summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/vc4/vc4_bo.c
blob: 88ebd681d7ebb491c647b54a798072bf10151125 (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
1070
1071
1072
1073
1074
/*
 *  Copyright © 2015 Broadcom
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 as
 * published by the Free Software Foundation.
 */

/**
 * DOC: VC4 GEM BO management support
 *
 * The VC4 GPU architecture (both scanout and rendering) has direct
 * access to system memory with no MMU in between.  To support it, we
 * use the GEM CMA helper functions to allocate contiguous ranges of
 * physical memory for our BOs.
 *
 * Since the CMA allocator is very slow, we keep a cache of recently
 * freed BOs around so that the kernel's allocation of objects for 3D
 * rendering can return quickly.
 */

#include <linux/dma-buf.h>

#include "vc4_drv.h"
#include "uapi/drm/vc4_drm.h"

static const char * const bo_type_names[] = {
	"kernel",
	"V3D",
	"V3D shader",
	"dumb",
	"binner",
	"RCL",
	"BCL",
	"kernel BO cache",
};

static bool is_user_label(int label)
{
	return label >= VC4_BO_TYPE_COUNT;
}

static void vc4_bo_stats_print(struct drm_printer *p, struct vc4_dev *vc4)
{
	int i;

	for (i = 0; i < vc4->num_labels; i++) {
		if (!vc4->bo_labels[i].num_allocated)
			continue;

		drm_printf(p, "%30s: %6dkb BOs (%d)\n",
			   vc4->bo_labels[i].name,
			   vc4->bo_labels[i].size_allocated / 1024,
			   vc4->bo_labels[i].num_allocated);
	}

	mutex_lock(&vc4->purgeable.lock);
	if (vc4->purgeable.num)
		drm_printf(p, "%30s: %6zdkb BOs (%d)\n", "userspace BO cache",
			   vc4->purgeable.size / 1024, vc4->purgeable.num);

	if (vc4->purgeable.purged_num)
		drm_printf(p, "%30s: %6zdkb BOs (%d)\n", "total purged BO",
			   vc4->purgeable.purged_size / 1024,
			   vc4->purgeable.purged_num);
	mutex_unlock(&vc4->purgeable.lock);
}

static int vc4_bo_stats_debugfs(struct seq_file *m, void *unused)
{
	struct drm_info_node *node = (struct drm_info_node *)m->private;
	struct drm_device *dev = node->minor->dev;
	struct vc4_dev *vc4 = to_vc4_dev(dev);
	struct drm_printer p = drm_seq_file_printer(m);

	vc4_bo_stats_print(&p, vc4);

	return 0;
}

/* Takes ownership of *name and returns the appropriate slot for it in
 * the bo_labels[] array, extending it as necessary.
 *
 * This is inefficient and could use a hash table instead of walking
 * an array and strcmp()ing.  However, the assumption is that user
 * labeling will be infrequent (scanout buffers and other long-lived
 * objects, or debug driver builds), so we can live with it for now.
 */
static int vc4_get_user_label(struct vc4_dev *vc4, const char *name)
{
	int i;
	int free_slot = -1;

	for (i = 0; i < vc4->num_labels; i++) {
		if (!vc4->bo_labels[i].name) {
			free_slot = i;
		} else if (strcmp(vc4->bo_labels[i].name, name) == 0) {
			kfree(name);
			return i;
		}
	}

	if (free_slot != -1) {
		WARN_ON(vc4->bo_labels[free_slot].num_allocated != 0);
		vc4->bo_labels[free_slot].name = name;
		return free_slot;
	} else {
		u32 new_label_count = vc4->num_labels + 1;
		struct vc4_label *new_labels =
			krealloc(vc4->bo_labels,
				 new_label_count * sizeof(*new_labels),
				 GFP_KERNEL);

		if (!new_labels) {
			kfree(name);
			return -1;
		}

		free_slot = vc4->num_labels;
		vc4->bo_labels = new_labels;
		vc4->num_labels = new_label_count;

		vc4->bo_labels[free_slot].name = name;
		vc4->bo_labels[free_slot].num_allocated = 0;
		vc4->bo_labels[free_slot].size_allocated = 0;

		return free_slot;
	}
}

static void vc4_bo_set_label(struct drm_gem_object *gem_obj, int label)
{
	struct vc4_bo *bo = to_vc4_bo(gem_obj);
	struct vc4_dev *vc4 = to_vc4_dev(gem_obj->dev);

	lockdep_assert_held(&vc4->bo_lock);

	if (label != -1) {
		vc4->bo_labels[label].num_allocated++;
		vc4->bo_labels[label].size_allocated += gem_obj->size;
	}

	vc4->bo_labels[bo->label].num_allocated--;
	vc4->bo_labels[bo->label].size_allocated -= gem_obj->size;

	if (vc4->bo_labels[bo->label].num_allocated == 0 &&
	    is_user_label(bo->label)) {
		/* Free user BO label slots on last unreference.
		 * Slots are just where we track the stats for a given
		 * name, and once a name is unused we can reuse that
		 * slot.
		 */
		kfree(vc4->bo_labels[bo->label].name);
		vc4->bo_labels[bo->label].name = NULL;
	}

	bo->label = label;
}

static uint32_t bo_page_index(size_t size)
{
	return (size / PAGE_SIZE) - 1;
}

static void vc4_bo_destroy(struct vc4_bo *bo)
{
	struct drm_gem_object *obj = &bo->base.base;
	struct vc4_dev *vc4 = to_vc4_dev(obj->dev);

	lockdep_assert_held(&vc4->bo_lock);

	vc4_bo_set_label(obj, -1);

	if (bo->validated_shader) {
		kfree(bo->validated_shader->uniform_addr_offsets);
		kfree(bo->validated_shader->texture_samples);
		kfree(bo->validated_shader);
		bo->validated_shader = NULL;
	}

	drm_gem_cma_free_object(obj);
}

static void vc4_bo_remove_from_cache(struct vc4_bo *bo)
{
	struct vc4_dev *vc4 = to_vc4_dev(bo->base.base.dev);

	lockdep_assert_held(&vc4->bo_lock);
	list_del(&bo->unref_head);
	list_del(&bo->size_head);
}

static struct list_head *vc4_get_cache_list_for_size(struct drm_device *dev,
						     size_t size)
{
	struct vc4_dev *vc4 = to_vc4_dev(dev);
	uint32_t page_index = bo_page_index(size);

	if (vc4->bo_cache.size_list_size <= page_index) {
		uint32_t new_size = max(vc4->bo_cache.size_list_size * 2,
					page_index + 1);
		struct list_head *new_list;
		uint32_t i;

		new_list = kmalloc_array(new_size, sizeof(struct list_head),
					 GFP_KERNEL);
		if (!new_list)
			return NULL;

		/* Rebase the old cached BO lists to their new list
		 * head locations.
		 */
		for (i = 0; i < vc4->bo_cache.size_list_size; i++) {
			struct list_head *old_list =
				&vc4->bo_cache.size_list[i];

			if (list_empty(old_list))
				INIT_LIST_HEAD(&new_list[i]);
			else
				list_replace(old_list, &new_list[i]);
		}
		/* And initialize the brand new BO list heads. */
		for (i = vc4->bo_cache.size_list_size; i < new_size; i++)
			INIT_LIST_HEAD(&new_list[i]);

		kfree(vc4->bo_cache.size_list);
		vc4->bo_cache.size_list = new_list;
		vc4->bo_cache.size_list_size = new_size;
	}

	return &vc4->bo_cache.size_list[page_index];
}

static void vc4_bo_cache_purge(struct drm_device *dev)
{
	struct vc4_dev *vc4 = to_vc4_dev(dev);

	mutex_lock(&vc4->bo_lock);
	while (!list_empty(&vc4->bo_cache.time_list)) {
		struct vc4_bo *bo = list_last_entry(&vc4->bo_cache.time_list,
						    struct vc4_bo, unref_head);
		vc4_bo_remove_from_cache(bo);
		vc4_bo_destroy(bo);
	}
	mutex_unlock(&vc4->bo_lock);
}

void vc4_bo_add_to_purgeable_pool(struct vc4_bo *bo)
{
	struct vc4_dev *vc4 = to_vc4_dev(bo->base.base.dev);

	mutex_lock(&vc4->purgeable.lock);
	list_add_tail(&bo->size_head, &vc4->purgeable.list);
	vc4->purgeable.num++;
	vc4->purgeable.size += bo->base.base.size;
	mutex_unlock(&vc4->purgeable.lock);
}

static void vc4_bo_remove_from_purgeable_pool_locked(struct vc4_bo *bo)
{
	struct vc4_dev *vc4 = to_vc4_dev(bo->base.base.dev);

	/* list_del_init() is used here because the caller might release
	 * the purgeable lock in order to acquire the madv one and update the
	 * madv status.
	 * During this short period of time a user might decide to mark
	 * the BO as unpurgeable, and if bo->madv is set to
	 * VC4_MADV_DONTNEED it will try to remove the BO from the
	 * purgeable list which will fail if the ->next/prev fields
	 * are set to LIST_POISON1/LIST_POISON2 (which is what
	 * list_del() does).
	 * Re-initializing the list element guarantees that list_del()
	 * will work correctly even if it's a NOP.
	 */
	list_del_init(&bo->size_head);
	vc4->purgeable.num--;
	vc4->purgeable.size -= bo->base.base.size;
}

void vc4_bo_remove_from_purgeable_pool(struct vc4_bo *bo)
{
	struct vc4_dev *vc4 = to_vc4_dev(bo->base.base.dev);

	mutex_lock(&vc4->purgeable.lock);
	vc4_bo_remove_from_purgeable_pool_locked(bo);
	mutex_unlock(&vc4->purgeable.lock);
}

static void vc4_bo_purge(struct drm_gem_object *obj)
{
	struct vc4_bo *bo = to_vc4_bo(obj);
	struct drm_device *dev = obj->dev;

	WARN_ON(!mutex_is_locked(&bo->madv_lock));
	WARN_ON(bo->madv != VC4_MADV_DONTNEED);

	drm_vma_node_unmap(&obj->vma_node, dev->anon_inode->i_mapping);

	dma_free_wc(dev->dev, obj->size, bo->base.vaddr, bo->base.paddr);
	bo->base.vaddr = NULL;
	bo->madv = __VC4_MADV_PURGED;
}

static void vc4_bo_userspace_cache_purge(struct drm_device *dev)
{
	struct vc4_dev *vc4 = to_vc4_dev(dev);

	mutex_lock(&vc4->purgeable.lock);
	while (!list_empty(&vc4->purgeable.list)) {
		struct vc4_bo *bo = list_first_entry(&vc4->purgeable.list,
						     struct vc4_bo, size_head);
		struct drm_gem_object *obj = &bo->base.base;
		size_t purged_size = 0;

		vc4_bo_remove_from_purgeable_pool_locked(bo);

		/* Release the purgeable lock while we're purging the BO so
		 * that other people can continue inserting things in the
		 * purgeable pool without having to wait for all BOs to be
		 * purged.
		 */
		mutex_unlock(&vc4->purgeable.lock);
		mutex_lock(&bo->madv_lock);

		/* Since we released the purgeable pool lock before acquiring
		 * the BO madv one, the user may have marked the BO as WILLNEED
		 * and re-used it in the meantime.
		 * Before purging the BO we need to make sure
		 * - it is still marked as DONTNEED
		 * - it has not been re-inserted in the purgeable list
		 * - it is not used by HW blocks
		 * If one of these conditions is not met, just skip the entry.
		 */
		if (bo->madv == VC4_MADV_DONTNEED &&
		    list_empty(&bo->size_head) &&
		    !refcount_read(&bo->usecnt)) {
			purged_size = bo->base.base.size;
			vc4_bo_purge(obj);
		}
		mutex_unlock(&bo->madv_lock);
		mutex_lock(&vc4->purgeable.lock);

		if (purged_size) {
			vc4->purgeable.purged_size += purged_size;
			vc4->purgeable.purged_num++;
		}
	}
	mutex_unlock(&vc4->purgeable.lock);
}

static struct vc4_bo *vc4_bo_get_from_cache(struct drm_device *dev,
					    uint32_t size,
					    enum vc4_kernel_bo_type type)
{
	struct vc4_dev *vc4 = to_vc4_dev(dev);
	uint32_t page_index = bo_page_index(size);
	struct vc4_bo *bo = NULL;

	size = roundup(size, PAGE_SIZE);

	mutex_lock(&vc4->bo_lock);
	if (page_index >= vc4->bo_cache.size_list_size)
		goto out;

	if (list_empty(&vc4->bo_cache.size_list[page_index]))
		goto out;

	bo = list_first_entry(&vc4->bo_cache.size_list[page_index],
			      struct vc4_bo, size_head);
	vc4_bo_remove_from_cache(bo);
	kref_init(&bo->base.base.refcount);

out:
	if (bo)
		vc4_bo_set_label(&bo->base.base, type);
	mutex_unlock(&vc4->bo_lock);
	return bo;
}

/**
 * vc4_gem_create_object - Implementation of driver->gem_create_object.
 * @dev: DRM device
 * @size: Size in bytes of the memory the object will reference
 *
 * This lets the CMA helpers allocate object structs for us, and keep
 * our BO stats correct.
 */
struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size)
{
	struct vc4_dev *vc4 = to_vc4_dev(dev);
	struct vc4_bo *bo;

	bo = kzalloc(sizeof(*bo), GFP_KERNEL);
	if (!bo)
		return ERR_PTR(-ENOMEM);

	bo->madv = VC4_MADV_WILLNEED;
	refcount_set(&bo->usecnt, 0);
	mutex_init(&bo->madv_lock);
	mutex_lock(&vc4->bo_lock);
	bo->label = VC4_BO_TYPE_KERNEL;
	vc4->bo_labels[VC4_BO_TYPE_KERNEL].num_allocated++;
	vc4->bo_labels[VC4_BO_TYPE_KERNEL].size_allocated += size;
	mutex_unlock(&vc4->bo_lock);

	return &bo->base.base;
}

struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t unaligned_size,
			     bool allow_unzeroed, enum vc4_kernel_bo_type type)
{
	size_t size = roundup(unaligned_size, PAGE_SIZE);
	struct vc4_dev *vc4 = to_vc4_dev(dev);
	struct drm_gem_cma_object *cma_obj;
	struct vc4_bo *bo;

	if (size == 0)
		return ERR_PTR(-EINVAL);

	/* First, try to get a vc4_bo from the kernel BO cache. */
	bo = vc4_bo_get_from_cache(dev, size, type);
	if (bo) {
		if (!allow_unzeroed)
			memset(bo->base.vaddr, 0, bo->base.base.size);
		return bo;
	}

	cma_obj = drm_gem_cma_create(dev, size);
	if (IS_ERR(cma_obj)) {
		/*
		 * If we've run out of CMA memory, kill the cache of
		 * CMA allocations we've got laying around and try again.
		 */
		vc4_bo_cache_purge(dev);
		cma_obj = drm_gem_cma_create(dev, size);
	}

	if (IS_ERR(cma_obj)) {
		/*
		 * Still not enough CMA memory, purge the userspace BO
		 * cache and retry.
		 * This is sub-optimal since we purge the whole userspace
		 * BO cache which forces user that want to re-use the BO to
		 * restore its initial content.
		 * Ideally, we should purge entries one by one and retry
		 * after each to see if CMA allocation succeeds. Or even
		 * better, try to find an entry with at least the same
		 * size.
		 */
		vc4_bo_userspace_cache_purge(dev);
		cma_obj = drm_gem_cma_create(dev, size);
	}

	if (IS_ERR(cma_obj)) {
		struct drm_printer p = drm_info_printer(vc4->dev->dev);
		DRM_ERROR("Failed to allocate from CMA:\n");
		vc4_bo_stats_print(&p, vc4);
		return ERR_PTR(-ENOMEM);
	}
	bo = to_vc4_bo(&cma_obj->base);

	/* By default, BOs do not support the MADV ioctl. This will be enabled
	 * only on BOs that are exposed to userspace (V3D, V3D_SHADER and DUMB
	 * BOs).
	 */
	bo->madv = __VC4_MADV_NOTSUPP;

	mutex_lock(&vc4->bo_lock);
	vc4_bo_set_label(&cma_obj->base, type);
	mutex_unlock(&vc4->bo_lock);

	return bo;
}

int vc4_dumb_create(struct drm_file *file_priv,
		    struct drm_device *dev,
		    struct drm_mode_create_dumb *args)
{
	int min_pitch = DIV_ROUND_UP(args->width * args->bpp, 8);
	struct vc4_bo *bo = NULL;
	int ret;

	if (args->pitch < min_pitch)
		args->pitch = min_pitch;

	if (args->size < args->pitch * args->height)
		args->size = args->pitch * args->height;

	bo = vc4_bo_create(dev, args->size, false, VC4_BO_TYPE_DUMB);
	if (IS_ERR(bo))
		return PTR_ERR(bo);

	bo->madv = VC4_MADV_WILLNEED;

	ret = drm_gem_handle_create(file_priv, &bo->base.base, &args->handle);
	drm_gem_object_put_unlocked(&bo->base.base);

	return ret;
}

static void vc4_bo_cache_free_old(struct drm_device *dev)
{
	struct vc4_dev *vc4 = to_vc4_dev(dev);
	unsigned long expire_time = jiffies - msecs_to_jiffies(1000);

	lockdep_assert_held(&vc4->bo_lock);

	while (!list_empty(&vc4->bo_cache.time_list)) {
		struct vc4_bo *bo = list_last_entry(&vc4->bo_cache.time_list,
						    struct vc4_bo, unref_head);
		if (time_before(expire_time, bo->free_time)) {
			mod_timer(&vc4->bo_cache.time_timer,
				  round_jiffies_up(jiffies +
						   msecs_to_jiffies(1000)));
			return;
		}

		vc4_bo_remove_from_cache(bo);
		vc4_bo_destroy(bo);
	}
}

/* Called on the last userspace/kernel unreference of the BO.  Returns
 * it to the BO cache if possible, otherwise frees it.
 */
void vc4_free_object(struct drm_gem_object *gem_bo)
{
	struct drm_device *dev = gem_bo->dev;
	struct vc4_dev *vc4 = to_vc4_dev(dev);
	struct vc4_bo *bo = to_vc4_bo(gem_bo);
	struct list_head *cache_list;

	/* Remove the BO from the purgeable list. */
	mutex_lock(&bo->madv_lock);
	if (bo->madv == VC4_MADV_DONTNEED && !refcount_read(&bo->usecnt))
		vc4_bo_remove_from_purgeable_pool(bo);
	mutex_unlock(&bo->madv_lock);

	mutex_lock(&vc4->bo_lock);
	/* If the object references someone else's memory, we can't cache it.
	 */
	if (gem_bo->import_attach) {
		vc4_bo_destroy(bo);
		goto out;
	}

	/* Don't cache if it was publicly named. */
	if (gem_bo->name) {
		vc4_bo_destroy(bo);
		goto out;
	}

	/* If this object was partially constructed but CMA allocation
	 * had failed, just free it. Can also happen when the BO has been
	 * purged.
	 */
	if (!bo->base.vaddr) {
		vc4_bo_destroy(bo);
		goto out;
	}

	cache_list = vc4_get_cache_list_for_size(dev, gem_bo->size);
	if (!cache_list) {
		vc4_bo_destroy(bo);
		goto out;
	}

	if (bo->validated_shader) {
		kfree(bo->validated_shader->uniform_addr_offsets);
		kfree(bo->validated_shader->texture_samples);
		kfree(bo->validated_shader);
		bo->validated_shader = NULL;
	}

	/* Reset madv and usecnt before adding the BO to the cache. */
	bo->madv = __VC4_MADV_NOTSUPP;
	refcount_set(&bo->usecnt, 0);

	bo->t_format = false;
	bo->free_time = jiffies;
	list_add(&bo->size_head, cache_list);
	list_add(&bo->unref_head, &vc4->bo_cache.time_list);

	vc4_bo_set_label(&bo->base.base, VC4_BO_TYPE_KERNEL_CACHE);

	vc4_bo_cache_free_old(dev);

out:
	mutex_unlock(&vc4->bo_lock);
}

static void vc4_bo_cache_time_work(struct work_struct *work)
{
	struct vc4_dev *vc4 =
		container_of(work, struct vc4_dev, bo_cache.time_work);
	struct drm_device *dev = vc4->dev;

	mutex_lock(&vc4->bo_lock);
	vc4_bo_cache_free_old(dev);
	mutex_unlock(&vc4->bo_lock);
}

int vc4_bo_inc_usecnt(struct vc4_bo *bo)
{
	int ret;

	/* Fast path: if the BO is already retained by someone, no need to
	 * check the madv status.
	 */
	if (refcount_inc_not_zero(&bo->usecnt))
		return 0;

	mutex_lock(&bo->madv_lock);
	switch (bo->madv) {
	case VC4_MADV_WILLNEED:
		if (!refcount_inc_not_zero(&bo->usecnt))
			refcount_set(&bo->usecnt, 1);
		ret = 0;
		break;
	case VC4_MADV_DONTNEED:
		/* We shouldn't use a BO marked as purgeable if at least
		 * someone else retained its content by incrementing usecnt.
		 * Luckily the BO hasn't been purged yet, but something wrong
		 * is happening here. Just throw an error instead of
		 * authorizing this use case.
		 */
	case __VC4_MADV_PURGED:
		/* We can't use a purged BO. */
	default:
		/* Invalid madv value. */
		ret = -EINVAL;
		break;
	}
	mutex_unlock(&bo->madv_lock);

	return ret;
}

void vc4_bo_dec_usecnt(struct vc4_bo *bo)
{
	/* Fast path: if the BO is still retained by someone, no need to test
	 * the madv value.
	 */
	if (refcount_dec_not_one(&bo->usecnt))
		return;

	mutex_lock(&bo->madv_lock);
	if (refcount_dec_and_test(&bo->usecnt) &&
	    bo->madv == VC4_MADV_DONTNEED)
		vc4_bo_add_to_purgeable_pool(bo);
	mutex_unlock(&bo->madv_lock);
}

static void vc4_bo_cache_time_timer(struct timer_list *t)
{
	struct vc4_dev *vc4 = from_timer(vc4, t, bo_cache.time_timer);

	schedule_work(&vc4->bo_cache.time_work);
}

struct dma_buf *
vc4_prime_export(struct drm_device *dev, struct drm_gem_object *obj, int flags)
{
	struct vc4_bo *bo = to_vc4_bo(obj);
	struct dma_buf *dmabuf;
	int ret;

	if (bo->validated_shader) {
		DRM_DEBUG("Attempting to export shader BO\n");
		return ERR_PTR(-EINVAL);
	}

	/* Note: as soon as the BO is exported it becomes unpurgeable, because
	 * noone ever decrements the usecnt even if the reference held by the
	 * exported BO is released. This shouldn't be a problem since we don't
	 * expect exported BOs to be marked as purgeable.
	 */
	ret = vc4_bo_inc_usecnt(bo);
	if (ret) {
		DRM_ERROR("Failed to increment BO usecnt\n");
		return ERR_PTR(ret);
	}

	dmabuf = drm_gem_prime_export(dev, obj, flags);
	if (IS_ERR(dmabuf))
		vc4_bo_dec_usecnt(bo);

	return dmabuf;
}

vm_fault_t vc4_fault(struct vm_fault *vmf)
{
	struct vm_area_struct *vma = vmf->vma;
	struct drm_gem_object *obj = vma->vm_private_data;
	struct vc4_bo *bo = to_vc4_bo(obj);

	/* The only reason we would end up here is when user-space accesses
	 * BO's memory after it's been purged.
	 */
	mutex_lock(&bo->madv_lock);
	WARN_ON(bo->madv != __VC4_MADV_PURGED);
	mutex_unlock(&bo->madv_lock);

	return VM_FAULT_SIGBUS;
}

int vc4_mmap(struct file *filp, struct vm_area_struct *vma)
{
	struct drm_gem_object *gem_obj;
	unsigned long vm_pgoff;
	struct vc4_bo *bo;
	int ret;

	ret = drm_gem_mmap(filp, vma);
	if (ret)
		return ret;

	gem_obj = vma->vm_private_data;
	bo = to_vc4_bo(gem_obj);

	if (bo->validated_shader && (vma->vm_flags & VM_WRITE)) {
		DRM_DEBUG("mmaping of shader BOs for writing not allowed.\n");
		return -EINVAL;
	}

	if (bo->madv != VC4_MADV_WILLNEED) {
		DRM_DEBUG("mmaping of %s BO not allowed\n",
			  bo->madv == VC4_MADV_DONTNEED ?
			  "purgeable" : "purged");
		return -EINVAL;
	}

	/*
	 * Clear the VM_PFNMAP flag that was set by drm_gem_mmap(), and set the
	 * vm_pgoff (used as a fake buffer offset by DRM) to 0 as we want to map
	 * the whole buffer.
	 */
	vma->vm_flags &= ~VM_PFNMAP;

	/* This ->vm_pgoff dance is needed to make all parties happy:
	 * - dma_mmap_wc() uses ->vm_pgoff as an offset within the allocated
	 *   mem-region, hence the need to set it to zero (the value set by
	 *   the DRM core is a virtual offset encoding the GEM object-id)
	 * - the mmap() core logic needs ->vm_pgoff to be restored to its
	 *   initial value before returning from this function because it
	 *   encodes the  offset of this GEM in the dev->anon_inode pseudo-file
	 *   and this information will be used when we invalidate userspace
	 *   mappings  with drm_vma_node_unmap() (called from vc4_gem_purge()).
	 */
	vm_pgoff = vma->vm_pgoff;
	vma->vm_pgoff = 0;
	ret = dma_mmap_wc(bo->base.base.dev->dev, vma, bo->base.vaddr,
			  bo->base.paddr, vma->vm_end - vma->vm_start);
	vma->vm_pgoff = vm_pgoff;

	if (ret)
		drm_gem_vm_close(vma);

	return ret;
}

int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)
{
	struct vc4_bo *bo = to_vc4_bo(obj);

	if (bo->validated_shader && (vma->vm_flags & VM_WRITE)) {
		DRM_DEBUG("mmaping of shader BOs for writing not allowed.\n");
		return -EINVAL;
	}

	return drm_gem_cma_prime_mmap(obj, vma);
}

void *vc4_prime_vmap(struct drm_gem_object *obj)
{
	struct vc4_bo *bo = to_vc4_bo(obj);

	if (bo->validated_shader) {
		DRM_DEBUG("mmaping of shader BOs not allowed.\n");
		return ERR_PTR(-EINVAL);
	}

	return drm_gem_cma_prime_vmap(obj);
}

struct drm_gem_object *
vc4_prime_import_sg_table(struct drm_device *dev,
			  struct dma_buf_attachment *attach,
			  struct sg_table *sgt)
{
	struct drm_gem_object *obj;

	obj = drm_gem_cma_prime_import_sg_table(dev, attach, sgt);
	if (IS_ERR(obj))
		return obj;

	obj->resv = attach->dmabuf->resv;

	return obj;
}

int vc4_create_bo_ioctl(struct drm_device *dev, void *data,
			struct drm_file *file_priv)
{
	struct drm_vc4_create_bo *args = data;
	struct vc4_bo *bo = NULL;
	int ret;

	/*
	 * We can't allocate from the BO cache, because the BOs don't
	 * get zeroed, and that might leak data between users.
	 */
	bo = vc4_bo_create(dev, args->size, false, VC4_BO_TYPE_V3D);
	if (IS_ERR(bo))
		return PTR_ERR(bo);

	bo->madv = VC4_MADV_WILLNEED;

	ret = drm_gem_handle_create(file_priv, &bo->base.base, &args->handle);
	drm_gem_object_put_unlocked(&bo->base.base);

	return ret;
}

int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
		      struct drm_file *file_priv)
{
	struct drm_vc4_mmap_bo *args = data;
	struct drm_gem_object *gem_obj;

	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
	if (!gem_obj) {
		DRM_DEBUG("Failed to look up GEM BO %d\n", args->handle);
		return -EINVAL;
	}

	/* The mmap offset was set up at BO allocation time. */
	args->offset = drm_vma_node_offset_addr(&gem_obj->vma_node);

	drm_gem_object_put_unlocked(gem_obj);
	return 0;
}

int
vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data,
			   struct drm_file *file_priv)
{
	struct drm_vc4_create_shader_bo *args = data;
	struct vc4_bo *bo = NULL;
	int ret;

	if (args->size == 0)
		return -EINVAL;

	if (args->size % sizeof(u64) != 0)
		return -EINVAL;

	if (args->flags != 0) {
		DRM_INFO("Unknown flags set: 0x%08x\n", args->flags);
		return -EINVAL;
	}

	if (args->pad != 0) {
		DRM_INFO("Pad set: 0x%08x\n", args->pad);
		return -EINVAL;
	}

	bo = vc4_bo_create(dev, args->size, true, VC4_BO_TYPE_V3D_SHADER);
	if (IS_ERR(bo))
		return PTR_ERR(bo);

	bo->madv = VC4_MADV_WILLNEED;

	if (copy_from_user(bo->base.vaddr,
			     (void __user *)(uintptr_t)args->data,
			     args->size)) {
		ret = -EFAULT;
		goto fail;
	}
	/* Clear the rest of the memory from allocating from the BO
	 * cache.
	 */
	memset(bo->base.vaddr + args->size, 0,
	       bo->base.base.size - args->size);

	bo->validated_shader = vc4_validate_shader(&bo->base);
	if (!bo->validated_shader) {
		ret = -EINVAL;
		goto fail;
	}

	/* We have to create the handle after validation, to avoid
	 * races for users to do doing things like mmap the shader BO.
	 */
	ret = drm_gem_handle_create(file_priv, &bo->base.base, &args->handle);

 fail:
	drm_gem_object_put_unlocked(&bo->base.base);

	return ret;
}

/**
 * vc4_set_tiling_ioctl() - Sets the tiling modifier for a BO.
 * @dev: DRM device
 * @data: ioctl argument
 * @file_priv: DRM file for this fd
 *
 * The tiling state of the BO decides the default modifier of an fb if
 * no specific modifier was set by userspace, and the return value of
 * vc4_get_tiling_ioctl() (so that userspace can treat a BO it
 * received from dmabuf as the same tiling format as the producer
 * used).
 */
int vc4_set_tiling_ioctl(struct drm_device *dev, void *data,
			 struct drm_file *file_priv)
{
	struct drm_vc4_set_tiling *args = data;
	struct drm_gem_object *gem_obj;
	struct vc4_bo *bo;
	bool t_format;

	if (args->flags != 0)
		return -EINVAL;

	switch (args->modifier) {
	case DRM_FORMAT_MOD_NONE:
		t_format = false;
		break;
	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
		t_format = true;
		break;
	default:
		return -EINVAL;
	}

	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
	if (!gem_obj) {
		DRM_DEBUG("Failed to look up GEM BO %d\n", args->handle);
		return -ENOENT;
	}
	bo = to_vc4_bo(gem_obj);
	bo->t_format = t_format;

	drm_gem_object_put_unlocked(gem_obj);

	return 0;
}

/**
 * vc4_get_tiling_ioctl() - Gets the tiling modifier for a BO.
 * @dev: DRM device
 * @data: ioctl argument
 * @file_priv: DRM file for this fd
 *
 * Returns the tiling modifier for a BO as set by vc4_set_tiling_ioctl().
 */
int vc4_get_tiling_ioctl(struct drm_device *dev, void *data,
			 struct drm_file *file_priv)
{
	struct drm_vc4_get_tiling *args = data;
	struct drm_gem_object *gem_obj;
	struct vc4_bo *bo;

	if (args->flags != 0 || args->modifier != 0)
		return -EINVAL;

	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
	if (!gem_obj) {
		DRM_DEBUG("Failed to look up GEM BO %d\n", args->handle);
		return -ENOENT;
	}
	bo = to_vc4_bo(gem_obj);

	if (bo->t_format)
		args->modifier = DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED;
	else
		args->modifier = DRM_FORMAT_MOD_NONE;

	drm_gem_object_put_unlocked(gem_obj);

	return 0;
}

int vc4_bo_cache_init(struct drm_device *dev)
{
	struct vc4_dev *vc4 = to_vc4_dev(dev);
	int i;

	/* Create the initial set of BO labels that the kernel will
	 * use.  This lets us avoid a bunch of string reallocation in
	 * the kernel's draw and BO allocation paths.
	 */
	vc4->bo_labels = kcalloc(VC4_BO_TYPE_COUNT, sizeof(*vc4->bo_labels),
				 GFP_KERNEL);
	if (!vc4->bo_labels)
		return -ENOMEM;
	vc4->num_labels = VC4_BO_TYPE_COUNT;

	BUILD_BUG_ON(ARRAY_SIZE(bo_type_names) != VC4_BO_TYPE_COUNT);
	for (i = 0; i < VC4_BO_TYPE_COUNT; i++)
		vc4->bo_labels[i].name = bo_type_names[i];

	mutex_init(&vc4->bo_lock);

	vc4_debugfs_add_file(dev, "bo_stats", vc4_bo_stats_debugfs, NULL);

	INIT_LIST_HEAD(&vc4->bo_cache.time_list);

	INIT_WORK(&vc4->bo_cache.time_work, vc4_bo_cache_time_work);
	timer_setup(&vc4->bo_cache.time_timer, vc4_bo_cache_time_timer, 0);

	return 0;
}

void vc4_bo_cache_destroy(struct drm_device *dev)
{
	struct vc4_dev *vc4 = to_vc4_dev(dev);
	int i;

	del_timer(&vc4->bo_cache.time_timer);
	cancel_work_sync(&vc4->bo_cache.time_work);

	vc4_bo_cache_purge(dev);

	for (i = 0; i < vc4->num_labels; i++) {
		if (vc4->bo_labels[i].num_allocated) {
			DRM_ERROR("Destroying BO cache with %d %s "
				  "BOs still allocated\n",
				  vc4->bo_labels[i].num_allocated,
				  vc4->bo_labels[i].name);
		}

		if (is_user_label(i))
			kfree(vc4->bo_labels[i].name);
	}
	kfree(vc4->bo_labels);
}

int vc4_label_bo_ioctl(struct drm_device *dev, void *data,
		       struct drm_file *file_priv)
{
	struct vc4_dev *vc4 = to_vc4_dev(dev);
	struct drm_vc4_label_bo *args = data;
	char *name;
	struct drm_gem_object *gem_obj;
	int ret = 0, label;

	if (!args->len)
		return -EINVAL;

	name = strndup_user(u64_to_user_ptr(args->name), args->len + 1);
	if (IS_ERR(name))
		return PTR_ERR(name);

	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
	if (!gem_obj) {
		DRM_ERROR("Failed to look up GEM BO %d\n", args->handle);
		kfree(name);
		return -ENOENT;
	}

	mutex_lock(&vc4->bo_lock);
	label = vc4_get_user_label(vc4, name);
	if (label != -1)
		vc4_bo_set_label(gem_obj, label);
	else
		ret = -ENOMEM;
	mutex_unlock(&vc4->bo_lock);

	drm_gem_object_put_unlocked(gem_obj);

	return ret;
}