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_hash_table.h"
  32 #include "util/u_memory.h"
  33 #include "util/ralloc.h"
  34
  35 #include "vc4_context.h"
  36 #include "vc4_screen.h"
  37
  38 #ifdef HAVE_VALGRIND
  39 #include <valgrind.h>
  40 #include <memcheck.h>
  41 #define VG(x) x
  42 #else
  43 #define VG(x)
  44 #endif
  45
  46 static bool dump_stats = false;
  47
  48 static void
  49 vc4_bo_cache_free_all(struct vc4_bo_cache *cache);
  50
  51 static void
  52 vc4_bo_dump_stats(struct vc4_screen *screen)
  53 {
  54         struct vc4_bo_cache *cache = &screen->bo_cache;
  55
  56         fprintf(stderr, "  BOs allocated:   %d\n", screen->bo_count);
  57         fprintf(stderr, "  BOs size:        %dkb\n", screen->bo_size / 1024);
  58         fprintf(stderr, "  BOs cached:      %d\n", cache->bo_count);
  59         fprintf(stderr, "  BOs cached size: %dkb\n", cache->bo_size / 1024);
  60
  61         if (!list_empty(&cache->time_list)) {
  62                 struct vc4_bo *first = LIST_ENTRY(struct vc4_bo,
  63                                                   cache->time_list.next,
  64                                                   time_list);
  65                 struct vc4_bo *last = LIST_ENTRY(struct vc4_bo,
  66                                                   cache->time_list.prev,
  67                                                   time_list);
  68
  69                 fprintf(stderr, "  oldest cache time: %ld\n",
  70                         (long)first->free_time);
  71                 fprintf(stderr, "  newest cache time: %ld\n",
  72                         (long)last->free_time);
  73
  74                 struct timespec time;
  75                 clock_gettime(CLOCK_MONOTONIC, &time);
  76                 fprintf(stderr, "  now:               %ld\n",
  77                         time.tv_sec);
  78         }
  79 }
  80
  81 static void
  82 vc4_bo_remove_from_cache(struct vc4_bo_cache *cache, struct vc4_bo *bo)
  83 {
  84         list_del(&bo->time_list);
  85         list_del(&bo->size_list);
  86         cache->bo_count--;
  87         cache->bo_size -= bo->size;
  88 }
  89
  90 static struct vc4_bo *
  91 vc4_bo_from_cache(struct vc4_screen *screen, uint32_t size, const char *name)
  92 {
  93         struct vc4_bo_cache *cache = &screen->bo_cache;
  94         uint32_t page_index = size / 4096 - 1;
  95
  96         if (cache->size_list_size <= page_index)
  97                 return NULL;
  98
  99         struct vc4_bo *bo = NULL;
 100         pipe_mutex_lock(cache->lock);
 101         if (!list_empty(&cache->size_list[page_index])) {
 102                 bo = LIST_ENTRY(struct vc4_bo, cache->size_list[page_index].next,
 103                                 size_list);
 104
 105                 /* Check that the BO has gone idle.  If not, then we want to
 106                  * allocate something new instead, since we assume that the
 107                  * user will proceed to CPU map it and fill it with stuff.
 108                  */
 109                 if (!vc4_bo_wait(bo, 0, NULL)) {
 110                         pipe_mutex_unlock(cache->lock);
 111                         return NULL;
 112                 }
 113
 114                 pipe_reference_init(&bo->reference, 1);
 115                 vc4_bo_remove_from_cache(cache, bo);
 116
 117                 bo->name = name;
 118         }
 119         pipe_mutex_unlock(cache->lock);
 120         return bo;
 121 }
 122
 123 struct vc4_bo *
 124 vc4_bo_alloc(struct vc4_screen *screen, uint32_t size, const char *name)
 125 {
 126         struct vc4_bo *bo;
 127         int ret;
 128
 129         size = align(size, 4096);
 130
 131         bo = vc4_bo_from_cache(screen, size, name);
 132         if (bo) {
 133                 if (dump_stats) {
 134                         fprintf(stderr, "Allocated %s %dkb from cache:\n",
 135                                 name, size / 1024);
 136                         vc4_bo_dump_stats(screen);
 137                 }
 138                 return bo;
 139         }
 140
 141         bo = CALLOC_STRUCT(vc4_bo);
 142         if (!bo)
 143                 return NULL;
 144
 145         pipe_reference_init(&bo->reference, 1);
 146         bo->screen = screen;
 147         bo->size = size;
 148         bo->name = name;
 149         bo->private = true;
 150
 151  retry:
 152         ;
 153
 154         bool cleared_and_retried = false;
 155         struct drm_vc4_create_bo create = {
 156                 .size = size
 157         };
 158
 159         ret = vc4_ioctl(screen->fd, DRM_IOCTL_VC4_CREATE_BO, &create);
 160         bo->handle = create.handle;
 161
 162         if (ret != 0) {
 163                 if (!list_empty(&screen->bo_cache.time_list) &&
 164                     !cleared_and_retried) {
 165                         cleared_and_retried = true;
 166                         vc4_bo_cache_free_all(&screen->bo_cache);
 167                         goto retry;
 168                 }
 169
 170                 free(bo);
 171                 return NULL;
 172         }
 173
 174         screen->bo_count++;
 175         screen->bo_size += bo->size;
 176         if (dump_stats) {
 177                 fprintf(stderr, "Allocated %s %dkb:\n", name, size / 1024);
 178                 vc4_bo_dump_stats(screen);
 179         }
 180
 181         return bo;
 182 }
 183
 184 void
 185 vc4_bo_last_unreference(struct vc4_bo *bo)
 186 {
 187         struct vc4_screen *screen = bo->screen;
 188
 189         struct timespec time;
 190         clock_gettime(CLOCK_MONOTONIC, &time);
 191         pipe_mutex_lock(screen->bo_cache.lock);
 192         vc4_bo_last_unreference_locked_timed(bo, time.tv_sec);
 193         pipe_mutex_unlock(screen->bo_cache.lock);
 194 }
 195
 196 static void
 197 vc4_bo_free(struct vc4_bo *bo)
 198 {
 199         struct vc4_screen *screen = bo->screen;
 200
 201         if (bo->map) {
 202                 if (using_vc4_simulator && bo->name &&
 203                     strcmp(bo->name, "winsys") == 0) {
 204                         free(bo->map);
 205                 } else {
 206                         munmap(bo->map, bo->size);
 207                         VG(VALGRIND_FREELIKE_BLOCK(bo->map, 0));
 208                 }
 209         }
 210
 211         struct drm_gem_close c;
 212         memset(&c, 0, sizeof(c));
 213         c.handle = bo->handle;
 214         int ret = vc4_ioctl(screen->fd, DRM_IOCTL_GEM_CLOSE, &c);
 215         if (ret != 0)
 216                 fprintf(stderr, "close object %d: %s\n", bo->handle, strerror(errno));
 217
 218         screen->bo_count--;
 219         screen->bo_size -= bo->size;
 220
 221         if (dump_stats) {
 222                 fprintf(stderr, "Freed %s%s%dkb:\n",
 223                         bo->name ? bo->name : "",
 224                         bo->name ? " " : "",
 225                         bo->size / 1024);
 226                 vc4_bo_dump_stats(screen);
 227         }
 228
 229         free(bo);
 230 }
 231
 232 static void
 233 free_stale_bos(struct vc4_screen *screen, time_t time)
 234 {
 235         struct vc4_bo_cache *cache = &screen->bo_cache;
 236         bool freed_any = false;
 237
 238         list_for_each_entry_safe(struct vc4_bo, bo, &cache->time_list,
 239                                  time_list) {
 240                 if (dump_stats && !freed_any) {
 241                         fprintf(stderr, "Freeing stale BOs:\n");
 242                         vc4_bo_dump_stats(screen);
 243                         freed_any = true;
 244                 }
 245
 246                 /* If it's more than a second old, free it. */
 247                 if (time - bo->free_time > 2) {
 248                         vc4_bo_remove_from_cache(cache, bo);
 249                         vc4_bo_free(bo);
 250                 } else {
 251                         break;
 252                 }
 253         }
 254
 255         if (dump_stats && freed_any) {
 256                 fprintf(stderr, "Freed stale BOs:\n");
 257                 vc4_bo_dump_stats(screen);
 258         }
 259 }
 260
 261 static void
 262 vc4_bo_cache_free_all(struct vc4_bo_cache *cache)
 263 {
 264         pipe_mutex_lock(cache->lock);
 265         list_for_each_entry_safe(struct vc4_bo, bo, &cache->time_list,
 266                                  time_list) {
 267                 vc4_bo_remove_from_cache(cache, bo);
 268                 vc4_bo_free(bo);
 269         }
 270         pipe_mutex_unlock(cache->lock);
 271 }
 272
 273 void
 274 vc4_bo_last_unreference_locked_timed(struct vc4_bo *bo, time_t time)
 275 {
 276         struct vc4_screen *screen = bo->screen;
 277         struct vc4_bo_cache *cache = &screen->bo_cache;
 278         uint32_t page_index = bo->size / 4096 - 1;
 279
 280         if (!bo->private) {
 281                 vc4_bo_free(bo);
 282                 return;
 283         }
 284
 285         if (cache->size_list_size <= page_index) {
 286                 struct list_head *new_list =
 287                         ralloc_array(screen, struct list_head, page_index + 1);
 288
 289                 /* Move old list contents over (since the array has moved, and
 290                  * therefore the pointers to the list heads have to change).
 291                  */
 292                 for (int i = 0; i < cache->size_list_size; i++) {
 293                         struct list_head *old_head = &cache->size_list[i];
 294                         if (list_empty(old_head))
 295                                 list_inithead(&new_list[i]);
 296                         else {
 297                                 new_list[i].next = old_head->next;
 298                                 new_list[i].prev = old_head->prev;
 299                                 new_list[i].next->prev = &new_list[i];
 300                                 new_list[i].prev->next = &new_list[i];
 301                         }
 302                 }
 303                 for (int i = cache->size_list_size; i < page_index + 1; i++)
 304                         list_inithead(&new_list[i]);
 305
 306                 cache->size_list = new_list;
 307                 cache->size_list_size = page_index + 1;
 308         }
 309
 310         bo->free_time = time;
 311         list_addtail(&bo->size_list, &cache->size_list[page_index]);
 312         list_addtail(&bo->time_list, &cache->time_list);
 313         cache->bo_count++;
 314         cache->bo_size += bo->size;
 315         if (dump_stats) {
 316                 fprintf(stderr, "Freed %s %dkb to cache:\n",
 317                         bo->name, bo->size / 1024);
 318                 vc4_bo_dump_stats(screen);
 319         }
 320         bo->name = NULL;
 321
 322         free_stale_bos(screen, time);
 323 }
 324
 325 static struct vc4_bo *
 326 vc4_bo_open_handle(struct vc4_screen *screen,
 327                    uint32_t winsys_stride,
 328                    uint32_t handle, uint32_t size)
 329 {
 330         struct vc4_bo *bo;
 331
 332         assert(size);
 333
 334         pipe_mutex_lock(screen->bo_handles_mutex);
 335
 336         bo = util_hash_table_get(screen->bo_handles, (void*)(uintptr_t)handle);
 337         if (bo) {
 338                 pipe_reference(NULL, &bo->reference);
 339                 goto done;
 340         }
 341
 342         bo = CALLOC_STRUCT(vc4_bo);
 343         pipe_reference_init(&bo->reference, 1);
 344         bo->screen = screen;
 345         bo->handle = handle;
 346         bo->size = size;
 347         bo->name = "winsys";
 348         bo->private = false;
 349
 350 #ifdef USE_VC4_SIMULATOR
 351         vc4_simulator_open_from_handle(screen->fd, winsys_stride,
 352                                        bo->handle, bo->size);
 353         bo->map = malloc(bo->size);
 354 #endif
 355
 356         util_hash_table_set(screen->bo_handles, (void *)(uintptr_t)handle, bo);
 357
 358 done:
 359         pipe_mutex_unlock(screen->bo_handles_mutex);
 360         return bo;
 361 }
 362
 363 struct vc4_bo *
 364 vc4_bo_open_name(struct vc4_screen *screen, uint32_t name,
 365                  uint32_t winsys_stride)
 366 {
 367         struct drm_gem_open o = {
 368                 .name = name
 369         };
 370         int ret = vc4_ioctl(screen->fd, DRM_IOCTL_GEM_OPEN, &o);
 371         if (ret) {
 372                 fprintf(stderr, "Failed to open bo %d: %s\n",
 373                         name, strerror(errno));
 374                 return NULL;
 375         }
 376
 377         return vc4_bo_open_handle(screen, winsys_stride, o.handle, o.size);
 378 }
 379
 380 struct vc4_bo *
 381 vc4_bo_open_dmabuf(struct vc4_screen *screen, int fd, uint32_t winsys_stride)
 382 {
 383         uint32_t handle;
 384         int ret = drmPrimeFDToHandle(screen->fd, fd, &handle);
 385         int size;
 386         if (ret) {
 387                 fprintf(stderr, "Failed to get vc4 handle for dmabuf %d\n", fd);
 388                 return NULL;
 389         }
 390
 391         /* Determine the size of the bo we were handed. */
 392         size = lseek(fd, 0, SEEK_END);
 393         if (size == -1) {
 394                 fprintf(stderr, "Couldn't get size of dmabuf fd %d.\n", fd);
 395                 return NULL;
 396         }
 397
 398         return vc4_bo_open_handle(screen, winsys_stride, handle, size);
 399 }
 400
 401 int
 402 vc4_bo_get_dmabuf(struct vc4_bo *bo)
 403 {
 404         int fd;
 405         int ret = drmPrimeHandleToFD(bo->screen->fd, bo->handle,
 406                                      O_CLOEXEC, &fd);
 407         if (ret != 0) {
 408                 fprintf(stderr, "Failed to export gem bo %d to dmabuf\n",
 409                         bo->handle);
 410                 return -1;
 411         }
 412
 413         pipe_mutex_lock(bo->screen->bo_handles_mutex);
 414         bo->private = false;
 415         util_hash_table_set(bo->screen->bo_handles, (void *)(uintptr_t)bo->handle, bo);
 416         pipe_mutex_unlock(bo->screen->bo_handles_mutex);
 417
 418         return fd;
 419 }
 420
 421 struct vc4_bo *
 422 vc4_bo_alloc_shader(struct vc4_screen *screen, const void *data, uint32_t size)
 423 {
 424         struct vc4_bo *bo;
 425         int ret;
 426
 427         bo = CALLOC_STRUCT(vc4_bo);
 428         if (!bo)
 429                 return NULL;
 430
 431         pipe_reference_init(&bo->reference, 1);
 432         bo->screen = screen;
 433         bo->size = align(size, 4096);
 434         bo->name = "code";
 435         bo->private = false; /* Make sure it doesn't go back to the cache. */
 436
 437         struct drm_vc4_create_shader_bo create = {
 438                 .size = size,
 439                 .data = (uintptr_t)data,
 440         };
 441
 442         ret = vc4_ioctl(screen->fd, DRM_IOCTL_VC4_CREATE_SHADER_BO,
 443                         &create);
 444         bo->handle = create.handle;
 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 = vc4_ioctl(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         struct drm_vc4_wait_seqno wait = {
 484                 .seqno = seqno,
 485                 .timeout_ns = timeout_ns,
 486         };
 487         int ret = vc4_ioctl(fd, DRM_IOCTL_VC4_WAIT_SEQNO, &wait);
 488         if (ret == -1)
 489                 return -errno;
 490         else
 491                 return 0;
 492
 493 }
 494
 495 bool
 496 vc4_wait_seqno(struct vc4_screen *screen, uint64_t seqno, uint64_t timeout_ns,
 497                const char *reason)
 498 {
 499         if (screen->finished_seqno >= seqno)
 500                 return true;
 501
 502         if (unlikely(vc4_debug & VC4_DEBUG_PERF) && timeout_ns && reason) {
 503                 if (vc4_wait_seqno_ioctl(screen->fd, seqno, 0) == -ETIME) {
 504                         fprintf(stderr, "Blocking on seqno %lld for %s\n",
 505                                 (long long)seqno, reason);
 506                 }
 507         }
 508
 509         int ret = vc4_wait_seqno_ioctl(screen->fd, seqno, timeout_ns);
 510         if (ret) {
 511                 if (ret != -ETIME) {
 512                         fprintf(stderr, "wait failed: %d\n", ret);
 513                         abort();
 514                 }
 515
 516                 return false;
 517         }
 518
 519         screen->finished_seqno = seqno;
 520         return true;
 521 }
 522
 523 static int vc4_wait_bo_ioctl(int fd, uint32_t handle, uint64_t timeout_ns)
 524 {
 525         struct drm_vc4_wait_bo wait = {
 526                 .handle = handle,
 527                 .timeout_ns = timeout_ns,
 528         };
 529         int ret = vc4_ioctl(fd, DRM_IOCTL_VC4_WAIT_BO, &wait);
 530         if (ret == -1)
 531                 return -errno;
 532         else
 533                 return 0;
 534
 535 }
 536
 537 bool
 538 vc4_bo_wait(struct vc4_bo *bo, uint64_t timeout_ns, const char *reason)
 539 {
 540         struct vc4_screen *screen = bo->screen;
 541
 542         if (unlikely(vc4_debug & VC4_DEBUG_PERF) && timeout_ns && reason) {
 543                 if (vc4_wait_bo_ioctl(screen->fd, bo->handle, 0) == -ETIME) {
 544                         fprintf(stderr, "Blocking on %s BO for %s\n",
 545                                 bo->name, reason);
 546                 }
 547         }
 548
 549         int ret = vc4_wait_bo_ioctl(screen->fd, bo->handle, timeout_ns);
 550         if (ret) {
 551                 if (ret != -ETIME) {
 552                         fprintf(stderr, "wait failed: %d\n", ret);
 553                         abort();
 554                 }
 555
 556                 return false;
 557         }
 558
 559         return true;
 560 }
 561
 562 void *
 563 vc4_bo_map_unsynchronized(struct vc4_bo *bo)
 564 {
 565         uint64_t offset;
 566         int ret;
 567
 568         if (bo->map)
 569                 return bo->map;
 570
 571         struct drm_vc4_mmap_bo map;
 572         memset(&map, 0, sizeof(map));
 573         map.handle = bo->handle;
 574         ret = vc4_ioctl(bo->screen->fd, DRM_IOCTL_VC4_MMAP_BO, &map);
 575         offset = map.offset;
 576         if (ret != 0) {
 577                 fprintf(stderr, "map ioctl failure\n");
 578                 abort();
 579         }
 580
 581         bo->map = mmap(NULL, bo->size, PROT_READ | PROT_WRITE, MAP_SHARED,
 582                        bo->screen->fd, offset);
 583         if (bo->map == MAP_FAILED) {
 584                 fprintf(stderr, "mmap of bo %d (offset 0x%016llx, size %d) failed\n",
 585                         bo->handle, (long long)offset, bo->size);
 586                 abort();
 587         }
 588         VG(VALGRIND_MALLOCLIKE_BLOCK(bo->map, bo->size, 0, false));
 589
 590         return bo->map;
 591 }
 592
 593 void *
 594 vc4_bo_map(struct vc4_bo *bo)
 595 {
 596         void *map = vc4_bo_map_unsynchronized(bo);
 597
 598         bool ok = vc4_bo_wait(bo, PIPE_TIMEOUT_INFINITE, "bo map");
 599         if (!ok) {
 600                 fprintf(stderr, "BO wait for map failed\n");
 601                 abort();
 602         }
 603
 604         return map;
 605 }
 606
 607 void
 608 vc4_bufmgr_destroy(struct pipe_screen *pscreen)
 609 {
 610         struct vc4_screen *screen = vc4_screen(pscreen);
 611         struct vc4_bo_cache *cache = &screen->bo_cache;
 612
 613         vc4_bo_cache_free_all(cache);
 614
 615         if (dump_stats) {
 616                 fprintf(stderr, "BO stats after screen destroy:\n");
 617                 vc4_bo_dump_stats(screen);
 618         }
 619 }