summaryrefslogtreecommitdiff
path: root/sys/uvm/uvm_pager.c
blob: b5ddf3914897933277eec15fd9677a184a782005 (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
/*	$OpenBSD: uvm_pager.c,v 1.47 2009/03/20 15:19:04 oga Exp $	*/
/*	$NetBSD: uvm_pager.c,v 1.36 2000/11/27 18:26:41 chs Exp $	*/

/*
 *
 * Copyright (c) 1997 Charles D. Cranor and Washington University.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *      This product includes software developed by Charles D. Cranor and
 *      Washington University.
 * 4. The name of the author may not be used to endorse or promote products
 *    derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 * from: Id: uvm_pager.c,v 1.1.2.23 1998/02/02 20:38:06 chuck Exp
 */

/*
 * uvm_pager.c: generic functions used to assist the pagers.
 */

#define UVM_PAGER
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/malloc.h>
#include <sys/pool.h>
#include <sys/vnode.h>
#include <sys/buf.h>

#include <uvm/uvm.h>

struct pool *uvm_aiobuf_pool;

struct uvm_pagerops *uvmpagerops[] = {
	&aobj_pager,
	&uvm_deviceops,
	&uvm_vnodeops,
};

/*
 * the pager map: provides KVA for I/O
 *
 * Each uvm_pseg has room for MAX_PAGERMAP_SEGS pager io space of
 * MAXBSIZE bytes.
 *
 * The number of uvm_pseg instances is dynamic using an array segs.
 * At most UVM_PSEG_COUNT instances can exist.
 *
 * psegs[0] always exists (so that the pager can always map in pages).
 * psegs[0] element 0 is always reserved for the pagedaemon.
 *
 * Any other pseg is automatically created when no space is available
 * and automatically destroyed when it is no longer in use.
 */
#define MAX_PAGER_SEGS	16
#define PSEG_NUMSEGS	(PAGER_MAP_SIZE / MAX_PAGER_SEGS / MAXBSIZE)
struct uvm_pseg {
	/* Start of virtual space; 0 if not inited. */
	vaddr_t	start;
	/* Bitmap of the segments in use in this pseg. */
	int	use;
};
struct	mutex uvm_pseg_lck;
struct	uvm_pseg psegs[PSEG_NUMSEGS];

#define UVM_PSEG_FULL(pseg)	((pseg)->use == (1 << MAX_PAGER_SEGS) - 1)
#define UVM_PSEG_EMPTY(pseg)	((pseg)->use == 0)
#define UVM_PSEG_INUSE(pseg,id)	(((pseg)->use & (1 << (id))) != 0)

void	uvm_pseg_init(struct uvm_pseg *);
void	uvm_pseg_destroy(struct uvm_pseg *);
vaddr_t	uvm_pseg_get(int);
void	uvm_pseg_release(vaddr_t);

/*
 * uvm_pager_init: init pagers (at boot time)
 */

void
uvm_pager_init(void)
{
	int lcv;

	/*
	 * init pager map
	 */

	uvm_pseg_init(&psegs[0]);
	mtx_init(&uvm_pseg_lck, IPL_VM);

	/*
	 * init ASYNC I/O queue
	 */

	TAILQ_INIT(&uvm.aio_done);

	/*
	 * call pager init functions
	 */
	for (lcv = 0 ; lcv < sizeof(uvmpagerops)/sizeof(struct uvm_pagerops *);
	    lcv++) {
		if (uvmpagerops[lcv]->pgo_init)
			uvmpagerops[lcv]->pgo_init();
	}
}

/*
 * Initialize a uvm_pseg.
 *
 * May fail, in which case seg->start == 0.
 *
 * Caller locks uvm_pseg_lck.
 */
void
uvm_pseg_init(struct uvm_pseg *pseg)
{
	KASSERT(pseg->start == 0);
	KASSERT(pseg->use == 0);
	pseg->start = uvm_km_valloc(kernel_map, MAX_PAGER_SEGS * MAXBSIZE);
}

/*
 * Destroy a uvm_pseg.
 *
 * Never fails.
 *
 * Requires that seg != &psegs[0]
 *
 * Caller locks uvm_pseg_lck.
 */
void
uvm_pseg_destroy(struct uvm_pseg *pseg)
{
	KASSERT(pseg != &psegs[0]);
	KASSERT(pseg->start != 0);
	KASSERT(pseg->use == 0);
	uvm_km_free(kernel_map, pseg->start, MAX_PAGER_SEGS * MAXBSIZE);
	pseg->start = 0;
}

/*
 * Acquire a pager map segment.
 *
 * Returns a vaddr for paging. 0 on failure.
 *
 * Caller does not lock.
 */
vaddr_t
uvm_pseg_get(int flags)
{
	int i;
	struct uvm_pseg *pseg;

	mtx_enter(&uvm_pseg_lck);

pager_seg_restart:
	/* Find first pseg that has room. */
	for (pseg = &psegs[0]; pseg != &psegs[PSEG_NUMSEGS]; pseg++) {
		if (UVM_PSEG_FULL(pseg))
			continue;

		if (pseg->start == 0) {
			/* Need initialization. */
			uvm_pseg_init(pseg);
			if (pseg->start == 0)
				goto pager_seg_fail;
		}

		/* Keep index 0 reserved for pagedaemon. */
		if (pseg == &psegs[0] && curproc != uvm.pagedaemon_proc)
			i = 1;
		else
			i = 0;

		for (; i < MAX_PAGER_SEGS; i++) {
			if (!UVM_PSEG_INUSE(pseg, i)) {
				pseg->use |= 1 << i;
				mtx_leave(&uvm_pseg_lck);
				return pseg->start + i * MAXBSIZE;
			}
		}
	}

pager_seg_fail:
	if ((flags & UVMPAGER_MAPIN_WAITOK) != 0) {
		msleep(&psegs, &uvm_pseg_lck, PVM, "pagerseg", 0);
		goto pager_seg_restart;
	}

	mtx_leave(&uvm_pseg_lck);
	return 0;
}

/*
 * Release a pager map segment.
 *
 * Caller does not lock.
 *
 * Deallocates pseg if it is no longer in use.
 */
void
uvm_pseg_release(vaddr_t segaddr)
{
	int id;
	struct uvm_pseg *pseg;

	for (pseg = &psegs[0]; pseg != &psegs[PSEG_NUMSEGS]; pseg++) {
		if (pseg->start <= segaddr &&
		    segaddr < pseg->start + MAX_PAGER_SEGS * MAXBSIZE)
			break;
	}
	KASSERT(pseg != &psegs[PSEG_NUMSEGS]);

	id = (segaddr - pseg->start) / MAXBSIZE;
	KASSERT(id >= 0 && id < MAX_PAGER_SEGS);

	/* test for no remainder */
	KDASSERT(segaddr == pseg->start + id * MAXBSIZE);

	mtx_enter(&uvm_pseg_lck);

	KASSERT(UVM_PSEG_INUSE(pseg, id));

	pseg->use &= ~(1 << id);
	wakeup(&psegs);

	if (pseg != &psegs[0] && UVM_PSEG_EMPTY(pseg))
		uvm_pseg_destroy(pseg);

	mtx_leave(&uvm_pseg_lck);
}

/*
 * uvm_pagermapin: map pages into KVA for I/O that needs mappings
 *
 * We basically just km_valloc a blank map entry to reserve the space in the
 * kernel map and then use pmap_enter() to put the mappings in by hand.
 */
vaddr_t
uvm_pagermapin(struct vm_page **pps, int npages, int flags)
{
	vaddr_t kva, cva;
	vm_prot_t prot;
	vsize_t size;
	struct vm_page *pp;

	UVMHIST_FUNC("uvm_pagermapin"); UVMHIST_CALLED(maphist);

	UVMHIST_LOG(maphist,"(pps=%p, npages=%ld, flags=%d)",
	    pps, npages, flags,0);

	prot = VM_PROT_READ;
	if (flags & UVMPAGER_MAPIN_READ)
		prot |= VM_PROT_WRITE;
	size = ptoa(npages);

	KASSERT(size <= MAXBSIZE);

	kva = uvm_pseg_get(flags);
	if (kva == 0) {
		UVMHIST_LOG(maphist,"<- NOWAIT failed", 0,0,0,0);
		return 0;
	}

	for (cva = kva ; size != 0 ; size -= PAGE_SIZE, cva += PAGE_SIZE) {
		pp = *pps++;
		KASSERT(pp);
		KASSERT(pp->pg_flags & PG_BUSY);
		/* Allow pmap_enter to fail. */
		if (pmap_enter(pmap_kernel(), cva, VM_PAGE_TO_PHYS(pp),
		    prot, PMAP_WIRED | PMAP_CANFAIL | prot) != 0) {
			pmap_remove(pmap_kernel(), kva, cva);
			pmap_update(pmap_kernel());
			uvm_pseg_release(kva);
			UVMHIST_LOG(maphist,"<- pmap_enter failed", 0,0,0,0);
			return 0;
		}
	}
	pmap_update(pmap_kernel());
	UVMHIST_LOG(maphist, "<- done (KVA=0x%lx)", kva,0,0,0);
	return kva;
}

/*
 * uvm_pagermapout: remove KVA mapping
 *
 * We remove our mappings by hand and then remove the mapping.
 */
void
uvm_pagermapout(vaddr_t kva, int npages)
{
	UVMHIST_FUNC("uvm_pagermapout"); UVMHIST_CALLED(maphist);

	UVMHIST_LOG(maphist, " (kva=0x%lx, npages=%ld)", kva, npages,0,0);

	pmap_remove(pmap_kernel(), kva, kva + (npages << PAGE_SHIFT));
	pmap_update(pmap_kernel());
	uvm_pseg_release(kva);

	UVMHIST_LOG(maphist,"<- done",0,0,0,0);
}

/*
 * uvm_mk_pcluster
 *
 * generic "make 'pager put' cluster" function.  a pager can either
 * [1] set pgo_mk_pcluster to NULL (never cluster), [2] set it to this
 * generic function, or [3] set it to a pager specific function.
 *
 * => caller must lock object _and_ pagequeues (since we need to look
 *    at active vs. inactive bits, etc.)
 * => caller must make center page busy and write-protect it
 * => we mark all cluster pages busy for the caller
 * => the caller must unbusy all pages (and check wanted/released
 *    status if it drops the object lock)
 * => flags:
 *      PGO_ALLPAGES:  all pages in object are valid targets
 *      !PGO_ALLPAGES: use "lo" and "hi" to limit range of cluster
 *      PGO_DOACTCLUST: include active pages in cluster.
 *        NOTE: the caller should clear PG_CLEANCHK bits if PGO_DOACTCLUST.
 *              PG_CLEANCHK is only a hint, but clearing will help reduce
 *		the number of calls we make to the pmap layer.
 */

struct vm_page **
uvm_mk_pcluster(uobj, pps, npages, center, flags, mlo, mhi)
	struct uvm_object *uobj;	/* IN */
	struct vm_page **pps, *center;  /* IN/OUT, IN */
	int *npages, flags;		/* IN/OUT, IN */
	voff_t mlo, mhi;		/* IN (if !PGO_ALLPAGES) */
{
	struct vm_page **ppsp, *pclust;
	voff_t lo, hi, curoff;
	int center_idx, forward, incr;
	UVMHIST_FUNC("uvm_mk_pcluster"); UVMHIST_CALLED(maphist);

	/* 
	 * center page should already be busy and write protected.  XXX:
	 * suppose page is wired?  if we lock, then a process could
	 * fault/block on it.  if we don't lock, a process could write the
	 * pages in the middle of an I/O.  (consider an msync()).  let's
	 * lock it for now (better to delay than corrupt data?).
	 */

	/*
	 * get cluster boundaries, check sanity, and apply our limits as well.
	 */

	uobj->pgops->pgo_cluster(uobj, center->offset, &lo, &hi);
	if ((flags & PGO_ALLPAGES) == 0) {
		if (lo < mlo)
			lo = mlo;
		if (hi > mhi)
			hi = mhi;
	}
	if ((hi - lo) >> PAGE_SHIFT > *npages) { /* pps too small, bail out! */
		pps[0] = center;
		*npages = 1;
		return(pps);
	}

	/*
	 * now determine the center and attempt to cluster around the
	 * edges
	 */

	center_idx = (center->offset - lo) >> PAGE_SHIFT;
	pps[center_idx] = center;	/* plug in the center page */
	ppsp = &pps[center_idx];
	*npages = 1;

	/*
	 * attempt to cluster around the left [backward], and then 
	 * the right side [forward].    
	 *
	 * note that for inactive pages (pages that have been deactivated)
	 * there are no valid mappings and PG_CLEAN should be up to date.
	 * [i.e. there is no need to query the pmap with pmap_is_modified
	 * since there are no mappings].
	 */

	for (forward  = 0 ; forward <= 1 ; forward++) {
		incr = forward ? PAGE_SIZE : -PAGE_SIZE;
		curoff = center->offset + incr;
		for ( ;(forward == 0 && curoff >= lo) ||
		       (forward && curoff < hi);
		      curoff += incr) {

			pclust = uvm_pagelookup(uobj, curoff); /* lookup page */
			if (pclust == NULL) {
				break;			/* no page */
			}
			/* handle active pages */
			/* NOTE: inactive pages don't have pmap mappings */
			if ((pclust->pg_flags & PQ_INACTIVE) == 0) {
				if ((flags & PGO_DOACTCLUST) == 0) {
					/* dont want mapped pages at all */
					break;
				}

				/* make sure "clean" bit is sync'd */
				if ((pclust->pg_flags & PG_CLEANCHK) == 0) {
					if ((pclust->pg_flags & (PG_CLEAN|PG_BUSY))
					   == PG_CLEAN &&
					   pmap_is_modified(pclust))
						atomic_clearbits_int(
						    &pclust->pg_flags,
						    PG_CLEAN);
					/* now checked */
					atomic_setbits_int(&pclust->pg_flags,
					    PG_CLEANCHK);
				}
			}

			/* is page available for cleaning and does it need it */
			if ((pclust->pg_flags & (PG_CLEAN|PG_BUSY)) != 0) {
				break;	/* page is already clean or is busy */
			}

			/* yes!   enroll the page in our array */
			atomic_setbits_int(&pclust->pg_flags, PG_BUSY);
			UVM_PAGE_OWN(pclust, "uvm_mk_pcluster");

			/* XXX: protect wired page?   see above comment. */
			pmap_page_protect(pclust, VM_PROT_READ);
			if (!forward) {
				ppsp--;			/* back up one page */
				*ppsp = pclust;
			} else {
				/* move forward one page */
				ppsp[*npages] = pclust;
			}
			(*npages)++;
		}
	}
	
	/*
	 * done!  return the cluster array to the caller!!!
	 */

	UVMHIST_LOG(maphist, "<- done",0,0,0,0);
	return(ppsp);
}

/*
 * uvm_pager_put: high level pageout routine
 *
 * we want to pageout page "pg" to backing store, clustering if
 * possible.
 *
 * => page queues must be locked by caller
 * => if page is not swap-backed, then "uobj" points to the object
 *	backing it.   this object should be locked by the caller.
 * => if page is swap-backed, then "uobj" should be NULL.
 * => "pg" should be PG_BUSY (by caller), and !PG_CLEAN
 *    for swap-backed memory, "pg" can be NULL if there is no page
 *    of interest [sometimes the case for the pagedaemon]
 * => "ppsp_ptr" should point to an array of npages vm_page pointers
 *	for possible cluster building
 * => flags (first two for non-swap-backed pages)
 *	PGO_ALLPAGES: all pages in uobj are valid targets
 *	PGO_DOACTCLUST: include "PQ_ACTIVE" pages as valid targets
 *	PGO_SYNCIO: do SYNC I/O (no async)
 *	PGO_PDFREECLUST: pagedaemon: drop cluster on successful I/O
 * => start/stop: if (uobj && !PGO_ALLPAGES) limit targets to this range
 *		  if (!uobj) start is the (daddr64_t) of the starting swapblk
 * => return state:
 *	1. we return the VM_PAGER status code of the pageout
 *	2. we return with the page queues unlocked
 *	3. if (uobj != NULL) [!swap_backed] we return with
 *		uobj locked _only_ if PGO_PDFREECLUST is set 
 *		AND result != VM_PAGER_PEND.   in all other cases
 *		we return with uobj unlocked.   [this is a hack
 *		that allows the pagedaemon to save one lock/unlock
 *		pair in the !swap_backed case since we have to
 *		lock the uobj to drop the cluster anyway]
 *	4. on errors we always drop the cluster.   thus, if we return
 *		!PEND, !OK, then the caller only has to worry about
 *		un-busying the main page (not the cluster pages).
 *	5. on success, if !PGO_PDFREECLUST, we return the cluster
 *		with all pages busy (caller must un-busy and check
 *		wanted/released flags).
 */

int
uvm_pager_put(uobj, pg, ppsp_ptr, npages, flags, start, stop)
	struct uvm_object *uobj;	/* IN */
	struct vm_page *pg, ***ppsp_ptr;/* IN, IN/OUT */
	int *npages;			/* IN/OUT */
	int flags;			/* IN */
	voff_t start, stop;		/* IN, IN */
{
	int result;
	daddr64_t swblk;
	struct vm_page **ppsp = *ppsp_ptr;
	UVMHIST_FUNC("uvm_pager_put"); UVMHIST_CALLED(pdhist);

	/*
	 * note that uobj is null  if we are doing a swap-backed pageout.
	 * note that uobj is !null if we are doing normal object pageout.
	 * note that the page queues must be locked to cluster.
	 */

	if (uobj) {	/* if !swap-backed */

		/*
		 * attempt to build a cluster for pageout using its
		 * make-put-cluster function (if it has one).
		 */

		if (uobj->pgops->pgo_mk_pcluster) {
			ppsp = uobj->pgops->pgo_mk_pcluster(uobj, ppsp,
			    npages, pg, flags, start, stop);
			*ppsp_ptr = ppsp;  /* update caller's pointer */
		} else {
			ppsp[0] = pg;
			*npages = 1;
		}

		swblk = 0;		/* XXX: keep gcc happy */

	} else {

		/*
		 * for swap-backed pageout, the caller (the pagedaemon) has
		 * already built the cluster for us.   the starting swap
		 * block we are writing to has been passed in as "start."
		 * "pg" could be NULL if there is no page we are especially
		 * interested in (in which case the whole cluster gets dropped
		 * in the event of an error or a sync "done").
		 */
		swblk = (daddr64_t) start;
		/* ppsp and npages should be ok */
	}

	/* now that we've clustered we can unlock the page queues */
	uvm_unlock_pageq();

	/*
	 * now attempt the I/O.   if we have a failure and we are
	 * clustered, we will drop the cluster and try again.
	 */

ReTry:
	if (uobj) {
		/* object is locked */
		result = uobj->pgops->pgo_put(uobj, ppsp, *npages, flags);
		UVMHIST_LOG(pdhist, "put -> %ld", result, 0,0,0);
		/* object is now unlocked */
	} else {
		/* nothing locked */
		/* XXX daddr64_t -> int */
		result = uvm_swap_put(swblk, ppsp, *npages, flags);
		/* nothing locked */
	}

	/*
	 * we have attempted the I/O.
	 *
	 * if the I/O was a success then:
	 * 	if !PGO_PDFREECLUST, we return the cluster to the 
	 *		caller (who must un-busy all pages)
	 *	else we un-busy cluster pages for the pagedaemon
	 *
	 * if I/O is pending (async i/o) then we return the pending code.
	 * [in this case the async i/o done function must clean up when
	 *  i/o is done...]
	 */

	if (result == VM_PAGER_PEND || result == VM_PAGER_OK) {
		if (result == VM_PAGER_OK && (flags & PGO_PDFREECLUST)) {
			/*
			 * drop cluster and relock object (only if I/O is
			 * not pending)
			 */
			if (uobj)
				/* required for dropcluster */
				simple_lock(&uobj->vmobjlock);
			if (*npages > 1 || pg == NULL)
				uvm_pager_dropcluster(uobj, pg, ppsp, npages,
				    PGO_PDFREECLUST);
			/* if (uobj): object still locked, as per
			 * return-state item #3 */
		}
		return (result);
	}

	/*
	 * a pager error occured (even after dropping the cluster, if there
	 * was one).  give up! the caller only has one page ("pg")
	 * to worry about.
	 */

	if (*npages > 1 || pg == NULL) {
		if (uobj) {
			simple_lock(&uobj->vmobjlock);
		}
		uvm_pager_dropcluster(uobj, pg, ppsp, npages, PGO_REALLOCSWAP);

		/*
		 * for failed swap-backed pageouts with a "pg",
		 * we need to reset pg's swslot to either:
		 * "swblk" (for transient errors, so we can retry),
		 * or 0 (for hard errors).
		 */

		if (uobj == NULL && pg != NULL) {
			/* XXX daddr64_t -> int */
			int nswblk = (result == VM_PAGER_AGAIN) ? swblk : 0;
			if (pg->pg_flags & PQ_ANON) {
				simple_lock(&pg->uanon->an_lock);
				pg->uanon->an_swslot = nswblk;
				simple_unlock(&pg->uanon->an_lock);
			} else {
				simple_lock(&pg->uobject->vmobjlock);
				uao_set_swslot(pg->uobject,
					       pg->offset >> PAGE_SHIFT,
					       nswblk);
				simple_unlock(&pg->uobject->vmobjlock);
			}
		}
		if (result == VM_PAGER_AGAIN) {

			/*
			 * for transient failures, free all the swslots that
			 * we're not going to retry with.
			 */

			if (uobj == NULL) {
				if (pg) {
					/* XXX daddr64_t -> int */
					uvm_swap_free(swblk + 1, *npages - 1);
				} else {
					/* XXX daddr64_t -> int */
					uvm_swap_free(swblk, *npages);
				}
			}
			if (pg) {
				ppsp[0] = pg;
				*npages = 1;
				goto ReTry;
			}
		} else if (uobj == NULL) {

			/*
			 * for hard errors on swap-backed pageouts,
			 * mark the swslots as bad.  note that we do not
			 * free swslots that we mark bad.
			 */

			/* XXX daddr64_t -> int */
			uvm_swap_markbad(swblk, *npages);
		}
	}

	/*
	 * a pager error occurred (even after dropping the cluster, if there
	 * was one).    give up!   the caller only has one page ("pg")
	 * to worry about.
	 */
	
	if (uobj && (flags & PGO_PDFREECLUST) != 0)
		simple_lock(&uobj->vmobjlock);
	return(result);
}

/*
 * uvm_pager_dropcluster: drop a cluster we have built (because we 
 * got an error, or, if PGO_PDFREECLUST we are un-busying the
 * cluster pages on behalf of the pagedaemon).
 *
 * => uobj, if non-null, is a non-swap-backed object that is 
 *	locked by the caller.   we return with this object still
 *	locked.
 * => page queues are not locked
 * => pg is our page of interest (the one we clustered around, can be null)
 * => ppsp/npages is our current cluster
 * => flags: PGO_PDFREECLUST: pageout was a success: un-busy cluster
 *	pages on behalf of the pagedaemon.
 *           PGO_REALLOCSWAP: drop previously allocated swap slots for 
 *		clustered swap-backed pages (except for "pg" if !NULL)
 *		"swblk" is the start of swap alloc (e.g. for ppsp[0])
 *		[only meaningful if swap-backed (uobj == NULL)]
 */

void
uvm_pager_dropcluster(uobj, pg, ppsp, npages, flags)
	struct uvm_object *uobj;	/* IN */
	struct vm_page *pg, **ppsp;	/* IN, IN/OUT */
	int *npages;			/* IN/OUT */
	int flags;
{
	int lcv;
	boolean_t obj_is_alive; 
	struct uvm_object *saved_uobj;

	/*
	 * drop all pages but "pg"
	 */

	for (lcv = 0 ; lcv < *npages ; lcv++) {

		/* skip "pg" or empty slot */
		if (ppsp[lcv] == pg || ppsp[lcv] == NULL)
			continue;
	
		/*
		 * if swap-backed, gain lock on object that owns page.  note
		 * that PQ_ANON bit can't change as long as we are holding
		 * the PG_BUSY bit (so there is no need to lock the page
		 * queues to test it).
		 *
		 * once we have the lock, dispose of the pointer to swap, if
		 * requested
		 */
		if (!uobj) {
			if (ppsp[lcv]->pg_flags & PQ_ANON) {
				simple_lock(&ppsp[lcv]->uanon->an_lock);
				if (flags & PGO_REALLOCSWAP)
					  /* zap swap block */
					  ppsp[lcv]->uanon->an_swslot = 0;
			} else {
				simple_lock(&ppsp[lcv]->uobject->vmobjlock);
				if (flags & PGO_REALLOCSWAP)
					uao_set_swslot(ppsp[lcv]->uobject,
					    ppsp[lcv]->offset >> PAGE_SHIFT, 0);
			}
		}

		/* did someone want the page while we had it busy-locked? */
		if (ppsp[lcv]->pg_flags & PG_WANTED) {
			/* still holding obj lock */
			wakeup(ppsp[lcv]);
		}

		/* if page was released, release it.  otherwise un-busy it */
		if (ppsp[lcv]->pg_flags & PG_RELEASED) {

			if (ppsp[lcv]->pg_flags & PQ_ANON) {
				/* so that anfree will free */
				atomic_clearbits_int(&ppsp[lcv]->pg_flags,
				    PG_BUSY);
				UVM_PAGE_OWN(ppsp[lcv], NULL);

				pmap_page_protect(ppsp[lcv], VM_PROT_NONE);
				simple_unlock(&ppsp[lcv]->uanon->an_lock);
				/* kills anon and frees pg */
				uvm_anfree(ppsp[lcv]->uanon);

				continue;
			}

			/*
			 * pgo_releasepg will dump the page for us
			 */

			saved_uobj = ppsp[lcv]->uobject;
			obj_is_alive =
			    saved_uobj->pgops->pgo_releasepg(ppsp[lcv], NULL);
			
			/* for normal objects, "pg" is still PG_BUSY by us,
			 * so obj can't die */
			KASSERT(!uobj || obj_is_alive);

			/* only unlock the object if it is still alive...  */
			if (obj_is_alive && saved_uobj != uobj)
				simple_unlock(&saved_uobj->vmobjlock);

			/*
			 * XXXCDC: suppose uobj died in the pgo_releasepg?
			 * how pass that
			 * info up to caller.  we are currently ignoring it...
			 */

			continue;		/* next page */
		} else {
			atomic_clearbits_int(&ppsp[lcv]->pg_flags,
			    PG_BUSY|PG_WANTED|PG_FAKE);
			UVM_PAGE_OWN(ppsp[lcv], NULL);
		}

		/*
		 * if we are operating on behalf of the pagedaemon and we 
		 * had a successful pageout update the page!
		 */
		if (flags & PGO_PDFREECLUST) {
			pmap_clear_reference(ppsp[lcv]);
			pmap_clear_modify(ppsp[lcv]);
			atomic_setbits_int(&ppsp[lcv]->pg_flags, PG_CLEAN);
		}

		/* if anonymous cluster, unlock object and move on */
		if (!uobj) {
			if (ppsp[lcv]->pg_flags & PQ_ANON)
				simple_unlock(&ppsp[lcv]->uanon->an_lock);
			else
				simple_unlock(&ppsp[lcv]->uobject->vmobjlock);
		}
	}
}

#ifdef UBC
/*
 * interrupt-context iodone handler for nested i/o bufs.
 *
 * => must be at splbio().
 */

void
uvm_aio_biodone1(bp)
	struct buf *bp;
{
	struct buf *mbp = bp->b_private;

	splassert(IPL_BIO);

	KASSERT(mbp != bp);
	if (bp->b_flags & B_ERROR) {
		mbp->b_flags |= B_ERROR;
		mbp->b_error = bp->b_error;
	}
	mbp->b_resid -= bp->b_bcount;
	pool_put(&bufpool, bp);
	if (mbp->b_resid == 0) {
		biodone(mbp);
	}
}
#endif

/*
 * interrupt-context iodone handler for single-buf i/os
 * or the top-level buf of a nested-buf i/o.
 *
 * => must be at splbio().
 */

void
uvm_aio_biodone(bp)
	struct buf *bp;
{
	splassert(IPL_BIO);

	/* reset b_iodone for when this is a single-buf i/o. */
	bp->b_iodone = uvm_aio_aiodone;

	simple_lock(&uvm.aiodoned_lock);	/* locks uvm.aio_done */
	TAILQ_INSERT_TAIL(&uvm.aio_done, bp, b_freelist);
	wakeup(&uvm.aiodoned);
	simple_unlock(&uvm.aiodoned_lock);
}

/*
 * uvm_aio_aiodone: do iodone processing for async i/os.
 * this should be called in thread context, not interrupt context.
 */

void
uvm_aio_aiodone(bp)
	struct buf *bp;
{
	int npages = bp->b_bufsize >> PAGE_SHIFT;
	struct vm_page *pg, *pgs[MAXPHYS >> PAGE_SHIFT];
	struct uvm_object *uobj;
	int i, error;
	boolean_t write, swap;
	UVMHIST_FUNC("uvm_aio_aiodone"); UVMHIST_CALLED(pdhist);
	UVMHIST_LOG(pdhist, "bp %p", bp, 0,0,0);

	KASSERT(npages <= MAXPHYS >> PAGE_SHIFT);
	splassert(IPL_BIO);

	error = (bp->b_flags & B_ERROR) ? (bp->b_error ? bp->b_error : EIO) : 0;
	write = (bp->b_flags & B_READ) == 0;
#ifdef UBC
	/* XXXUBC B_NOCACHE is for swap pager, should be done differently */
	if (write && !(bp->b_flags & B_NOCACHE) && bioops.io_pageiodone) {
		(*bioops.io_pageiodone)(bp);
	}
#endif

	uobj = NULL;
	for (i = 0; i < npages; i++) {
		pgs[i] = uvm_pageratop((vaddr_t)bp->b_data + (i << PAGE_SHIFT));
		UVMHIST_LOG(pdhist, "pgs[%ld] = %p", i, pgs[i],0,0);
	}
	uvm_pagermapout((vaddr_t)bp->b_data, npages);
#ifdef UVM_SWAP_ENCRYPT
	/*
	 * XXX - assumes that we only get ASYNC writes. used to be above.
	 */
	if (pgs[0]->pg_flags & PQ_ENCRYPT) {
		uvm_swap_freepages(pgs, npages);
		goto freed;
	}
#endif /* UVM_SWAP_ENCRYPT */
	for (i = 0; i < npages; i++) {
		pg = pgs[i];

		if (i == 0) {
			swap = (pg->pg_flags & PQ_SWAPBACKED) != 0;
			if (!swap) {
				uobj = pg->uobject;
				simple_lock(&uobj->vmobjlock);
			}
		}
		KASSERT(swap || pg->uobject == uobj);
		if (swap) {
			if (pg->pg_flags & PQ_ANON) {
				simple_lock(&pg->uanon->an_lock);
			} else {
				simple_lock(&pg->uobject->vmobjlock);
			}
		}

		/*
		 * if this is a read and we got an error, mark the pages
		 * PG_RELEASED so that uvm_page_unbusy() will free them.
		 */
		if (!write && error) {
			atomic_setbits_int(&pg->pg_flags, PG_RELEASED);
			continue;
		}
		KASSERT(!write || (pgs[i]->pg_flags & PG_FAKE) == 0);

		/*
		 * if this is a read and the page is PG_FAKE,
		 * or this was a successful write,
		 * mark the page PG_CLEAN and not PG_FAKE.
		 */

		if ((pgs[i]->pg_flags & PG_FAKE) || (write && error != ENOMEM)) {
			pmap_clear_reference(pgs[i]);
			pmap_clear_modify(pgs[i]);
			atomic_setbits_int(&pgs[i]->pg_flags, PG_CLEAN);
			atomic_clearbits_int(&pgs[i]->pg_flags, PG_FAKE);
		}
		if (swap) {
			if (pg->pg_flags & PQ_ANON) {
				simple_unlock(&pg->uanon->an_lock);
			} else {
				simple_unlock(&pg->uobject->vmobjlock);
			}
		}
	}
	uvm_page_unbusy(pgs, npages);
	if (!swap) {
		simple_unlock(&uobj->vmobjlock);
	}

#ifdef UVM_SWAP_ENCRYPT
freed:
#endif
	if (write && (bp->b_flags & B_AGE) != 0 && bp->b_vp != NULL) {
		vwakeup(bp->b_vp);
	}
	pool_put(&bufpool, bp);
}