summaryrefslogtreecommitdiffstats
path: root/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
blob: 55784a9f26c4c83b17008a766130c234df8ecbaf (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
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
/*
 * Copyright 2018 Advanced Micro Devices, Inc.
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 */

#include <linux/io-64-nonatomic-lo-hi.h>
#ifdef CONFIG_X86
#include <asm/hypervisor.h>
#endif

#include "amdgpu.h"
#include "amdgpu_gmc.h"
#include "amdgpu_ras.h"
#include "amdgpu_reset.h"
#include "amdgpu_xgmi.h"

#include <drm/drm_drv.h>
#include <drm/ttm/ttm_tt.h>

/**
 * amdgpu_gmc_pdb0_alloc - allocate vram for pdb0
 *
 * @adev: amdgpu_device pointer
 *
 * Allocate video memory for pdb0 and map it for CPU access
 * Returns 0 for success, error for failure.
 */
int amdgpu_gmc_pdb0_alloc(struct amdgpu_device *adev)
{
	int r;
	struct amdgpu_bo_param bp;
	u64 vram_size = adev->gmc.xgmi.node_segment_size * adev->gmc.xgmi.num_physical_nodes;
	uint32_t pde0_page_shift = adev->gmc.vmid0_page_table_block_size + 21;
	uint32_t npdes = (vram_size + (1ULL << pde0_page_shift) -1) >> pde0_page_shift;

	memset(&bp, 0, sizeof(bp));
	bp.size = PAGE_ALIGN((npdes + 1) * 8);
	bp.byte_align = PAGE_SIZE;
	bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
	bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
		AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
	bp.type = ttm_bo_type_kernel;
	bp.resv = NULL;
	bp.bo_ptr_size = sizeof(struct amdgpu_bo);

	r = amdgpu_bo_create(adev, &bp, &adev->gmc.pdb0_bo);
	if (r)
		return r;

	r = amdgpu_bo_reserve(adev->gmc.pdb0_bo, false);
	if (unlikely(r != 0))
		goto bo_reserve_failure;

	r = amdgpu_bo_pin(adev->gmc.pdb0_bo, AMDGPU_GEM_DOMAIN_VRAM);
	if (r)
		goto bo_pin_failure;
	r = amdgpu_bo_kmap(adev->gmc.pdb0_bo, &adev->gmc.ptr_pdb0);
	if (r)
		goto bo_kmap_failure;

	amdgpu_bo_unreserve(adev->gmc.pdb0_bo);
	return 0;

bo_kmap_failure:
	amdgpu_bo_unpin(adev->gmc.pdb0_bo);
bo_pin_failure:
	amdgpu_bo_unreserve(adev->gmc.pdb0_bo);
bo_reserve_failure:
	amdgpu_bo_unref(&adev->gmc.pdb0_bo);
	return r;
}

/**
 * amdgpu_gmc_get_pde_for_bo - get the PDE for a BO
 *
 * @bo: the BO to get the PDE for
 * @level: the level in the PD hirarchy
 * @addr: resulting addr
 * @flags: resulting flags
 *
 * Get the address and flags to be used for a PDE (Page Directory Entry).
 */
void amdgpu_gmc_get_pde_for_bo(struct amdgpu_bo *bo, int level,
			       uint64_t *addr, uint64_t *flags)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);

	switch (bo->tbo.resource->mem_type) {
	case TTM_PL_TT:
		*addr = bo->tbo.ttm->dma_address[0];
		break;
	case TTM_PL_VRAM:
		*addr = amdgpu_bo_gpu_offset(bo);
		break;
	default:
		*addr = 0;
		break;
	}
	*flags = amdgpu_ttm_tt_pde_flags(bo->tbo.ttm, bo->tbo.resource);
	amdgpu_gmc_get_vm_pde(adev, level, addr, flags);
}

/*
 * amdgpu_gmc_pd_addr - return the address of the root directory
 */
uint64_t amdgpu_gmc_pd_addr(struct amdgpu_bo *bo)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
	uint64_t pd_addr;

	/* TODO: move that into ASIC specific code */
	if (adev->asic_type >= CHIP_VEGA10) {
		uint64_t flags = AMDGPU_PTE_VALID;

		amdgpu_gmc_get_pde_for_bo(bo, -1, &pd_addr, &flags);
		pd_addr |= flags;
	} else {
		pd_addr = amdgpu_bo_gpu_offset(bo);
	}
	return pd_addr;
}

/**
 * amdgpu_gmc_set_pte_pde - update the page tables using CPU
 *
 * @adev: amdgpu_device pointer
 * @cpu_pt_addr: cpu address of the page table
 * @gpu_page_idx: entry in the page table to update
 * @addr: dst addr to write into pte/pde
 * @flags: access flags
 *
 * Update the page tables using CPU.
 */
int amdgpu_gmc_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
				uint32_t gpu_page_idx, uint64_t addr,
				uint64_t flags)
{
	void __iomem *ptr = (void *)cpu_pt_addr;
	uint64_t value;

	/*
	 * The following is for PTE only. GART does not have PDEs.
	*/
	value = addr & 0x0000FFFFFFFFF000ULL;
	value |= flags;
	writeq(value, ptr + (gpu_page_idx * 8));

	return 0;
}

/**
 * amdgpu_gmc_agp_addr - return the address in the AGP address space
 *
 * @bo: TTM BO which needs the address, must be in GTT domain
 *
 * Tries to figure out how to access the BO through the AGP aperture. Returns
 * AMDGPU_BO_INVALID_OFFSET if that is not possible.
 */
uint64_t amdgpu_gmc_agp_addr(struct ttm_buffer_object *bo)
{
	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);

	if (!bo->ttm)
		return AMDGPU_BO_INVALID_OFFSET;

	if (bo->ttm->num_pages != 1 || bo->ttm->caching == ttm_cached)
		return AMDGPU_BO_INVALID_OFFSET;

	if (bo->ttm->dma_address[0] + PAGE_SIZE >= adev->gmc.agp_size)
		return AMDGPU_BO_INVALID_OFFSET;

	return adev->gmc.agp_start + bo->ttm->dma_address[0];
}

/**
 * amdgpu_gmc_vram_location - try to find VRAM location
 *
 * @adev: amdgpu device structure holding all necessary information
 * @mc: memory controller structure holding memory information
 * @base: base address at which to put VRAM
 *
 * Function will try to place VRAM at base address provided
 * as parameter.
 */
void amdgpu_gmc_vram_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc,
			      u64 base)
{
	uint64_t vis_limit = (uint64_t)amdgpu_vis_vram_limit << 20;
	uint64_t limit = (uint64_t)amdgpu_vram_limit << 20;

	mc->vram_start = base;
	mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
	if (limit < mc->real_vram_size)
		mc->real_vram_size = limit;

	if (vis_limit && vis_limit < mc->visible_vram_size)
		mc->visible_vram_size = vis_limit;

	if (mc->real_vram_size < mc->visible_vram_size)
		mc->visible_vram_size = mc->real_vram_size;

	if (mc->xgmi.num_physical_nodes == 0) {
		mc->fb_start = mc->vram_start;
		mc->fb_end = mc->vram_end;
	}
	dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
			mc->mc_vram_size >> 20, mc->vram_start,
			mc->vram_end, mc->real_vram_size >> 20);
}

/** amdgpu_gmc_sysvm_location - place vram and gart in sysvm aperture
 *
 * @adev: amdgpu device structure holding all necessary information
 * @mc: memory controller structure holding memory information
 *
 * This function is only used if use GART for FB translation. In such
 * case, we use sysvm aperture (vmid0 page tables) for both vram
 * and gart (aka system memory) access.
 *
 * GPUVM (and our organization of vmid0 page tables) require sysvm
 * aperture to be placed at a location aligned with 8 times of native
 * page size. For example, if vm_context0_cntl.page_table_block_size
 * is 12, then native page size is 8G (2M*2^12), sysvm should start
 * with a 64G aligned address. For simplicity, we just put sysvm at
 * address 0. So vram start at address 0 and gart is right after vram.
 */
void amdgpu_gmc_sysvm_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc)
{
	u64 hive_vram_start = 0;
	u64 hive_vram_end = mc->xgmi.node_segment_size * mc->xgmi.num_physical_nodes - 1;
	mc->vram_start = mc->xgmi.node_segment_size * mc->xgmi.physical_node_id;
	mc->vram_end = mc->vram_start + mc->xgmi.node_segment_size - 1;
	mc->gart_start = hive_vram_end + 1;
	mc->gart_end = mc->gart_start + mc->gart_size - 1;
	mc->fb_start = hive_vram_start;
	mc->fb_end = hive_vram_end;
	dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
			mc->mc_vram_size >> 20, mc->vram_start,
			mc->vram_end, mc->real_vram_size >> 20);
	dev_info(adev->dev, "GART: %lluM 0x%016llX - 0x%016llX\n",
			mc->gart_size >> 20, mc->gart_start, mc->gart_end);
}

/**
 * amdgpu_gmc_gart_location - try to find GART location
 *
 * @adev: amdgpu device structure holding all necessary information
 * @mc: memory controller structure holding memory information
 * @gart_placement: GART placement policy with respect to VRAM
 *
 * Function will place try to place GART before or after VRAM.
 * If GART size is bigger than space left then we ajust GART size.
 * Thus function will never fails.
 */
void amdgpu_gmc_gart_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc,
			      enum amdgpu_gart_placement gart_placement)
{
	const uint64_t four_gb = 0x100000000ULL;
	u64 size_af, size_bf;
	/*To avoid the hole, limit the max mc address to AMDGPU_GMC_HOLE_START*/
	u64 max_mc_address = min(adev->gmc.mc_mask, AMDGPU_GMC_HOLE_START - 1);

	/* VCE doesn't like it when BOs cross a 4GB segment, so align
	 * the GART base on a 4GB boundary as well.
	 */
	size_bf = mc->fb_start;
	size_af = max_mc_address + 1 - ALIGN(mc->fb_end + 1, four_gb);

	if (mc->gart_size > max(size_bf, size_af)) {
		dev_warn(adev->dev, "limiting GART\n");
		mc->gart_size = max(size_bf, size_af);
	}

	switch (gart_placement) {
	case AMDGPU_GART_PLACEMENT_HIGH:
		mc->gart_start = max_mc_address - mc->gart_size + 1;
		break;
	case AMDGPU_GART_PLACEMENT_LOW:
		mc->gart_start = 0;
		break;
	case AMDGPU_GART_PLACEMENT_BEST_FIT:
	default:
		if ((size_bf >= mc->gart_size && size_bf < size_af) ||
		    (size_af < mc->gart_size))
			mc->gart_start = 0;
		else
			mc->gart_start = max_mc_address - mc->gart_size + 1;
		break;
	}

	mc->gart_start &= ~(four_gb - 1);
	mc->gart_end = mc->gart_start + mc->gart_size - 1;
	dev_info(adev->dev, "GART: %lluM 0x%016llX - 0x%016llX\n",
			mc->gart_size >> 20, mc->gart_start, mc->gart_end);
}

/**
 * amdgpu_gmc_agp_location - try to find AGP location
 * @adev: amdgpu device structure holding all necessary information
 * @mc: memory controller structure holding memory information
 *
 * Function will place try to find a place for the AGP BAR in the MC address
 * space.
 *
 * AGP BAR will be assigned the largest available hole in the address space.
 * Should be called after VRAM and GART locations are setup.
 */
void amdgpu_gmc_agp_location(struct amdgpu_device *adev, struct amdgpu_gmc *mc)
{
	const uint64_t sixteen_gb = 1ULL << 34;
	const uint64_t sixteen_gb_mask = ~(sixteen_gb - 1);
	u64 size_af, size_bf;

	if (mc->fb_start > mc->gart_start) {
		size_bf = (mc->fb_start & sixteen_gb_mask) -
			ALIGN(mc->gart_end + 1, sixteen_gb);
		size_af = mc->mc_mask + 1 - ALIGN(mc->fb_end + 1, sixteen_gb);
	} else {
		size_bf = mc->fb_start & sixteen_gb_mask;
		size_af = (mc->gart_start & sixteen_gb_mask) -
			ALIGN(mc->fb_end + 1, sixteen_gb);
	}

	if (size_bf > size_af) {
		mc->agp_start = (mc->fb_start - size_bf) & sixteen_gb_mask;
		mc->agp_size = size_bf;
	} else {
		mc->agp_start = ALIGN(mc->fb_end + 1, sixteen_gb);
		mc->agp_size = size_af;
	}

	mc->agp_end = mc->agp_start + mc->agp_size - 1;
	dev_info(adev->dev, "AGP: %lluM 0x%016llX - 0x%016llX\n",
			mc->agp_size >> 20, mc->agp_start, mc->agp_end);
}

/**
 * amdgpu_gmc_set_agp_default - Set the default AGP aperture value.
 * @adev: amdgpu device structure holding all necessary information
 * @mc: memory controller structure holding memory information
 *
 * To disable the AGP aperture, you need to set the start to a larger
 * value than the end.  This function sets the default value which
 * can then be overridden using amdgpu_gmc_agp_location() if you want
 * to enable the AGP aperture on a specific chip.
 *
 */
void amdgpu_gmc_set_agp_default(struct amdgpu_device *adev,
				struct amdgpu_gmc *mc)
{
	mc->agp_start = 0xffffffffffff;
	mc->agp_end = 0;
	mc->agp_size = 0;
}

/**
 * amdgpu_gmc_fault_key - get hask key from vm fault address and pasid
 *
 * @addr: 48 bit physical address, page aligned (36 significant bits)
 * @pasid: 16 bit process address space identifier
 */
static inline uint64_t amdgpu_gmc_fault_key(uint64_t addr, uint16_t pasid)
{
	return addr << 4 | pasid;
}

/**
 * amdgpu_gmc_filter_faults - filter VM faults
 *
 * @adev: amdgpu device structure
 * @ih: interrupt ring that the fault received from
 * @addr: address of the VM fault
 * @pasid: PASID of the process causing the fault
 * @timestamp: timestamp of the fault
 *
 * Returns:
 * True if the fault was filtered and should not be processed further.
 * False if the fault is a new one and needs to be handled.
 */
bool amdgpu_gmc_filter_faults(struct amdgpu_device *adev,
			      struct amdgpu_ih_ring *ih, uint64_t addr,
			      uint16_t pasid, uint64_t timestamp)
{
	struct amdgpu_gmc *gmc = &adev->gmc;
	uint64_t stamp, key = amdgpu_gmc_fault_key(addr, pasid);
	struct amdgpu_gmc_fault *fault;
	uint32_t hash;

	/* Stale retry fault if timestamp goes backward */
	if (amdgpu_ih_ts_after(timestamp, ih->processed_timestamp))
		return true;

	/* If we don't have space left in the ring buffer return immediately */
	stamp = max(timestamp, AMDGPU_GMC_FAULT_TIMEOUT + 1) -
		AMDGPU_GMC_FAULT_TIMEOUT;
	if (gmc->fault_ring[gmc->last_fault].timestamp >= stamp)
		return true;

	/* Try to find the fault in the hash */
	hash = hash_64(key, AMDGPU_GMC_FAULT_HASH_ORDER);
	fault = &gmc->fault_ring[gmc->fault_hash[hash].idx];
	while (fault->timestamp >= stamp) {
		uint64_t tmp;

		if (atomic64_read(&fault->key) == key) {
			/*
			 * if we get a fault which is already present in
			 * the fault_ring and the timestamp of
			 * the fault is after the expired timestamp,
			 * then this is a new fault that needs to be added
			 * into the fault ring.
			 */
			if (fault->timestamp_expiry != 0 &&
			    amdgpu_ih_ts_after(fault->timestamp_expiry,
					       timestamp))
				break;
			else
				return true;
		}

		tmp = fault->timestamp;
		fault = &gmc->fault_ring[fault->next];

		/* Check if the entry was reused */
		if (fault->timestamp >= tmp)
			break;
	}

	/* Add the fault to the ring */
	fault = &gmc->fault_ring[gmc->last_fault];
	atomic64_set(&fault->key, key);
	fault->timestamp = timestamp;

	/* And update the hash */
	fault->next = gmc->fault_hash[hash].idx;
	gmc->fault_hash[hash].idx = gmc->last_fault++;
	return false;
}

/**
 * amdgpu_gmc_filter_faults_remove - remove address from VM faults filter
 *
 * @adev: amdgpu device structure
 * @addr: address of the VM fault
 * @pasid: PASID of the process causing the fault
 *
 * Remove the address from fault filter, then future vm fault on this address
 * will pass to retry fault handler to recover.
 */
void amdgpu_gmc_filter_faults_remove(struct amdgpu_device *adev, uint64_t addr,
				     uint16_t pasid)
{
	struct amdgpu_gmc *gmc = &adev->gmc;
	uint64_t key = amdgpu_gmc_fault_key(addr, pasid);
	struct amdgpu_ih_ring *ih;
	struct amdgpu_gmc_fault *fault;
	uint32_t last_wptr;
	uint64_t last_ts;
	uint32_t hash;
	uint64_t tmp;

	if (adev->irq.retry_cam_enabled)
		return;

	ih = &adev->irq.ih1;
	/* Get the WPTR of the last entry in IH ring */
	last_wptr = amdgpu_ih_get_wptr(adev, ih);
	/* Order wptr with ring data. */
	rmb();
	/* Get the timetamp of the last entry in IH ring */
	last_ts = amdgpu_ih_decode_iv_ts(adev, ih, last_wptr, -1);

	hash = hash_64(key, AMDGPU_GMC_FAULT_HASH_ORDER);
	fault = &gmc->fault_ring[gmc->fault_hash[hash].idx];
	do {
		if (atomic64_read(&fault->key) == key) {
			/*
			 * Update the timestamp when this fault
			 * expired.
			 */
			fault->timestamp_expiry = last_ts;
			break;
		}

		tmp = fault->timestamp;
		fault = &gmc->fault_ring[fault->next];
	} while (fault->timestamp < tmp);
}

int amdgpu_gmc_ras_sw_init(struct amdgpu_device *adev)
{
	int r;

	/* umc ras block */
	r = amdgpu_umc_ras_sw_init(adev);
	if (r)
		return r;

	/* mmhub ras block */
	r = amdgpu_mmhub_ras_sw_init(adev);
	if (r)
		return r;

	/* hdp ras block */
	r = amdgpu_hdp_ras_sw_init(adev);
	if (r)
		return r;

	/* mca.x ras block */
	r = amdgpu_mca_mp0_ras_sw_init(adev);
	if (r)
		return r;

	r = amdgpu_mca_mp1_ras_sw_init(adev);
	if (r)
		return r;

	r = amdgpu_mca_mpio_ras_sw_init(adev);
	if (r)
		return r;

	/* xgmi ras block */
	r = amdgpu_xgmi_ras_sw_init(adev);
	if (r)
		return r;

	return 0;
}

int amdgpu_gmc_ras_late_init(struct amdgpu_device *adev)
{
	return 0;
}

void amdgpu_gmc_ras_fini(struct amdgpu_device *adev)
{

}

	/*
	 * The latest engine allocation on gfx9/10 is:
	 * Engine 2, 3: firmware
	 * Engine 0, 1, 4~16: amdgpu ring,
	 *                    subject to change when ring number changes
	 * Engine 17: Gart flushes
	 */
#define AMDGPU_VMHUB_INV_ENG_BITMAP		0x1FFF3

int amdgpu_gmc_allocate_vm_inv_eng(struct amdgpu_device *adev)
{
	struct amdgpu_ring *ring;
	unsigned vm_inv_engs[AMDGPU_MAX_VMHUBS] = {0};
	unsigned i;
	unsigned vmhub, inv_eng;

	/* init the vm inv eng for all vmhubs */
	for_each_set_bit(i, adev->vmhubs_mask, AMDGPU_MAX_VMHUBS) {
		vm_inv_engs[i] = AMDGPU_VMHUB_INV_ENG_BITMAP;
		/* reserve engine 5 for firmware */
		if (adev->enable_mes)
			vm_inv_engs[i] &= ~(1 << 5);
		/* reserve mmhub engine 3 for firmware */
		if (adev->enable_umsch_mm)
			vm_inv_engs[i] &= ~(1 << 3);
	}

	for (i = 0; i < adev->num_rings; ++i) {
		ring = adev->rings[i];
		vmhub = ring->vm_hub;

		if (ring == &adev->mes.ring ||
		    ring == &adev->umsch_mm.ring)
			continue;

		inv_eng = ffs(vm_inv_engs[vmhub]);
		if (!inv_eng) {
			dev_err(adev->dev, "no VM inv eng for ring %s\n",
				ring->name);
			return -EINVAL;
		}

		ring->vm_inv_eng = inv_eng - 1;
		vm_inv_engs[vmhub] &= ~(1 << ring->vm_inv_eng);

		dev_info(adev->dev, "ring %s uses VM inv eng %u on hub %u\n",
			 ring->name, ring->vm_inv_eng, ring->vm_hub);
	}

	return 0;
}

void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
			      uint32_t vmhub, uint32_t flush_type)
{
	struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
	struct amdgpu_vmhub *hub = &adev->vmhub[vmhub];
	struct dma_fence *fence;
	struct amdgpu_job *job;
	int r;

	if (!hub->sdma_invalidation_workaround || vmid ||
	    !adev->mman.buffer_funcs_enabled ||
	    !adev->ib_pool_ready || amdgpu_in_reset(adev) ||
	    !ring->sched.ready) {

		/*
		 * A GPU reset should flush all TLBs anyway, so no need to do
		 * this while one is ongoing.
		 */
		if (!down_read_trylock(&adev->reset_domain->sem))
			return;

		if (adev->gmc.flush_tlb_needs_extra_type_2)
			adev->gmc.gmc_funcs->flush_gpu_tlb(adev, vmid,
							   vmhub, 2);

		if (adev->gmc.flush_tlb_needs_extra_type_0 && flush_type == 2)
			adev->gmc.gmc_funcs->flush_gpu_tlb(adev, vmid,
							   vmhub, 0);

		adev->gmc.gmc_funcs->flush_gpu_tlb(adev, vmid, vmhub,
						   flush_type);
		up_read(&adev->reset_domain->sem);
		return;
	}

	/* The SDMA on Navi 1x has a bug which can theoretically result in memory
	 * corruption if an invalidation happens at the same time as an VA
	 * translation. Avoid this by doing the invalidation from the SDMA
	 * itself at least for GART.
	 */
	mutex_lock(&adev->mman.gtt_window_lock);
	r = amdgpu_job_alloc_with_ib(ring->adev, &adev->mman.high_pr,
				     AMDGPU_FENCE_OWNER_UNDEFINED,
				     16 * 4, AMDGPU_IB_POOL_IMMEDIATE,
				     &job);
	if (r)
		goto error_alloc;

	job->vm_pd_addr = amdgpu_gmc_pd_addr(adev->gart.bo);
	job->vm_needs_flush = true;
	job->ibs->ptr[job->ibs->length_dw++] = ring->funcs->nop;
	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
	fence = amdgpu_job_submit(job);
	mutex_unlock(&adev->mman.gtt_window_lock);

	dma_fence_wait(fence, false);
	dma_fence_put(fence);

	return;

error_alloc:
	mutex_unlock(&adev->mman.gtt_window_lock);
	dev_err(adev->dev, "Error flushing GPU TLB using the SDMA (%d)!\n", r);
}

int amdgpu_gmc_flush_gpu_tlb_pasid(struct amdgpu_device *adev, uint16_t pasid,
				   uint32_t flush_type, bool all_hub,
				   uint32_t inst)
{
	u32 usec_timeout = amdgpu_sriov_vf(adev) ? SRIOV_USEC_TIMEOUT :
		adev->usec_timeout;
	struct amdgpu_ring *ring = &adev->gfx.kiq[inst].ring;
	struct amdgpu_kiq *kiq = &adev->gfx.kiq[inst];
	unsigned int ndw;
	signed long r;
	uint32_t seq;

	if (!adev->gmc.flush_pasid_uses_kiq || !ring->sched.ready ||
	    !down_read_trylock(&adev->reset_domain->sem)) {

		if (adev->gmc.flush_tlb_needs_extra_type_2)
			adev->gmc.gmc_funcs->flush_gpu_tlb_pasid(adev, pasid,
								 2, all_hub,
								 inst);

		if (adev->gmc.flush_tlb_needs_extra_type_0 && flush_type == 2)
			adev->gmc.gmc_funcs->flush_gpu_tlb_pasid(adev, pasid,
								 0, all_hub,
								 inst);

		adev->gmc.gmc_funcs->flush_gpu_tlb_pasid(adev, pasid,
							 flush_type, all_hub,
							 inst);
		return 0;
	}

	/* 2 dwords flush + 8 dwords fence */
	ndw = kiq->pmf->invalidate_tlbs_size + 8;

	if (adev->gmc.flush_tlb_needs_extra_type_2)
		ndw += kiq->pmf->invalidate_tlbs_size;

	if (adev->gmc.flush_tlb_needs_extra_type_0)
		ndw += kiq->pmf->invalidate_tlbs_size;

	spin_lock(&adev->gfx.kiq[inst].ring_lock);
	amdgpu_ring_alloc(ring, ndw);
	if (adev->gmc.flush_tlb_needs_extra_type_2)
		kiq->pmf->kiq_invalidate_tlbs(ring, pasid, 2, all_hub);

	if (flush_type == 2 && adev->gmc.flush_tlb_needs_extra_type_0)
		kiq->pmf->kiq_invalidate_tlbs(ring, pasid, 0, all_hub);

	kiq->pmf->kiq_invalidate_tlbs(ring, pasid, flush_type, all_hub);
	r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
	if (r) {
		amdgpu_ring_undo(ring);
		spin_unlock(&adev->gfx.kiq[inst].ring_lock);
		goto error_unlock_reset;
	}

	amdgpu_ring_commit(ring);
	spin_unlock(&adev->gfx.kiq[inst].ring_lock);
	r = amdgpu_fence_wait_polling(ring, seq, usec_timeout);
	if (r < 1) {
		dev_err(adev->dev, "wait for kiq fence error: %ld.\n", r);
		r = -ETIME;
		goto error_unlock_reset;
	}
	r = 0;

error_unlock_reset:
	up_read(&adev->reset_domain->sem);
	return r;
}

/**
 * amdgpu_gmc_tmz_set -- check and set if a device supports TMZ
 * @adev: amdgpu_device pointer
 *
 * Check and set if an the device @adev supports Trusted Memory
 * Zones (TMZ).
 */
void amdgpu_gmc_tmz_set(struct amdgpu_device *adev)
{
	switch (amdgpu_ip_version(adev, GC_HWIP, 0)) {
	/* RAVEN */
	case IP_VERSION(9, 2, 2):
	case IP_VERSION(9, 1, 0):
	/* RENOIR looks like RAVEN */
	case IP_VERSION(9, 3, 0):
	/* GC 10.3.7 */
	case IP_VERSION(10, 3, 7):
	/* GC 11.0.1 */
	case IP_VERSION(11, 0, 1):
		if (amdgpu_tmz == 0) {
			adev->gmc.tmz_enabled = false;
			dev_info(adev->dev,
				 "Trusted Memory Zone (TMZ) feature disabled (cmd line)\n");
		} else {
			adev->gmc.tmz_enabled = true;
			dev_info(adev->dev,
				 "Trusted Memory Zone (TMZ) feature enabled\n");
		}
		break;
	case IP_VERSION(10, 1, 10):
	case IP_VERSION(10, 1, 1):
	case IP_VERSION(10, 1, 2):
	case IP_VERSION(10, 1, 3):
	case IP_VERSION(10, 3, 0):
	case IP_VERSION(10, 3, 2):
	case IP_VERSION(10, 3, 4):
	case IP_VERSION(10, 3, 5):
	case IP_VERSION(10, 3, 6):
	/* VANGOGH */
	case IP_VERSION(10, 3, 1):
	/* YELLOW_CARP*/
	case IP_VERSION(10, 3, 3):
	case IP_VERSION(11, 0, 4):
	case IP_VERSION(11, 5, 0):
		/* Don't enable it by default yet.
		 */
		if (amdgpu_tmz < 1) {
			adev->gmc.tmz_enabled = false;
			dev_info(adev->dev,
				 "Trusted Memory Zone (TMZ) feature disabled as experimental (default)\n");
		} else {
			adev->gmc.tmz_enabled = true;
			dev_info(adev->dev,
				 "Trusted Memory Zone (TMZ) feature enabled as experimental (cmd line)\n");
		}
		break;
	default:
		adev->gmc.tmz_enabled = false;
		dev_info(adev->dev,
			 "Trusted Memory Zone (TMZ) feature not supported\n");
		break;
	}
}

/**
 * amdgpu_gmc_noretry_set -- set per asic noretry defaults
 * @adev: amdgpu_device pointer
 *
 * Set a per asic default for the no-retry parameter.
 *
 */
void amdgpu_gmc_noretry_set(struct amdgpu_device *adev)
{
	struct amdgpu_gmc *gmc = &adev->gmc;
	uint32_t gc_ver = amdgpu_ip_version(adev, GC_HWIP, 0);
	bool noretry_default = (gc_ver == IP_VERSION(9, 0, 1) ||
				gc_ver == IP_VERSION(9, 3, 0) ||
				gc_ver == IP_VERSION(9, 4, 0) ||
				gc_ver == IP_VERSION(9, 4, 1) ||
				gc_ver == IP_VERSION(9, 4, 2) ||
				gc_ver == IP_VERSION(9, 4, 3) ||
				gc_ver >= IP_VERSION(10, 3, 0));

	if (!amdgpu_sriov_xnack_support(adev))
		gmc->noretry = 1;
	else
		gmc->noretry = (amdgpu_noretry == -1) ? noretry_default : amdgpu_noretry;
}

void amdgpu_gmc_set_vm_fault_masks(struct amdgpu_device *adev, int hub_type,
				   bool enable)
{
	struct amdgpu_vmhub *hub;
	u32 tmp, reg, i;

	hub = &adev->vmhub[hub_type];
	for (i = 0; i < 16; i++) {
		reg = hub->vm_context0_cntl + hub->ctx_distance * i;

		tmp = (hub_type == AMDGPU_GFXHUB(0)) ?
			RREG32_SOC15_IP(GC, reg) :
			RREG32_SOC15_IP(MMHUB, reg);

		if (enable)
			tmp |= hub->vm_cntx_cntl_vm_fault;
		else
			tmp &= ~hub->vm_cntx_cntl_vm_fault;

		(hub_type == AMDGPU_GFXHUB(0)) ?
			WREG32_SOC15_IP(GC, reg, tmp) :
			WREG32_SOC15_IP(MMHUB, reg, tmp);
	}
}

void amdgpu_gmc_get_vbios_allocations(struct amdgpu_device *adev)
{
	unsigned size;

	/*
	 * Some ASICs need to reserve a region of video memory to avoid access
	 * from driver
	 */
	adev->mman.stolen_reserved_offset = 0;
	adev->mman.stolen_reserved_size = 0;

	/*
	 * TODO:
	 * Currently there is a bug where some memory client outside
	 * of the driver writes to first 8M of VRAM on S3 resume,
	 * this overrides GART which by default gets placed in first 8M and
	 * causes VM_FAULTS once GTT is accessed.
	 * Keep the stolen memory reservation until the while this is not solved.
	 */
	switch (adev->asic_type) {
	case CHIP_VEGA10:
		adev->mman.keep_stolen_vga_memory = true;
		/*
		 * VEGA10 SRIOV VF with MS_HYPERV host needs some firmware reserved area.
		 */
#ifdef CONFIG_X86
		if (amdgpu_sriov_vf(adev) && hypervisor_is_type(X86_HYPER_MS_HYPERV)) {
			adev->mman.stolen_reserved_offset = 0x500000;
			adev->mman.stolen_reserved_size = 0x200000;
		}
#endif
		break;
	case CHIP_RAVEN:
	case CHIP_RENOIR:
		adev->mman.keep_stolen_vga_memory = true;
		break;
	default:
		adev->mman.keep_stolen_vga_memory = false;
		break;
	}

	if (amdgpu_sriov_vf(adev) ||
	    !amdgpu_device_has_display_hardware(adev)) {
		size = 0;
	} else {
		size = amdgpu_gmc_get_vbios_fb_size(adev);

		if (adev->mman.keep_stolen_vga_memory)
			size = max(size, (unsigned)AMDGPU_VBIOS_VGA_ALLOCATION);
	}

	/* set to 0 if the pre-OS buffer uses up most of vram */
	if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
		size = 0;

	if (size > AMDGPU_VBIOS_VGA_ALLOCATION) {
		adev->mman.stolen_vga_size = AMDGPU_VBIOS_VGA_ALLOCATION;
		adev->mman.stolen_extended_size = size - adev->mman.stolen_vga_size;
	} else {
		adev->mman.stolen_vga_size = size;
		adev->mman.stolen_extended_size = 0;
	}
}

/**
 * amdgpu_gmc_init_pdb0 - initialize PDB0
 *
 * @adev: amdgpu_device pointer
 *
 * This function is only used when GART page table is used
 * for FB address translatioin. In such a case, we construct
 * a 2-level system VM page table: PDB0->PTB, to cover both
 * VRAM of the hive and system memory.
 *
 * PDB0 is static, initialized once on driver initialization.
 * The first n entries of PDB0 are used as PTE by setting
 * P bit to 1, pointing to VRAM. The n+1'th entry points
 * to a big PTB covering system memory.
 *
 */
void amdgpu_gmc_init_pdb0(struct amdgpu_device *adev)
{
	int i;
	uint64_t flags = adev->gart.gart_pte_flags; //TODO it is UC. explore NC/RW?
	/* Each PDE0 (used as PTE) covers (2^vmid0_page_table_block_size)*2M
	 */
	u64 vram_size = adev->gmc.xgmi.node_segment_size * adev->gmc.xgmi.num_physical_nodes;
	u64 pde0_page_size = (1ULL<<adev->gmc.vmid0_page_table_block_size)<<21;
	u64 vram_addr = adev->vm_manager.vram_base_offset -
		adev->gmc.xgmi.physical_node_id * adev->gmc.xgmi.node_segment_size;
	u64 vram_end = vram_addr + vram_size;
	u64 gart_ptb_gpu_pa = amdgpu_gmc_vram_pa(adev, adev->gart.bo);
	int idx;

	if (!drm_dev_enter(adev_to_drm(adev), &idx))
		return;

	flags |= AMDGPU_PTE_VALID | AMDGPU_PTE_READABLE;
	flags |= AMDGPU_PTE_WRITEABLE;
	flags |= AMDGPU_PTE_SNOOPED;
	flags |= AMDGPU_PTE_FRAG((adev->gmc.vmid0_page_table_block_size + 9*1));
	flags |= AMDGPU_PDE_PTE;

	/* The first n PDE0 entries are used as PTE,
	 * pointing to vram
	 */
	for (i = 0; vram_addr < vram_end; i++, vram_addr += pde0_page_size)
		amdgpu_gmc_set_pte_pde(adev, adev->gmc.ptr_pdb0, i, vram_addr, flags);

	/* The n+1'th PDE0 entry points to a huge
	 * PTB who has more than 512 entries each
	 * pointing to a 4K system page
	 */
	flags = AMDGPU_PTE_VALID;
	flags |= AMDGPU_PDE_BFS(0) | AMDGPU_PTE_SNOOPED;
	/* Requires gart_ptb_gpu_pa to be 4K aligned */
	amdgpu_gmc_set_pte_pde(adev, adev->gmc.ptr_pdb0, i, gart_ptb_gpu_pa, flags);
	drm_dev_exit(idx);
}

/**
 * amdgpu_gmc_vram_mc2pa - calculate vram buffer's physical address from MC
 * address
 *
 * @adev: amdgpu_device pointer
 * @mc_addr: MC address of buffer
 */
uint64_t amdgpu_gmc_vram_mc2pa(struct amdgpu_device *adev, uint64_t mc_addr)
{
	return mc_addr - adev->gmc.vram_start + adev->vm_manager.vram_base_offset;
}

/**
 * amdgpu_gmc_vram_pa - calculate vram buffer object's physical address from
 * GPU's view
 *
 * @adev: amdgpu_device pointer
 * @bo: amdgpu buffer object
 */
uint64_t amdgpu_gmc_vram_pa(struct amdgpu_device *adev, struct amdgpu_bo *bo)
{
	return amdgpu_gmc_vram_mc2pa(adev, amdgpu_bo_gpu_offset(bo));
}

/**
 * amdgpu_gmc_vram_cpu_pa - calculate vram buffer object's physical address
 * from CPU's view
 *
 * @adev: amdgpu_device pointer
 * @bo: amdgpu buffer object
 */
uint64_t amdgpu_gmc_vram_cpu_pa(struct amdgpu_device *adev, struct amdgpu_bo *bo)
{
	return amdgpu_bo_gpu_offset(bo) - adev->gmc.vram_start + adev->gmc.aper_base;
}

int amdgpu_gmc_vram_checking(struct amdgpu_device *adev)
{
	struct amdgpu_bo *vram_bo = NULL;
	uint64_t vram_gpu = 0;
	void *vram_ptr = NULL;

	int ret, size = 0x100000;
	uint8_t cptr[10];

	ret = amdgpu_bo_create_kernel(adev, size, PAGE_SIZE,
				AMDGPU_GEM_DOMAIN_VRAM,
				&vram_bo,
				&vram_gpu,
				&vram_ptr);
	if (ret)
		return ret;

	memset(vram_ptr, 0x86, size);
	memset(cptr, 0x86, 10);

	/**
	 * Check the start, the mid, and the end of the memory if the content of
	 * each byte is the pattern "0x86". If yes, we suppose the vram bo is
	 * workable.
	 *
	 * Note: If check the each byte of whole 1M bo, it will cost too many
	 * seconds, so here, we just pick up three parts for emulation.
	 */
	ret = memcmp(vram_ptr, cptr, 10);
	if (ret) {
		ret = -EIO;
		goto release_buffer;
	}

	ret = memcmp(vram_ptr + (size / 2), cptr, 10);
	if (ret) {
		ret = -EIO;
		goto release_buffer;
	}

	ret = memcmp(vram_ptr + size - 10, cptr, 10);
	if (ret) {
		ret = -EIO;
		goto release_buffer;
	}

release_buffer:
	amdgpu_bo_free_kernel(&vram_bo, &vram_gpu,
			&vram_ptr);

	return ret;
}

static ssize_t current_memory_partition_show(
	struct device *dev, struct device_attribute *addr, char *buf)
{
	struct drm_device *ddev = dev_get_drvdata(dev);
	struct amdgpu_device *adev = drm_to_adev(ddev);
	enum amdgpu_memory_partition mode;

	mode = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
	switch (mode) {
	case AMDGPU_NPS1_PARTITION_MODE:
		return sysfs_emit(buf, "NPS1\n");
	case AMDGPU_NPS2_PARTITION_MODE:
		return sysfs_emit(buf, "NPS2\n");
	case AMDGPU_NPS3_PARTITION_MODE:
		return sysfs_emit(buf, "NPS3\n");
	case AMDGPU_NPS4_PARTITION_MODE:
		return sysfs_emit(buf, "NPS4\n");
	case AMDGPU_NPS6_PARTITION_MODE:
		return sysfs_emit(buf, "NPS6\n");
	case AMDGPU_NPS8_PARTITION_MODE:
		return sysfs_emit(buf, "NPS8\n");
	default:
		return sysfs_emit(buf, "UNKNOWN\n");
	}

	return sysfs_emit(buf, "UNKNOWN\n");
}

static DEVICE_ATTR_RO(current_memory_partition);

int amdgpu_gmc_sysfs_init(struct amdgpu_device *adev)
{
	if (!adev->gmc.gmc_funcs->query_mem_partition_mode)
		return 0;

	return device_create_file(adev->dev,
				  &dev_attr_current_memory_partition);
}

void amdgpu_gmc_sysfs_fini(struct amdgpu_device *adev)
{
	device_remove_file(adev->dev, &dev_attr_current_memory_partition);
}