
 446         }
 447         if (ret != 0) {
 448                 fprintf(stderr, "create shader ioctl failure\n");
 449                 abort();
 450         }
 451
 452         screen->bo_count++;
 453         screen->bo_size += bo->size;
 454         if (dump_stats) {
 455                 fprintf(stderr, "Allocated shader %dkb:\n", bo->size / 1024);
 456                 vc4_bo_dump_stats(screen);
 457         }
 458
 459         return bo;
 460 }
 461
 462 bool
 463 vc4_bo_flink(struct vc4_bo *bo, uint32_t *name)
 464 {
 465         struct drm_gem_flink flink = {
 466                 .handle = bo->handle,
 467         };
 468         int ret = drmIoctl(bo->screen->fd, DRM_IOCTL_GEM_FLINK, &flink);
 469         if (ret) {
 470                 fprintf(stderr, "Failed to flink bo %d: %s\n",
 471                         bo->handle, strerror(errno));
 472                 free(bo);
 473                 return false;
 474         }
 475
 476         bo->private = false;
 477         *name = flink.name;
 478
 479         return true;
 480 }
 481
 482 static int vc4_wait_seqno_ioctl(int fd, uint64_t seqno, uint64_t timeout_ns)
 483 {
 484         if (using_vc4_simulator)
 485                 return 0;
 486
 487         struct drm_vc4_wait_seqno wait = {
 488                 .seqno = seqno,
 489                 .timeout_ns = timeout_ns,
 490         };
 491         int ret = drmIoctl(fd, DRM_IOCTL_VC4_WAIT_SEQNO, &wait);
 492         if (ret == -1)
 493                 return -errno;
 494         else
 495                 return 0;
 496
 497 }
 498
 499 bool
 500 vc4_wait_seqno(struct vc4_screen *screen, uint64_t seqno, uint64_t timeout_ns,
 501                const char *reason)
 502 {
 503         if (screen->finished_seqno >= seqno)
 504                 return true;
 505
 506         if (unlikely(vc4_debug & VC4_DEBUG_PERF) && timeout_ns && reason) {
 507                 if (vc4_wait_seqno_ioctl(screen->fd, seqno, 0) == -ETIME) {
 508                         fprintf(stderr, "Blocking on seqno %lld for %s\n",
 509                                 (long long)seqno, reason);
 510                 }
 511         }
 512
 513         int ret = vc4_wait_seqno_ioctl(screen->fd, seqno, timeout_ns);
 514         if (ret) {
 515                 if (ret != -ETIME) {
 516                         fprintf(stderr, "wait failed: %d\n", ret);
 517                         abort();
 518                 }
 519
 520                 return false;
 521         }
 522
 523         screen->finished_seqno = seqno;
 524         return true;
 525 }
 526
 527 static int vc4_wait_bo_ioctl(int fd, uint32_t handle, uint64_t timeout_ns)
 528 {
 529         if (using_vc4_simulator)
 530                 return 0;
 531
 532         struct drm_vc4_wait_bo wait = {
 533                 .handle = handle,
 534                 .timeout_ns = timeout_ns,
 535         };
 536         int ret = drmIoctl(fd, DRM_IOCTL_VC4_WAIT_BO, &wait);
 537         if (ret == -1)
 538                 return -errno;
 539         else
 540                 return 0;
 541
 542 }
 543
 544 bool
 545 vc4_bo_wait(struct vc4_bo *bo, uint64_t timeout_ns, const char *reason)
 546 {
 547         struct vc4_screen *screen = bo->screen;
 548
 549         if (unlikely(vc4_debug & VC4_DEBUG_PERF) && timeout_ns && reason) {
 550                 if (vc4_wait_bo_ioctl(screen->fd, bo->handle, 0) == -ETIME) {
 551                         fprintf(stderr, "Blocking on %s BO for %s\n",
 552                                 bo->name, reason);
 553                 }
 554         }
 555
 556         int ret = vc4_wait_bo_ioctl(screen->fd, bo->handle, timeout_ns);
 557         if (ret) {
 558                 if (ret != -ETIME) {
 559                         fprintf(stderr, "wait failed: %d\n", ret);
 560                         abort();
 561                 }
 562
 563                 return false;
 564         }
 565
 566         return true;
 567 }
 568
 569 void *
 570 vc4_bo_map_unsynchronized(struct vc4_bo *bo)
 571 {
 572         uint64_t offset;
 573         int ret;
 574
 575         if (bo->map)
 576                 return bo->map;
 577
 578         if (!using_vc4_simulator) {
 579                 struct drm_vc4_mmap_bo map;
 580                 memset(&map, 0, sizeof(map));
 581                 map.handle = bo->handle;
 582                 ret = drmIoctl(bo->screen->fd, DRM_IOCTL_VC4_MMAP_BO, &map);
 583                 offset = map.offset;
 584         } else {
 585                 struct drm_mode_map_dumb map;
 586                 memset(&map, 0, sizeof(map));
 587                 map.handle = bo->handle;
 588                 ret = drmIoctl(bo->screen->fd, DRM_IOCTL_MODE_MAP_DUMB, &map);
 589                 offset = map.offset;
 590         }
 591         if (ret != 0) {
 592                 fprintf(stderr, "map ioctl failure\n");
 593                 abort();
 594         }
 595
 596         bo->map = mmap(NULL, bo->size, PROT_READ | PROT_WRITE, MAP_SHARED,
 597                        bo->screen->fd, offset);
 598         if (bo->map == MAP_FAILED) {
 599                 fprintf(stderr, "mmap of bo %d (offset 0x%016llx, size %d) failed\n",
 600                         bo->handle, (long long)offset, bo->size);
 601                 abort();
 602         }
 603
 604         return bo->map;
 605 }
 606
 607 void *
 608 vc4_bo_map(struct vc4_bo *bo)
 609 {
 610         void *map = vc4_bo_map_unsynchronized(bo);
 611
 612         bool ok = vc4_bo_wait(bo, PIPE_TIMEOUT_INFINITE, "bo map");
 613         if (!ok) {
 614                 fprintf(stderr, "BO wait for map failed\n");
 615                 abort();
 616         }
 617
 618         return map;
 619 }
 620
 621 void
 622 vc4_bufmgr_destroy(struct pipe_screen *pscreen)
 623 {
 624         struct vc4_screen *screen = vc4_screen(pscreen);
 625         struct vc4_bo_cache *cache = &screen->bo_cache;
 626
 627         vc4_bo_cache_free_all(cache);
 628
 629         if (dump_stats) {
 630                 fprintf(stderr, "BO stats after screen destroy:\n");
 631                 vc4_bo_dump_stats(screen);
 632         }
 633 }