summaryrefslogtreecommitdiff
path: root/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
blob: ba087ac0f34dec2a790ed6672b9043fa32687869 (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
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
/**************************************************************************
 *
 * Copyright 2007-2010 VMware, 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 above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 * 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 VMWARE 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.
 *
 **************************************************************************/

/**
 * \file
 * Implementation of fenced buffers.
 *
 * \author Jose Fonseca <jfonseca-at-vmware-dot-com>
 * \author Thomas Hellström <thellstrom-at-vmware-dot-com>
 */


#include "pipe/p_config.h"

#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS)
#include <unistd.h>
#include <sched.h>
#endif

#include "pipe/p_compiler.h"
#include "pipe/p_defines.h"
#include "util/u_debug.h"
#include "pipe/p_thread.h"
#include "util/u_memory.h"
#include "util/u_double_list.h"

#include "pb_buffer.h"
#include "pb_buffer_fenced.h"
#include "pb_bufmgr.h"



/**
 * Convenience macro (type safe).
 */
#define SUPER(__derived) (&(__derived)->base)


struct fenced_manager
{
   struct pb_manager base;
   struct pb_manager *provider;
   struct pb_fence_ops *ops;

   /**
    * Maximum buffer size that can be safely allocated.
    */
   pb_size max_buffer_size;

   /**
    * Maximum cpu memory we can allocate before we start waiting for the
    * GPU to idle.
    */
   pb_size max_cpu_total_size;

   /**
    * Following members are mutable and protected by this mutex.
    */
   pipe_mutex mutex;

   /**
    * Fenced buffer list.
    *
    * All fenced buffers are placed in this listed, ordered from the oldest
    * fence to the newest fence.
    */
   struct list_head fenced;
   pb_size num_fenced;

   struct list_head unfenced;
   pb_size num_unfenced;

   /**
    * How much temporary CPU memory is being used to hold unvalidated buffers.
    */
   pb_size cpu_total_size;
};


/**
 * Fenced buffer.
 *
 * Wrapper around a pipe buffer which adds fencing and reference counting.
 */
struct fenced_buffer
{
   /*
    * Immutable members.
    */

   struct pb_buffer base;
   struct fenced_manager *mgr;

   /*
    * Following members are mutable and protected by fenced_manager::mutex.
    */

   struct list_head head;

   /**
    * Buffer with storage.
    */
   struct pb_buffer *buffer;
   pb_size size;
   struct pb_desc desc;

   /**
    * Temporary CPU storage data. Used when there isn't enough GPU memory to
    * store the buffer.
    */
   void *data;

   /**
    * A bitmask of PIPE_BUFFER_USAGE_CPU/GPU_READ/WRITE describing the current
    * buffer usage.
    */
   unsigned flags;

   unsigned mapcount;

   struct pb_validate *vl;
   unsigned validation_flags;

   struct pipe_fence_handle *fence;
};


static INLINE struct fenced_manager *
fenced_manager(struct pb_manager *mgr)
{
   assert(mgr);
   return (struct fenced_manager *)mgr;
}


static INLINE struct fenced_buffer *
fenced_buffer(struct pb_buffer *buf)
{
   assert(buf);
   return (struct fenced_buffer *)buf;
}


static void
fenced_buffer_destroy_cpu_storage_locked(struct fenced_buffer *fenced_buf);

static enum pipe_error
fenced_buffer_create_cpu_storage_locked(struct fenced_manager *fenced_mgr,
                                        struct fenced_buffer *fenced_buf);

static void
fenced_buffer_destroy_gpu_storage_locked(struct fenced_buffer *fenced_buf);

static enum pipe_error
fenced_buffer_create_gpu_storage_locked(struct fenced_manager *fenced_mgr,
                                        struct fenced_buffer *fenced_buf,
                                        boolean wait);

static enum pipe_error
fenced_buffer_copy_storage_to_gpu_locked(struct fenced_buffer *fenced_buf);

static enum pipe_error
fenced_buffer_copy_storage_to_cpu_locked(struct fenced_buffer *fenced_buf);


/**
 * Dump the fenced buffer list.
 *
 * Useful to understand failures to allocate buffers.
 */
static void
fenced_manager_dump_locked(struct fenced_manager *fenced_mgr)
{
#ifdef DEBUG
   struct pb_fence_ops *ops = fenced_mgr->ops;
   struct list_head *curr, *next;
   struct fenced_buffer *fenced_buf;

   debug_printf("%10s %7s %8s %7s %10s %s\n",
                "buffer", "size", "refcount", "storage", "fence", "signalled");

   curr = fenced_mgr->unfenced.next;
   next = curr->next;
   while(curr != &fenced_mgr->unfenced) {
      fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
      assert(!fenced_buf->fence);
      debug_printf("%10p %7u %8u %7s\n",
                   (void *) fenced_buf,
                   fenced_buf->base.base.size,
                   p_atomic_read(&fenced_buf->base.base.reference.count),
                   fenced_buf->buffer ? "gpu" : (fenced_buf->data ? "cpu" : "none"));
      curr = next;
      next = curr->next;
   }

   curr = fenced_mgr->fenced.next;
   next = curr->next;
   while(curr != &fenced_mgr->fenced) {
      int signaled;
      fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
      assert(fenced_buf->buffer);
      signaled = ops->fence_signalled(ops, fenced_buf->fence, 0);
      debug_printf("%10p %7u %8u %7s %10p %s\n",
                   (void *) fenced_buf,
                   fenced_buf->base.base.size,
                   p_atomic_read(&fenced_buf->base.base.reference.count),
                   "gpu",
                   (void *) fenced_buf->fence,
                   signaled == 0 ? "y" : "n");
      curr = next;
      next = curr->next;
   }
#else
   (void)fenced_mgr;
#endif
}


static INLINE void
fenced_buffer_destroy_locked(struct fenced_manager *fenced_mgr,
                             struct fenced_buffer *fenced_buf)
{
   assert(!pipe_is_referenced(&fenced_buf->base.base.reference));

   assert(!fenced_buf->fence);
   assert(fenced_buf->head.prev);
   assert(fenced_buf->head.next);
   LIST_DEL(&fenced_buf->head);
   assert(fenced_mgr->num_unfenced);
   --fenced_mgr->num_unfenced;

   fenced_buffer_destroy_gpu_storage_locked(fenced_buf);
   fenced_buffer_destroy_cpu_storage_locked(fenced_buf);

   FREE(fenced_buf);
}


/**
 * Add the buffer to the fenced list.
 *
 * Reference count should be incremented before calling this function.
 */
static INLINE void
fenced_buffer_add_locked(struct fenced_manager *fenced_mgr,
                         struct fenced_buffer *fenced_buf)
{
   assert(pipe_is_referenced(&fenced_buf->base.base.reference));
   assert(fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE);
   assert(fenced_buf->fence);

   p_atomic_inc(&fenced_buf->base.base.reference.count);

   LIST_DEL(&fenced_buf->head);
   assert(fenced_mgr->num_unfenced);
   --fenced_mgr->num_unfenced;
   LIST_ADDTAIL(&fenced_buf->head, &fenced_mgr->fenced);
   ++fenced_mgr->num_fenced;
}


/**
 * Remove the buffer from the fenced list, and potentially destroy the buffer
 * if the reference count reaches zero.
 *
 * Returns TRUE if the buffer was detroyed.
 */
static INLINE boolean
fenced_buffer_remove_locked(struct fenced_manager *fenced_mgr,
                            struct fenced_buffer *fenced_buf)
{
   struct pb_fence_ops *ops = fenced_mgr->ops;

   assert(fenced_buf->fence);
   assert(fenced_buf->mgr == fenced_mgr);

   ops->fence_reference(ops, &fenced_buf->fence, NULL);
   fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE;

   assert(fenced_buf->head.prev);
   assert(fenced_buf->head.next);

   LIST_DEL(&fenced_buf->head);
   assert(fenced_mgr->num_fenced);
   --fenced_mgr->num_fenced;

   LIST_ADDTAIL(&fenced_buf->head, &fenced_mgr->unfenced);
   ++fenced_mgr->num_unfenced;

   if (p_atomic_dec_zero(&fenced_buf->base.base.reference.count)) {
      fenced_buffer_destroy_locked(fenced_mgr, fenced_buf);
      return TRUE;
   }

   return FALSE;
}


/**
 * Wait for the fence to expire, and remove it from the fenced list.
 *
 * This function will release and re-aquire the mutex, so any copy of mutable
 * state must be discarded after calling it.
 */
static INLINE enum pipe_error
fenced_buffer_finish_locked(struct fenced_manager *fenced_mgr,
                            struct fenced_buffer *fenced_buf)
{
   struct pb_fence_ops *ops = fenced_mgr->ops;
   enum pipe_error ret = PIPE_ERROR;

#if 0
   debug_warning("waiting for GPU");
#endif

   assert(pipe_is_referenced(&fenced_buf->base.base.reference));
   assert(fenced_buf->fence);

   if(fenced_buf->fence) {
      struct pipe_fence_handle *fence = NULL;
      int finished;
      boolean proceed;

      ops->fence_reference(ops, &fence, fenced_buf->fence);

      pipe_mutex_unlock(fenced_mgr->mutex);

      finished = ops->fence_finish(ops, fenced_buf->fence, 0);

      pipe_mutex_lock(fenced_mgr->mutex);

      assert(pipe_is_referenced(&fenced_buf->base.base.reference));

      /*
       * Only proceed if the fence object didn't change in the meanwhile.
       * Otherwise assume the work has been already carried out by another
       * thread that re-aquired the lock before us.
       */
      proceed = fence == fenced_buf->fence ? TRUE : FALSE;

      ops->fence_reference(ops, &fence, NULL);

      if(proceed && finished == 0) {
         /*
          * Remove from the fenced list
          */

         boolean destroyed;

         destroyed = fenced_buffer_remove_locked(fenced_mgr, fenced_buf);

         /* TODO: remove consequents buffers with the same fence? */

         assert(!destroyed);

         fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE;

         ret = PIPE_OK;
      }
   }

   return ret;
}


/**
 * Remove as many fenced buffers from the fenced list as possible.
 *
 * Returns TRUE if at least one buffer was removed.
 */
static boolean
fenced_manager_check_signalled_locked(struct fenced_manager *fenced_mgr,
                                      boolean wait)
{
   struct pb_fence_ops *ops = fenced_mgr->ops;
   struct list_head *curr, *next;
   struct fenced_buffer *fenced_buf;
   struct pipe_fence_handle *prev_fence = NULL;
   boolean ret = FALSE;

   curr = fenced_mgr->fenced.next;
   next = curr->next;
   while(curr != &fenced_mgr->fenced) {
      fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);

      if(fenced_buf->fence != prev_fence) {
	 int signaled;

	 if (wait) {
	    signaled = ops->fence_finish(ops, fenced_buf->fence, 0);

	    /*
	     * Don't return just now. Instead preemptively check if the
	     * following buffers' fences already expired, without further waits.
	     */
	    wait = FALSE;
	 }
	 else {
	    signaled = ops->fence_signalled(ops, fenced_buf->fence, 0);
	 }

	 if (signaled != 0) {
	    return ret;
         }

	 prev_fence = fenced_buf->fence;
      }
      else {
         /* This buffer's fence object is identical to the previous buffer's
          * fence object, so no need to check the fence again.
          */
	 assert(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0);
      }

      fenced_buffer_remove_locked(fenced_mgr, fenced_buf);

      ret = TRUE;

      curr = next;
      next = curr->next;
   }

   return ret;
}


/**
 * Try to free some GPU memory by backing it up into CPU memory.
 *
 * Returns TRUE if at least one buffer was freed.
 */
static boolean
fenced_manager_free_gpu_storage_locked(struct fenced_manager *fenced_mgr)
{
   struct list_head *curr, *next;
   struct fenced_buffer *fenced_buf;

   curr = fenced_mgr->unfenced.next;
   next = curr->next;
   while(curr != &fenced_mgr->unfenced) {
      fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);

      /*
       * We can only move storage if the buffer is not mapped and not
       * validated.
       */
      if(fenced_buf->buffer &&
         !fenced_buf->mapcount &&
         !fenced_buf->vl) {
         enum pipe_error ret;

         ret = fenced_buffer_create_cpu_storage_locked(fenced_mgr, fenced_buf);
         if(ret == PIPE_OK) {
            ret = fenced_buffer_copy_storage_to_cpu_locked(fenced_buf);
            if(ret == PIPE_OK) {
               fenced_buffer_destroy_gpu_storage_locked(fenced_buf);
               return TRUE;
            }
            fenced_buffer_destroy_cpu_storage_locked(fenced_buf);
         }
      }

      curr = next;
      next = curr->next;
   }

   return FALSE;
}


/**
 * Destroy CPU storage for this buffer.
 */
static void
fenced_buffer_destroy_cpu_storage_locked(struct fenced_buffer *fenced_buf)
{
   if(fenced_buf->data) {
      align_free(fenced_buf->data);
      fenced_buf->data = NULL;
      assert(fenced_buf->mgr->cpu_total_size >= fenced_buf->size);
      fenced_buf->mgr->cpu_total_size -= fenced_buf->size;
   }
}


/**
 * Create CPU storage for this buffer.
 */
static enum pipe_error
fenced_buffer_create_cpu_storage_locked(struct fenced_manager *fenced_mgr,
                                        struct fenced_buffer *fenced_buf)
{
   assert(!fenced_buf->data);
   if(fenced_buf->data)
      return PIPE_OK;

   if (fenced_mgr->cpu_total_size + fenced_buf->size > fenced_mgr->max_cpu_total_size)
      return PIPE_ERROR_OUT_OF_MEMORY;

   fenced_buf->data = align_malloc(fenced_buf->size, fenced_buf->desc.alignment);
   if(!fenced_buf->data)
      return PIPE_ERROR_OUT_OF_MEMORY;

   fenced_mgr->cpu_total_size += fenced_buf->size;

   return PIPE_OK;
}


/**
 * Destroy the GPU storage.
 */
static void
fenced_buffer_destroy_gpu_storage_locked(struct fenced_buffer *fenced_buf)
{
   if(fenced_buf->buffer) {
      pb_reference(&fenced_buf->buffer, NULL);
   }
}


/**
 * Try to create GPU storage for this buffer.
 *
 * This function is a shorthand around pb_manager::create_buffer for
 * fenced_buffer_create_gpu_storage_locked()'s benefit.
 */
static INLINE boolean
fenced_buffer_try_create_gpu_storage_locked(struct fenced_manager *fenced_mgr,
                                            struct fenced_buffer *fenced_buf)
{
   struct pb_manager *provider = fenced_mgr->provider;

   assert(!fenced_buf->buffer);

   fenced_buf->buffer = provider->create_buffer(fenced_mgr->provider,
                                                fenced_buf->size,
                                                &fenced_buf->desc);
   return fenced_buf->buffer ? TRUE : FALSE;
}


/**
 * Create GPU storage for this buffer.
 */
static enum pipe_error
fenced_buffer_create_gpu_storage_locked(struct fenced_manager *fenced_mgr,
                                        struct fenced_buffer *fenced_buf,
                                        boolean wait)
{
   assert(!fenced_buf->buffer);

   /*
    * Check for signaled buffers before trying to allocate.
    */
   fenced_manager_check_signalled_locked(fenced_mgr, FALSE);

   fenced_buffer_try_create_gpu_storage_locked(fenced_mgr, fenced_buf);

   /*
    * Keep trying while there is some sort of progress:
    * - fences are expiring,
    * - or buffers are being being swapped out from GPU memory into CPU memory.
    */
   while(!fenced_buf->buffer &&
         (fenced_manager_check_signalled_locked(fenced_mgr, FALSE) ||
          fenced_manager_free_gpu_storage_locked(fenced_mgr))) {
      fenced_buffer_try_create_gpu_storage_locked(fenced_mgr, fenced_buf);
   }

   if(!fenced_buf->buffer && wait) {
      /*
       * Same as before, but this time around, wait to free buffers if
       * necessary.
       */
      while(!fenced_buf->buffer &&
            (fenced_manager_check_signalled_locked(fenced_mgr, TRUE) ||
             fenced_manager_free_gpu_storage_locked(fenced_mgr))) {
         fenced_buffer_try_create_gpu_storage_locked(fenced_mgr, fenced_buf);
      }
   }

   if(!fenced_buf->buffer) {
      if(0)
         fenced_manager_dump_locked(fenced_mgr);

      /* give up */
      return PIPE_ERROR_OUT_OF_MEMORY;
   }

   return PIPE_OK;
}


static enum pipe_error
fenced_buffer_copy_storage_to_gpu_locked(struct fenced_buffer *fenced_buf)
{
   uint8_t *map;

   assert(fenced_buf->data);
   assert(fenced_buf->buffer);

   map = pb_map(fenced_buf->buffer, PIPE_BUFFER_USAGE_CPU_WRITE);
   if(!map)
      return PIPE_ERROR;

   memcpy(map, fenced_buf->data, fenced_buf->size);

   pb_unmap(fenced_buf->buffer);

   return PIPE_OK;
}


static enum pipe_error
fenced_buffer_copy_storage_to_cpu_locked(struct fenced_buffer *fenced_buf)
{
   const uint8_t *map;

   assert(fenced_buf->data);
   assert(fenced_buf->buffer);

   map = pb_map(fenced_buf->buffer, PIPE_BUFFER_USAGE_CPU_READ);
   if(!map)
      return PIPE_ERROR;

   memcpy(fenced_buf->data, map, fenced_buf->size);

   pb_unmap(fenced_buf->buffer);

   return PIPE_OK;
}


static void
fenced_buffer_destroy(struct pb_buffer *buf)
{
   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
   struct fenced_manager *fenced_mgr = fenced_buf->mgr;

   assert(!pipe_is_referenced(&fenced_buf->base.base.reference));

   pipe_mutex_lock(fenced_mgr->mutex);

   fenced_buffer_destroy_locked(fenced_mgr, fenced_buf);

   pipe_mutex_unlock(fenced_mgr->mutex);
}


static void *
fenced_buffer_map(struct pb_buffer *buf,
                  unsigned flags)
{
   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
   struct fenced_manager *fenced_mgr = fenced_buf->mgr;
   struct pb_fence_ops *ops = fenced_mgr->ops;
   void *map = NULL;

   pipe_mutex_lock(fenced_mgr->mutex);

   assert(!(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE));

   /*
    * Serialize writes.
    */
   while((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_WRITE) ||
         ((fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ) &&
          (flags & PIPE_BUFFER_USAGE_CPU_WRITE))) {

      /* 
       * Don't wait for the GPU to finish accessing it, if blocking is forbidden.
       */
      if((flags & PIPE_BUFFER_USAGE_DONTBLOCK) &&
          ops->fence_signalled(ops, fenced_buf->fence, 0) == 0) {
         goto done;
      }

      if (flags & PIPE_BUFFER_USAGE_UNSYNCHRONIZED) {
         break;
      }

      /*
       * Wait for the GPU to finish accessing. This will release and re-acquire
       * the mutex, so all copies of mutable state must be discarded.
       */
      fenced_buffer_finish_locked(fenced_mgr, fenced_buf);
   }

   if(fenced_buf->buffer) {
      map = pb_map(fenced_buf->buffer, flags);
   }
   else {
      assert(fenced_buf->data);
      map = fenced_buf->data;
   }

   if(map) {
      ++fenced_buf->mapcount;
      fenced_buf->flags |= flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE;
   }

done:
   pipe_mutex_unlock(fenced_mgr->mutex);

   return map;
}


static void
fenced_buffer_unmap(struct pb_buffer *buf)
{
   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
   struct fenced_manager *fenced_mgr = fenced_buf->mgr;

   pipe_mutex_lock(fenced_mgr->mutex);

   assert(fenced_buf->mapcount);
   if(fenced_buf->mapcount) {
      if (fenced_buf->buffer)
         pb_unmap(fenced_buf->buffer);
      --fenced_buf->mapcount;
      if(!fenced_buf->mapcount)
	 fenced_buf->flags &= ~PIPE_BUFFER_USAGE_CPU_READ_WRITE;
   }

   pipe_mutex_unlock(fenced_mgr->mutex);
}


static enum pipe_error
fenced_buffer_validate(struct pb_buffer *buf,
                       struct pb_validate *vl,
                       unsigned flags)
{
   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
   struct fenced_manager *fenced_mgr = fenced_buf->mgr;
   enum pipe_error ret;

   pipe_mutex_lock(fenced_mgr->mutex);

   if(!vl) {
      /* invalidate */
      fenced_buf->vl = NULL;
      fenced_buf->validation_flags = 0;
      ret = PIPE_OK;
      goto done;
   }

   assert(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE);
   assert(!(flags & ~PIPE_BUFFER_USAGE_GPU_READ_WRITE));
   flags &= PIPE_BUFFER_USAGE_GPU_READ_WRITE;

   /* Buffer cannot be validated in two different lists */
   if(fenced_buf->vl && fenced_buf->vl != vl) {
      ret = PIPE_ERROR_RETRY;
      goto done;
   }

   if(fenced_buf->vl == vl &&
      (fenced_buf->validation_flags & flags) == flags) {
      /* Nothing to do -- buffer already validated */
      ret = PIPE_OK;
      goto done;
   }

   /*
    * Create and update GPU storage.
    */
   if(!fenced_buf->buffer) {
      assert(!fenced_buf->mapcount);

      ret = fenced_buffer_create_gpu_storage_locked(fenced_mgr, fenced_buf, TRUE);
      if(ret != PIPE_OK) {
         goto done;
      }

      ret = fenced_buffer_copy_storage_to_gpu_locked(fenced_buf);
      if(ret != PIPE_OK) {
         fenced_buffer_destroy_gpu_storage_locked(fenced_buf);
         goto done;
      }

      if(fenced_buf->mapcount) {
         debug_printf("warning: validating a buffer while it is still mapped\n");
      }
      else {
         fenced_buffer_destroy_cpu_storage_locked(fenced_buf);
      }
   }

   ret = pb_validate(fenced_buf->buffer, vl, flags);
   if (ret != PIPE_OK)
      goto done;

   fenced_buf->vl = vl;
   fenced_buf->validation_flags |= flags;

done:
   pipe_mutex_unlock(fenced_mgr->mutex);

   return ret;
}


static void
fenced_buffer_fence(struct pb_buffer *buf,
                    struct pipe_fence_handle *fence)
{
   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
   struct fenced_manager *fenced_mgr = fenced_buf->mgr;
   struct pb_fence_ops *ops = fenced_mgr->ops;

   pipe_mutex_lock(fenced_mgr->mutex);

   assert(pipe_is_referenced(&fenced_buf->base.base.reference));
   assert(fenced_buf->buffer);

   if(fence != fenced_buf->fence) {
      assert(fenced_buf->vl);
      assert(fenced_buf->validation_flags);

      if (fenced_buf->fence) {
         boolean destroyed;
         destroyed = fenced_buffer_remove_locked(fenced_mgr, fenced_buf);
         assert(!destroyed);
      }
      if (fence) {
         ops->fence_reference(ops, &fenced_buf->fence, fence);
         fenced_buf->flags |= fenced_buf->validation_flags;
         fenced_buffer_add_locked(fenced_mgr, fenced_buf);
      }

      pb_fence(fenced_buf->buffer, fence);

      fenced_buf->vl = NULL;
      fenced_buf->validation_flags = 0;
   }

   pipe_mutex_unlock(fenced_mgr->mutex);
}


static void
fenced_buffer_get_base_buffer(struct pb_buffer *buf,
                              struct pb_buffer **base_buf,
                              pb_size *offset)
{
   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
   struct fenced_manager *fenced_mgr = fenced_buf->mgr;

   pipe_mutex_lock(fenced_mgr->mutex);

   /*
    * This should only be called when the buffer is validated. Typically
    * when processing relocations.
    */
   assert(fenced_buf->vl);
   assert(fenced_buf->buffer);

   if(fenced_buf->buffer)
      pb_get_base_buffer(fenced_buf->buffer, base_buf, offset);
   else {
      *base_buf = buf;
      *offset = 0;
   }

   pipe_mutex_unlock(fenced_mgr->mutex);
}


static const struct pb_vtbl
fenced_buffer_vtbl = {
      fenced_buffer_destroy,
      fenced_buffer_map,
      fenced_buffer_unmap,
      fenced_buffer_validate,
      fenced_buffer_fence,
      fenced_buffer_get_base_buffer
};


/**
 * Wrap a buffer in a fenced buffer.
 */
static struct pb_buffer *
fenced_bufmgr_create_buffer(struct pb_manager *mgr,
                            pb_size size,
                            const struct pb_desc *desc)
{
   struct fenced_manager *fenced_mgr = fenced_manager(mgr);
   struct fenced_buffer *fenced_buf;
   enum pipe_error ret;

   /*
    * Don't stall the GPU, waste time evicting buffers, or waste memory
    * trying to create a buffer that will most likely never fit into the
    * graphics aperture.
    */
   if(size > fenced_mgr->max_buffer_size) {
      goto no_buffer;
   }

   fenced_buf = CALLOC_STRUCT(fenced_buffer);
   if(!fenced_buf)
      goto no_buffer;

   pipe_reference_init(&fenced_buf->base.base.reference, 1);
   fenced_buf->base.base.alignment = desc->alignment;
   fenced_buf->base.base.usage = desc->usage;
   fenced_buf->base.base.size = size;
   fenced_buf->size = size;
   fenced_buf->desc = *desc;

   fenced_buf->base.vtbl = &fenced_buffer_vtbl;
   fenced_buf->mgr = fenced_mgr;

   pipe_mutex_lock(fenced_mgr->mutex);

   /*
    * Try to create GPU storage without stalling,
    */
   ret = fenced_buffer_create_gpu_storage_locked(fenced_mgr, fenced_buf, FALSE);

   /*
    * Attempt to use CPU memory to avoid stalling the GPU.
    */
   if(ret != PIPE_OK) {
      ret = fenced_buffer_create_cpu_storage_locked(fenced_mgr, fenced_buf);
   }

   /*
    * Create GPU storage, waiting for some to be available.
    */
   if(ret != PIPE_OK) {
      ret = fenced_buffer_create_gpu_storage_locked(fenced_mgr, fenced_buf, TRUE);
   }

   /*
    * Give up.
    */
   if(ret != PIPE_OK) {
      goto no_storage;
   }

   assert(fenced_buf->buffer || fenced_buf->data);

   LIST_ADDTAIL(&fenced_buf->head, &fenced_mgr->unfenced);
   ++fenced_mgr->num_unfenced;
   pipe_mutex_unlock(fenced_mgr->mutex);

   return &fenced_buf->base;

no_storage:
   pipe_mutex_unlock(fenced_mgr->mutex);
   FREE(fenced_buf);
no_buffer:
   return NULL;
}


static void
fenced_bufmgr_flush(struct pb_manager *mgr)
{
   struct fenced_manager *fenced_mgr = fenced_manager(mgr);

   pipe_mutex_lock(fenced_mgr->mutex);
   while(fenced_manager_check_signalled_locked(fenced_mgr, TRUE))
      ;
   pipe_mutex_unlock(fenced_mgr->mutex);

   assert(fenced_mgr->provider->flush);
   if(fenced_mgr->provider->flush)
      fenced_mgr->provider->flush(fenced_mgr->provider);
}


static void
fenced_bufmgr_destroy(struct pb_manager *mgr)
{
   struct fenced_manager *fenced_mgr = fenced_manager(mgr);

   pipe_mutex_lock(fenced_mgr->mutex);

   /* Wait on outstanding fences */
   while (fenced_mgr->num_fenced) {
      pipe_mutex_unlock(fenced_mgr->mutex);
#if defined(PIPE_OS_LINUX) || defined(PIPE_OS_BSD) || defined(PIPE_OS_SOLARIS)
      sched_yield();
#endif
      pipe_mutex_lock(fenced_mgr->mutex);
      while(fenced_manager_check_signalled_locked(fenced_mgr, TRUE))
         ;
   }

#ifdef DEBUG
   /*assert(!fenced_mgr->num_unfenced);*/
#endif

   pipe_mutex_unlock(fenced_mgr->mutex);
   pipe_mutex_destroy(fenced_mgr->mutex);

   if(fenced_mgr->provider)
      fenced_mgr->provider->destroy(fenced_mgr->provider);

   fenced_mgr->ops->destroy(fenced_mgr->ops);

   FREE(fenced_mgr);
}


struct pb_manager *
fenced_bufmgr_create(struct pb_manager *provider,
                     struct pb_fence_ops *ops,
                     pb_size max_buffer_size,
                     pb_size max_cpu_total_size)
{
   struct fenced_manager *fenced_mgr;

   if(!provider)
      return NULL;

   fenced_mgr = CALLOC_STRUCT(fenced_manager);
   if (!fenced_mgr)
      return NULL;

   fenced_mgr->base.destroy = fenced_bufmgr_destroy;
   fenced_mgr->base.create_buffer = fenced_bufmgr_create_buffer;
   fenced_mgr->base.flush = fenced_bufmgr_flush;

   fenced_mgr->provider = provider;
   fenced_mgr->ops = ops;
   fenced_mgr->max_buffer_size = max_buffer_size;
   fenced_mgr->max_cpu_total_size = max_cpu_total_size;

   LIST_INITHEAD(&fenced_mgr->fenced);
   fenced_mgr->num_fenced = 0;

   LIST_INITHEAD(&fenced_mgr->unfenced);
   fenced_mgr->num_unfenced = 0;

   pipe_mutex_init(fenced_mgr->mutex);

   return &fenced_mgr->base;
}