aboutsummaryrefslogtreecommitdiffstats
path: root/src/gallium/winsys/amdgpu/drm/amdgpu_bo.c
blob: 036301ef089de5bae793b8219289ba28282d7014 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
/*
 * Copyright © 2011 Marek Olšák <maraeo@gmail.com>
 * Copyright © 2015 Advanced Micro Devices, Inc.
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining
 * a copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS
 * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 */
/*
 * Authors:
 *      Marek Olšák <maraeo@gmail.com>
 */

#include "amdgpu_cs.h"

#include "os/os_time.h"
#include "state_tracker/drm_driver.h"
#include <amdgpu_drm.h>
#include <xf86drm.h>
#include <stdio.h>
#include <inttypes.h>

static bool amdgpu_bo_wait(struct pb_buffer *_buf, uint64_t timeout,
                           enum radeon_bo_usage usage)
{
   struct amdgpu_winsys_bo *bo = amdgpu_winsys_bo(_buf);
   struct amdgpu_winsys *ws = bo->ws;
   int i;

   if (bo->is_shared) {
      /* We can't use user fences for shared buffers, because user fences
       * are local to this process only. If we want to wait for all buffer
       * uses in all processes, we have to use amdgpu_bo_wait_for_idle.
       */
      bool buffer_busy = true;
      int r;

      r = amdgpu_bo_wait_for_idle(bo->bo, timeout, &buffer_busy);
      if (r)
         fprintf(stderr, "%s: amdgpu_bo_wait_for_idle failed %i\n", __func__,
                 r);
      return !buffer_busy;
   }

   if (timeout == 0) {
      /* Timeout == 0 is quite simple. */
      pipe_mutex_lock(ws->bo_fence_lock);
      for (i = 0; i < RING_LAST; i++)
         if (bo->fence[i]) {
            if (amdgpu_fence_wait(bo->fence[i], 0, false)) {
               /* Release the idle fence to avoid checking it again later. */
               amdgpu_fence_reference(&bo->fence[i], NULL);
            } else {
               pipe_mutex_unlock(ws->bo_fence_lock);
               return false;
            }
         }
      pipe_mutex_unlock(ws->bo_fence_lock);
      return true;

   } else {
      struct pipe_fence_handle *fence[RING_LAST] = {};
      bool fence_idle[RING_LAST] = {};
      bool buffer_idle = true;
      int64_t abs_timeout = os_time_get_absolute_timeout(timeout);

      /* Take references to all fences, so that we can wait for them
       * without the lock. */
      pipe_mutex_lock(ws->bo_fence_lock);
      for (i = 0; i < RING_LAST; i++)
         amdgpu_fence_reference(&fence[i], bo->fence[i]);
      pipe_mutex_unlock(ws->bo_fence_lock);

      /* Now wait for the fences. */
      for (i = 0; i < RING_LAST; i++) {
         if (fence[i]) {
            if (amdgpu_fence_wait(fence[i], abs_timeout, true))
               fence_idle[i] = true;
            else
               buffer_idle = false;
         }
      }

      /* Release idle fences to avoid checking them again later. */
      pipe_mutex_lock(ws->bo_fence_lock);
      for (i = 0; i < RING_LAST; i++) {
         if (fence[i] == bo->fence[i] && fence_idle[i])
            amdgpu_fence_reference(&bo->fence[i], NULL);

         amdgpu_fence_reference(&fence[i], NULL);
      }
      pipe_mutex_unlock(ws->bo_fence_lock);

      return buffer_idle;
   }
}

static enum radeon_bo_domain amdgpu_bo_get_initial_domain(
      struct pb_buffer *buf)
{
   return ((struct amdgpu_winsys_bo*)buf)->initial_domain;
}

void amdgpu_bo_destroy(struct pb_buffer *_buf)
{
   struct amdgpu_winsys_bo *bo = amdgpu_winsys_bo(_buf);
   int i;

   pipe_mutex_lock(bo->ws->global_bo_list_lock);
   LIST_DEL(&bo->global_list_item);
   bo->ws->num_buffers--;
   pipe_mutex_unlock(bo->ws->global_bo_list_lock);

   amdgpu_bo_va_op(bo->bo, 0, bo->base.size, bo->va, 0, AMDGPU_VA_OP_UNMAP);
   amdgpu_va_range_free(bo->va_handle);
   amdgpu_bo_free(bo->bo);

   for (i = 0; i < RING_LAST; i++)
      amdgpu_fence_reference(&bo->fence[i], NULL);

   if (bo->initial_domain & RADEON_DOMAIN_VRAM)
      bo->ws->allocated_vram -= align64(bo->base.size, bo->ws->gart_page_size);
   else if (bo->initial_domain & RADEON_DOMAIN_GTT)
      bo->ws->allocated_gtt -= align64(bo->base.size, bo->ws->gart_page_size);
   FREE(bo);
}

static void amdgpu_bo_destroy_or_cache(struct pb_buffer *_buf)
{
   struct amdgpu_winsys_bo *bo = amdgpu_winsys_bo(_buf);

   if (bo->use_reusable_pool)
      pb_cache_add_buffer(&bo->cache_entry);
   else
      amdgpu_bo_destroy(_buf);
}

static void *amdgpu_bo_map(struct pb_buffer *buf,
                           struct radeon_winsys_cs *rcs,
                           enum pipe_transfer_usage usage)
{
   struct amdgpu_winsys_bo *bo = (struct amdgpu_winsys_bo*)buf;
   struct amdgpu_cs *cs = (struct amdgpu_cs*)rcs;
   int r;
   void *cpu = NULL;

   /* If it's not unsynchronized bo_map, flush CS if needed and then wait. */
   if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
      /* DONTBLOCK doesn't make sense with UNSYNCHRONIZED. */
      if (usage & PIPE_TRANSFER_DONTBLOCK) {
         if (!(usage & PIPE_TRANSFER_WRITE)) {
            /* Mapping for read.
             *
             * Since we are mapping for read, we don't need to wait
             * if the GPU is using the buffer for read too
             * (neither one is changing it).
             *
             * Only check whether the buffer is being used for write. */
            if (cs && amdgpu_bo_is_referenced_by_cs_with_usage(cs, bo,
                                                               RADEON_USAGE_WRITE)) {
               cs->flush_cs(cs->flush_data, RADEON_FLUSH_ASYNC, NULL);
               return NULL;
            }

            if (!amdgpu_bo_wait((struct pb_buffer*)bo, 0,
                                RADEON_USAGE_WRITE)) {
               return NULL;
            }
         } else {
            if (cs && amdgpu_bo_is_referenced_by_cs(cs, bo)) {
               cs->flush_cs(cs->flush_data, RADEON_FLUSH_ASYNC, NULL);
               return NULL;
            }

            if (!amdgpu_bo_wait((struct pb_buffer*)bo, 0,
                                RADEON_USAGE_READWRITE)) {
               return NULL;
            }
         }
      } else {
         uint64_t time = os_time_get_nano();

         if (!(usage & PIPE_TRANSFER_WRITE)) {
            /* Mapping for read.
             *
             * Since we are mapping for read, we don't need to wait
             * if the GPU is using the buffer for read too
             * (neither one is changing it).
             *
             * Only check whether the buffer is being used for write. */
            if (cs && amdgpu_bo_is_referenced_by_cs_with_usage(cs, bo,
                                                               RADEON_USAGE_WRITE)) {
               cs->flush_cs(cs->flush_data, 0, NULL);
            }
            amdgpu_bo_wait((struct pb_buffer*)bo, PIPE_TIMEOUT_INFINITE,
                           RADEON_USAGE_WRITE);
         } else {
            /* Mapping for write. */
            if (cs && amdgpu_bo_is_referenced_by_cs(cs, bo))
               cs->flush_cs(cs->flush_data, 0, NULL);

            amdgpu_bo_wait((struct pb_buffer*)bo, PIPE_TIMEOUT_INFINITE,
                           RADEON_USAGE_READWRITE);
         }

         bo->ws->buffer_wait_time += os_time_get_nano() - time;
      }
   }

   /* If the buffer is created from user memory, return the user pointer. */
   if (bo->user_ptr)
       return bo->user_ptr;

   r = amdgpu_bo_cpu_map(bo->bo, &cpu);
   if (r) {
      /* Clear the cache and try again. */
      pb_cache_release_all_buffers(&bo->ws->bo_cache);
      r = amdgpu_bo_cpu_map(bo->bo, &cpu);
   }
   return r ? NULL : cpu;
}

static void amdgpu_bo_unmap(struct pb_buffer *buf)
{
   struct amdgpu_winsys_bo *bo = (struct amdgpu_winsys_bo*)buf;

   amdgpu_bo_cpu_unmap(bo->bo);
}

static const struct pb_vtbl amdgpu_winsys_bo_vtbl = {
   amdgpu_bo_destroy_or_cache
   /* other functions are never called */
};

static void amdgpu_add_buffer_to_global_list(struct amdgpu_winsys_bo *bo)
{
   struct amdgpu_winsys *ws = bo->ws;

   pipe_mutex_lock(ws->global_bo_list_lock);
   LIST_ADDTAIL(&bo->global_list_item, &ws->global_bo_list);
   ws->num_buffers++;
   pipe_mutex_unlock(ws->global_bo_list_lock);
}

static struct amdgpu_winsys_bo *amdgpu_create_bo(struct amdgpu_winsys *ws,
                                                 uint64_t size,
                                                 unsigned alignment,
                                                 unsigned usage,
                                                 enum radeon_bo_domain initial_domain,
                                                 unsigned flags)
{
   struct amdgpu_bo_alloc_request request = {0};
   amdgpu_bo_handle buf_handle;
   uint64_t va = 0;
   struct amdgpu_winsys_bo *bo;
   amdgpu_va_handle va_handle;
   int r;

   assert(initial_domain & RADEON_DOMAIN_VRAM_GTT);
   bo = CALLOC_STRUCT(amdgpu_winsys_bo);
   if (!bo) {
      return NULL;
   }

   pb_cache_init_entry(&ws->bo_cache, &bo->cache_entry, &bo->base);
   request.alloc_size = size;
   request.phys_alignment = alignment;

   if (initial_domain & RADEON_DOMAIN_VRAM)
      request.preferred_heap |= AMDGPU_GEM_DOMAIN_VRAM;
   if (initial_domain & RADEON_DOMAIN_GTT)
      request.preferred_heap |= AMDGPU_GEM_DOMAIN_GTT;

   if (flags & RADEON_FLAG_CPU_ACCESS)
      request.flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
   if (flags & RADEON_FLAG_NO_CPU_ACCESS)
      request.flags |= AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
   if (flags & RADEON_FLAG_GTT_WC)
      request.flags |= AMDGPU_GEM_CREATE_CPU_GTT_USWC;

   r = amdgpu_bo_alloc(ws->dev, &request, &buf_handle);
   if (r) {
      fprintf(stderr, "amdgpu: Failed to allocate a buffer:\n");
      fprintf(stderr, "amdgpu:    size      : %"PRIu64" bytes\n", size);
      fprintf(stderr, "amdgpu:    alignment : %u bytes\n", alignment);
      fprintf(stderr, "amdgpu:    domains   : %u\n", initial_domain);
      goto error_bo_alloc;
   }

   r = amdgpu_va_range_alloc(ws->dev, amdgpu_gpu_va_range_general,
                             size, alignment, 0, &va, &va_handle, 0);
   if (r)
      goto error_va_alloc;

   r = amdgpu_bo_va_op(buf_handle, 0, size, va, 0, AMDGPU_VA_OP_MAP);
   if (r)
      goto error_va_map;

   pipe_reference_init(&bo->base.reference, 1);
   bo->base.alignment = alignment;
   bo->base.usage = usage;
   bo->base.size = size;
   bo->base.vtbl = &amdgpu_winsys_bo_vtbl;
   bo->ws = ws;
   bo->bo = buf_handle;
   bo->va = va;
   bo->va_handle = va_handle;
   bo->initial_domain = initial_domain;
   bo->unique_id = __sync_fetch_and_add(&ws->next_bo_unique_id, 1);

   if (initial_domain & RADEON_DOMAIN_VRAM)
      ws->allocated_vram += align64(size, ws->gart_page_size);
   else if (initial_domain & RADEON_DOMAIN_GTT)
      ws->allocated_gtt += align64(size, ws->gart_page_size);

   amdgpu_add_buffer_to_global_list(bo);

   return bo;

error_va_map:
   amdgpu_va_range_free(va_handle);

error_va_alloc:
   amdgpu_bo_free(buf_handle);

error_bo_alloc:
   FREE(bo);
   return NULL;
}

bool amdgpu_bo_can_reclaim(struct pb_buffer *_buf)
{
   struct amdgpu_winsys_bo *bo = amdgpu_winsys_bo(_buf);

   if (amdgpu_bo_is_referenced_by_any_cs(bo)) {
      return false;
   }

   return amdgpu_bo_wait(_buf, 0, RADEON_USAGE_READWRITE);
}

static unsigned eg_tile_split(unsigned tile_split)
{
   switch (tile_split) {
   case 0:     tile_split = 64;    break;
   case 1:     tile_split = 128;   break;
   case 2:     tile_split = 256;   break;
   case 3:     tile_split = 512;   break;
   default:
   case 4:     tile_split = 1024;  break;
   case 5:     tile_split = 2048;  break;
   case 6:     tile_split = 4096;  break;
   }
   return tile_split;
}

static unsigned eg_tile_split_rev(unsigned eg_tile_split)
{
   switch (eg_tile_split) {
   case 64:    return 0;
   case 128:   return 1;
   case 256:   return 2;
   case 512:   return 3;
   default:
   case 1024:  return 4;
   case 2048:  return 5;
   case 4096:  return 6;
   }
}

static void amdgpu_buffer_get_metadata(struct pb_buffer *_buf,
                                       struct radeon_bo_metadata *md)
{
   struct amdgpu_winsys_bo *bo = amdgpu_winsys_bo(_buf);
   struct amdgpu_bo_info info = {0};
   uint32_t tiling_flags;
   int r;

   r = amdgpu_bo_query_info(bo->bo, &info);
   if (r)
      return;

   tiling_flags = info.metadata.tiling_info;

   md->microtile = RADEON_LAYOUT_LINEAR;
   md->macrotile = RADEON_LAYOUT_LINEAR;

   if (AMDGPU_TILING_GET(tiling_flags, ARRAY_MODE) == 4)  /* 2D_TILED_THIN1 */
      md->macrotile = RADEON_LAYOUT_TILED;
   else if (AMDGPU_TILING_GET(tiling_flags, ARRAY_MODE) == 2) /* 1D_TILED_THIN1 */
      md->microtile = RADEON_LAYOUT_TILED;

   md->bankw = 1 << AMDGPU_TILING_GET(tiling_flags, BANK_WIDTH);
   md->bankh = 1 << AMDGPU_TILING_GET(tiling_flags, BANK_HEIGHT);
   md->tile_split = eg_tile_split(AMDGPU_TILING_GET(tiling_flags, TILE_SPLIT));
   md->mtilea = 1 << AMDGPU_TILING_GET(tiling_flags, MACRO_TILE_ASPECT);
   md->scanout = AMDGPU_TILING_GET(tiling_flags, MICRO_TILE_MODE) == 0; /* DISPLAY */

   md->size_metadata = info.metadata.size_metadata;
   memcpy(md->metadata, info.metadata.umd_metadata, sizeof(md->metadata));
}

static void amdgpu_buffer_set_metadata(struct pb_buffer *_buf,
                                       struct radeon_bo_metadata *md)
{
   struct amdgpu_winsys_bo *bo = amdgpu_winsys_bo(_buf);
   struct amdgpu_bo_metadata metadata = {0};
   uint32_t tiling_flags = 0;

   if (md->macrotile == RADEON_LAYOUT_TILED)
      tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 4); /* 2D_TILED_THIN1 */
   else if (md->microtile == RADEON_LAYOUT_TILED)
      tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 2); /* 1D_TILED_THIN1 */
   else
      tiling_flags |= AMDGPU_TILING_SET(ARRAY_MODE, 1); /* LINEAR_ALIGNED */

   tiling_flags |= AMDGPU_TILING_SET(PIPE_CONFIG, md->pipe_config);
   tiling_flags |= AMDGPU_TILING_SET(BANK_WIDTH, util_logbase2(md->bankw));
   tiling_flags |= AMDGPU_TILING_SET(BANK_HEIGHT, util_logbase2(md->bankh));
   if (md->tile_split)
      tiling_flags |= AMDGPU_TILING_SET(TILE_SPLIT, eg_tile_split_rev(md->tile_split));
   tiling_flags |= AMDGPU_TILING_SET(MACRO_TILE_ASPECT, util_logbase2(md->mtilea));
   tiling_flags |= AMDGPU_TILING_SET(NUM_BANKS, util_logbase2(md->num_banks)-1);

   if (md->scanout)
      tiling_flags |= AMDGPU_TILING_SET(MICRO_TILE_MODE, 0); /* DISPLAY_MICRO_TILING */
   else
      tiling_flags |= AMDGPU_TILING_SET(MICRO_TILE_MODE, 1); /* THIN_MICRO_TILING */

   metadata.tiling_info = tiling_flags;
   metadata.size_metadata = md->size_metadata;
   memcpy(metadata.umd_metadata, md->metadata, sizeof(md->metadata));

   amdgpu_bo_set_metadata(bo->bo, &metadata);
}

static struct pb_buffer *
amdgpu_bo_create(struct radeon_winsys *rws,
                 uint64_t size,
                 unsigned alignment,
                 boolean use_reusable_pool,
                 enum radeon_bo_domain domain,
                 enum radeon_bo_flag flags)
{
   struct amdgpu_winsys *ws = amdgpu_winsys(rws);
   struct amdgpu_winsys_bo *bo;
   unsigned usage = 0;

   /* Align size to page size. This is the minimum alignment for normal
    * BOs. Aligning this here helps the cached bufmgr. Especially small BOs,
    * like constant/uniform buffers, can benefit from better and more reuse.
    */
   size = align64(size, ws->gart_page_size);

   /* Only set one usage bit each for domains and flags, or the cache manager
    * might consider different sets of domains / flags compatible
    */
   if (domain == RADEON_DOMAIN_VRAM_GTT)
      usage = 1 << 2;
   else
      usage = domain >> 1;
   assert(flags < sizeof(usage) * 8 - 3);
   usage |= 1 << (flags + 3);

   /* Get a buffer from the cache. */
   if (use_reusable_pool) {
       bo = (struct amdgpu_winsys_bo*)
            pb_cache_reclaim_buffer(&ws->bo_cache, size, alignment,
                                    usage);
       if (bo)
          return &bo->base;
   }

   /* Create a new one. */
   bo = amdgpu_create_bo(ws, size, alignment, usage, domain, flags);
   if (!bo) {
      /* Clear the cache and try again. */
      pb_cache_release_all_buffers(&ws->bo_cache);
      bo = amdgpu_create_bo(ws, size, alignment, usage, domain, flags);
      if (!bo)
         return NULL;
   }

   bo->use_reusable_pool = use_reusable_pool;
   return &bo->base;
}

static struct pb_buffer *amdgpu_bo_from_handle(struct radeon_winsys *rws,
                                               struct winsys_handle *whandle,
                                               unsigned *stride,
                                               unsigned *offset)
{
   struct amdgpu_winsys *ws = amdgpu_winsys(rws);
   struct amdgpu_winsys_bo *bo;
   enum amdgpu_bo_handle_type type;
   struct amdgpu_bo_import_result result = {0};
   uint64_t va;
   amdgpu_va_handle va_handle;
   struct amdgpu_bo_info info = {0};
   enum radeon_bo_domain initial = 0;
   int r;

   /* Initialize the structure. */
   bo = CALLOC_STRUCT(amdgpu_winsys_bo);
   if (!bo) {
      return NULL;
   }

   switch (whandle->type) {
   case DRM_API_HANDLE_TYPE_SHARED:
      type = amdgpu_bo_handle_type_gem_flink_name;
      break;
   case DRM_API_HANDLE_TYPE_FD:
      type = amdgpu_bo_handle_type_dma_buf_fd;
      break;
   default:
      return NULL;
   }

   r = amdgpu_bo_import(ws->dev, type, whandle->handle, &result);
   if (r)
      goto error;

   /* Get initial domains. */
   r = amdgpu_bo_query_info(result.buf_handle, &info);
   if (r)
      goto error_query;

   r = amdgpu_va_range_alloc(ws->dev, amdgpu_gpu_va_range_general,
                             result.alloc_size, 1 << 20, 0, &va, &va_handle, 0);
   if (r)
      goto error_query;

   r = amdgpu_bo_va_op(result.buf_handle, 0, result.alloc_size, va, 0, AMDGPU_VA_OP_MAP);
   if (r)
      goto error_va_map;

   if (info.preferred_heap & AMDGPU_GEM_DOMAIN_VRAM)
      initial |= RADEON_DOMAIN_VRAM;
   if (info.preferred_heap & AMDGPU_GEM_DOMAIN_GTT)
      initial |= RADEON_DOMAIN_GTT;


   pipe_reference_init(&bo->base.reference, 1);
   bo->base.alignment = info.phys_alignment;
   bo->base.usage = PB_USAGE_GPU_WRITE | PB_USAGE_GPU_READ;
   bo->bo = result.buf_handle;
   bo->base.size = result.alloc_size;
   bo->base.vtbl = &amdgpu_winsys_bo_vtbl;
   bo->ws = ws;
   bo->va = va;
   bo->va_handle = va_handle;
   bo->initial_domain = initial;
   bo->unique_id = __sync_fetch_and_add(&ws->next_bo_unique_id, 1);
   bo->is_shared = true;

   if (stride)
      *stride = whandle->stride;
   if (offset)
      *offset = whandle->offset;

   if (bo->initial_domain & RADEON_DOMAIN_VRAM)
      ws->allocated_vram += align64(bo->base.size, ws->gart_page_size);
   else if (bo->initial_domain & RADEON_DOMAIN_GTT)
      ws->allocated_gtt += align64(bo->base.size, ws->gart_page_size);

   amdgpu_add_buffer_to_global_list(bo);

   return &bo->base;

error_va_map:
   amdgpu_va_range_free(va_handle);

error_query:
   amdgpu_bo_free(result.buf_handle);

error:
   FREE(bo);
   return NULL;
}

static boolean amdgpu_bo_get_handle(struct pb_buffer *buffer,
                                    unsigned stride, unsigned offset,
                                    unsigned slice_size,
                                    struct winsys_handle *whandle)
{
   struct amdgpu_winsys_bo *bo = amdgpu_winsys_bo(buffer);
   enum amdgpu_bo_handle_type type;
   int r;

   bo->use_reusable_pool = false;

   switch (whandle->type) {
   case DRM_API_HANDLE_TYPE_SHARED:
      type = amdgpu_bo_handle_type_gem_flink_name;
      break;
   case DRM_API_HANDLE_TYPE_FD:
      type = amdgpu_bo_handle_type_dma_buf_fd;
      break;
   case DRM_API_HANDLE_TYPE_KMS:
      type = amdgpu_bo_handle_type_kms;
      break;
   default:
      return FALSE;
   }

   r = amdgpu_bo_export(bo->bo, type, &whandle->handle);
   if (r)
      return FALSE;

   whandle->stride = stride;
   whandle->offset = offset;
   whandle->offset += slice_size * whandle->layer;
   bo->is_shared = true;
   return TRUE;
}

static struct pb_buffer *amdgpu_bo_from_ptr(struct radeon_winsys *rws,
					    void *pointer, uint64_t size)
{
    struct amdgpu_winsys *ws = amdgpu_winsys(rws);
    amdgpu_bo_handle buf_handle;
    struct amdgpu_winsys_bo *bo;
    uint64_t va;
    amdgpu_va_handle va_handle;

    bo = CALLOC_STRUCT(amdgpu_winsys_bo);
    if (!bo)
        return NULL;

    if (amdgpu_create_bo_from_user_mem(ws->dev, pointer, size, &buf_handle))
        goto error;

    if (amdgpu_va_range_alloc(ws->dev, amdgpu_gpu_va_range_general,
                              size, 1 << 12, 0, &va, &va_handle, 0))
        goto error_va_alloc;

    if (amdgpu_bo_va_op(buf_handle, 0, size, va, 0, AMDGPU_VA_OP_MAP))
        goto error_va_map;

    /* Initialize it. */
    pipe_reference_init(&bo->base.reference, 1);
    bo->bo = buf_handle;
    bo->base.alignment = 0;
    bo->base.usage = PB_USAGE_GPU_WRITE | PB_USAGE_GPU_READ;
    bo->base.size = size;
    bo->base.vtbl = &amdgpu_winsys_bo_vtbl;
    bo->ws = ws;
    bo->user_ptr = pointer;
    bo->va = va;
    bo->va_handle = va_handle;
    bo->initial_domain = RADEON_DOMAIN_GTT;
    bo->unique_id = __sync_fetch_and_add(&ws->next_bo_unique_id, 1);

    ws->allocated_gtt += align64(bo->base.size, ws->gart_page_size);

    amdgpu_add_buffer_to_global_list(bo);

    return (struct pb_buffer*)bo;

error_va_map:
    amdgpu_va_range_free(va_handle);

error_va_alloc:
    amdgpu_bo_free(buf_handle);

error:
    FREE(bo);
    return NULL;
}

static bool amdgpu_bo_is_user_ptr(struct pb_buffer *buf)
{
   return ((struct amdgpu_winsys_bo*)buf)->user_ptr != NULL;
}

static uint64_t amdgpu_bo_get_va(struct pb_buffer *buf)
{
   return ((struct amdgpu_winsys_bo*)buf)->va;
}

void amdgpu_bo_init_functions(struct amdgpu_winsys *ws)
{
   ws->base.buffer_set_metadata = amdgpu_buffer_set_metadata;
   ws->base.buffer_get_metadata = amdgpu_buffer_get_metadata;
   ws->base.buffer_map = amdgpu_bo_map;
   ws->base.buffer_unmap = amdgpu_bo_unmap;
   ws->base.buffer_wait = amdgpu_bo_wait;
   ws->base.buffer_create = amdgpu_bo_create;
   ws->base.buffer_from_handle = amdgpu_bo_from_handle;
   ws->base.buffer_from_ptr = amdgpu_bo_from_ptr;
   ws->base.buffer_is_user_ptr = amdgpu_bo_is_user_ptr;
   ws->base.buffer_get_handle = amdgpu_bo_get_handle;
   ws->base.buffer_get_virtual_address = amdgpu_bo_get_va;
   ws->base.buffer_get_initial_domain = amdgpu_bo_get_initial_domain;
}