summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
blob: f52d0ba91a770a4b437a1137aa88cf6a7113cf53 (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
/*
 * Copyright 2009 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, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 */
/*
 * Authors:
 *    Jerome Glisse <glisse@freedesktop.org>
 *    Dave Airlie
 */
#include <linux/seq_file.h>
#include <linux/atomic.h>
#include <linux/wait.h>
#include <linux/kref.h>
#include <linux/slab.h>
#include <linux/firmware.h>
#include <linux/pm_runtime.h>

#include <drm/drm_drv.h>
#include "amdgpu.h"
#include "amdgpu_trace.h"
#include "amdgpu_reset.h"

/*
 * Fences
 * Fences mark an event in the GPUs pipeline and are used
 * for GPU/CPU synchronization.  When the fence is written,
 * it is expected that all buffers associated with that fence
 * are no longer in use by the associated ring on the GPU and
 * that the relevant GPU caches have been flushed.
 */

struct amdgpu_fence {
	struct dma_fence base;

	/* RB, DMA, etc. */
	struct amdgpu_ring		*ring;
	ktime_t				start_timestamp;
};

static struct kmem_cache *amdgpu_fence_slab;

int amdgpu_fence_slab_init(void)
{
	amdgpu_fence_slab = kmem_cache_create(
		"amdgpu_fence", sizeof(struct amdgpu_fence), 0,
		SLAB_HWCACHE_ALIGN, NULL);
	if (!amdgpu_fence_slab)
		return -ENOMEM;
	return 0;
}

void amdgpu_fence_slab_fini(void)
{
	rcu_barrier();
	kmem_cache_destroy(amdgpu_fence_slab);
}
/*
 * Cast helper
 */
static const struct dma_fence_ops amdgpu_fence_ops;
static const struct dma_fence_ops amdgpu_job_fence_ops;
static inline struct amdgpu_fence *to_amdgpu_fence(struct dma_fence *f)
{
	struct amdgpu_fence *__f = container_of(f, struct amdgpu_fence, base);

	if (__f->base.ops == &amdgpu_fence_ops ||
	    __f->base.ops == &amdgpu_job_fence_ops)
		return __f;

	return NULL;
}

/**
 * amdgpu_fence_write - write a fence value
 *
 * @ring: ring the fence is associated with
 * @seq: sequence number to write
 *
 * Writes a fence value to memory (all asics).
 */
static void amdgpu_fence_write(struct amdgpu_ring *ring, u32 seq)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;

	if (drv->cpu_addr)
		*drv->cpu_addr = cpu_to_le32(seq);
}

/**
 * amdgpu_fence_read - read a fence value
 *
 * @ring: ring the fence is associated with
 *
 * Reads a fence value from memory (all asics).
 * Returns the value of the fence read from memory.
 */
static u32 amdgpu_fence_read(struct amdgpu_ring *ring)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	u32 seq = 0;

	if (drv->cpu_addr)
		seq = le32_to_cpu(*drv->cpu_addr);
	else
		seq = atomic_read(&drv->last_seq);

	return seq;
}

/**
 * amdgpu_fence_emit - emit a fence on the requested ring
 *
 * @ring: ring the fence is associated with
 * @f: resulting fence object
 * @job: job the fence is embedded in
 * @flags: flags to pass into the subordinate .emit_fence() call
 *
 * Emits a fence command on the requested ring (all asics).
 * Returns 0 on success, -ENOMEM on failure.
 */
int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amdgpu_job *job,
		      unsigned flags)
{
	struct amdgpu_device *adev = ring->adev;
	struct dma_fence *fence;
	struct amdgpu_fence *am_fence;
	struct dma_fence __rcu **ptr;
	uint32_t seq;
	int r;

	if (job == NULL) {
		/* create a sperate hw fence */
		am_fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_ATOMIC);
		if (am_fence == NULL)
			return -ENOMEM;
		fence = &am_fence->base;
		am_fence->ring = ring;
	} else {
		/* take use of job-embedded fence */
		fence = &job->hw_fence;
	}

	seq = ++ring->fence_drv.sync_seq;
	if (job && job->job_run_counter) {
		/* reinit seq for resubmitted jobs */
		fence->seqno = seq;
		/* TO be inline with external fence creation and other drivers */
		dma_fence_get(fence);
	} else {
		if (job) {
			dma_fence_init(fence, &amdgpu_job_fence_ops,
				       &ring->fence_drv.lock,
				       adev->fence_context + ring->idx, seq);
			/* Against remove in amdgpu_job_{free, free_cb} */
			dma_fence_get(fence);
		}
		else
			dma_fence_init(fence, &amdgpu_fence_ops,
				       &ring->fence_drv.lock,
				       adev->fence_context + ring->idx, seq);
	}

	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
			       seq, flags | AMDGPU_FENCE_FLAG_INT);
	pm_runtime_get_noresume(adev_to_drm(adev)->dev);
	ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask];
	if (unlikely(rcu_dereference_protected(*ptr, 1))) {
		struct dma_fence *old;

		rcu_read_lock();
		old = dma_fence_get_rcu_safe(ptr);
		rcu_read_unlock();

		if (old) {
			r = dma_fence_wait(old, false);
			dma_fence_put(old);
			if (r)
				return r;
		}
	}

	to_amdgpu_fence(fence)->start_timestamp = ktime_get();

	/* This function can't be called concurrently anyway, otherwise
	 * emitting the fence would mess up the hardware ring buffer.
	 */
	rcu_assign_pointer(*ptr, dma_fence_get(fence));

	*f = fence;

	return 0;
}

/**
 * amdgpu_fence_emit_polling - emit a fence on the requeste ring
 *
 * @ring: ring the fence is associated with
 * @s: resulting sequence number
 * @timeout: the timeout for waiting in usecs
 *
 * Emits a fence command on the requested ring (all asics).
 * Used For polling fence.
 * Returns 0 on success, -ENOMEM on failure.
 */
int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s,
			      uint32_t timeout)
{
	uint32_t seq;
	signed long r;

	if (!s)
		return -EINVAL;

	seq = ++ring->fence_drv.sync_seq;
	r = amdgpu_fence_wait_polling(ring,
				      seq - ring->fence_drv.num_fences_mask,
				      timeout);
	if (r < 1)
		return -ETIMEDOUT;

	amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr,
			       seq, 0);

	*s = seq;

	return 0;
}

/**
 * amdgpu_fence_schedule_fallback - schedule fallback check
 *
 * @ring: pointer to struct amdgpu_ring
 *
 * Start a timer as fallback to our interrupts.
 */
static void amdgpu_fence_schedule_fallback(struct amdgpu_ring *ring)
{
	mod_timer(&ring->fence_drv.fallback_timer,
		  jiffies + AMDGPU_FENCE_JIFFIES_TIMEOUT);
}

/**
 * amdgpu_fence_process - check for fence activity
 *
 * @ring: pointer to struct amdgpu_ring
 *
 * Checks the current fence value and calculates the last
 * signalled fence value. Wakes the fence queue if the
 * sequence number has increased.
 *
 * Returns true if fence was processed
 */
bool amdgpu_fence_process(struct amdgpu_ring *ring)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	struct amdgpu_device *adev = ring->adev;
	uint32_t seq, last_seq;

	do {
		last_seq = atomic_read(&ring->fence_drv.last_seq);
		seq = amdgpu_fence_read(ring);

	} while (atomic_cmpxchg(&drv->last_seq, last_seq, seq) != last_seq);

	if (del_timer(&ring->fence_drv.fallback_timer) &&
	    seq != ring->fence_drv.sync_seq)
		amdgpu_fence_schedule_fallback(ring);

	if (unlikely(seq == last_seq))
		return false;

	last_seq &= drv->num_fences_mask;
	seq &= drv->num_fences_mask;

	do {
		struct dma_fence *fence, **ptr;

		++last_seq;
		last_seq &= drv->num_fences_mask;
		ptr = &drv->fences[last_seq];

		/* There is always exactly one thread signaling this fence slot */
		fence = rcu_dereference_protected(*ptr, 1);
		RCU_INIT_POINTER(*ptr, NULL);

		if (!fence)
			continue;

		dma_fence_signal(fence);
		dma_fence_put(fence);
		pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
	} while (last_seq != seq);

	return true;
}

/**
 * amdgpu_fence_fallback - fallback for hardware interrupts
 *
 * @t: timer context used to obtain the pointer to ring structure
 *
 * Checks for fence activity.
 */
static void amdgpu_fence_fallback(struct timer_list *t)
{
	struct amdgpu_ring *ring = from_timer(ring, t,
					      fence_drv.fallback_timer);

	if (amdgpu_fence_process(ring))
		DRM_WARN("Fence fallback timer expired on ring %s\n", ring->name);
}

/**
 * amdgpu_fence_wait_empty - wait for all fences to signal
 *
 * @ring: ring index the fence is associated with
 *
 * Wait for all fences on the requested ring to signal (all asics).
 * Returns 0 if the fences have passed, error for all other cases.
 */
int amdgpu_fence_wait_empty(struct amdgpu_ring *ring)
{
	uint64_t seq = READ_ONCE(ring->fence_drv.sync_seq);
	struct dma_fence *fence, **ptr;
	int r;

	if (!seq)
		return 0;

	ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask];
	rcu_read_lock();
	fence = rcu_dereference(*ptr);
	if (!fence || !dma_fence_get_rcu(fence)) {
		rcu_read_unlock();
		return 0;
	}
	rcu_read_unlock();

	r = dma_fence_wait(fence, false);
	dma_fence_put(fence);
	return r;
}

/**
 * amdgpu_fence_wait_polling - busy wait for givn sequence number
 *
 * @ring: ring index the fence is associated with
 * @wait_seq: sequence number to wait
 * @timeout: the timeout for waiting in usecs
 *
 * Wait for all fences on the requested ring to signal (all asics).
 * Returns left time if no timeout, 0 or minus if timeout.
 */
signed long amdgpu_fence_wait_polling(struct amdgpu_ring *ring,
				      uint32_t wait_seq,
				      signed long timeout)
{
	uint32_t seq;

	do {
		seq = amdgpu_fence_read(ring);
		udelay(5);
		timeout -= 5;
	} while ((int32_t)(wait_seq - seq) > 0 && timeout > 0);

	return timeout > 0 ? timeout : 0;
}
/**
 * amdgpu_fence_count_emitted - get the count of emitted fences
 *
 * @ring: ring the fence is associated with
 *
 * Get the number of fences emitted on the requested ring (all asics).
 * Returns the number of emitted fences on the ring.  Used by the
 * dynpm code to ring track activity.
 */
unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring)
{
	uint64_t emitted;

	/* We are not protected by ring lock when reading the last sequence
	 * but it's ok to report slightly wrong fence count here.
	 */
	emitted = 0x100000000ull;
	emitted -= atomic_read(&ring->fence_drv.last_seq);
	emitted += READ_ONCE(ring->fence_drv.sync_seq);
	return lower_32_bits(emitted);
}

/**
 * amdgpu_fence_last_unsignaled_time_us - the time fence emitted until now
 * @ring: ring the fence is associated with
 *
 * Find the earliest fence unsignaled until now, calculate the time delta
 * between the time fence emitted and now.
 */
u64 amdgpu_fence_last_unsignaled_time_us(struct amdgpu_ring *ring)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	struct dma_fence *fence;
	uint32_t last_seq, sync_seq;

	last_seq = atomic_read(&ring->fence_drv.last_seq);
	sync_seq = READ_ONCE(ring->fence_drv.sync_seq);
	if (last_seq == sync_seq)
		return 0;

	++last_seq;
	last_seq &= drv->num_fences_mask;
	fence = drv->fences[last_seq];
	if (!fence)
		return 0;

	return ktime_us_delta(ktime_get(),
		to_amdgpu_fence(fence)->start_timestamp);
}

/**
 * amdgpu_fence_update_start_timestamp - update the timestamp of the fence
 * @ring: ring the fence is associated with
 * @seq: the fence seq number to update.
 * @timestamp: the start timestamp to update.
 *
 * The function called at the time the fence and related ib is about to
 * resubmit to gpu in MCBP scenario. Thus we do not consider race condition
 * with amdgpu_fence_process to modify the same fence.
 */
void amdgpu_fence_update_start_timestamp(struct amdgpu_ring *ring, uint32_t seq, ktime_t timestamp)
{
	struct amdgpu_fence_driver *drv = &ring->fence_drv;
	struct dma_fence *fence;

	seq &= drv->num_fences_mask;
	fence = drv->fences[seq];
	if (!fence)
		return;

	to_amdgpu_fence(fence)->start_timestamp = timestamp;
}

/**
 * amdgpu_fence_driver_start_ring - make the fence driver
 * ready for use on the requested ring.
 *
 * @ring: ring to start the fence driver on
 * @irq_src: interrupt source to use for this ring
 * @irq_type: interrupt type to use for this ring
 *
 * Make the fence driver ready for processing (all asics).
 * Not all asics have all rings, so each asic will only
 * start the fence driver on the rings it has.
 * Returns 0 for success, errors for failure.
 */
int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
				   struct amdgpu_irq_src *irq_src,
				   unsigned irq_type)
{
	struct amdgpu_device *adev = ring->adev;
	uint64_t index;

	if (ring->funcs->type != AMDGPU_RING_TYPE_UVD) {
		ring->fence_drv.cpu_addr = ring->fence_cpu_addr;
		ring->fence_drv.gpu_addr = ring->fence_gpu_addr;
	} else {
		/* put fence directly behind firmware */
		index = ALIGN(adev->uvd.fw->size, 8);
		ring->fence_drv.cpu_addr = adev->uvd.inst[ring->me].cpu_addr + index;
		ring->fence_drv.gpu_addr = adev->uvd.inst[ring->me].gpu_addr + index;
	}
	amdgpu_fence_write(ring, atomic_read(&ring->fence_drv.last_seq));

	ring->fence_drv.irq_src = irq_src;
	ring->fence_drv.irq_type = irq_type;
	ring->fence_drv.initialized = true;

	DRM_DEV_DEBUG(adev->dev, "fence driver on ring %s use gpu addr 0x%016llx\n",
		      ring->name, ring->fence_drv.gpu_addr);
	return 0;
}

/**
 * amdgpu_fence_driver_init_ring - init the fence driver
 * for the requested ring.
 *
 * @ring: ring to init the fence driver on
 *
 * Init the fence driver for the requested ring (all asics).
 * Helper function for amdgpu_fence_driver_init().
 */
int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring)
{
	struct amdgpu_device *adev = ring->adev;

	if (!adev)
		return -EINVAL;

	if (!is_power_of_2(ring->num_hw_submission))
		return -EINVAL;

	ring->fence_drv.cpu_addr = NULL;
	ring->fence_drv.gpu_addr = 0;
	ring->fence_drv.sync_seq = 0;
	atomic_set(&ring->fence_drv.last_seq, 0);
	ring->fence_drv.initialized = false;

	timer_setup(&ring->fence_drv.fallback_timer, amdgpu_fence_fallback, 0);

	ring->fence_drv.num_fences_mask = ring->num_hw_submission * 2 - 1;
	spin_lock_init(&ring->fence_drv.lock);
	ring->fence_drv.fences = kcalloc(ring->num_hw_submission * 2, sizeof(void *),
					 GFP_KERNEL);

	if (!ring->fence_drv.fences)
		return -ENOMEM;

	return 0;
}

/**
 * amdgpu_fence_driver_sw_init - init the fence driver
 * for all possible rings.
 *
 * @adev: amdgpu device pointer
 *
 * Init the fence driver for all possible rings (all asics).
 * Not all asics have all rings, so each asic will only
 * start the fence driver on the rings it has using
 * amdgpu_fence_driver_start_ring().
 * Returns 0 for success.
 */
int amdgpu_fence_driver_sw_init(struct amdgpu_device *adev)
{
	return 0;
}

/**
 * amdgpu_fence_driver_hw_fini - tear down the fence driver
 * for all possible rings.
 *
 * @adev: amdgpu device pointer
 *
 * Tear down the fence driver for all possible rings (all asics).
 */
void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev)
{
	int i, r;

	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];

		if (!ring || !ring->fence_drv.initialized)
			continue;

		/* You can't wait for HW to signal if it's gone */
		if (!drm_dev_is_unplugged(adev_to_drm(adev)))
			r = amdgpu_fence_wait_empty(ring);
		else
			r = -ENODEV;
		/* no need to trigger GPU reset as we are unloading */
		if (r)
			amdgpu_fence_driver_force_completion(ring);

		if (ring->fence_drv.irq_src)
			amdgpu_irq_put(adev, ring->fence_drv.irq_src,
				       ring->fence_drv.irq_type);

		del_timer_sync(&ring->fence_drv.fallback_timer);
	}
}

/* Will either stop and flush handlers for amdgpu interrupt or reanble it */
void amdgpu_fence_driver_isr_toggle(struct amdgpu_device *adev, bool stop)
{
	int i;

	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];

		if (!ring || !ring->fence_drv.initialized || !ring->fence_drv.irq_src)
			continue;

		if (stop)
			disable_irq(adev->irq.irq);
		else
			enable_irq(adev->irq.irq);
	}
}

void amdgpu_fence_driver_sw_fini(struct amdgpu_device *adev)
{
	unsigned int i, j;

	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];

		if (!ring || !ring->fence_drv.initialized)
			continue;

		/*
		 * Notice we check for sched.ops since there's some
		 * override on the meaning of sched.ready by amdgpu.
		 * The natural check would be sched.ready, which is
		 * set as drm_sched_init() finishes...
		 */
		if (ring->sched.ops)
			drm_sched_fini(&ring->sched);

		for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j)
			dma_fence_put(ring->fence_drv.fences[j]);
		kfree(ring->fence_drv.fences);
		ring->fence_drv.fences = NULL;
		ring->fence_drv.initialized = false;
	}
}

/**
 * amdgpu_fence_driver_hw_init - enable the fence driver
 * for all possible rings.
 *
 * @adev: amdgpu device pointer
 *
 * Enable the fence driver for all possible rings (all asics).
 * Not all asics have all rings, so each asic will only
 * start the fence driver on the rings it has using
 * amdgpu_fence_driver_start_ring().
 * Returns 0 for success.
 */
void amdgpu_fence_driver_hw_init(struct amdgpu_device *adev)
{
	int i;

	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
		struct amdgpu_ring *ring = adev->rings[i];
		if (!ring || !ring->fence_drv.initialized)
			continue;

		/* enable the interrupt */
		if (ring->fence_drv.irq_src)
			amdgpu_irq_get(adev, ring->fence_drv.irq_src,
				       ring->fence_drv.irq_type);
	}
}

/**
 * amdgpu_fence_driver_clear_job_fences - clear job embedded fences of ring
 *
 * @ring: fence of the ring to be cleared
 *
 */
void amdgpu_fence_driver_clear_job_fences(struct amdgpu_ring *ring)
{
	int i;
	struct dma_fence *old, **ptr;

	for (i = 0; i <= ring->fence_drv.num_fences_mask; i++) {
		ptr = &ring->fence_drv.fences[i];
		old = rcu_dereference_protected(*ptr, 1);
		if (old && old->ops == &amdgpu_job_fence_ops) {
			struct amdgpu_job *job;

			/* For non-scheduler bad job, i.e. failed ib test, we need to signal
			 * it right here or we won't be able to track them in fence_drv
			 * and they will remain unsignaled during sa_bo free.
			 */
			job = container_of(old, struct amdgpu_job, hw_fence);
			if (!job->base.s_fence && !dma_fence_is_signaled(old))
				dma_fence_signal(old);
			RCU_INIT_POINTER(*ptr, NULL);
			dma_fence_put(old);
		}
	}
}

/**
 * amdgpu_fence_driver_force_completion - force signal latest fence of ring
 *
 * @ring: fence of the ring to signal
 *
 */
void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring)
{
	amdgpu_fence_write(ring, ring->fence_drv.sync_seq);
	amdgpu_fence_process(ring);
}

/*
 * Common fence implementation
 */

static const char *amdgpu_fence_get_driver_name(struct dma_fence *fence)
{
	return "amdgpu";
}

static const char *amdgpu_fence_get_timeline_name(struct dma_fence *f)
{
	return (const char *)to_amdgpu_fence(f)->ring->name;
}

static const char *amdgpu_job_fence_get_timeline_name(struct dma_fence *f)
{
	struct amdgpu_job *job = container_of(f, struct amdgpu_job, hw_fence);

	return (const char *)to_amdgpu_ring(job->base.sched)->name;
}

/**
 * amdgpu_fence_enable_signaling - enable signalling on fence
 * @f: fence
 *
 * This function is called with fence_queue lock held, and adds a callback
 * to fence_queue that checks if this fence is signaled, and if so it
 * signals the fence and removes itself.
 */
static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
{
	if (!timer_pending(&to_amdgpu_fence(f)->ring->fence_drv.fallback_timer))
		amdgpu_fence_schedule_fallback(to_amdgpu_fence(f)->ring);

	return true;
}

/**
 * amdgpu_job_fence_enable_signaling - enable signalling on job fence
 * @f: fence
 *
 * This is the simliar function with amdgpu_fence_enable_signaling above, it
 * only handles the job embedded fence.
 */
static bool amdgpu_job_fence_enable_signaling(struct dma_fence *f)
{
	struct amdgpu_job *job = container_of(f, struct amdgpu_job, hw_fence);

	if (!timer_pending(&to_amdgpu_ring(job->base.sched)->fence_drv.fallback_timer))
		amdgpu_fence_schedule_fallback(to_amdgpu_ring(job->base.sched));

	return true;
}

/**
 * amdgpu_fence_free - free up the fence memory
 *
 * @rcu: RCU callback head
 *
 * Free up the fence memory after the RCU grace period.
 */
static void amdgpu_fence_free(struct rcu_head *rcu)
{
	struct dma_fence *f = container_of(rcu, struct dma_fence, rcu);

	/* free fence_slab if it's separated fence*/
	kmem_cache_free(amdgpu_fence_slab, to_amdgpu_fence(f));
}

/**
 * amdgpu_job_fence_free - free up the job with embedded fence
 *
 * @rcu: RCU callback head
 *
 * Free up the job with embedded fence after the RCU grace period.
 */
static void amdgpu_job_fence_free(struct rcu_head *rcu)
{
	struct dma_fence *f = container_of(rcu, struct dma_fence, rcu);

	/* free job if fence has a parent job */
	kfree(container_of(f, struct amdgpu_job, hw_fence));
}

/**
 * amdgpu_fence_release - callback that fence can be freed
 *
 * @f: fence
 *
 * This function is called when the reference count becomes zero.
 * It just RCU schedules freeing up the fence.
 */
static void amdgpu_fence_release(struct dma_fence *f)
{
	call_rcu(&f->rcu, amdgpu_fence_free);
}

/**
 * amdgpu_job_fence_release - callback that job embedded fence can be freed
 *
 * @f: fence
 *
 * This is the simliar function with amdgpu_fence_release above, it
 * only handles the job embedded fence.
 */
static void amdgpu_job_fence_release(struct dma_fence *f)
{
	call_rcu(&f->rcu, amdgpu_job_fence_free);
}

static const struct dma_fence_ops amdgpu_fence_ops = {
	.get_driver_name = amdgpu_fence_get_driver_name,
	.get_timeline_name = amdgpu_fence_get_timeline_name,
	.enable_signaling = amdgpu_fence_enable_signaling,
	.release = amdgpu_fence_release,
};

static const struct dma_fence_ops amdgpu_job_fence_ops = {
	.get_driver_name = amdgpu_fence_get_driver_name,
	.get_timeline_name = amdgpu_job_fence_get_timeline_name,
	.enable_signaling = amdgpu_job_fence_enable_signaling,
	.release = amdgpu_job_fence_release,
};

/*
 * Fence debugfs
 */
#if defined(CONFIG_DEBUG_FS)
static int amdgpu_debugfs_fence_info_show(struct seq_file *m, void *unused)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
	int i;

	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
		struct amdgpu_ring *ring = adev->rings[i];
		if (!ring || !ring->fence_drv.initialized)
			continue;

		amdgpu_fence_process(ring);

		seq_printf(m, "--- ring %d (%s) ---\n", i, ring->name);
		seq_printf(m, "Last signaled fence          0x%08x\n",
			   atomic_read(&ring->fence_drv.last_seq));
		seq_printf(m, "Last emitted                 0x%08x\n",
			   ring->fence_drv.sync_seq);

		if (ring->funcs->type == AMDGPU_RING_TYPE_GFX ||
		    ring->funcs->type == AMDGPU_RING_TYPE_SDMA) {
			seq_printf(m, "Last signaled trailing fence 0x%08x\n",
				   le32_to_cpu(*ring->trail_fence_cpu_addr));
			seq_printf(m, "Last emitted                 0x%08x\n",
				   ring->trail_seq);
		}

		if (ring->funcs->type != AMDGPU_RING_TYPE_GFX)
			continue;

		/* set in CP_VMID_PREEMPT and preemption occurred */
		seq_printf(m, "Last preempted               0x%08x\n",
			   le32_to_cpu(*(ring->fence_drv.cpu_addr + 2)));
		/* set in CP_VMID_RESET and reset occurred */
		seq_printf(m, "Last reset                   0x%08x\n",
			   le32_to_cpu(*(ring->fence_drv.cpu_addr + 4)));
		/* Both preemption and reset occurred */
		seq_printf(m, "Last both                    0x%08x\n",
			   le32_to_cpu(*(ring->fence_drv.cpu_addr + 6)));
	}
	return 0;
}

/*
 * amdgpu_debugfs_gpu_recover - manually trigger a gpu reset & recover
 *
 * Manually trigger a gpu reset at the next fence wait.
 */
static int gpu_recover_get(void *data, u64 *val)
{
	struct amdgpu_device *adev = (struct amdgpu_device *)data;
	struct drm_device *dev = adev_to_drm(adev);
	int r;

	r = pm_runtime_get_sync(dev->dev);
	if (r < 0) {
		pm_runtime_put_autosuspend(dev->dev);
		return 0;
	}

	if (amdgpu_reset_domain_schedule(adev->reset_domain, &adev->reset_work))
		flush_work(&adev->reset_work);

	*val = atomic_read(&adev->reset_domain->reset_res);

	pm_runtime_mark_last_busy(dev->dev);
	pm_runtime_put_autosuspend(dev->dev);

	return 0;
}

DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_fence_info);
DEFINE_DEBUGFS_ATTRIBUTE(amdgpu_debugfs_gpu_recover_fops, gpu_recover_get, NULL,
			 "%lld\n");

static void amdgpu_debugfs_reset_work(struct work_struct *work)
{
	struct amdgpu_device *adev = container_of(work, struct amdgpu_device,
						  reset_work);

	struct amdgpu_reset_context reset_context;
	memset(&reset_context, 0, sizeof(reset_context));

	reset_context.method = AMD_RESET_METHOD_NONE;
	reset_context.reset_req_dev = adev;
	set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags);

	amdgpu_device_gpu_recover(adev, NULL, &reset_context);
}

#endif

void amdgpu_debugfs_fence_init(struct amdgpu_device *adev)
{
#if defined(CONFIG_DEBUG_FS)
	struct drm_minor *minor = adev_to_drm(adev)->primary;
	struct dentry *root = minor->debugfs_root;

	debugfs_create_file("amdgpu_fence_info", 0444, root, adev,
			    &amdgpu_debugfs_fence_info_fops);

	if (!amdgpu_sriov_vf(adev)) {

		INIT_WORK(&adev->reset_work, amdgpu_debugfs_reset_work);
		debugfs_create_file("amdgpu_gpu_recover", 0444, root, adev,
				    &amdgpu_debugfs_gpu_recover_fops);
	}
#endif
}