summaryrefslogtreecommitdiffstats
path: root/drivers/gpu/nvgpu/common/mm/page_allocator.c
blob: 9c35f528adc38065575cb04180182104ea7234b4 (plain) (blame)
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
/*
 * Copyright (c) 2016-2017, NVIDIA CORPORATION.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms and conditions of the GNU General Public License,
 * version 2, as published by the Free Software Foundation.
 *
 * This program is distributed in the hope it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */

#include <nvgpu/bitops.h>
#include <nvgpu/allocator.h>
#include <nvgpu/page_allocator.h>
#include <nvgpu/kmem.h>
#include <nvgpu/bug.h>
#include <nvgpu/log2.h>

#include "buddy_allocator_priv.h"

#define palloc_dbg(a, fmt, arg...)			\
	alloc_dbg(palloc_owner(a), fmt, ##arg)

/*
 * Since some Linux headers are still leaked into common code this is necessary
 * for some builds.
 */
#ifdef PAGE_SIZE
#undef PAGE_SIZE
#undef PAGE_ALIGN
#endif

/*
 * VIDMEM page size is 4k.
 */
#define PAGE_SIZE		0x1000
#define PAGE_ALIGN(addr)	((addr + (PAGE_SIZE - 1)) &		\
				 ((typeof(addr)) ~(PAGE_SIZE - 1)))

/*
 * Handle the book-keeping for these operations.
 */
static inline void add_slab_page_to_empty(struct page_alloc_slab *slab,
					  struct page_alloc_slab_page *page)
{
	BUG_ON(page->state != SP_NONE);
	nvgpu_list_add(&page->list_entry, &slab->empty);
	slab->nr_empty++;
	page->state = SP_EMPTY;
}
static inline void add_slab_page_to_partial(struct page_alloc_slab *slab,
					    struct page_alloc_slab_page *page)
{
	BUG_ON(page->state != SP_NONE);
	nvgpu_list_add(&page->list_entry, &slab->partial);
	slab->nr_partial++;
	page->state = SP_PARTIAL;
}
static inline void add_slab_page_to_full(struct page_alloc_slab *slab,
					 struct page_alloc_slab_page *page)
{
	BUG_ON(page->state != SP_NONE);
	nvgpu_list_add(&page->list_entry, &slab->full);
	slab->nr_full++;
	page->state = SP_FULL;
}

static inline void del_slab_page_from_empty(struct page_alloc_slab *slab,
					    struct page_alloc_slab_page *page)
{
	nvgpu_list_del(&page->list_entry);
	slab->nr_empty--;
	page->state = SP_NONE;
}
static inline void del_slab_page_from_partial(struct page_alloc_slab *slab,
					      struct page_alloc_slab_page *page)
{
	nvgpu_list_del(&page->list_entry);
	slab->nr_partial--;
	page->state = SP_NONE;
}
static inline void del_slab_page_from_full(struct page_alloc_slab *slab,
					   struct page_alloc_slab_page *page)
{
	nvgpu_list_del(&page->list_entry);
	slab->nr_full--;
	page->state = SP_NONE;
}

static u64 nvgpu_page_alloc_length(struct nvgpu_allocator *a)
{
	struct nvgpu_page_allocator *va = a->priv;

	return nvgpu_alloc_length(&va->source_allocator);
}

static u64 nvgpu_page_alloc_base(struct nvgpu_allocator *a)
{
	struct nvgpu_page_allocator *va = a->priv;

	return nvgpu_alloc_base(&va->source_allocator);
}

static int nvgpu_page_alloc_inited(struct nvgpu_allocator *a)
{
	struct nvgpu_page_allocator *va = a->priv;

	return nvgpu_alloc_initialized(&va->source_allocator);
}

static u64 nvgpu_page_alloc_end(struct nvgpu_allocator *a)
{
	struct nvgpu_page_allocator *va = a->priv;

	return nvgpu_alloc_end(&va->source_allocator);
}

static u64 nvgpu_page_alloc_space(struct nvgpu_allocator *a)
{
	struct nvgpu_page_allocator *va = a->priv;

	return nvgpu_alloc_space(&va->source_allocator);
}

static int nvgpu_page_reserve_co(struct nvgpu_allocator *a,
				 struct nvgpu_alloc_carveout *co)
{
	struct nvgpu_page_allocator *va = a->priv;

	return nvgpu_alloc_reserve_carveout(&va->source_allocator, co);
}

static void nvgpu_page_release_co(struct nvgpu_allocator *a,
				  struct nvgpu_alloc_carveout *co)
{
	struct nvgpu_page_allocator *va = a->priv;

	nvgpu_alloc_release_carveout(&va->source_allocator, co);
}

static void *nvgpu_page_alloc_sgl_next(void *sgl)
{
	struct nvgpu_mem_sgl *nvgpu_sgl = sgl;

	return nvgpu_sgl->next;
}

static u64 nvgpu_page_alloc_sgl_phys(void *sgl)
{
	struct nvgpu_mem_sgl *nvgpu_sgl = sgl;

	return nvgpu_sgl->phys;
}

static u64 nvgpu_page_alloc_sgl_dma(void *sgl)
{
	struct nvgpu_mem_sgl *nvgpu_sgl = sgl;

	return nvgpu_sgl->dma;
}

static u64 nvgpu_page_alloc_sgl_length(void *sgl)
{
	struct nvgpu_mem_sgl *nvgpu_sgl = sgl;

	return nvgpu_sgl->length;
}

static u64 nvgpu_page_alloc_sgl_gpu_addr(struct gk20a *g, void *sgl,
					 struct nvgpu_gmmu_attrs *attrs)
{
	struct nvgpu_mem_sgl *nvgpu_sgl = sgl;

	return nvgpu_sgl->phys;
}

static void nvgpu_page_alloc_sgt_free(struct gk20a *g, struct nvgpu_sgt *sgt)
{
	/*
	 * No-op here. The free is handled by the page_alloc free() functions.
	 */
}

/*
 * These implement the generic scatter gather ops for pages allocated
 * by the page allocator. however, the primary aim for this, is of course,
 * vidmem.
 */
static const struct nvgpu_sgt_ops page_alloc_sgl_ops = {
	.sgl_next = nvgpu_page_alloc_sgl_next,
	.sgl_phys = nvgpu_page_alloc_sgl_phys,
	.sgl_dma = nvgpu_page_alloc_sgl_dma,
	.sgl_length = nvgpu_page_alloc_sgl_length,
	.sgl_gpu_addr = nvgpu_page_alloc_sgl_gpu_addr,
	.sgt_free = nvgpu_page_alloc_sgt_free,
};

/*
 * This actually frees the sgl memory. Used by the page_alloc free() functions.
 */
static void nvgpu_page_alloc_sgl_proper_free(struct gk20a *g,
					     struct nvgpu_mem_sgl *sgl)
{
	struct nvgpu_mem_sgl *next;

	while (sgl) {
		next = sgl->next;
		nvgpu_kfree(g, sgl);
		sgl = next;
	}
}

static void __nvgpu_free_pages(struct nvgpu_page_allocator *a,
			       struct nvgpu_page_alloc *alloc,
			       bool free_buddy_alloc)
{
	struct nvgpu_mem_sgl *sgl = alloc->sgt.sgl;

	if (free_buddy_alloc) {
		while (sgl) {
			nvgpu_free(&a->source_allocator,
				   nvgpu_sgt_get_phys(&alloc->sgt, sgl));
			sgl = nvgpu_sgt_get_next(&alloc->sgt, sgl);
		}
	}

	nvgpu_page_alloc_sgl_proper_free(a->owner->g, sgl);
	nvgpu_kmem_cache_free(a->alloc_cache, alloc);
}

static int __insert_page_alloc(struct nvgpu_page_allocator *a,
			       struct nvgpu_page_alloc *alloc)
{
	alloc->tree_entry.key_start = alloc->base;
	alloc->tree_entry.key_end = alloc->base + alloc->length;

	nvgpu_rbtree_insert(&alloc->tree_entry, &a->allocs);
	return 0;
}

static struct nvgpu_page_alloc *__find_page_alloc(
	struct nvgpu_page_allocator *a,
	u64 addr)
{
	struct nvgpu_page_alloc *alloc;
	struct nvgpu_rbtree_node *node = NULL;

	nvgpu_rbtree_search(addr, &node, a->allocs);
	if (!node)
		return NULL;

	alloc = nvgpu_page_alloc_from_rbtree_node(node);

	nvgpu_rbtree_unlink(node, &a->allocs);

	return alloc;
}

static struct page_alloc_slab_page *alloc_slab_page(
	struct nvgpu_page_allocator *a,
	struct page_alloc_slab *slab)
{
	struct page_alloc_slab_page *slab_page;

	slab_page = nvgpu_kmem_cache_alloc(a->slab_page_cache);
	if (!slab_page) {
		palloc_dbg(a, "OOM: unable to alloc slab_page struct!\n");
		return NULL;
	}

	memset(slab_page, 0, sizeof(*slab_page));

	slab_page->page_addr = nvgpu_alloc(&a->source_allocator, a->page_size);
	if (!slab_page->page_addr) {
		nvgpu_kmem_cache_free(a->slab_page_cache, slab_page);
		palloc_dbg(a, "OOM: vidmem is full!\n");
		return NULL;
	}

	nvgpu_init_list_node(&slab_page->list_entry);
	slab_page->slab_size = slab->slab_size;
	slab_page->nr_objects = (u32)a->page_size / slab->slab_size;
	slab_page->nr_objects_alloced = 0;
	slab_page->owner = slab;
	slab_page->state = SP_NONE;

	a->pages_alloced++;

	palloc_dbg(a, "Allocated new slab page @ 0x%012llx size=%u\n",
		   slab_page->page_addr, slab_page->slab_size);

	return slab_page;
}

static void free_slab_page(struct nvgpu_page_allocator *a,
			   struct page_alloc_slab_page *slab_page)
{
	palloc_dbg(a, "Freeing slab page @ 0x%012llx\n", slab_page->page_addr);

	BUG_ON((slab_page->state != SP_NONE && slab_page->state != SP_EMPTY) ||
	       slab_page->nr_objects_alloced != 0 ||
	       slab_page->bitmap != 0);

	nvgpu_free(&a->source_allocator, slab_page->page_addr);
	a->pages_freed++;

	nvgpu_kmem_cache_free(a->slab_page_cache, slab_page);
}

/*
 * This expects @alloc to have 1 empty sgl_entry ready for usage.
 */
static int __do_slab_alloc(struct nvgpu_page_allocator *a,
			   struct page_alloc_slab *slab,
			   struct nvgpu_page_alloc *alloc)
{
	struct page_alloc_slab_page *slab_page = NULL;
	struct nvgpu_mem_sgl *sgl;
	unsigned long offs;

	/*
	 * Check the partial and empty lists to see if we have some space
	 * readily available. Take the slab_page out of what ever list it
	 * was in since it may be put back into a different list later.
	 */
	if (!nvgpu_list_empty(&slab->partial)) {
		slab_page = nvgpu_list_first_entry(&slab->partial,
					     page_alloc_slab_page,
					     list_entry);
		del_slab_page_from_partial(slab, slab_page);
	} else if (!nvgpu_list_empty(&slab->empty)) {
		slab_page = nvgpu_list_first_entry(&slab->empty,
					     page_alloc_slab_page,
					     list_entry);
		del_slab_page_from_empty(slab, slab_page);
	}

	if (!slab_page) {
		slab_page = alloc_slab_page(a, slab);
		if (!slab_page)
			return -ENOMEM;
	}

	/*
	 * We now have a slab_page. Do the alloc.
	 */
	offs = bitmap_find_next_zero_area(&slab_page->bitmap,
					  slab_page->nr_objects,
					  0, 1, 0);
	if (offs >= slab_page->nr_objects) {
		WARN(1, "Empty/partial slab with no free objects?");

		/* Add the buggy page to the full list... This isn't ideal. */
		add_slab_page_to_full(slab, slab_page);
		return -ENOMEM;
	}

	bitmap_set(&slab_page->bitmap, offs, 1);
	slab_page->nr_objects_alloced++;

	if (slab_page->nr_objects_alloced < slab_page->nr_objects)
		add_slab_page_to_partial(slab, slab_page);
	else if (slab_page->nr_objects_alloced == slab_page->nr_objects)
		add_slab_page_to_full(slab, slab_page);
	else
		BUG(); /* Should be impossible to hit this. */

	/*
	 * Handle building the nvgpu_page_alloc struct. We expect one sgl
	 * to be present.
	 */
	alloc->slab_page = slab_page;
	alloc->nr_chunks = 1;
	alloc->length = slab_page->slab_size;
	alloc->base = slab_page->page_addr + (offs * slab_page->slab_size);

	sgl         = alloc->sgt.sgl;
	sgl->phys   = alloc->base;
	sgl->dma    = alloc->base;
	sgl->length = alloc->length;
	sgl->next   = NULL;

	return 0;
}

/*
 * Allocate from a slab instead of directly from the page allocator.
 */
static struct nvgpu_page_alloc *__nvgpu_alloc_slab(
	struct nvgpu_page_allocator *a, u64 len)
{
	int err, slab_nr;
	struct page_alloc_slab *slab;
	struct nvgpu_page_alloc *alloc = NULL;
	struct nvgpu_mem_sgl *sgl = NULL;

	/*
	 * Align the length to a page and then divide by the page size (4k for
	 * this code). ilog2() of that then gets us the correct slab to use.
	 */
	slab_nr = (int)ilog2(PAGE_ALIGN(len) >> 12);
	slab = &a->slabs[slab_nr];

	alloc = nvgpu_kmem_cache_alloc(a->alloc_cache);
	if (!alloc) {
		palloc_dbg(a, "OOM: could not alloc page_alloc struct!\n");
		goto fail;
	}

	alloc->sgt.ops = &page_alloc_sgl_ops;

	sgl = nvgpu_kzalloc(a->owner->g, sizeof(*sgl));
	if (!sgl) {
		palloc_dbg(a, "OOM: could not alloc sgl struct!\n");
		goto fail;
	}

	alloc->sgt.sgl = sgl;
	err = __do_slab_alloc(a, slab, alloc);
	if (err)
		goto fail;

	palloc_dbg(a, "Alloc 0x%04llx sr=%d id=0x%010llx [slab]\n",
		   len, slab_nr, alloc->base);
	a->nr_slab_allocs++;

	return alloc;

fail:
	if (alloc)
		nvgpu_kmem_cache_free(a->alloc_cache, alloc);
	if (sgl)
		nvgpu_kfree(a->owner->g, sgl);
	return NULL;
}

static void __nvgpu_free_slab(struct nvgpu_page_allocator *a,
			      struct nvgpu_page_alloc *alloc)
{
	struct page_alloc_slab_page *slab_page = alloc->slab_page;
	struct page_alloc_slab *slab = slab_page->owner;
	enum slab_page_state new_state;
	int offs;

	offs = (u32)(alloc->base - slab_page->page_addr) / slab_page->slab_size;
	bitmap_clear(&slab_page->bitmap, offs, 1);

	slab_page->nr_objects_alloced--;

	if (slab_page->nr_objects_alloced == 0)
		new_state = SP_EMPTY;
	else
		new_state = SP_PARTIAL;

	/*
	 * Need to migrate the page to a different list.
	 */
	if (new_state != slab_page->state) {
		/* Delete - can't be in empty. */
		if (slab_page->state == SP_PARTIAL)
			del_slab_page_from_partial(slab, slab_page);
		else
			del_slab_page_from_full(slab, slab_page);

		/* And add. */
		if (new_state == SP_EMPTY) {
			if (nvgpu_list_empty(&slab->empty))
				add_slab_page_to_empty(slab, slab_page);
			else
				free_slab_page(a, slab_page);
		} else {
			add_slab_page_to_partial(slab, slab_page);
		}
	}

	/*
	 * Now handle the page_alloc.
	 */
	__nvgpu_free_pages(a, alloc, false);
	a->nr_slab_frees++;

	return;
}

/*
 * Allocate physical pages. Since the underlying allocator is a buddy allocator
 * the returned pages are always contiguous. However, since there could be
 * fragmentation in the space this allocator will collate smaller non-contiguous
 * allocations together if necessary.
 */
static struct nvgpu_page_alloc *__do_nvgpu_alloc_pages(
	struct nvgpu_page_allocator *a, u64 pages)
{
	struct nvgpu_page_alloc *alloc;
	struct nvgpu_mem_sgl *sgl, *prev_sgl = NULL;
	u64 max_chunk_len = pages << a->page_shift;
	int i = 0;

	alloc = nvgpu_kmem_cache_alloc(a->alloc_cache);
	if (!alloc)
		goto fail;

	memset(alloc, 0, sizeof(*alloc));

	alloc->length = pages << a->page_shift;
	alloc->sgt.ops = &page_alloc_sgl_ops;

	while (pages) {
		u64 chunk_addr = 0;
		u64 chunk_pages = (u64)1 << __fls(pages);
		u64 chunk_len = chunk_pages << a->page_shift;

		/*
		 * Take care of the possibility that the allocation must be
		 * contiguous. If this is not the first iteration then that
		 * means the first iteration failed to alloc the entire
		 * requested size. The buddy allocator guarantees any given
		 * single alloc is contiguous.
		 */
		if (a->flags & GPU_ALLOC_FORCE_CONTIG && i != 0)
			goto fail_cleanup;

		if (chunk_len > max_chunk_len)
			chunk_len = max_chunk_len;

		/*
		 * Keep attempting to allocate in smaller chunks until the alloc
		 * either succeeds or is smaller than the page_size of the
		 * allocator (i.e the allocator is OOM).
		 */
		do {
			chunk_addr = nvgpu_alloc(&a->source_allocator,
						 chunk_len);

			/* Divide by 2 and try again */
			if (!chunk_addr) {
				palloc_dbg(a, "balloc failed: 0x%llx\n",
					   chunk_len);
				chunk_len >>= 1;
				max_chunk_len = chunk_len;
			}
		} while (!chunk_addr && chunk_len >= a->page_size);

		chunk_pages = chunk_len >> a->page_shift;

		if (!chunk_addr) {
			palloc_dbg(a, "bailing @ 0x%llx\n", chunk_len);
			goto fail_cleanup;
		}

		sgl = nvgpu_kzalloc(a->owner->g, sizeof(*sgl));
		if (!sgl) {
			nvgpu_free(&a->source_allocator, chunk_addr);
			goto fail_cleanup;
		}

		pages -= chunk_pages;

		sgl->phys   = chunk_addr;
		sgl->dma    = chunk_addr;
		sgl->length = chunk_len;

		/*
		 * Build the singly linked list with a head node that is part of
		 * the list.
		 */
		if (prev_sgl)
			prev_sgl->next = sgl;
		else
			alloc->sgt.sgl = sgl;

		prev_sgl = sgl;

		i++;
	}

	alloc->nr_chunks = i;
	alloc->base = ((struct nvgpu_mem_sgl *)alloc->sgt.sgl)->phys;

	return alloc;

fail_cleanup:
	sgl = alloc->sgt.sgl;
	while (sgl) {
		struct nvgpu_mem_sgl *next = sgl->next;

		nvgpu_free(&a->source_allocator, sgl->phys);
		nvgpu_kfree(a->owner->g, sgl);

		sgl = next;
	}

	nvgpu_kmem_cache_free(a->alloc_cache, alloc);
fail:
	return NULL;
}

static struct nvgpu_page_alloc *__nvgpu_alloc_pages(
	struct nvgpu_page_allocator *a, u64 len)
{
	struct nvgpu_page_alloc *alloc = NULL;
	struct nvgpu_mem_sgl *sgl;
	u64 pages;
	int i = 0;

	pages = ALIGN(len, a->page_size) >> a->page_shift;

	alloc = __do_nvgpu_alloc_pages(a, pages);
	if (!alloc) {
		palloc_dbg(a, "Alloc 0x%llx (%llu) (failed)\n",
			   pages << a->page_shift, pages);
		return NULL;
	}

	palloc_dbg(a, "Alloc 0x%llx (%llu) id=0x%010llx\n",
		   pages << a->page_shift, pages, alloc->base);
	sgl = alloc->sgt.sgl;
	while (sgl) {
		palloc_dbg(a, "  Chunk %2d: 0x%010llx + 0x%llx\n",
			   i++,
			   nvgpu_sgt_get_phys(&alloc->sgt, sgl),
			   nvgpu_sgt_get_length(&alloc->sgt, sgl));
		sgl = nvgpu_sgt_get_next(&alloc->sgt, sgl);
	}
	palloc_dbg(a, "Alloc done\n");

	return alloc;
}

/*
 * Allocate enough pages to satisfy @len. Page size is determined at
 * initialization of the allocator.
 *
 * The return is actually a pointer to a struct nvgpu_page_alloc pointer. This
 * is because it doesn't make a lot of sense to return the address of the first
 * page in the list of pages (since they could be discontiguous). This has
 * precedent in the dma_alloc APIs, though, it's really just an annoying
 * artifact of the fact that the nvgpu_alloc() API requires a u64 return type.
 */
static u64 nvgpu_page_alloc(struct nvgpu_allocator *__a, u64 len)
{
	struct nvgpu_page_allocator *a = page_allocator(__a);
	struct nvgpu_page_alloc *alloc = NULL;
	u64 real_len;

	/*
	 * If we want contig pages we have to round up to a power of two. It's
	 * easier to do that here than in the buddy allocator.
	 */
	real_len = a->flags & GPU_ALLOC_FORCE_CONTIG ?
		roundup_pow_of_two(len) : len;

	alloc_lock(__a);
	if (a->flags & GPU_ALLOC_4K_VIDMEM_PAGES &&
	    real_len <= (a->page_size / 2))
		alloc = __nvgpu_alloc_slab(a, real_len);
	else
		alloc = __nvgpu_alloc_pages(a, real_len);

	if (!alloc) {
		alloc_unlock(__a);
		return 0;
	}

	__insert_page_alloc(a, alloc);

	a->nr_allocs++;
	if (real_len > a->page_size / 2)
		a->pages_alloced += alloc->length >> a->page_shift;
	alloc_unlock(__a);

	if (a->flags & GPU_ALLOC_NO_SCATTER_GATHER)
		return alloc->base;
	else
		return (u64) (uintptr_t) alloc;
}

/*
 * Note: this will remove the nvgpu_page_alloc struct from the RB tree
 * if it's found.
 */
static void nvgpu_page_free(struct nvgpu_allocator *__a, u64 base)
{
	struct nvgpu_page_allocator *a = page_allocator(__a);
	struct nvgpu_page_alloc *alloc;

	alloc_lock(__a);

	if (a->flags & GPU_ALLOC_NO_SCATTER_GATHER)
		alloc = __find_page_alloc(a, base);
	else
		alloc = __find_page_alloc(a,
			((struct nvgpu_page_alloc *)(uintptr_t)base)->base);

	if (!alloc) {
		palloc_dbg(a, "Hrm, found no alloc?\n");
		goto done;
	}

	a->nr_frees++;

	palloc_dbg(a, "Free  0x%llx id=0x%010llx\n",
		   alloc->length, alloc->base);

	/*
	 * Frees *alloc.
	 */
	if (alloc->slab_page) {
		__nvgpu_free_slab(a, alloc);
	} else {
		a->pages_freed += (alloc->length >> a->page_shift);
		__nvgpu_free_pages(a, alloc, true);
	}

done:
	alloc_unlock(__a);
}

static struct nvgpu_page_alloc *__nvgpu_alloc_pages_fixed(
	struct nvgpu_page_allocator *a, u64 base, u64 length, u32 unused)
{
	struct nvgpu_page_alloc *alloc;
	struct nvgpu_mem_sgl *sgl;

	alloc = nvgpu_kmem_cache_alloc(a->alloc_cache);
	sgl = nvgpu_kzalloc(a->owner->g, sizeof(*sgl));
	if (!alloc || !sgl)
		goto fail;

	alloc->sgt.ops = &page_alloc_sgl_ops;
	alloc->base = nvgpu_alloc_fixed(&a->source_allocator, base, length, 0);
	if (!alloc->base) {
		WARN(1, "nvgpu: failed to fixed alloc pages @ 0x%010llx", base);
		goto fail;
	}

	alloc->nr_chunks = 1;
	alloc->length = length;
	alloc->sgt.sgl = sgl;

	sgl->phys   = alloc->base;
	sgl->dma    = alloc->base;
	sgl->length = length;
	sgl->next   = NULL;

	return alloc;

fail:
	if (sgl)
		nvgpu_kfree(a->owner->g, sgl);
	if (alloc)
		nvgpu_kmem_cache_free(a->alloc_cache, alloc);
	return NULL;
}

/*
 * @page_size is ignored.
 */
static u64 nvgpu_page_alloc_fixed(struct nvgpu_allocator *__a,
				  u64 base, u64 len, u32 page_size)
{
	struct nvgpu_page_allocator *a = page_allocator(__a);
	struct nvgpu_page_alloc *alloc = NULL;
	struct nvgpu_mem_sgl *sgl;
	u64 aligned_len, pages;
	int i = 0;

	aligned_len = ALIGN(len, a->page_size);
	pages = aligned_len >> a->page_shift;

	alloc_lock(__a);

	alloc = __nvgpu_alloc_pages_fixed(a, base, aligned_len, 0);
	if (!alloc) {
		alloc_unlock(__a);
		return 0;
	}

	__insert_page_alloc(a, alloc);
	alloc_unlock(__a);

	palloc_dbg(a, "Alloc [fixed] @ 0x%010llx + 0x%llx (%llu)\n",
		   alloc->base, aligned_len, pages);
	sgl = alloc->sgt.sgl;
	while (sgl) {
		palloc_dbg(a, "  Chunk %2d: 0x%010llx + 0x%llx\n",
			   i++,
			   nvgpu_sgt_get_phys(&alloc->sgt, sgl),
			   nvgpu_sgt_get_length(&alloc->sgt, sgl));
		sgl = nvgpu_sgt_get_next(&alloc->sgt, sgl);
	}

	a->nr_fixed_allocs++;
	a->pages_alloced += pages;

	if (a->flags & GPU_ALLOC_NO_SCATTER_GATHER)
		return alloc->base;
	else
		return (u64) (uintptr_t) alloc;
}

static void nvgpu_page_free_fixed(struct nvgpu_allocator *__a,
				  u64 base, u64 len)
{
	struct nvgpu_page_allocator *a = page_allocator(__a);
	struct nvgpu_page_alloc *alloc;

	alloc_lock(__a);

	if (a->flags & GPU_ALLOC_NO_SCATTER_GATHER) {
		alloc = __find_page_alloc(a, base);
		if (!alloc)
			goto done;
	} else {
		alloc = (struct nvgpu_page_alloc *) (uintptr_t) base;
	}

	palloc_dbg(a, "Free  [fixed] 0x%010llx + 0x%llx\n",
		   alloc->base, alloc->length);

	a->nr_fixed_frees++;
	a->pages_freed += (alloc->length >> a->page_shift);

	/*
	 * This works for the time being since the buddy allocator
	 * uses the same free function for both fixed and regular
	 * allocs. This would have to be updated if the underlying
	 * allocator were to change.
	 */
	__nvgpu_free_pages(a, alloc, true);

done:
	alloc_unlock(__a);
}

static void nvgpu_page_allocator_destroy(struct nvgpu_allocator *__a)
{
	struct nvgpu_page_allocator *a = page_allocator(__a);

	alloc_lock(__a);
	nvgpu_kfree(nvgpu_alloc_to_gpu(__a), a);
	__a->priv = NULL;
	alloc_unlock(__a);
}

#ifdef __KERNEL__
static void nvgpu_page_print_stats(struct nvgpu_allocator *__a,
				   struct seq_file *s, int lock)
{
	struct nvgpu_page_allocator *a = page_allocator(__a);
	int i;

	if (lock)
		alloc_lock(__a);

	__alloc_pstat(s, __a, "Page allocator:\n");
	__alloc_pstat(s, __a, "  allocs         %lld\n", a->nr_allocs);
	__alloc_pstat(s, __a, "  frees          %lld\n", a->nr_frees);
	__alloc_pstat(s, __a, "  fixed_allocs   %lld\n", a->nr_fixed_allocs);
	__alloc_pstat(s, __a, "  fixed_frees    %lld\n", a->nr_fixed_frees);
	__alloc_pstat(s, __a, "  slab_allocs    %lld\n", a->nr_slab_allocs);
	__alloc_pstat(s, __a, "  slab_frees     %lld\n", a->nr_slab_frees);
	__alloc_pstat(s, __a, "  pages alloced  %lld\n", a->pages_alloced);
	__alloc_pstat(s, __a, "  pages freed    %lld\n", a->pages_freed);
	__alloc_pstat(s, __a, "\n");

	__alloc_pstat(s, __a, "Page size:       %lld KB\n",
		      a->page_size >> 10);
	__alloc_pstat(s, __a, "Total pages:     %lld (%lld MB)\n",
		      a->length / a->page_size,
		      a->length >> 20);
	__alloc_pstat(s, __a, "Available pages: %lld (%lld MB)\n",
		      nvgpu_alloc_space(&a->source_allocator) / a->page_size,
		      nvgpu_alloc_space(&a->source_allocator) >> 20);
	__alloc_pstat(s, __a, "\n");

	/*
	 * Slab info.
	 */
	if (a->flags & GPU_ALLOC_4K_VIDMEM_PAGES) {
		__alloc_pstat(s, __a, "Slabs:\n");
		__alloc_pstat(s, __a, "  size      empty     partial   full\n");
		__alloc_pstat(s, __a, "  ----      -----     -------   ----\n");

		for (i = 0; i < a->nr_slabs; i++) {
			struct page_alloc_slab *slab = &a->slabs[i];

			__alloc_pstat(s, __a, "  %-9u %-9d %-9u %u\n",
				      slab->slab_size,
				      slab->nr_empty, slab->nr_partial,
				      slab->nr_full);
		}
		__alloc_pstat(s, __a, "\n");
	}

	__alloc_pstat(s, __a, "Source alloc: %s\n",
		      a->source_allocator.name);
	nvgpu_alloc_print_stats(&a->source_allocator, s, lock);

	if (lock)
		alloc_unlock(__a);
}
#endif

static const struct nvgpu_allocator_ops page_ops = {
	.alloc		= nvgpu_page_alloc,
	.free		= nvgpu_page_free,

	.alloc_fixed	= nvgpu_page_alloc_fixed,
	.free_fixed	= nvgpu_page_free_fixed,

	.reserve_carveout	= nvgpu_page_reserve_co,
	.release_carveout	= nvgpu_page_release_co,

	.base		= nvgpu_page_alloc_base,
	.length		= nvgpu_page_alloc_length,
	.end		= nvgpu_page_alloc_end,
	.inited		= nvgpu_page_alloc_inited,
	.space		= nvgpu_page_alloc_space,

	.fini		= nvgpu_page_allocator_destroy,

#ifdef __KERNEL__
	.print_stats	= nvgpu_page_print_stats,
#endif
};

/*
 * nr_slabs is computed as follows: divide page_size by 4096 to get number of
 * 4k pages in page_size. Then take the base 2 log of that to get number of
 * slabs. For 64k page_size that works on like:
 *
 *   1024*64 / 1024*4 = 16
 *   ilog2(16) = 4
 *
 * That gives buckets of 1, 2, 4, and 8 pages (i.e 4k, 8k, 16k, 32k).
 */
static int nvgpu_page_alloc_init_slabs(struct nvgpu_page_allocator *a)
{
	size_t nr_slabs = ilog2(a->page_size >> 12);
	unsigned int i;

	a->slabs = nvgpu_kcalloc(nvgpu_alloc_to_gpu(a->owner),
				 nr_slabs,
				 sizeof(struct page_alloc_slab));
	if (!a->slabs)
		return -ENOMEM;
	a->nr_slabs = nr_slabs;

	for (i = 0; i < nr_slabs; i++) {
		struct page_alloc_slab *slab = &a->slabs[i];

		slab->slab_size = SZ_4K * (1 << i);
		nvgpu_init_list_node(&slab->empty);
		nvgpu_init_list_node(&slab->partial);
		nvgpu_init_list_node(&slab->full);
		slab->nr_empty = 0;
		slab->nr_partial = 0;
		slab->nr_full = 0;
	}

	return 0;
}

int nvgpu_page_allocator_init(struct gk20a *g, struct nvgpu_allocator *__a,
			      const char *name, u64 base, u64 length,
			      u64 blk_size, u64 flags)
{
	struct nvgpu_page_allocator *a;
	char buddy_name[sizeof(__a->name)];
	int err;

	if (blk_size < SZ_4K)
		return -EINVAL;

	a = nvgpu_kzalloc(g, sizeof(struct nvgpu_page_allocator));
	if (!a)
		return -ENOMEM;

	err = __nvgpu_alloc_common_init(__a, g, name, a, false, &page_ops);
	if (err)
		goto fail;

	a->alloc_cache = nvgpu_kmem_cache_create(g,
					sizeof(struct nvgpu_page_alloc));
	a->slab_page_cache = nvgpu_kmem_cache_create(g,
					sizeof(struct page_alloc_slab_page));
	if (!a->alloc_cache || !a->slab_page_cache) {
		err = -ENOMEM;
		goto fail;
	}

	a->base = base;
	a->length = length;
	a->page_size = blk_size;
	a->page_shift = __ffs(blk_size);
	a->allocs = NULL;
	a->owner = __a;
	a->flags = flags;

	if (flags & GPU_ALLOC_4K_VIDMEM_PAGES && blk_size > SZ_4K) {
		err = nvgpu_page_alloc_init_slabs(a);
		if (err)
			goto fail;
	}

	snprintf(buddy_name, sizeof(buddy_name), "%s-src", name);

	err = nvgpu_buddy_allocator_init(g, &a->source_allocator, buddy_name,
					 base, length, blk_size, 0);
	if (err)
		goto fail;

#ifdef CONFIG_DEBUG_FS
	nvgpu_init_alloc_debug(g, __a);
#endif
	palloc_dbg(a, "New allocator: type      page\n");
	palloc_dbg(a, "               base      0x%llx\n", a->base);
	palloc_dbg(a, "               size      0x%llx\n", a->length);
	palloc_dbg(a, "               page_size 0x%llx\n", a->page_size);
	palloc_dbg(a, "               flags     0x%llx\n", a->flags);
	palloc_dbg(a, "               slabs:    %d\n", a->nr_slabs);

	return 0;

fail:
	if (a->alloc_cache)
		nvgpu_kmem_cache_destroy(a->alloc_cache);
	if (a->slab_page_cache)
		nvgpu_kmem_cache_destroy(a->slab_page_cache);
	nvgpu_kfree(g, a);
	return err;
}