/
uvm_pmemrange.c
2157 lines (1928 loc) · 56.3 KB
/
uvm_pmemrange.c
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
/* $OpenBSD: uvm_pmemrange.c,v 1.59 2020/02/18 12:13:40 mpi Exp $ */
/*
* Copyright (c) 2009, 2010 Ariane van der Steldt <ariane@stack.nl>
*
* Permission to use, copy, modify, and distribute this software for any
* purpose with or without fee is hereby granted, provided that the above
* copyright notice and this permission notice appear in all copies.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
*/
#include <sys/param.h>
#include <sys/systm.h>
#include <uvm/uvm.h>
#include <sys/malloc.h>
#include <sys/kernel.h>
#include <sys/proc.h>
#include <sys/mount.h>
/*
* 2 trees: addr tree and size tree.
*
* The allocator keeps chunks of free pages (called a range).
* Two pages are part of the same range if:
* - all pages in between are part of that range,
* - they are of the same memory type (zeroed or non-zeroed),
* - they are part of the same pmemrange.
* A pmemrange is a range of memory which is part of the same vm_physseg
* and has a use-count.
*
* addr tree is vm_page[0].objt
* size tree is vm_page[1].objt
*
* The size tree is not used for memory ranges of 1 page, instead,
* single queue is vm_page[0].pageq
*
* vm_page[0].fpgsz describes the length of a free range. Two adjecent ranges
* are joined, unless:
* - they have pages in between them which are not free
* - they belong to different memtypes (zeroed vs dirty memory)
* - they are in different pmemrange areas (ISA vs non-ISA memory for instance)
* - they are not a continuation of the same array
* The latter issue is caused by vm_physseg ordering and splitting from the
* MD initialization machinery. The MD code is dependant on freelists and
* happens to split ISA memory from non-ISA memory.
* (Note: freelists die die die!)
*
* uvm_page_init guarantees that every vm_physseg contains an array of
* struct vm_page. Also, uvm_page_physload allocates an array of struct
* vm_page. This code depends on that array. The array may break across
* vm_physsegs boundaries.
*/
/*
* Validate the flags of the page. (Used in asserts.)
* Any free page must have the PQ_FREE flag set.
* Free pages may be zeroed.
* Pmap flags are left untouched.
*
* The PQ_FREE flag is not checked here: by not checking, we can easily use
* this check in pages which are freed.
*/
#define VALID_FLAGS(pg_flags) \
(((pg_flags) & ~(PQ_FREE|PG_ZERO|PG_PMAPMASK)) == 0x0)
/* Tree comparators. */
int uvm_pmemrange_addr_cmp(const struct uvm_pmemrange *,
const struct uvm_pmemrange *);
int uvm_pmemrange_use_cmp(struct uvm_pmemrange *, struct uvm_pmemrange *);
int uvm_pmr_pg_to_memtype(struct vm_page *);
#ifdef DDB
void uvm_pmr_print(void);
#endif
/*
* Memory types. The page flags are used to derive what the current memory
* type of a page is.
*/
int
uvm_pmr_pg_to_memtype(struct vm_page *pg)
{
if (pg->pg_flags & PG_ZERO)
return UVM_PMR_MEMTYPE_ZERO;
/* Default: dirty memory. */
return UVM_PMR_MEMTYPE_DIRTY;
}
/* Trees. */
RBT_GENERATE(uvm_pmr_addr, vm_page, objt, uvm_pmr_addr_cmp);
RBT_GENERATE(uvm_pmr_size, vm_page, objt, uvm_pmr_size_cmp);
RBT_GENERATE(uvm_pmemrange_addr, uvm_pmemrange, pmr_addr,
uvm_pmemrange_addr_cmp);
/* Validation. */
#ifdef DEBUG
void uvm_pmr_assertvalid(struct uvm_pmemrange *pmr);
#else
#define uvm_pmr_assertvalid(pmr) do {} while (0)
#endif
psize_t uvm_pmr_get1page(psize_t, int, struct pglist *,
paddr_t, paddr_t, int);
struct uvm_pmemrange *uvm_pmr_allocpmr(void);
struct vm_page *uvm_pmr_nfindsz(struct uvm_pmemrange *, psize_t, int);
struct vm_page *uvm_pmr_nextsz(struct uvm_pmemrange *,
struct vm_page *, int);
void uvm_pmr_pnaddr(struct uvm_pmemrange *pmr,
struct vm_page *pg, struct vm_page **pg_prev,
struct vm_page **pg_next);
struct vm_page *uvm_pmr_findnextsegment(struct uvm_pmemrange *,
struct vm_page *, paddr_t);
struct vm_page *uvm_pmr_findprevsegment(struct uvm_pmemrange *,
struct vm_page *, paddr_t);
psize_t uvm_pmr_remove_1strange(struct pglist *, paddr_t,
struct vm_page **, int);
psize_t uvm_pmr_remove_1strange_reverse(struct pglist *,
paddr_t *);
void uvm_pmr_split(paddr_t);
struct uvm_pmemrange *uvm_pmemrange_find(paddr_t);
struct uvm_pmemrange *uvm_pmemrange_use_insert(struct uvm_pmemrange_use *,
struct uvm_pmemrange *);
psize_t pow2divide(psize_t, psize_t);
struct vm_page *uvm_pmr_rootupdate(struct uvm_pmemrange *,
struct vm_page *, paddr_t, paddr_t, int);
/*
* Computes num/denom and rounds it up to the next power-of-2.
*
* This is a division function which calculates an approximation of
* num/denom, with result =~ num/denom. It is meant to be fast and doesn't
* have to be accurate.
*
* Providing too large a value makes the allocator slightly faster, at the
* risk of hitting the failure case more often. Providing too small a value
* makes the allocator a bit slower, but less likely to hit a failure case.
*/
psize_t
pow2divide(psize_t num, psize_t denom)
{
int rshift;
for (rshift = 0; num > denom; rshift++, denom <<= 1)
;
return (paddr_t)1 << rshift;
}
/*
* Predicate: lhs is a subrange or rhs.
*
* If rhs_low == 0: don't care about lower bound.
* If rhs_high == 0: don't care about upper bound.
*/
#define PMR_IS_SUBRANGE_OF(lhs_low, lhs_high, rhs_low, rhs_high) \
(((rhs_low) == 0 || (lhs_low) >= (rhs_low)) && \
((rhs_high) == 0 || (lhs_high) <= (rhs_high)))
/*
* Predicate: lhs intersects with rhs.
*
* If rhs_low == 0: don't care about lower bound.
* If rhs_high == 0: don't care about upper bound.
* Ranges don't intersect if they don't have any page in common, array
* semantics mean that < instead of <= should be used here.
*/
#define PMR_INTERSECTS_WITH(lhs_low, lhs_high, rhs_low, rhs_high) \
(((rhs_low) == 0 || (rhs_low) < (lhs_high)) && \
((rhs_high) == 0 || (lhs_low) < (rhs_high)))
/*
* Align to power-of-2 alignment.
*/
#define PMR_ALIGN(pgno, align) \
(((pgno) + ((align) - 1)) & ~((align) - 1))
#define PMR_ALIGN_DOWN(pgno, align) \
((pgno) & ~((align) - 1))
/*
* Comparator: sort by address ascending.
*/
int
uvm_pmemrange_addr_cmp(const struct uvm_pmemrange *lhs,
const struct uvm_pmemrange *rhs)
{
return lhs->low < rhs->low ? -1 : lhs->low > rhs->low;
}
/*
* Comparator: sort by use ascending.
*
* The higher the use value of a range, the more devices need memory in
* this range. Therefore allocate from the range with the lowest use first.
*/
int
uvm_pmemrange_use_cmp(struct uvm_pmemrange *lhs, struct uvm_pmemrange *rhs)
{
int result;
result = lhs->use < rhs->use ? -1 : lhs->use > rhs->use;
if (result == 0)
result = uvm_pmemrange_addr_cmp(lhs, rhs);
return result;
}
int
uvm_pmr_addr_cmp(const struct vm_page *lhs, const struct vm_page *rhs)
{
paddr_t lhs_addr, rhs_addr;
lhs_addr = VM_PAGE_TO_PHYS(lhs);
rhs_addr = VM_PAGE_TO_PHYS(rhs);
return (lhs_addr < rhs_addr ? -1 : lhs_addr > rhs_addr);
}
int
uvm_pmr_size_cmp(const struct vm_page *lhs, const struct vm_page *rhs)
{
psize_t lhs_size, rhs_size;
int cmp;
/* Using second tree, so we receive pg[1] instead of pg[0]. */
lhs_size = (lhs - 1)->fpgsz;
rhs_size = (rhs - 1)->fpgsz;
cmp = (lhs_size < rhs_size ? -1 : lhs_size > rhs_size);
if (cmp == 0)
cmp = uvm_pmr_addr_cmp(lhs - 1, rhs - 1);
return cmp;
}
/*
* Find the first range of free pages that is at least sz pages long.
*/
struct vm_page *
uvm_pmr_nfindsz(struct uvm_pmemrange *pmr, psize_t sz, int mti)
{
struct vm_page *node, *best;
KASSERT(sz >= 1);
if (sz == 1 && !TAILQ_EMPTY(&pmr->single[mti]))
return TAILQ_FIRST(&pmr->single[mti]);
node = RBT_ROOT(uvm_pmr_size, &pmr->size[mti]);
best = NULL;
while (node != NULL) {
if ((node - 1)->fpgsz >= sz) {
best = (node - 1);
node = RBT_LEFT(uvm_objtree, node);
} else
node = RBT_RIGHT(uvm_objtree, node);
}
return best;
}
/*
* Finds the next range. The next range has a size >= pg->fpgsz.
* Returns NULL if no more ranges are available.
*/
struct vm_page *
uvm_pmr_nextsz(struct uvm_pmemrange *pmr, struct vm_page *pg, int mt)
{
struct vm_page *npg;
KASSERT(pmr != NULL && pg != NULL);
if (pg->fpgsz == 1) {
if (TAILQ_NEXT(pg, pageq) != NULL)
return TAILQ_NEXT(pg, pageq);
else
npg = RBT_MIN(uvm_pmr_size, &pmr->size[mt]);
} else
npg = RBT_NEXT(uvm_pmr_size, pg + 1);
return npg == NULL ? NULL : npg - 1;
}
/*
* Finds the previous and next ranges relative to the (uninserted) pg range.
*
* *pg_prev == NULL if no previous range is available, that can join with
* pg.
* *pg_next == NULL if no next range is available, that can join with
* pg.
*/
void
uvm_pmr_pnaddr(struct uvm_pmemrange *pmr, struct vm_page *pg,
struct vm_page **pg_prev, struct vm_page **pg_next)
{
KASSERT(pg_prev != NULL && pg_next != NULL);
*pg_next = RBT_NFIND(uvm_pmr_addr, &pmr->addr, pg);
if (*pg_next == NULL)
*pg_prev = RBT_MAX(uvm_pmr_addr, &pmr->addr);
else
*pg_prev = RBT_PREV(uvm_pmr_addr, *pg_next);
KDASSERT(*pg_next == NULL ||
VM_PAGE_TO_PHYS(*pg_next) > VM_PAGE_TO_PHYS(pg));
KDASSERT(*pg_prev == NULL ||
VM_PAGE_TO_PHYS(*pg_prev) < VM_PAGE_TO_PHYS(pg));
/* Reset if not contig. */
if (*pg_prev != NULL &&
(atop(VM_PAGE_TO_PHYS(*pg_prev)) + (*pg_prev)->fpgsz
!= atop(VM_PAGE_TO_PHYS(pg)) ||
*pg_prev + (*pg_prev)->fpgsz != pg || /* Array broke. */
uvm_pmr_pg_to_memtype(*pg_prev) != uvm_pmr_pg_to_memtype(pg)))
*pg_prev = NULL;
if (*pg_next != NULL &&
(atop(VM_PAGE_TO_PHYS(pg)) + pg->fpgsz
!= atop(VM_PAGE_TO_PHYS(*pg_next)) ||
pg + pg->fpgsz != *pg_next || /* Array broke. */
uvm_pmr_pg_to_memtype(*pg_next) != uvm_pmr_pg_to_memtype(pg)))
*pg_next = NULL;
return;
}
/*
* Remove a range from the address tree.
* Address tree maintains pmr counters.
*/
void
uvm_pmr_remove_addr(struct uvm_pmemrange *pmr, struct vm_page *pg)
{
KDASSERT(RBT_FIND(uvm_pmr_addr, &pmr->addr, pg) == pg);
KDASSERT(pg->pg_flags & PQ_FREE);
RBT_REMOVE(uvm_pmr_addr, &pmr->addr, pg);
pmr->nsegs--;
}
/*
* Remove a range from the size tree.
*/
void
uvm_pmr_remove_size(struct uvm_pmemrange *pmr, struct vm_page *pg)
{
int memtype;
#ifdef DEBUG
struct vm_page *i;
#endif
KDASSERT(pg->fpgsz >= 1);
KDASSERT(pg->pg_flags & PQ_FREE);
memtype = uvm_pmr_pg_to_memtype(pg);
if (pg->fpgsz == 1) {
#ifdef DEBUG
TAILQ_FOREACH(i, &pmr->single[memtype], pageq) {
if (i == pg)
break;
}
KDASSERT(i == pg);
#endif
TAILQ_REMOVE(&pmr->single[memtype], pg, pageq);
} else {
KDASSERT(RBT_FIND(uvm_pmr_size, &pmr->size[memtype],
pg + 1) == pg + 1);
RBT_REMOVE(uvm_pmr_size, &pmr->size[memtype], pg + 1);
}
}
/* Remove from both trees. */
void
uvm_pmr_remove(struct uvm_pmemrange *pmr, struct vm_page *pg)
{
uvm_pmr_assertvalid(pmr);
uvm_pmr_remove_size(pmr, pg);
uvm_pmr_remove_addr(pmr, pg);
uvm_pmr_assertvalid(pmr);
}
/*
* Insert the range described in pg.
* Returns the range thus created (which may be joined with the previous and
* next ranges).
* If no_join, the caller guarantees that the range cannot possibly join
* with adjecent ranges.
*/
struct vm_page *
uvm_pmr_insert_addr(struct uvm_pmemrange *pmr, struct vm_page *pg, int no_join)
{
struct vm_page *prev, *next;
#ifdef DEBUG
struct vm_page *i;
int mt;
#endif
KDASSERT(pg->pg_flags & PQ_FREE);
KDASSERT(pg->fpgsz >= 1);
#ifdef DEBUG
for (mt = 0; mt < UVM_PMR_MEMTYPE_MAX; mt++) {
TAILQ_FOREACH(i, &pmr->single[mt], pageq)
KDASSERT(i != pg);
if (pg->fpgsz > 1) {
KDASSERT(RBT_FIND(uvm_pmr_size, &pmr->size[mt],
pg + 1) == NULL);
}
KDASSERT(RBT_FIND(uvm_pmr_addr, &pmr->addr, pg) == NULL);
}
#endif
if (!no_join) {
uvm_pmr_pnaddr(pmr, pg, &prev, &next);
if (next != NULL) {
uvm_pmr_remove_size(pmr, next);
uvm_pmr_remove_addr(pmr, next);
pg->fpgsz += next->fpgsz;
next->fpgsz = 0;
}
if (prev != NULL) {
uvm_pmr_remove_size(pmr, prev);
prev->fpgsz += pg->fpgsz;
pg->fpgsz = 0;
return prev;
}
}
RBT_INSERT(uvm_pmr_addr, &pmr->addr, pg);
pmr->nsegs++;
return pg;
}
/*
* Insert the range described in pg.
* Returns the range thus created (which may be joined with the previous and
* next ranges).
* Page must already be in the address tree.
*/
void
uvm_pmr_insert_size(struct uvm_pmemrange *pmr, struct vm_page *pg)
{
int memtype;
#ifdef DEBUG
struct vm_page *i;
int mti;
#endif
KDASSERT(pg->fpgsz >= 1);
KDASSERT(pg->pg_flags & PQ_FREE);
memtype = uvm_pmr_pg_to_memtype(pg);
#ifdef DEBUG
for (mti = 0; mti < UVM_PMR_MEMTYPE_MAX; mti++) {
TAILQ_FOREACH(i, &pmr->single[mti], pageq)
KDASSERT(i != pg);
if (pg->fpgsz > 1) {
KDASSERT(RBT_FIND(uvm_pmr_size, &pmr->size[mti],
pg + 1) == NULL);
}
KDASSERT(RBT_FIND(uvm_pmr_addr, &pmr->addr, pg) == pg);
}
for (i = pg; i < pg + pg->fpgsz; i++)
KASSERT(uvm_pmr_pg_to_memtype(i) == memtype);
#endif
if (pg->fpgsz == 1)
TAILQ_INSERT_TAIL(&pmr->single[memtype], pg, pageq);
else
RBT_INSERT(uvm_pmr_size, &pmr->size[memtype], pg + 1);
}
/* Insert in both trees. */
struct vm_page *
uvm_pmr_insert(struct uvm_pmemrange *pmr, struct vm_page *pg, int no_join)
{
uvm_pmr_assertvalid(pmr);
pg = uvm_pmr_insert_addr(pmr, pg, no_join);
uvm_pmr_insert_size(pmr, pg);
uvm_pmr_assertvalid(pmr);
return pg;
}
/*
* Find the last page that is part of this segment.
* => pg: the range at which to start the search.
* => boundary: the page number boundary specification (0 = no boundary).
* => pmr: the pmemrange of the page.
*
* This function returns 1 before the next range, so if you want to have the
* next range, you need to run TAILQ_NEXT(result, pageq) after calling.
* The reason is that this way, the length of the segment is easily
* calculated using: atop(result) - atop(pg) + 1.
* Hence this function also never returns NULL.
*/
struct vm_page *
uvm_pmr_findnextsegment(struct uvm_pmemrange *pmr,
struct vm_page *pg, paddr_t boundary)
{
paddr_t first_boundary;
struct vm_page *next;
struct vm_page *prev;
KDASSERT(pmr->low <= atop(VM_PAGE_TO_PHYS(pg)) &&
pmr->high > atop(VM_PAGE_TO_PHYS(pg)));
if (boundary != 0) {
first_boundary =
PMR_ALIGN(atop(VM_PAGE_TO_PHYS(pg)) + 1, boundary);
} else
first_boundary = 0;
/*
* Increase next until it hits the first page of the next segment.
*
* While loop checks the following:
* - next != NULL we have not reached the end of pgl
* - boundary == 0 || next < first_boundary
* we do not cross a boundary
* - atop(prev) + 1 == atop(next)
* still in the same segment
* - low <= last
* - high > last still in the same memory range
* - memtype is equal allocator is unable to view different memtypes
* as part of the same segment
* - prev + 1 == next no array breakage occurs
*/
prev = pg;
next = TAILQ_NEXT(prev, pageq);
while (next != NULL &&
(boundary == 0 || atop(VM_PAGE_TO_PHYS(next)) < first_boundary) &&
atop(VM_PAGE_TO_PHYS(prev)) + 1 == atop(VM_PAGE_TO_PHYS(next)) &&
pmr->low <= atop(VM_PAGE_TO_PHYS(next)) &&
pmr->high > atop(VM_PAGE_TO_PHYS(next)) &&
uvm_pmr_pg_to_memtype(prev) == uvm_pmr_pg_to_memtype(next) &&
prev + 1 == next) {
prev = next;
next = TAILQ_NEXT(prev, pageq);
}
/*
* End of this segment.
*/
return prev;
}
/*
* Find the first page that is part of this segment.
* => pg: the range at which to start the search.
* => boundary: the page number boundary specification (0 = no boundary).
* => pmr: the pmemrange of the page.
*
* This function returns 1 after the previous range, so if you want to have the
* previous range, you need to run TAILQ_NEXT(result, pageq) after calling.
* The reason is that this way, the length of the segment is easily
* calculated using: atop(pg) - atop(result) + 1.
* Hence this function also never returns NULL.
*/
struct vm_page *
uvm_pmr_findprevsegment(struct uvm_pmemrange *pmr,
struct vm_page *pg, paddr_t boundary)
{
paddr_t first_boundary;
struct vm_page *next;
struct vm_page *prev;
KDASSERT(pmr->low <= atop(VM_PAGE_TO_PHYS(pg)) &&
pmr->high > atop(VM_PAGE_TO_PHYS(pg)));
if (boundary != 0) {
first_boundary =
PMR_ALIGN_DOWN(atop(VM_PAGE_TO_PHYS(pg)), boundary);
} else
first_boundary = 0;
/*
* Increase next until it hits the first page of the previous segment.
*
* While loop checks the following:
* - next != NULL we have not reached the end of pgl
* - boundary == 0 || next >= first_boundary
* we do not cross a boundary
* - atop(prev) - 1 == atop(next)
* still in the same segment
* - low <= last
* - high > last still in the same memory range
* - memtype is equal allocator is unable to view different memtypes
* as part of the same segment
* - prev - 1 == next no array breakage occurs
*/
prev = pg;
next = TAILQ_NEXT(prev, pageq);
while (next != NULL &&
(boundary == 0 || atop(VM_PAGE_TO_PHYS(next)) >= first_boundary) &&
atop(VM_PAGE_TO_PHYS(prev)) - 1 == atop(VM_PAGE_TO_PHYS(next)) &&
pmr->low <= atop(VM_PAGE_TO_PHYS(next)) &&
pmr->high > atop(VM_PAGE_TO_PHYS(next)) &&
uvm_pmr_pg_to_memtype(prev) == uvm_pmr_pg_to_memtype(next) &&
prev - 1 == next) {
prev = next;
next = TAILQ_NEXT(prev, pageq);
}
/*
* Start of this segment.
*/
return prev;
}
/*
* Remove the first segment of contiguous pages from pgl.
* A segment ends if it crosses boundary (unless boundary = 0) or
* if it would enter a different uvm_pmemrange.
*
* Work: the page range that the caller is currently working with.
* May be null.
*
* If is_desperate is non-zero, the smallest segment is erased. Otherwise,
* the first segment is erased (which, if called by uvm_pmr_getpages(),
* probably is the smallest or very close to it).
*/
psize_t
uvm_pmr_remove_1strange(struct pglist *pgl, paddr_t boundary,
struct vm_page **work, int is_desperate)
{
struct vm_page *start, *end, *iter, *iter_end, *inserted, *lowest;
psize_t count;
struct uvm_pmemrange *pmr, *pmr_iter;
KASSERT(!TAILQ_EMPTY(pgl));
/*
* Initialize to first page.
* Unless desperate scan finds a better candidate, this is what'll be
* erased.
*/
start = TAILQ_FIRST(pgl);
pmr = uvm_pmemrange_find(atop(VM_PAGE_TO_PHYS(start)));
end = uvm_pmr_findnextsegment(pmr, start, boundary);
/*
* If we are desperate, we _really_ want to get rid of the smallest
* element (rather than a close match to the smallest element).
*/
if (is_desperate) {
/* Linear search for smallest segment. */
pmr_iter = pmr;
for (iter = TAILQ_NEXT(end, pageq);
iter != NULL && start != end;
iter = TAILQ_NEXT(iter_end, pageq)) {
/*
* Only update pmr if it doesn't match current
* iteration.
*/
if (pmr->low > atop(VM_PAGE_TO_PHYS(iter)) ||
pmr->high <= atop(VM_PAGE_TO_PHYS(iter))) {
pmr_iter = uvm_pmemrange_find(atop(
VM_PAGE_TO_PHYS(iter)));
}
iter_end = uvm_pmr_findnextsegment(pmr_iter, iter,
boundary);
/*
* Current iteration is smaller than best match so
* far; update.
*/
if (VM_PAGE_TO_PHYS(iter_end) - VM_PAGE_TO_PHYS(iter) <
VM_PAGE_TO_PHYS(end) - VM_PAGE_TO_PHYS(start)) {
start = iter;
end = iter_end;
pmr = pmr_iter;
}
}
}
/*
* Calculate count and end of the list.
*/
count = atop(VM_PAGE_TO_PHYS(end) - VM_PAGE_TO_PHYS(start)) + 1;
lowest = start;
end = TAILQ_NEXT(end, pageq);
/*
* Actually remove the range of pages.
*
* Sadly, this cannot be done using pointer iteration:
* vm_physseg is not guaranteed to be sorted on address, hence
* uvm_page_init() may not have initialized its array sorted by
* page number.
*/
for (iter = start; iter != end; iter = iter_end) {
iter_end = TAILQ_NEXT(iter, pageq);
TAILQ_REMOVE(pgl, iter, pageq);
}
lowest->fpgsz = count;
inserted = uvm_pmr_insert(pmr, lowest, 0);
/*
* If the caller was working on a range and this function modified
* that range, update the pointer.
*/
if (work != NULL && *work != NULL &&
atop(VM_PAGE_TO_PHYS(inserted)) <= atop(VM_PAGE_TO_PHYS(*work)) &&
atop(VM_PAGE_TO_PHYS(inserted)) + inserted->fpgsz >
atop(VM_PAGE_TO_PHYS(*work)))
*work = inserted;
return count;
}
/*
* Remove the first segment of contiguous pages from a pgl
* with the list elements in reverse order of physaddr.
*
* A segment ends if it would enter a different uvm_pmemrange.
*
* Stores starting physical address of the segment in pstart.
*/
psize_t
uvm_pmr_remove_1strange_reverse(struct pglist *pgl, paddr_t *pstart)
{
struct vm_page *start, *end, *iter, *iter_end, *lowest;
psize_t count;
struct uvm_pmemrange *pmr;
KASSERT(!TAILQ_EMPTY(pgl));
start = TAILQ_FIRST(pgl);
pmr = uvm_pmemrange_find(atop(VM_PAGE_TO_PHYS(start)));
end = uvm_pmr_findprevsegment(pmr, start, 0);
KASSERT(end <= start);
/*
* Calculate count and end of the list.
*/
count = atop(VM_PAGE_TO_PHYS(start) - VM_PAGE_TO_PHYS(end)) + 1;
lowest = end;
end = TAILQ_NEXT(end, pageq);
/*
* Actually remove the range of pages.
*
* Sadly, this cannot be done using pointer iteration:
* vm_physseg is not guaranteed to be sorted on address, hence
* uvm_page_init() may not have initialized its array sorted by
* page number.
*/
for (iter = start; iter != end; iter = iter_end) {
iter_end = TAILQ_NEXT(iter, pageq);
TAILQ_REMOVE(pgl, iter, pageq);
}
lowest->fpgsz = count;
(void) uvm_pmr_insert(pmr, lowest, 0);
*pstart = VM_PAGE_TO_PHYS(lowest);
return count;
}
/*
* Extract a number of pages from a segment of free pages.
* Called by uvm_pmr_getpages.
*
* Returns the segment that was created from pages left over at the tail
* of the remove set of pages, or NULL if no pages were left at the tail.
*/
struct vm_page *
uvm_pmr_extract_range(struct uvm_pmemrange *pmr, struct vm_page *pg,
paddr_t start, paddr_t end, struct pglist *result)
{
struct vm_page *after, *pg_i;
psize_t before_sz, after_sz;
#ifdef DEBUG
psize_t i;
#endif
KDASSERT(end > start);
KDASSERT(pmr->low <= atop(VM_PAGE_TO_PHYS(pg)));
KDASSERT(pmr->high >= atop(VM_PAGE_TO_PHYS(pg)) + pg->fpgsz);
KDASSERT(atop(VM_PAGE_TO_PHYS(pg)) <= start);
KDASSERT(atop(VM_PAGE_TO_PHYS(pg)) + pg->fpgsz >= end);
before_sz = start - atop(VM_PAGE_TO_PHYS(pg));
after_sz = atop(VM_PAGE_TO_PHYS(pg)) + pg->fpgsz - end;
KDASSERT(before_sz + after_sz + (end - start) == pg->fpgsz);
uvm_pmr_assertvalid(pmr);
uvm_pmr_remove_size(pmr, pg);
if (before_sz == 0)
uvm_pmr_remove_addr(pmr, pg);
after = pg + before_sz + (end - start);
/* Add selected pages to result. */
for (pg_i = pg + before_sz; pg_i != after; pg_i++) {
KASSERT(pg_i->pg_flags & PQ_FREE);
pg_i->fpgsz = 0;
TAILQ_INSERT_TAIL(result, pg_i, pageq);
}
/* Before handling. */
if (before_sz > 0) {
pg->fpgsz = before_sz;
uvm_pmr_insert_size(pmr, pg);
}
/* After handling. */
if (after_sz > 0) {
#ifdef DEBUG
for (i = 0; i < after_sz; i++) {
KASSERT(!uvm_pmr_isfree(after + i));
}
#endif
KDASSERT(atop(VM_PAGE_TO_PHYS(after)) == end);
after->fpgsz = after_sz;
after = uvm_pmr_insert_addr(pmr, after, 1);
uvm_pmr_insert_size(pmr, after);
}
uvm_pmr_assertvalid(pmr);
return (after_sz > 0 ? after : NULL);
}
/*
* Indicate to the page daemon that a nowait call failed and it should
* recover at least some memory in the most restricted region (assumed
* to be dma_constraint).
*/
extern volatile int uvm_nowait_failed;
/*
* Acquire a number of pages.
*
* count: the number of pages returned
* start: lowest page number
* end: highest page number +1
* (start = end = 0: no limitation)
* align: power-of-2 alignment constraint (align = 1: no alignment)
* boundary: power-of-2 boundary (boundary = 0: no boundary)
* maxseg: maximum number of segments to return
* flags: UVM_PLA_* flags
* result: returned pages storage (uses pageq)
*/
int
uvm_pmr_getpages(psize_t count, paddr_t start, paddr_t end, paddr_t align,
paddr_t boundary, int maxseg, int flags, struct pglist *result)
{
struct uvm_pmemrange *pmr; /* Iterate memory ranges. */
struct vm_page *found, *f_next; /* Iterate chunks. */
psize_t fcount; /* Current found pages. */
int fnsegs; /* Current segment counter. */
int try, start_try;
psize_t search[3];
paddr_t fstart, fend; /* Pages to be taken from found. */
int memtype; /* Requested memtype. */
int memtype_init; /* Best memtype. */
int desperate; /* True if allocation failed. */
#ifdef DIAGNOSTIC
struct vm_page *diag_prev; /* Used during validation. */
#endif /* DIAGNOSTIC */
/*
* Validate arguments.
*/
KASSERT(count > 0);
KASSERT(start == 0 || end == 0 || start < end);
KASSERT(align >= 1);
KASSERT(powerof2(align));
KASSERT(maxseg > 0);
KASSERT(boundary == 0 || powerof2(boundary));
KASSERT(boundary == 0 || maxseg * boundary >= count);
KASSERT(TAILQ_EMPTY(result));
/*
* TRYCONTIG is a noop if you only want a single segment.
* Remove it if that's the case: otherwise it'll deny the fast
* allocation.
*/
if (maxseg == 1 || count == 1)
flags &= ~UVM_PLA_TRYCONTIG;
/*
* Configure search.
*
* search[0] is one segment, only used in UVM_PLA_TRYCONTIG case.
* search[1] is multiple segments, chosen to fulfill the search in
* approximately even-sized segments.
* This is a good trade-off between slightly reduced allocation speed
* and less fragmentation.
* search[2] is the worst case, in which all segments are evaluated.
* This provides the least fragmentation, but makes the search
* possibly longer (although in the case it is selected, that no
* longer matters most).
*
* The exception is when maxseg == 1: since we can only fulfill that
* with one segment of size pages, only a single search type has to
* be attempted.
*/
if (maxseg == 1 || count == 1) {
start_try = 2;
search[2] = count;
} else if (maxseg >= count && (flags & UVM_PLA_TRYCONTIG) == 0) {
start_try = 2;
search[2] = 1;
} else {
start_try = 0;
search[0] = count;
search[1] = pow2divide(count, maxseg);
search[2] = 1;
if ((flags & UVM_PLA_TRYCONTIG) == 0)
start_try = 1;
if (search[1] >= search[0]) {
search[1] = search[0];
start_try = 1;
}
if (search[2] >= search[start_try]) {
start_try = 2;
}
}
/*
* Memory type: if zeroed memory is requested, traverse the zero set.
* Otherwise, traverse the dirty set.
*
* The memtype iterator is reinitialized to memtype_init on entrance
* of a pmemrange.
*/
if (flags & UVM_PLA_ZERO)
memtype_init = UVM_PMR_MEMTYPE_ZERO;
else
memtype_init = UVM_PMR_MEMTYPE_DIRTY;
/*
* Initially, we're not desperate.
*
* Note that if we return from a sleep, we are still desperate.
* Chances are that memory pressure is still high, so resetting
* seems over-optimistic to me.
*/
desperate = 0;
uvm_lock_fpageq();
retry: /* Return point after sleeping. */
fcount = 0;
fnsegs = 0;
retry_desperate:
/*
* If we just want any page(s), go for the really fast option.
*/
if (count <= maxseg && align == 1 && boundary == 0 &&
(flags & UVM_PLA_TRYCONTIG) == 0) {
fcount += uvm_pmr_get1page(count - fcount, memtype_init,
result, start, end, 0);
/*
* If we found sufficient pages, go to the succes exit code.
*
* Otherwise, go immediately to fail, since we collected
* all we could anyway.
*/
if (fcount == count)
goto out;
else
goto fail;
}
/*
* The heart of the contig case.
*
* The code actually looks like this:
*
* foreach (struct pmemrange) {
* foreach (memtype) {
* foreach(try) {
* foreach (free range of memtype in pmemrange,
* starting at search[try]) {
* while (range has space left)
* take from range
* }
* }
* }
*
* if next pmemrange has higher usecount than current:
* enter desperate case (which will drain the pmemranges
* until empty prior to moving to the next one)
* }
*
* When desperate is activated, try always starts at the highest
* value. The memtype loop is using a goto ReScanMemtype.
* The try loop is using a goto ReScan.
* The 'range has space left' loop uses label DrainFound.
*
* Writing them all as loops would take up a lot of screen space in
* the form of indentation and some parts are easier to express
* using the labels.
*/
TAILQ_FOREACH(pmr, &uvm.pmr_control.use, pmr_use) {
/* Empty range. */