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 #ifdef USE_VC4_SIMULATOR
 203                 if (bo->simulator_winsys_map) {
 204                         free(bo->map);
 205                         bo->map = bo->simulator_winsys_map;
 206                 }
 207 #endif
 208                 munmap(bo->map, bo->size);
 209                 VG(VALGRIND_FREELIKE_BLOCK(bo->map, 0));
 210         }
 211
 212         struct drm_gem_close c;
 213         memset(&c, 0, sizeof(c));
 214         c.handle = bo->handle;
 215         int ret = vc4_ioctl(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;
 332
 333         assert(size);
 334
 335         pipe_mutex_lock(screen->bo_handles_mutex);
 336
 337         bo = util_hash_table_get(screen->bo_handles, (void*)(uintptr_t)handle);
 338         if (bo) {
 339                 pipe_reference(NULL, &bo->reference);
 340                 goto done;
 341         }
 342
 343         bo = CALLOC_STRUCT(vc4_bo);
 344         pipe_reference_init(&bo->reference, 1);
 345         bo->screen = screen;
 346         bo->handle = handle;
 347         bo->size = size;
 348         bo->name = "winsys";
 349         bo->private = false;
 350
 351 #ifdef USE_VC4_SIMULATOR
 352         vc4_bo_map(bo);
 353         bo->simulator_winsys_map = bo->map;
 354         bo->simulator_winsys_stride = winsys_stride;
 355         bo->map = malloc(bo->size);
 356 #endif
 357
 358         util_hash_table_set(screen->bo_handles, (void *)(uintptr_t)handle, bo);
 359
 360 done:
 361         pipe_mutex_unlock(screen->bo_handles_mutex);
 362         return bo;
 363 }
 364
 365 struct vc4_bo *
 366 vc4_bo_open_name(struct vc4_screen *screen, uint32_t name,
 367                  uint32_t winsys_stride)
 368 {
 369         struct drm_gem_open o = {
 370                 .name = name
 371         };
 372         int ret = vc4_ioctl(screen->fd, DRM_IOCTL_GEM_OPEN, &o);
 373         if (ret) {
 374                 fprintf(stderr, "Failed to open bo %d: %s\n",
 375                         name, strerror(errno));
 376                 return NULL;
 377         }
 378
 379         return vc4_bo_open_handle(screen, winsys_stride, o.handle, o.size);
 380 }
 381
 382 struct vc4_bo *
 383 vc4_bo_open_dmabuf(struct vc4_screen *screen, int fd, uint32_t winsys_stride)
 384 {
 385         uint32_t handle;
 386         int ret = drmPrimeFDToHandle(screen->fd, fd, &handle);
 387         int size;
 388         if (ret) {
 389                 fprintf(stderr, "Failed to get vc4 handle for dmabuf %d\n", fd);
 390                 return NULL;
 391         }
 392
 393         /* Determine the size of the bo we were handed. */
 394         size = lseek(fd, 0, SEEK_END);
 395         if (size == -1) {
 396                 fprintf(stderr, "Couldn't get size of dmabuf fd %d.\n", fd);
 397                 return NULL;
 398         }
 399
 400         return vc4_bo_open_handle(screen, winsys_stride, handle, size);
 401 }
 402
 403 int
 404 vc4_bo_get_dmabuf(struct vc4_bo *bo)
 405 {
 406         int fd;
 407         int ret = drmPrimeHandleToFD(bo->screen->fd, bo->handle,
 408                                      O_CLOEXEC, &fd);
 409         if (ret != 0) {
 410                 fprintf(stderr, "Failed to export gem bo %d to dmabuf\n",
 411                         bo->handle);
 412                 return -1;
 413         }
 414
 415         pipe_mutex_lock(bo->screen->bo_handles_mutex);
 416         bo->private = false;
 417         util_hash_table_set(bo->screen->bo_handles, (void *)(uintptr_t)bo->handle, bo);
 418         pipe_mutex_unlock(bo->screen->bo_handles_mutex);
 419
 420         return fd;
 421 }
 422
 423 struct vc4_bo *
 424 vc4_bo_alloc_shader(struct vc4_screen *screen, const void *data, uint32_t size)
 425 {
 426         struct vc4_bo *bo;
 427         int ret;
 428
 429         bo = CALLOC_STRUCT(vc4_bo);
 430         if (!bo)
 431                 return NULL;
 432
 433         pipe_reference_init(&bo->reference, 1);
 434         bo->screen = screen;
 435         bo->size = align(size, 4096);
 436         bo->name = "code";
 437         bo->private = false; /* Make sure it doesn't go back to the cache. */
 438
 439         struct drm_vc4_create_shader_bo create = {
 440                 .size = size,
 441                 .data = (uintptr_t)data,
 442         };
 443
 444         ret = vc4_ioctl(screen->fd, DRM_IOCTL_VC4_CREATE_SHADER_BO,
 445                         &create);
 446         bo->handle = create.handle;
 447
 448         if (ret != 0) {
 449                 fprintf(stderr, "create shader ioctl failure\n");
 450                 abort();
 451         }
 452
 453         screen->bo_count++;
 454         screen->bo_size += bo->size;
 455         if (dump_stats) {
 456                 fprintf(stderr, "Allocated shader %dkb:\n", bo->size / 1024);
 457                 vc4_bo_dump_stats(screen);
 458         }
 459
 460         return bo;
 461 }
 462
 463 bool
 464 vc4_bo_flink(struct vc4_bo *bo, uint32_t *name)
 465 {
 466         struct drm_gem_flink flink = {
 467                 .handle = bo->handle,
 468         };
 469         int ret = vc4_ioctl(bo->screen->fd, DRM_IOCTL_GEM_FLINK, &flink);
 470         if (ret) {
 471                 fprintf(stderr, "Failed to flink bo %d: %s\n",
 472                         bo->handle, strerror(errno));
 473                 free(bo);
 474                 return false;
 475         }
 476
 477         bo->private = false;
 478         *name = flink.name;
 479
 480         return true;
 481 }
 482
 483 static int vc4_wait_seqno_ioctl(int fd, uint64_t seqno, uint64_t timeout_ns)
 484 {
 485         struct drm_vc4_wait_seqno wait = {
 486                 .seqno = seqno,
 487                 .timeout_ns = timeout_ns,
 488         };
 489         int ret = vc4_ioctl(fd, DRM_IOCTL_VC4_WAIT_SEQNO, &wait);
 490         if (ret == -1)
 491                 return -errno;
 492         else
 493                 return 0;
 494
 495 }
 496
 497 bool
 498 vc4_wait_seqno(struct vc4_screen *screen, uint64_t seqno, uint64_t timeout_ns,
 499                const char *reason)
 500 {
 501         if (screen->finished_seqno >= seqno)
 502                 return true;
 503
 504         if (unlikely(vc4_debug & VC4_DEBUG_PERF) && timeout_ns && reason) {
 505                 if (vc4_wait_seqno_ioctl(screen->fd, seqno, 0) == -ETIME) {
 506                         fprintf(stderr, "Blocking on seqno %lld for %s\n",
 507                                 (long long)seqno, reason);
 508                 }
 509         }
 510
 511         int ret = vc4_wait_seqno_ioctl(screen->fd, seqno, timeout_ns);
 512         if (ret) {
 513                 if (ret != -ETIME) {
 514                         fprintf(stderr, "wait failed: %d\n", ret);
 515                         abort();
 516                 }
 517
 518                 return false;
 519         }
 520
 521         screen->finished_seqno = seqno;
 522         return true;
 523 }
 524
 525 static int vc4_wait_bo_ioctl(int fd, uint32_t handle, uint64_t timeout_ns)
 526 {
 527         struct drm_vc4_wait_bo wait = {
 528                 .handle = handle,
 529                 .timeout_ns = timeout_ns,
 530         };
 531         int ret = vc4_ioctl(fd, DRM_IOCTL_VC4_WAIT_BO, &wait);
 532         if (ret == -1)
 533                 return -errno;
 534         else
 535                 return 0;
 536
 537 }
 538
 539 bool
 540 vc4_bo_wait(struct vc4_bo *bo, uint64_t timeout_ns, const char *reason)
 541 {
 542         struct vc4_screen *screen = bo->screen;
 543
 544         if (unlikely(vc4_debug & VC4_DEBUG_PERF) && timeout_ns && reason) {
 545                 if (vc4_wait_bo_ioctl(screen->fd, bo->handle, 0) == -ETIME) {
 546                         fprintf(stderr, "Blocking on %s BO for %s\n",
 547                                 bo->name, reason);
 548                 }
 549         }
 550
 551         int ret = vc4_wait_bo_ioctl(screen->fd, bo->handle, timeout_ns);
 552         if (ret) {
 553                 if (ret != -ETIME) {
 554                         fprintf(stderr, "wait failed: %d\n", ret);
 555                         abort();
 556                 }
 557
 558                 return false;
 559         }
 560
 561         return true;
 562 }
 563
 564 void *
 565 vc4_bo_map_unsynchronized(struct vc4_bo *bo)
 566 {
 567         uint64_t offset;
 568         int ret;
 569
 570         if (bo->map)
 571                 return bo->map;
 572
 573         struct drm_vc4_mmap_bo map;
 574         memset(&map, 0, sizeof(map));
 575         map.handle = bo->handle;
 576         ret = vc4_ioctl(bo->screen->fd, DRM_IOCTL_VC4_MMAP_BO, &map);
 577         offset = map.offset;
 578         if (ret != 0) {
 579                 fprintf(stderr, "map ioctl failure\n");
 580                 abort();
 581         }
 582
 583         bo->map = mmap(NULL, bo->size, PROT_READ | PROT_WRITE, MAP_SHARED,
 584                        bo->screen->fd, offset);
 585         if (bo->map == MAP_FAILED) {
 586                 fprintf(stderr, "mmap of bo %d (offset 0x%016llx, size %d) failed\n",
 587                         bo->handle, (long long)offset, bo->size);
 588                 abort();
 589         }
 590         VG(VALGRIND_MALLOCLIKE_BLOCK(bo->map, bo->size, 0, false));
 591
 592         return bo->map;
 593 }
 594
 595 void *
 596 vc4_bo_map(struct vc4_bo *bo)
 597 {
 598         void *map = vc4_bo_map_unsynchronized(bo);
 599
 600         bool ok = vc4_bo_wait(bo, PIPE_TIMEOUT_INFINITE, "bo map");
 601         if (!ok) {
 602                 fprintf(stderr, "BO wait for map failed\n");
 603                 abort();
 604         }
 605
 606         return map;
 607 }
 608
 609 void
 610 vc4_bufmgr_destroy(struct pipe_screen *pscreen)
 611 {
 612         struct vc4_screen *screen = vc4_screen(pscreen);
 613         struct vc4_bo_cache *cache = &screen->bo_cache;
 614
 615         vc4_bo_cache_free_all(cache);
 616
 617         if (dump_stats) {
 618                 fprintf(stderr, "BO stats after screen destroy:\n");
 619                 vc4_bo_dump_stats(screen);
 620         }
 621 }