summaryrefslogtreecommitdiffstats
path: root/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
blob: 3b355aeb62fd353320fd47260bb05263e0e998ab (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
/*
 * Copyright 2008 Jerome Glisse.
 * 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, sublicense,
 * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
 * PRECISION INSIGHT 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.
 *
 * Authors:
 *    Jerome Glisse <glisse@freedesktop.org>
 */
#include <linux/list_sort.h>
#include <drm/drmP.h>
#include <drm/amdgpu_drm.h>
#include "amdgpu.h"
#include "amdgpu_trace.h"

#define AMDGPU_CS_MAX_PRIORITY		32u
#define AMDGPU_CS_NUM_BUCKETS		(AMDGPU_CS_MAX_PRIORITY + 1)

/* This is based on the bucket sort with O(n) time complexity.
 * An item with priority "i" is added to bucket[i]. The lists are then
 * concatenated in descending order.
 */
struct amdgpu_cs_buckets {
	struct list_head bucket[AMDGPU_CS_NUM_BUCKETS];
};

static void amdgpu_cs_buckets_init(struct amdgpu_cs_buckets *b)
{
	unsigned i;

	for (i = 0; i < AMDGPU_CS_NUM_BUCKETS; i++)
		INIT_LIST_HEAD(&b->bucket[i]);
}

static void amdgpu_cs_buckets_add(struct amdgpu_cs_buckets *b,
				  struct list_head *item, unsigned priority)
{
	/* Since buffers which appear sooner in the relocation list are
	 * likely to be used more often than buffers which appear later
	 * in the list, the sort mustn't change the ordering of buffers
	 * with the same priority, i.e. it must be stable.
	 */
	list_add_tail(item, &b->bucket[min(priority, AMDGPU_CS_MAX_PRIORITY)]);
}

static void amdgpu_cs_buckets_get_list(struct amdgpu_cs_buckets *b,
				       struct list_head *out_list)
{
	unsigned i;

	/* Connect the sorted buckets in the output list. */
	for (i = 0; i < AMDGPU_CS_NUM_BUCKETS; i++) {
		list_splice(&b->bucket[i], out_list);
	}
}

int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
		       u32 ip_instance, u32 ring,
		       struct amdgpu_ring **out_ring)
{
	/* Right now all IPs have only one instance - multiple rings. */
	if (ip_instance != 0) {
		DRM_ERROR("invalid ip instance: %d\n", ip_instance);
		return -EINVAL;
	}

	switch (ip_type) {
	default:
		DRM_ERROR("unknown ip type: %d\n", ip_type);
		return -EINVAL;
	case AMDGPU_HW_IP_GFX:
		if (ring < adev->gfx.num_gfx_rings) {
			*out_ring = &adev->gfx.gfx_ring[ring];
		} else {
			DRM_ERROR("only %d gfx rings are supported now\n",
				  adev->gfx.num_gfx_rings);
			return -EINVAL;
		}
		break;
	case AMDGPU_HW_IP_COMPUTE:
		if (ring < adev->gfx.num_compute_rings) {
			*out_ring = &adev->gfx.compute_ring[ring];
		} else {
			DRM_ERROR("only %d compute rings are supported now\n",
				  adev->gfx.num_compute_rings);
			return -EINVAL;
		}
		break;
	case AMDGPU_HW_IP_DMA:
		if (ring < 2) {
			*out_ring = &adev->sdma[ring].ring;
		} else {
			DRM_ERROR("only two SDMA rings are supported\n");
			return -EINVAL;
		}
		break;
	case AMDGPU_HW_IP_UVD:
		*out_ring = &adev->uvd.ring;
		break;
	case AMDGPU_HW_IP_VCE:
		if (ring < 2){
			*out_ring = &adev->vce.ring[ring];
		} else {
			DRM_ERROR("only two VCE rings are supported\n");
			return -EINVAL;
		}
		break;
	}
	return 0;
}

struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev,
                                               struct drm_file *filp,
                                               struct amdgpu_ctx *ctx,
                                               struct amdgpu_ib *ibs,
                                               uint32_t num_ibs)
{
	struct amdgpu_cs_parser *parser;
	int i;

	parser = kzalloc(sizeof(struct amdgpu_cs_parser), GFP_KERNEL);
	if (!parser)
		return NULL;

	parser->adev = adev;
	parser->filp = filp;
	parser->ctx = ctx;
	parser->ibs = ibs;
	parser->num_ibs = num_ibs;
	for (i = 0; i < num_ibs; i++)
		ibs[i].ctx = ctx;

	return parser;
}

int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
{
	union drm_amdgpu_cs *cs = data;
	uint64_t *chunk_array_user;
	uint64_t *chunk_array = NULL;
	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
	unsigned size, i;
	int r = 0;

	if (!cs->in.num_chunks)
		goto out;

	p->ctx = amdgpu_ctx_get(fpriv, cs->in.ctx_id);
	if (!p->ctx) {
		r = -EINVAL;
		goto out;
	}
	p->bo_list = amdgpu_bo_list_get(fpriv, cs->in.bo_list_handle);

	/* get chunks */
	INIT_LIST_HEAD(&p->validated);
	chunk_array = kmalloc_array(cs->in.num_chunks, sizeof(uint64_t), GFP_KERNEL);
	if (chunk_array == NULL) {
		r = -ENOMEM;
		goto out;
	}

	chunk_array_user = (uint64_t __user *)(cs->in.chunks);
	if (copy_from_user(chunk_array, chunk_array_user,
			   sizeof(uint64_t)*cs->in.num_chunks)) {
		r = -EFAULT;
		goto out;
	}

	p->nchunks = cs->in.num_chunks;
	p->chunks = kmalloc_array(p->nchunks, sizeof(struct amdgpu_cs_chunk),
			    GFP_KERNEL);
	if (p->chunks == NULL) {
		r = -ENOMEM;
		goto out;
	}

	for (i = 0; i < p->nchunks; i++) {
		struct drm_amdgpu_cs_chunk __user **chunk_ptr = NULL;
		struct drm_amdgpu_cs_chunk user_chunk;
		uint32_t __user *cdata;

		chunk_ptr = (void __user *)chunk_array[i];
		if (copy_from_user(&user_chunk, chunk_ptr,
				       sizeof(struct drm_amdgpu_cs_chunk))) {
			r = -EFAULT;
			goto out;
		}
		p->chunks[i].chunk_id = user_chunk.chunk_id;
		p->chunks[i].length_dw = user_chunk.length_dw;

		size = p->chunks[i].length_dw;
		cdata = (void __user *)user_chunk.chunk_data;
		p->chunks[i].user_ptr = cdata;

		p->chunks[i].kdata = drm_malloc_ab(size, sizeof(uint32_t));
		if (p->chunks[i].kdata == NULL) {
			r = -ENOMEM;
			goto out;
		}
		size *= sizeof(uint32_t);
		if (copy_from_user(p->chunks[i].kdata, cdata, size)) {
			r = -EFAULT;
			goto out;
		}

		switch (p->chunks[i].chunk_id) {
		case AMDGPU_CHUNK_ID_IB:
			p->num_ibs++;
			break;

		case AMDGPU_CHUNK_ID_FENCE:
			size = sizeof(struct drm_amdgpu_cs_chunk_fence);
			if (p->chunks[i].length_dw * sizeof(uint32_t) >= size) {
				uint32_t handle;
				struct drm_gem_object *gobj;
				struct drm_amdgpu_cs_chunk_fence *fence_data;

				fence_data = (void *)p->chunks[i].kdata;
				handle = fence_data->handle;
				gobj = drm_gem_object_lookup(p->adev->ddev,
							     p->filp, handle);
				if (gobj == NULL) {
					r = -EINVAL;
					goto out;
				}

				p->uf.bo = gem_to_amdgpu_bo(gobj);
				p->uf.offset = fence_data->offset;
			} else {
				r = -EINVAL;
				goto out;
			}
			break;

		case AMDGPU_CHUNK_ID_DEPENDENCIES:
			break;

		default:
			r = -EINVAL;
			goto out;
		}
	}


	p->ibs = kcalloc(p->num_ibs, sizeof(struct amdgpu_ib), GFP_KERNEL);
	if (!p->ibs)
		r = -ENOMEM;

out:
	kfree(chunk_array);
	return r;
}

/* Returns how many bytes TTM can move per IB.
 */
static u64 amdgpu_cs_get_threshold_for_moves(struct amdgpu_device *adev)
{
	u64 real_vram_size = adev->mc.real_vram_size;
	u64 vram_usage = atomic64_read(&adev->vram_usage);

	/* This function is based on the current VRAM usage.
	 *
	 * - If all of VRAM is free, allow relocating the number of bytes that
	 *   is equal to 1/4 of the size of VRAM for this IB.

	 * - If more than one half of VRAM is occupied, only allow relocating
	 *   1 MB of data for this IB.
	 *
	 * - From 0 to one half of used VRAM, the threshold decreases
	 *   linearly.
	 *         __________________
	 * 1/4 of -|\               |
	 * VRAM    | \              |
	 *         |  \             |
	 *         |   \            |
	 *         |    \           |
	 *         |     \          |
	 *         |      \         |
	 *         |       \________|1 MB
	 *         |----------------|
	 *    VRAM 0 %             100 %
	 *         used            used
	 *
	 * Note: It's a threshold, not a limit. The threshold must be crossed
	 * for buffer relocations to stop, so any buffer of an arbitrary size
	 * can be moved as long as the threshold isn't crossed before
	 * the relocation takes place. We don't want to disable buffer
	 * relocations completely.
	 *
	 * The idea is that buffers should be placed in VRAM at creation time
	 * and TTM should only do a minimum number of relocations during
	 * command submission. In practice, you need to submit at least
	 * a dozen IBs to move all buffers to VRAM if they are in GTT.
	 *
	 * Also, things can get pretty crazy under memory pressure and actual
	 * VRAM usage can change a lot, so playing safe even at 50% does
	 * consistently increase performance.
	 */

	u64 half_vram = real_vram_size >> 1;
	u64 half_free_vram = vram_usage >= half_vram ? 0 : half_vram - vram_usage;
	u64 bytes_moved_threshold = half_free_vram >> 1;
	return max(bytes_moved_threshold, 1024*1024ull);
}

int amdgpu_cs_list_validate(struct amdgpu_cs_parser *p)
{
	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
	struct amdgpu_vm *vm = &fpriv->vm;
	struct amdgpu_device *adev = p->adev;
	struct amdgpu_bo_list_entry *lobj;
	struct list_head duplicates;
	struct amdgpu_bo *bo;
	u64 bytes_moved = 0, initial_bytes_moved;
	u64 bytes_moved_threshold = amdgpu_cs_get_threshold_for_moves(adev);
	int r;

	INIT_LIST_HEAD(&duplicates);
	r = ttm_eu_reserve_buffers(&p->ticket, &p->validated, true, &duplicates);
	if (unlikely(r != 0)) {
		return r;
	}

	list_for_each_entry(lobj, &p->validated, tv.head) {
		bo = lobj->robj;
		if (!bo->pin_count) {
			u32 domain = lobj->prefered_domains;
			u32 current_domain =
				amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);

			/* Check if this buffer will be moved and don't move it
			 * if we have moved too many buffers for this IB already.
			 *
			 * Note that this allows moving at least one buffer of
			 * any size, because it doesn't take the current "bo"
			 * into account. We don't want to disallow buffer moves
			 * completely.
			 */
			if ((lobj->allowed_domains & current_domain) != 0 &&
			    (domain & current_domain) == 0 && /* will be moved */
			    bytes_moved > bytes_moved_threshold) {
				/* don't move it */
				domain = current_domain;
			}

		retry:
			amdgpu_ttm_placement_from_domain(bo, domain);
			initial_bytes_moved = atomic64_read(&adev->num_bytes_moved);
			r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
			bytes_moved += atomic64_read(&adev->num_bytes_moved) -
				       initial_bytes_moved;

			if (unlikely(r)) {
				if (r != -ERESTARTSYS && domain != lobj->allowed_domains) {
					domain = lobj->allowed_domains;
					goto retry;
				}
				ttm_eu_backoff_reservation(&p->ticket, &p->validated);
				return r;
			}
		}
		lobj->bo_va = amdgpu_vm_bo_find(vm, bo);
	}
	return 0;
}

static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
{
	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
	struct amdgpu_cs_buckets buckets;
	bool need_mmap_lock = false;
	int i, r;

	if (p->bo_list) {
		need_mmap_lock = p->bo_list->has_userptr;
		amdgpu_cs_buckets_init(&buckets);
		for (i = 0; i < p->bo_list->num_entries; i++)
			amdgpu_cs_buckets_add(&buckets, &p->bo_list->array[i].tv.head,
								  p->bo_list->array[i].priority);

		amdgpu_cs_buckets_get_list(&buckets, &p->validated);
	}

	p->vm_bos = amdgpu_vm_get_bos(p->adev, &fpriv->vm,
				      &p->validated);

	if (need_mmap_lock)
		down_read(&current->mm->mmap_sem);

	r = amdgpu_cs_list_validate(p);

	if (need_mmap_lock)
		up_read(&current->mm->mmap_sem);

	return r;
}

static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
{
	struct amdgpu_bo_list_entry *e;
	int r;

	list_for_each_entry(e, &p->validated, tv.head) {
		struct reservation_object *resv = e->robj->tbo.resv;
		r = amdgpu_sync_resv(p->adev, &p->ibs[0].sync, resv, p->filp);

		if (r)
			return r;
	}
	return 0;
}

static int cmp_size_smaller_first(void *priv, struct list_head *a,
				  struct list_head *b)
{
	struct amdgpu_bo_list_entry *la = list_entry(a, struct amdgpu_bo_list_entry, tv.head);
	struct amdgpu_bo_list_entry *lb = list_entry(b, struct amdgpu_bo_list_entry, tv.head);

	/* Sort A before B if A is smaller. */
	return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages;
}

static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int error, bool backoff)
{
	if (!error) {
		/* Sort the buffer list from the smallest to largest buffer,
		 * which affects the order of buffers in the LRU list.
		 * This assures that the smallest buffers are added first
		 * to the LRU list, so they are likely to be later evicted
		 * first, instead of large buffers whose eviction is more
		 * expensive.
		 *
		 * This slightly lowers the number of bytes moved by TTM
		 * per frame under memory pressure.
		 */
		list_sort(NULL, &parser->validated, cmp_size_smaller_first);

		ttm_eu_fence_buffer_objects(&parser->ticket,
				&parser->validated,
				&parser->ibs[parser->num_ibs-1].fence->base);
	} else if (backoff) {
		ttm_eu_backoff_reservation(&parser->ticket,
					   &parser->validated);
	}
}

static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser)
{
	unsigned i;
	if (parser->ctx)
		amdgpu_ctx_put(parser->ctx);
	if (parser->bo_list)
		amdgpu_bo_list_put(parser->bo_list);

	drm_free_large(parser->vm_bos);
	for (i = 0; i < parser->nchunks; i++)
		drm_free_large(parser->chunks[i].kdata);
	kfree(parser->chunks);
	if (!amdgpu_enable_scheduler)
	{
		if (parser->ibs)
			for (i = 0; i < parser->num_ibs; i++)
				amdgpu_ib_free(parser->adev, &parser->ibs[i]);
		kfree(parser->ibs);
		if (parser->uf.bo)
			drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
	}

	kfree(parser);
}

/**
 * cs_parser_fini() - clean parser states
 * @parser:	parser structure holding parsing context.
 * @error:	error number
 *
 * If error is set than unvalidate buffer, otherwise just free memory
 * used by parsing context.
 **/
static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
{
       amdgpu_cs_parser_fini_early(parser, error, backoff);
       amdgpu_cs_parser_fini_late(parser);
}

static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
				   struct amdgpu_vm *vm)
{
	struct amdgpu_device *adev = p->adev;
	struct amdgpu_bo_va *bo_va;
	struct amdgpu_bo *bo;
	int i, r;

	r = amdgpu_vm_update_page_directory(adev, vm);
	if (r)
		return r;

	r = amdgpu_sync_fence(adev, &p->ibs[0].sync, vm->page_directory_fence);
	if (r)
		return r;

	r = amdgpu_vm_clear_freed(adev, vm);
	if (r)
		return r;

	if (p->bo_list) {
		for (i = 0; i < p->bo_list->num_entries; i++) {
			struct fence *f;

			/* ignore duplicates */
			bo = p->bo_list->array[i].robj;
			if (!bo)
				continue;

			bo_va = p->bo_list->array[i].bo_va;
			if (bo_va == NULL)
				continue;

			r = amdgpu_vm_bo_update(adev, bo_va, &bo->tbo.mem);
			if (r)
				return r;

			f = bo_va->last_pt_update;
			r = amdgpu_sync_fence(adev, &p->ibs[0].sync, f);
			if (r)
				return r;
		}
	}

	return amdgpu_vm_clear_invalids(adev, vm, &p->ibs[0].sync);
}

static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
				 struct amdgpu_cs_parser *parser)
{
	struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
	struct amdgpu_vm *vm = &fpriv->vm;
	struct amdgpu_ring *ring;
	int i, r;

	if (parser->num_ibs == 0)
		return 0;

	/* Only for UVD/VCE VM emulation */
	for (i = 0; i < parser->num_ibs; i++) {
		ring = parser->ibs[i].ring;
		if (ring->funcs->parse_cs) {
			r = amdgpu_ring_parse_cs(ring, parser, i);
			if (r)
				return r;
		}
	}

	mutex_lock(&vm->mutex);
	r = amdgpu_bo_vm_update_pte(parser, vm);
	if (r) {
		goto out;
	}
	amdgpu_cs_sync_rings(parser);
	if (!amdgpu_enable_scheduler)
		r = amdgpu_ib_schedule(adev, parser->num_ibs, parser->ibs,
				       parser->filp);

out:
	mutex_unlock(&vm->mutex);
	return r;
}

static int amdgpu_cs_handle_lockup(struct amdgpu_device *adev, int r)
{
	if (r == -EDEADLK) {
		r = amdgpu_gpu_reset(adev);
		if (!r)
			r = -EAGAIN;
	}
	return r;
}

static int amdgpu_cs_ib_fill(struct amdgpu_device *adev,
			     struct amdgpu_cs_parser *parser)
{
	struct amdgpu_fpriv *fpriv = parser->filp->driver_priv;
	struct amdgpu_vm *vm = &fpriv->vm;
	int i, j;
	int r;

	for (i = 0, j = 0; i < parser->nchunks && j < parser->num_ibs; i++) {
		struct amdgpu_cs_chunk *chunk;
		struct amdgpu_ib *ib;
		struct drm_amdgpu_cs_chunk_ib *chunk_ib;
		struct amdgpu_ring *ring;

		chunk = &parser->chunks[i];
		ib = &parser->ibs[j];
		chunk_ib = (struct drm_amdgpu_cs_chunk_ib *)chunk->kdata;

		if (chunk->chunk_id != AMDGPU_CHUNK_ID_IB)
			continue;

		r = amdgpu_cs_get_ring(adev, chunk_ib->ip_type,
				       chunk_ib->ip_instance, chunk_ib->ring,
				       &ring);
		if (r)
			return r;

		if (ring->funcs->parse_cs) {
			struct amdgpu_bo_va_mapping *m;
			struct amdgpu_bo *aobj = NULL;
			uint64_t offset;
			uint8_t *kptr;

			m = amdgpu_cs_find_mapping(parser, chunk_ib->va_start,
						   &aobj);
			if (!aobj) {
				DRM_ERROR("IB va_start is invalid\n");
				return -EINVAL;
			}

			if ((chunk_ib->va_start + chunk_ib->ib_bytes) >
			    (m->it.last + 1) * AMDGPU_GPU_PAGE_SIZE) {
				DRM_ERROR("IB va_start+ib_bytes is invalid\n");
				return -EINVAL;
			}

			/* the IB should be reserved at this point */
			r = amdgpu_bo_kmap(aobj, (void **)&kptr);
			if (r) {
				return r;
			}

			offset = ((uint64_t)m->it.start) * AMDGPU_GPU_PAGE_SIZE;
			kptr += chunk_ib->va_start - offset;

			r =  amdgpu_ib_get(ring, NULL, chunk_ib->ib_bytes, ib);
			if (r) {
				DRM_ERROR("Failed to get ib !\n");
				return r;
			}

			memcpy(ib->ptr, kptr, chunk_ib->ib_bytes);
			amdgpu_bo_kunmap(aobj);
		} else {
			r =  amdgpu_ib_get(ring, vm, 0, ib);
			if (r) {
				DRM_ERROR("Failed to get ib !\n");
				return r;
			}

			ib->gpu_addr = chunk_ib->va_start;
		}

		ib->length_dw = chunk_ib->ib_bytes / 4;
		ib->flags = chunk_ib->flags;
		ib->ctx = parser->ctx;
		j++;
	}

	if (!parser->num_ibs)
		return 0;

	/* add GDS resources to first IB */
	if (parser->bo_list) {
		struct amdgpu_bo *gds = parser->bo_list->gds_obj;
		struct amdgpu_bo *gws = parser->bo_list->gws_obj;
		struct amdgpu_bo *oa = parser->bo_list->oa_obj;
		struct amdgpu_ib *ib = &parser->ibs[0];

		if (gds) {
			ib->gds_base = amdgpu_bo_gpu_offset(gds);
			ib->gds_size = amdgpu_bo_size(gds);
		}
		if (gws) {
			ib->gws_base = amdgpu_bo_gpu_offset(gws);
			ib->gws_size = amdgpu_bo_size(gws);
		}
		if (oa) {
			ib->oa_base = amdgpu_bo_gpu_offset(oa);
			ib->oa_size = amdgpu_bo_size(oa);
		}
	}
	/* wrap the last IB with user fence */
	if (parser->uf.bo) {
		struct amdgpu_ib *ib = &parser->ibs[parser->num_ibs - 1];

		/* UVD & VCE fw doesn't support user fences */
		if (ib->ring->type == AMDGPU_RING_TYPE_UVD ||
		    ib->ring->type == AMDGPU_RING_TYPE_VCE)
			return -EINVAL;

		ib->user = &parser->uf;
	}

	return 0;
}

static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
				  struct amdgpu_cs_parser *p)
{
	struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
	struct amdgpu_ib *ib;
	int i, j, r;

	if (!p->num_ibs)
		return 0;

	/* Add dependencies to first IB */
	ib = &p->ibs[0];
	for (i = 0; i < p->nchunks; ++i) {
		struct drm_amdgpu_cs_chunk_dep *deps;
		struct amdgpu_cs_chunk *chunk;
		unsigned num_deps;

		chunk = &p->chunks[i];

		if (chunk->chunk_id != AMDGPU_CHUNK_ID_DEPENDENCIES)
			continue;

		deps = (struct drm_amdgpu_cs_chunk_dep *)chunk->kdata;
		num_deps = chunk->length_dw * 4 /
			sizeof(struct drm_amdgpu_cs_chunk_dep);

		for (j = 0; j < num_deps; ++j) {
			struct amdgpu_ring *ring;
			struct amdgpu_ctx *ctx;
			struct fence *fence;

			r = amdgpu_cs_get_ring(adev, deps[j].ip_type,
					       deps[j].ip_instance,
					       deps[j].ring, &ring);
			if (r)
				return r;

			ctx = amdgpu_ctx_get(fpriv, deps[j].ctx_id);
			if (ctx == NULL)
				return -EINVAL;

			fence = amdgpu_ctx_get_fence(ctx, ring,
						     deps[j].handle);
			if (IS_ERR(fence)) {
				r = PTR_ERR(fence);
				amdgpu_ctx_put(ctx);
				return r;

			} else if (fence) {
				r = amdgpu_sync_fence(adev, &ib->sync, fence);
				fence_put(fence);
				amdgpu_ctx_put(ctx);
				if (r)
					return r;
			}
		}
	}

	return 0;
}

static int amdgpu_cs_free_job(struct amdgpu_job *sched_job)
{
	int i;
	if (sched_job->ibs)
		for (i = 0; i < sched_job->num_ibs; i++)
			amdgpu_ib_free(sched_job->adev, &sched_job->ibs[i]);
	kfree(sched_job->ibs);
	if (sched_job->uf.bo)
		drm_gem_object_unreference_unlocked(&sched_job->uf.bo->gem_base);
	return 0;
}

int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
{
	struct amdgpu_device *adev = dev->dev_private;
	union drm_amdgpu_cs *cs = data;
	struct amdgpu_cs_parser *parser;
	bool reserved_buffers = false;
	int i, r;

	down_read(&adev->exclusive_lock);
	if (!adev->accel_working) {
		up_read(&adev->exclusive_lock);
		return -EBUSY;
	}

	parser = amdgpu_cs_parser_create(adev, filp, NULL, NULL, 0);
	if (!parser)
		return -ENOMEM;
	r = amdgpu_cs_parser_init(parser, data);
	if (r) {
		DRM_ERROR("Failed to initialize parser !\n");
		amdgpu_cs_parser_fini(parser, r, false);
		up_read(&adev->exclusive_lock);
		r = amdgpu_cs_handle_lockup(adev, r);
		return r;
	}

	r = amdgpu_cs_parser_relocs(parser);
	if (r == -ENOMEM)
		DRM_ERROR("Not enough memory for command submission!\n");
	else if (r && r != -ERESTARTSYS)
		DRM_ERROR("Failed to process the buffer list %d!\n", r);
	else if (!r) {
		reserved_buffers = true;
		r = amdgpu_cs_ib_fill(adev, parser);
	}

	if (!r) {
		r = amdgpu_cs_dependencies(adev, parser);
		if (r)
			DRM_ERROR("Failed in the dependencies handling %d!\n", r);
	}

	if (r)
		goto out;

	for (i = 0; i < parser->num_ibs; i++)
		trace_amdgpu_cs(parser, i);

	r = amdgpu_cs_ib_vm_chunk(adev, parser);
	if (r)
		goto out;

	if (amdgpu_enable_scheduler && parser->num_ibs) {
		struct amdgpu_job *job;
		struct amdgpu_ring * ring =  parser->ibs->ring;
		job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
		if (!job)
			return -ENOMEM;
		job->base.sched = ring->scheduler;
		job->base.s_entity = &parser->ctx->rings[ring->idx].entity;
		job->adev = parser->adev;
		job->ibs = parser->ibs;
		job->num_ibs = parser->num_ibs;
		job->base.owner = parser->filp;
		mutex_init(&job->job_lock);
		if (job->ibs[job->num_ibs - 1].user) {
			memcpy(&job->uf,  &parser->uf,
			       sizeof(struct amdgpu_user_fence));
			job->ibs[job->num_ibs - 1].user = &job->uf;
		}

		job->free_job = amdgpu_cs_free_job;
		mutex_lock(&job->job_lock);
		r = amd_sched_entity_push_job((struct amd_sched_job *)job);
		if (r) {
			mutex_unlock(&job->job_lock);
			amdgpu_cs_free_job(job);
			kfree(job);
			goto out;
		}
		cs->out.handle =
			amdgpu_ctx_add_fence(parser->ctx, ring,
					     &job->base.s_fence->base);
		parser->ibs[parser->num_ibs - 1].sequence = cs->out.handle;

		list_sort(NULL, &parser->validated, cmp_size_smaller_first);
		ttm_eu_fence_buffer_objects(&parser->ticket,
				&parser->validated,
				&job->base.s_fence->base);

		mutex_unlock(&job->job_lock);
		amdgpu_cs_parser_fini_late(parser);
		up_read(&adev->exclusive_lock);
		return 0;
	}

	cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence;
out:
	amdgpu_cs_parser_fini(parser, r, reserved_buffers);
	up_read(&adev->exclusive_lock);
	r = amdgpu_cs_handle_lockup(adev, r);
	return r;
}

/**
 * amdgpu_cs_wait_ioctl - wait for a command submission to finish
 *
 * @dev: drm device
 * @data: data from userspace
 * @filp: file private
 *
 * Wait for the command submission identified by handle to finish.
 */
int amdgpu_cs_wait_ioctl(struct drm_device *dev, void *data,
			 struct drm_file *filp)
{
	union drm_amdgpu_wait_cs *wait = data;
	struct amdgpu_device *adev = dev->dev_private;
	unsigned long timeout = amdgpu_gem_timeout(wait->in.timeout);
	struct amdgpu_ring *ring = NULL;
	struct amdgpu_ctx *ctx;
	struct fence *fence;
	long r;

	r = amdgpu_cs_get_ring(adev, wait->in.ip_type, wait->in.ip_instance,
			       wait->in.ring, &ring);
	if (r)
		return r;

	ctx = amdgpu_ctx_get(filp->driver_priv, wait->in.ctx_id);
	if (ctx == NULL)
		return -EINVAL;

	fence = amdgpu_ctx_get_fence(ctx, ring, wait->in.handle);
	if (IS_ERR(fence))
		r = PTR_ERR(fence);
	else if (fence) {
		r = fence_wait_timeout(fence, true, timeout);
		fence_put(fence);
	} else
		r = 1;

	amdgpu_ctx_put(ctx);
	if (r < 0)
		return r;

	memset(wait, 0, sizeof(*wait));
	wait->out.status = (r == 0);

	return 0;
}

/**
 * amdgpu_cs_find_bo_va - find bo_va for VM address
 *
 * @parser: command submission parser context
 * @addr: VM address
 * @bo: resulting BO of the mapping found
 *
 * Search the buffer objects in the command submission context for a certain
 * virtual memory address. Returns allocation structure when found, NULL
 * otherwise.
 */
struct amdgpu_bo_va_mapping *
amdgpu_cs_find_mapping(struct amdgpu_cs_parser *parser,
		       uint64_t addr, struct amdgpu_bo **bo)
{
	struct amdgpu_bo_list_entry *reloc;
	struct amdgpu_bo_va_mapping *mapping;

	addr /= AMDGPU_GPU_PAGE_SIZE;

	list_for_each_entry(reloc, &parser->validated, tv.head) {
		if (!reloc->bo_va)
			continue;

		list_for_each_entry(mapping, &reloc->bo_va->valids, list) {
			if (mapping->it.start > addr ||
			    addr > mapping->it.last)
				continue;

			*bo = reloc->bo_va->bo;
			return mapping;
		}

		list_for_each_entry(mapping, &reloc->bo_va->invalids, list) {
			if (mapping->it.start > addr ||
			    addr > mapping->it.last)
				continue;

			*bo = reloc->bo_va->bo;
			return mapping;
		}
	}

	return NULL;
}