vm_reserv.c 39.7 KB
Newer Older
1
/*-
2
3
 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
 *
4
 * Copyright (c) 2002-2006 Rice University
Alan Cox's avatar
Alan Cox committed
5
 * Copyright (c) 2007-2011 Alan L. Cox <alc@cs.rice.edu>
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
 * All rights reserved.
 *
 * This software was developed for the FreeBSD Project by Alan L. Cox,
 * Olivier Crameri, Peter Druschel, Sitaram Iyer, and Juan Navarro.
 *
 * 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.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * ``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 COPYRIGHT
 * HOLDERS OR CONTRIBUTORS 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.
 */

/*
 *	Superpage reservation management module
36
37
38
 *
 * Any external functions defined by this module are only to be used by the
 * virtual memory system.
39
40
41
42
43
44
45
46
47
48
49
50
51
 */

#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");

#include "opt_vm.h"

#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mutex.h>
#include <sys/queue.h>
52
#include <sys/rwlock.h>
53
54
55
#include <sys/sbuf.h>
#include <sys/sysctl.h>
#include <sys/systm.h>
56
#include <sys/vmmeter.h>
57
58
59
60
61

#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
62
#include <vm/vm_pageout.h>
63
#include <vm/vm_phys.h>
64
#include <vm/vm_pagequeue.h>
65
#include <vm/vm_radix.h>
66
67
68
69
#include <vm/vm_reserv.h>

/*
 * The reservation system supports the speculative allocation of large physical
70
 * pages ("superpages").  Speculative allocation enables the fully automatic
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
 * utilization of superpages by the virtual memory system.  In other words, no
 * programmatic directives are required to use superpages.
 */

#if VM_NRESERVLEVEL > 0

/*
 * The number of small pages that are contained in a level 0 reservation
 */
#define	VM_LEVEL_0_NPAGES	(1 << VM_LEVEL_0_ORDER)

/*
 * The number of bits by which a physical address is shifted to obtain the
 * reservation number
 */
#define	VM_LEVEL_0_SHIFT	(VM_LEVEL_0_ORDER + PAGE_SHIFT)

/*
 * The size of a level 0 reservation in bytes
 */
#define	VM_LEVEL_0_SIZE		(1 << VM_LEVEL_0_SHIFT)

/*
 * Computes the index of the small page underlying the given (object, pindex)
 * within the reservation's array of small pages.
 */
#define	VM_RESERV_INDEX(object, pindex)	\
    (((object)->pg_color + (pindex)) & (VM_LEVEL_0_NPAGES - 1))

Alan Cox's avatar
Alan Cox committed
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
/*
 * The size of a population map entry
 */
typedef	u_long		popmap_t;

/*
 * The number of bits in a population map entry
 */
#define	NBPOPMAP	(NBBY * sizeof(popmap_t))

/*
 * The number of population map entries in a reservation
 */
#define	NPOPMAP		howmany(VM_LEVEL_0_NPAGES, NBPOPMAP)

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
/*
 * Clear a bit in the population map.
 */
static __inline void
popmap_clear(popmap_t popmap[], int i)
{

	popmap[i / NBPOPMAP] &= ~(1UL << (i % NBPOPMAP));
}

/*
 * Set a bit in the population map.
 */
static __inline void
popmap_set(popmap_t popmap[], int i)
{

	popmap[i / NBPOPMAP] |= 1UL << (i % NBPOPMAP);
}

/*
 * Is a bit in the population map clear?
 */
static __inline boolean_t
popmap_is_clear(popmap_t popmap[], int i)
{

	return ((popmap[i / NBPOPMAP] & (1UL << (i % NBPOPMAP))) == 0);
}

/*
 * Is a bit in the population map set?
 */
static __inline boolean_t
popmap_is_set(popmap_t popmap[], int i)
{

	return ((popmap[i / NBPOPMAP] & (1UL << (i % NBPOPMAP))) != 0);
}

155
156
157
158
159
160
161
162
/*
 * The reservation structure
 *
 * A reservation structure is constructed whenever a large physical page is
 * speculatively allocated to an object.  The reservation provides the small
 * physical pages for the range [pindex, pindex + VM_LEVEL_0_NPAGES) of offsets
 * within that object.  The reservation's "popcnt" tracks the number of these
 * small physical pages that are in use at any given time.  When and if the
163
 * reservation is not fully utilized, it appears in the queue of partially
164
165
166
 * populated reservations.  The reservation always appears on the containing
 * object's list of reservations.
 *
167
 * A partially populated reservation can be broken and reclaimed at any time.
168
169
170
171
 *
 * f - vm_domain_free_lock
 * o - vm_reserv_object_lock
 * c - constant after boot
172
173
 */
struct vm_reserv {
174
175
176
177
178
179
180
181
182
	TAILQ_ENTRY(vm_reserv) partpopq;	/* (f) per-domain queue. */
	LIST_ENTRY(vm_reserv) objq;		/* (o, f) object queue */
	vm_object_t	object;			/* (o, f) containing object */
	vm_pindex_t	pindex;			/* (o, f) offset in object */
	vm_page_t	pages;			/* (c) first page  */
	int		domain;			/* (c) NUMA domain. */
	int		popcnt;			/* (f) # of pages in use */
	char		inpartpopq;		/* (f) */
	popmap_t	popmap[NPOPMAP];	/* (f) bit vector, used pages */
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
};

/*
 * The reservation array
 *
 * This array is analoguous in function to vm_page_array.  It differs in the
 * respect that it may contain a greater number of useful reservation
 * structures than there are (physical) superpages.  These "invalid"
 * reservation structures exist to trade-off space for time in the
 * implementation of vm_reserv_from_page().  Invalid reservation structures are
 * distinguishable from "valid" reservation structures by inspecting the
 * reservation's "pages" field.  Invalid reservation structures have a NULL
 * "pages" field.
 *
 * vm_reserv_from_page() maps a small (physical) page to an element of this
 * array by computing a physical reservation number from the page's physical
 * address.  The physical reservation number is used as the array index.
 *
 * An "active" reservation is a valid reservation structure that has a non-NULL
 * "object" field and a non-zero "popcnt" field.  In other words, every active
 * reservation belongs to a particular object.  Moreover, every active
 * reservation has an entry in the containing object's list of reservations.  
 */
static vm_reserv_t vm_reserv_array;

/*
209
 * The partially populated reservation queue
210
 *
211
212
213
 * This queue enables the fast recovery of an unused free small page from a
 * partially populated reservation.  The reservation at the head of this queue
 * is the least recently changed, partially populated reservation.
214
215
216
 *
 * Access to this queue is synchronized by the free page queue lock.
 */
217
static TAILQ_HEAD(, vm_reserv) vm_rvq_partpop[MAXMEMDOM];
218
219
220
221
222
223
224
225
226
227
228

static SYSCTL_NODE(_vm, OID_AUTO, reserv, CTLFLAG_RD, 0, "Reservation Info");

static long vm_reserv_broken;
SYSCTL_LONG(_vm_reserv, OID_AUTO, broken, CTLFLAG_RD,
    &vm_reserv_broken, 0, "Cumulative number of broken reservations");

static long vm_reserv_freed;
SYSCTL_LONG(_vm_reserv, OID_AUTO, freed, CTLFLAG_RD,
    &vm_reserv_freed, 0, "Cumulative number of freed reservations");

229
230
231
232
233
static int sysctl_vm_reserv_fullpop(SYSCTL_HANDLER_ARGS);

SYSCTL_PROC(_vm_reserv, OID_AUTO, fullpop, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
    sysctl_vm_reserv_fullpop, "I", "Current number of full reservations");

234
235
236
static int sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS);

SYSCTL_OID(_vm_reserv, OID_AUTO, partpopq, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
237
    sysctl_vm_reserv_partpopq, "A", "Partially populated reservation queues");
238
239
240
241
242

static long vm_reserv_reclaimed;
SYSCTL_LONG(_vm_reserv, OID_AUTO, reclaimed, CTLFLAG_RD,
    &vm_reserv_reclaimed, 0, "Cumulative number of reclaimed reservations");

243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
/*
 * The object lock pool is used to synchronize the rvq.  We can not use a
 * pool mutex because it is required before malloc works.
 *
 * The "hash" function could be made faster without divide and modulo.
 */
#define	VM_RESERV_OBJ_LOCK_COUNT	MAXCPU

struct mtx_padalign vm_reserv_object_mtx[VM_RESERV_OBJ_LOCK_COUNT];

#define	vm_reserv_object_lock_idx(object)			\
	    (((uintptr_t)object / sizeof(*object)) % VM_RESERV_OBJ_LOCK_COUNT)
#define	vm_reserv_object_lock_ptr(object)			\
	    &vm_reserv_object_mtx[vm_reserv_object_lock_idx((object))]
#define	vm_reserv_object_lock(object)				\
	    mtx_lock(vm_reserv_object_lock_ptr((object)))
#define	vm_reserv_object_unlock(object)				\
	    mtx_unlock(vm_reserv_object_lock_ptr((object)))

Alan Cox's avatar
Alan Cox committed
262
263
static void		vm_reserv_break(vm_reserv_t rv, vm_page_t m);
static void		vm_reserv_depopulate(vm_reserv_t rv, int index);
264
265
266
static vm_reserv_t	vm_reserv_from_page(vm_page_t m);
static boolean_t	vm_reserv_has_pindex(vm_reserv_t rv,
			    vm_pindex_t pindex);
Alan Cox's avatar
Alan Cox committed
267
static void		vm_reserv_populate(vm_reserv_t rv, int index);
268
static void		vm_reserv_reclaim(vm_reserv_t rv);
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
/*
 * Returns the current number of full reservations.
 *
 * Since the number of full reservations is computed without acquiring the
 * free page queue lock, the returned value may be inexact.
 */
static int
sysctl_vm_reserv_fullpop(SYSCTL_HANDLER_ARGS)
{
	vm_paddr_t paddr;
	struct vm_phys_seg *seg;
	vm_reserv_t rv;
	int fullpop, segind;

	fullpop = 0;
	for (segind = 0; segind < vm_phys_nsegs; segind++) {
		seg = &vm_phys_segs[segind];
		paddr = roundup2(seg->start, VM_LEVEL_0_SIZE);
		while (paddr + VM_LEVEL_0_SIZE <= seg->end) {
			rv = &vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT];
			fullpop += rv->popcnt == VM_LEVEL_0_NPAGES;
			paddr += VM_LEVEL_0_SIZE;
		}
	}
	return (sysctl_handle_int(oidp, &fullpop, 0, req));
}

297
/*
298
 * Describes the current state of the partially populated reservation queue.
299
300
301
302
303
304
 */
static int
sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS)
{
	struct sbuf sbuf;
	vm_reserv_t rv;
305
	int counter, error, domain, level, unused_pages;
306

307
308
309
	error = sysctl_wire_old_buffer(req, 0);
	if (error != 0)
		return (error);
310
	sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
311
312
313
314
315
	sbuf_printf(&sbuf, "\nDOMAIN    LEVEL     SIZE  NUMBER\n\n");
	for (domain = 0; domain < vm_ndomains; domain++) {
		for (level = -1; level <= VM_NRESERVLEVEL - 2; level++) {
			counter = 0;
			unused_pages = 0;
316
			vm_domain_free_lock(VM_DOMAIN(domain));
317
318
319
320
			TAILQ_FOREACH(rv, &vm_rvq_partpop[domain], partpopq) {
				counter++;
				unused_pages += VM_LEVEL_0_NPAGES - rv->popcnt;
			}
321
			vm_domain_free_unlock(VM_DOMAIN(domain));
322
323
324
			sbuf_printf(&sbuf, "%6d, %7d, %6dK, %6d\n",
			    domain, level,
			    unused_pages * ((int)PAGE_SIZE / 1024), counter);
325
326
		}
	}
327
	error = sbuf_finish(&sbuf);
328
329
330
331
	sbuf_delete(&sbuf);
	return (error);
}

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
/*
 * Remove a reservation from the object's objq.
 */
static void
vm_reserv_remove(vm_reserv_t rv)
{
	vm_object_t object;

	KASSERT(rv->object != NULL,
	    ("vm_reserv_remove: reserv %p is free", rv));
	KASSERT(!rv->inpartpopq,
	    ("vm_reserv_remove: reserv %p's inpartpopq is TRUE", rv));
	object = rv->object;
	vm_reserv_object_lock(object);
	LIST_REMOVE(rv, objq);
	rv->object = NULL;
	vm_reserv_object_unlock(object);
}

/*
 * Insert a new reservation into the object's objq.
 */
static void
vm_reserv_insert(vm_reserv_t rv, vm_object_t object, vm_pindex_t pindex)
{
	int i;

	KASSERT(rv->object == NULL,
	    ("vm_reserv_insert: reserv %p isn't free", rv));
	KASSERT(rv->popcnt == 0,
	    ("vm_reserv_insert: reserv %p's popcnt is corrupted", rv));
	KASSERT(!rv->inpartpopq,
	    ("vm_reserv_insert: reserv %p's inpartpopq is TRUE", rv));
	for (i = 0; i < NPOPMAP; i++)
		KASSERT(rv->popmap[i] == 0,
		    ("vm_reserv_insert: reserv %p's popmap is corrupted", rv));
	vm_reserv_object_lock(object);
	rv->pindex = pindex;
	rv->object = object;
	LIST_INSERT_HEAD(&object->rvq, rv, objq);
	vm_reserv_object_unlock(object);
}

375
376
377
/*
 * Reduces the given reservation's population count.  If the population count
 * becomes zero, the reservation is destroyed.  Additionally, moves the
378
 * reservation to the tail of the partially populated reservation queue if the
379
380
381
382
383
 * population count is non-zero.
 *
 * The free page queue lock must be held.
 */
static void
Alan Cox's avatar
Alan Cox committed
384
vm_reserv_depopulate(vm_reserv_t rv, int index)
385
386
{

387
	vm_domain_free_assert_locked(VM_DOMAIN(rv->domain));
388
389
	KASSERT(rv->object != NULL,
	    ("vm_reserv_depopulate: reserv %p is free", rv));
390
	KASSERT(popmap_is_set(rv->popmap, index),
391
392
	    ("vm_reserv_depopulate: reserv %p's popmap[%d] is clear", rv,
	    index));
393
394
	KASSERT(rv->popcnt > 0,
	    ("vm_reserv_depopulate: reserv %p's popcnt is corrupted", rv));
395
396
397
	KASSERT(rv->domain >= 0 && rv->domain < vm_ndomains,
	    ("vm_reserv_depopulate: reserv %p's domain is corrupted %d",
	    rv, rv->domain));
398
	if (rv->inpartpopq) {
399
		TAILQ_REMOVE(&vm_rvq_partpop[rv->domain], rv, partpopq);
400
		rv->inpartpopq = FALSE;
401
402
403
404
405
	} else {
		KASSERT(rv->pages->psind == 1,
		    ("vm_reserv_depopulate: reserv %p is already demoted",
		    rv));
		rv->pages->psind = 0;
406
	}
407
	popmap_clear(rv->popmap, index);
408
409
	rv->popcnt--;
	if (rv->popcnt == 0) {
410
		vm_reserv_remove(rv);
411
412
413
414
		vm_phys_free_pages(rv->pages, VM_LEVEL_0_ORDER);
		vm_reserv_freed++;
	} else {
		rv->inpartpopq = TRUE;
415
		TAILQ_INSERT_TAIL(&vm_rvq_partpop[rv->domain], rv, partpopq);
416
417
418
419
420
421
422
423
424
425
426
427
428
	}
}

/*
 * Returns the reservation to which the given page might belong.
 */
static __inline vm_reserv_t
vm_reserv_from_page(vm_page_t m)
{

	return (&vm_reserv_array[VM_PAGE_TO_PHYS(m) >> VM_LEVEL_0_SHIFT]);
}

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
/*
 * Returns an existing reservation or NULL and initialized successor pointer.
 */
static vm_reserv_t
vm_reserv_from_object(vm_object_t object, vm_pindex_t pindex,
    vm_page_t mpred, vm_page_t *msuccp)
{
	vm_reserv_t rv;
	vm_page_t msucc;

	msucc = NULL;
	if (mpred != NULL) {
		KASSERT(mpred->object == object,
		    ("vm_reserv_from_object: object doesn't contain mpred"));
		KASSERT(mpred->pindex < pindex,
		    ("vm_reserv_from_object: mpred doesn't precede pindex"));
		rv = vm_reserv_from_page(mpred);
		if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
			goto found;
		msucc = TAILQ_NEXT(mpred, listq);
	} else
		msucc = TAILQ_FIRST(&object->memq);
	if (msucc != NULL) {
		KASSERT(msucc->pindex > pindex,
		    ("vm_reserv_from_object: msucc doesn't succeed pindex"));
		rv = vm_reserv_from_page(msucc);
		if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
			goto found;
	}
	rv = NULL;

found:
	*msuccp = msucc;

	return (rv);
}

466
467
468
469
470
471
472
473
474
475
476
477
478
/*
 * Returns TRUE if the given reservation contains the given page index and
 * FALSE otherwise.
 */
static __inline boolean_t
vm_reserv_has_pindex(vm_reserv_t rv, vm_pindex_t pindex)
{

	return (((pindex - rv->pindex) & ~(VM_LEVEL_0_NPAGES - 1)) == 0);
}

/*
 * Increases the given reservation's population count.  Moves the reservation
479
 * to the tail of the partially populated reservation queue.
480
481
482
483
 *
 * The free page queue must be locked.
 */
static void
Alan Cox's avatar
Alan Cox committed
484
vm_reserv_populate(vm_reserv_t rv, int index)
485
486
{

487
	vm_domain_free_assert_locked(VM_DOMAIN(rv->domain));
488
489
	KASSERT(rv->object != NULL,
	    ("vm_reserv_populate: reserv %p is free", rv));
490
	KASSERT(popmap_is_clear(rv->popmap, index),
491
492
	    ("vm_reserv_populate: reserv %p's popmap[%d] is set", rv,
	    index));
493
494
	KASSERT(rv->popcnt < VM_LEVEL_0_NPAGES,
	    ("vm_reserv_populate: reserv %p is already full", rv));
495
496
	KASSERT(rv->pages->psind == 0,
	    ("vm_reserv_populate: reserv %p is already promoted", rv));
497
498
499
	KASSERT(rv->domain >= 0 && rv->domain < vm_ndomains,
	    ("vm_reserv_populate: reserv %p's domain is corrupted %d",
	    rv, rv->domain));
500
	if (rv->inpartpopq) {
501
		TAILQ_REMOVE(&vm_rvq_partpop[rv->domain], rv, partpopq);
502
503
		rv->inpartpopq = FALSE;
	}
504
	popmap_set(rv->popmap, index);
505
506
507
	rv->popcnt++;
	if (rv->popcnt < VM_LEVEL_0_NPAGES) {
		rv->inpartpopq = TRUE;
508
		TAILQ_INSERT_TAIL(&vm_rvq_partpop[rv->domain], rv, partpopq);
509
510
	} else
		rv->pages->psind = 1;
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
/*
 * Allocates a contiguous set of physical pages of the given size "npages"
 * from existing or newly created reservations.  All of the physical pages
 * must be at or above the given physical address "low" and below the given
 * physical address "high".  The given value "alignment" determines the
 * alignment of the first physical page in the set.  If the given value
 * "boundary" is non-zero, then the set of physical pages cannot cross any
 * physical address boundary that is a multiple of that value.  Both
 * "alignment" and "boundary" must be a power of two.
 *
 * The page "mpred" must immediately precede the offset "pindex" within the
 * specified object.
 *
 * The object and free page queue must be locked.
 */
vm_page_t
vm_reserv_extend_contig(int req, vm_object_t object, vm_pindex_t pindex,
    int domain, u_long npages, vm_paddr_t low, vm_paddr_t high,
    u_long alignment, vm_paddr_t boundary, vm_page_t mpred)
{
	struct vm_domain *vmd;
	vm_paddr_t pa, size;
	vm_page_t m, msucc;
	vm_reserv_t rv;
	int i, index;

	VM_OBJECT_ASSERT_WLOCKED(object);
	KASSERT(npages != 0, ("vm_reserv_alloc_contig: npages is 0"));

	/*
	 * Is a reservation fundamentally impossible?
	 */
	if (pindex < VM_RESERV_INDEX(object, pindex) ||
	    pindex + npages > object->size || object->resident_page_count == 0)
		return (NULL);

	/*
	 * All reservations of a particular size have the same alignment.
	 * Assuming that the first page is allocated from a reservation, the
	 * least significant bits of its physical address can be determined
	 * from its offset from the beginning of the reservation and the size
	 * of the reservation.
	 *
	 * Could the specified index within a reservation of the smallest
	 * possible size satisfy the alignment and boundary requirements?
	 */
	pa = VM_RESERV_INDEX(object, pindex) << PAGE_SHIFT;
	if ((pa & (alignment - 1)) != 0)
		return (NULL);
	size = npages << PAGE_SHIFT;
	if (((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0)
		return (NULL);

	/*
	 * Look for an existing reservation.
	 */
	rv = vm_reserv_from_object(object, pindex, mpred, &msucc);
	if (rv == NULL)
		return (NULL);
	KASSERT(object != kernel_object || rv->domain == domain,
	    ("vm_reserv_extend_contig: Domain mismatch from reservation."));
	index = VM_RESERV_INDEX(object, pindex);
	/* Does the allocation fit within the reservation? */
	if (index + npages > VM_LEVEL_0_NPAGES)
		return (NULL);
	domain = rv->domain;
	vmd = VM_DOMAIN(domain);
	vm_domain_free_lock(vmd);
	if (rv->object != object || !vm_domain_available(vmd, req, npages)) {
		m = NULL;
		goto out;
	}
	m = &rv->pages[index];
	pa = VM_PAGE_TO_PHYS(m);
	if (pa < low || pa + size > high || (pa & (alignment - 1)) != 0 ||
	    ((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0) {
		m = NULL;
		goto out;
	}
	/* Handle vm_page_rename(m, new_object, ...). */
	for (i = 0; i < npages; i++) {
		if (popmap_is_set(rv->popmap, index + i)) {
			m = NULL;
			goto out;
		}
	}
	for (i = 0; i < npages; i++)
		vm_reserv_populate(rv, index + i);
	vm_domain_freecnt_adj(vmd, -npages);
out:
	vm_domain_free_unlock(vmd);
	return (m);
}

607
/*
608
 * Allocates a contiguous set of physical pages of the given size "npages"
609
 * from existing or newly created reservations.  All of the physical pages
610
611
612
613
614
615
 * must be at or above the given physical address "low" and below the given
 * physical address "high".  The given value "alignment" determines the
 * alignment of the first physical page in the set.  If the given value
 * "boundary" is non-zero, then the set of physical pages cannot cross any
 * physical address boundary that is a multiple of that value.  Both
 * "alignment" and "boundary" must be a power of two.
616
 *
617
618
619
 * The page "mpred" must immediately precede the offset "pindex" within the
 * specified object.
 *
620
621
622
 * The object and free page queue must be locked.
 */
vm_page_t
623
624
625
vm_reserv_alloc_contig(vm_object_t object, vm_pindex_t pindex, int domain,
    u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
    vm_paddr_t boundary, vm_page_t mpred)
626
{
627
	vm_paddr_t pa, size;
628
	vm_page_t m, m_ret, msucc;
629
630
	vm_pindex_t first, leftcap, rightcap;
	vm_reserv_t rv;
631
632
	u_long allocpages, maxpages, minpages;
	int i, index, n;
633

634
	vm_domain_free_assert_locked(VM_DOMAIN(domain));
635
	VM_OBJECT_ASSERT_WLOCKED(object);
636
	KASSERT(npages != 0, ("vm_reserv_alloc_contig: npages is 0"));
637
638

	/*
639
	 * Is a reservation fundamentally impossible?
640
641
	 */
	if (pindex < VM_RESERV_INDEX(object, pindex) ||
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
	    pindex + npages > object->size)
		return (NULL);

	/*
	 * All reservations of a particular size have the same alignment.
	 * Assuming that the first page is allocated from a reservation, the
	 * least significant bits of its physical address can be determined
	 * from its offset from the beginning of the reservation and the size
	 * of the reservation.
	 *
	 * Could the specified index within a reservation of the smallest
	 * possible size satisfy the alignment and boundary requirements?
	 */
	pa = VM_RESERV_INDEX(object, pindex) << PAGE_SHIFT;
	if ((pa & (alignment - 1)) != 0)
		return (NULL);
	size = npages << PAGE_SHIFT;
	if (((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0)
660
661
662
		return (NULL);

	/*
663
664
665
	 * Callers should've extended an existing reservation prior to
	 * calling this function.  If a reservation exists it is
	 * incompatible with the allocation.
666
	 */
667
668
669
	rv = vm_reserv_from_object(object, pindex, mpred, &msucc);
	if (rv != NULL)
		return (NULL);
670
671

	/*
672
	 * Could at least one reservation fit between the first index to the
673
674
	 * left that can be used ("leftcap") and the first index to the right
	 * that cannot be used ("rightcap")?
675
676
677
678
	 *
	 * We must synchronize with the reserv object lock to protect the
	 * pindex/object of the resulting reservations against rename while
	 * we are inspecting.
679
680
	 */
	first = pindex - VM_RESERV_INDEX(object, pindex);
681
682
683
684
	minpages = VM_RESERV_INDEX(object, pindex) + npages;
	maxpages = roundup2(minpages, VM_LEVEL_0_NPAGES);
	allocpages = maxpages;
	vm_reserv_object_lock(object);
685
686
687
688
689
	if (mpred != NULL) {
		if ((rv = vm_reserv_from_page(mpred))->object != object)
			leftcap = mpred->pindex + 1;
		else
			leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
690
691
		if (leftcap > first) {
			vm_reserv_object_unlock(object);
692
			return (NULL);
693
		}
694
695
696
697
698
699
700
	}
	if (msucc != NULL) {
		if ((rv = vm_reserv_from_page(msucc))->object != object)
			rightcap = msucc->pindex;
		else
			rightcap = rv->pindex;
		if (first + maxpages > rightcap) {
701
702
			if (maxpages == VM_LEVEL_0_NPAGES) {
				vm_reserv_object_unlock(object);
703
				return (NULL);
704
			}
705
706
707
708
709
710
711

			/*
			 * At least one reservation will fit between "leftcap"
			 * and "rightcap".  However, a reservation for the
			 * last of the requested pages will not fit.  Reduce
			 * the size of the upcoming allocation accordingly.
			 */
712
713
714
			allocpages = minpages;
		}
	}
715
	vm_reserv_object_unlock(object);
716
717

	/*
718
	 * Would the last new reservation extend past the end of the object?
719
	 */
720
	if (first + maxpages > object->size) {
721
		/*
722
723
		 * Don't allocate the last new reservation if the object is a
		 * vnode or backed by another object that is a vnode. 
724
725
726
		 */
		if (object->type == OBJT_VNODE ||
		    (object->backing_object != NULL &&
727
728
729
730
731
		    object->backing_object->type == OBJT_VNODE)) {
			if (maxpages == VM_LEVEL_0_NPAGES)
				return (NULL);
			allocpages = minpages;
		}
732
733
734
735
		/* Speculate that the object may grow. */
	}

	/*
736
737
738
739
740
	 * Allocate the physical pages.  The alignment and boundary specified
	 * for this allocation may be different from the alignment and
	 * boundary specified for the requested pages.  For instance, the
	 * specified index may not be the first page within the first new
	 * reservation.
741
	 */
742
	m = vm_phys_alloc_contig(domain, allocpages, low, high, ulmax(alignment,
743
744
745
	    VM_LEVEL_0_SIZE), boundary > VM_LEVEL_0_SIZE ? boundary : 0);
	if (m == NULL)
		return (NULL);
746
	KASSERT(vm_phys_domain(m) == domain,
747
	    ("vm_reserv_alloc_contig: Page domain does not match requested."));
748
749
750
751
752
753
754

	/*
	 * The allocated physical pages always begin at a reservation
	 * boundary, but they do not always end at a reservation boundary.
	 * Initialize every reservation that is completely covered by the
	 * allocated physical pages.
	 */
755
756
757
	m_ret = NULL;
	index = VM_RESERV_INDEX(object, pindex);
	do {
758
759
		rv = vm_reserv_from_page(m);
		KASSERT(rv->pages == m,
760
		    ("vm_reserv_alloc_contig: reserv %p's pages is corrupted",
761
		    rv));
762
		vm_reserv_insert(rv, object, first);
763
764
		n = ulmin(VM_LEVEL_0_NPAGES - index, npages);
		for (i = 0; i < n; i++)
Alan Cox's avatar
Alan Cox committed
765
			vm_reserv_populate(rv, index + i);
766
767
768
769
770
771
772
773
		npages -= n;
		if (m_ret == NULL) {
			m_ret = &rv->pages[index];
			index = 0;
		}
		m += VM_LEVEL_0_NPAGES;
		first += VM_LEVEL_0_NPAGES;
		allocpages -= VM_LEVEL_0_NPAGES;
774
	} while (allocpages >= VM_LEVEL_0_NPAGES);
775
	return (m_ret);
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
}

/*
 * Attempts to extend an existing reservation and allocate the page to the
 * object.
 *
 * The page "mpred" must immediately precede the offset "pindex" within the
 * specified object.
 *
 * The object must be locked.
 */
vm_page_t
vm_reserv_extend(int req, vm_object_t object, vm_pindex_t pindex, int domain,
    vm_page_t mpred)
{
	struct vm_domain *vmd;
	vm_page_t m, msucc;
	vm_reserv_t rv;
	int index, free_count;

	VM_OBJECT_ASSERT_WLOCKED(object);
797
798

	/*
799
	 * Could a reservation currently exist?
800
	 */
801
802
	if (pindex < VM_RESERV_INDEX(object, pindex) ||
	    pindex >= object->size || object->resident_page_count == 0)
803
		return (NULL);
804
805
806
807
808
809

	/*
	 * Look for an existing reservation.
	 */
	rv = vm_reserv_from_object(object, pindex, mpred, &msucc);
	if (rv == NULL)
810
		return (NULL);
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834

	KASSERT(object != kernel_object || rv->domain == domain,
	    ("vm_reserv_extend: Domain mismatch from reservation."));
	domain = rv->domain;
	vmd = VM_DOMAIN(domain);
	index = VM_RESERV_INDEX(object, pindex);
	m = &rv->pages[index];
	vm_domain_free_lock(vmd);
	if (vm_domain_available(vmd, req, 1) == 0 ||
	    /* Handle reclaim race. */
	    rv->object != object ||
	    /* Handle vm_page_rename(m, new_object, ...). */
	    popmap_is_set(rv->popmap, index))
		m = NULL;
	if (m != NULL) {
		vm_reserv_populate(rv, index);
		free_count = vm_domain_freecnt_adj(vmd, -1);
	} else
		free_count = vmd->vmd_free_count;
	vm_domain_free_unlock(vmd);

	if (vm_paging_needed(vmd, free_count))
		pagedaemon_wakeup(domain);

835
836
837
838
	return (m);
}

/*
839
 * Allocates a page from an existing reservation.
840
 *
841
842
843
 * The page "mpred" must immediately precede the offset "pindex" within the
 * specified object.
 *
844
845
846
 * The object and free page queue must be locked.
 */
vm_page_t
847
848
vm_reserv_alloc_page(vm_object_t object, vm_pindex_t pindex, int domain,
    vm_page_t mpred)
849
{
850
	vm_page_t m, msucc;
851
852
	vm_pindex_t first, leftcap, rightcap;
	vm_reserv_t rv;
853
	int index;
854

855
	vm_domain_free_assert_locked(VM_DOMAIN(domain));
856
	VM_OBJECT_ASSERT_WLOCKED(object);
857
858
859
860
861
862
863
864
865

	/*
	 * Is a reservation fundamentally impossible?
	 */
	if (pindex < VM_RESERV_INDEX(object, pindex) ||
	    pindex >= object->size)
		return (NULL);

	/*
866
867
868
	 * Callers should've extended an existing reservation prior to
	 * calling this function.  If a reservation exists it is
	 * incompatible with the allocation.
869
	 */
870
871
872
	rv = vm_reserv_from_object(object, pindex, mpred, &msucc);
	if (rv != NULL)
		return (NULL);
873
874
875
876

	/*
	 * Could a reservation fit between the first index to the left that
	 * can be used and the first index to the right that cannot be used?
877
878
879
880
	 *
	 * We must synchronize with the reserv object lock to protect the
	 * pindex/object of the resulting reservations against rename while
	 * we are inspecting.
881
882
	 */
	first = pindex - VM_RESERV_INDEX(object, pindex);
883
	vm_reserv_object_lock(object);
884
885
886
887
888
	if (mpred != NULL) {
		if ((rv = vm_reserv_from_page(mpred))->object != object)
			leftcap = mpred->pindex + 1;
		else
			leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
889
890
		if (leftcap > first) {
			vm_reserv_object_unlock(object);
891
			return (NULL);
892
		}
893
894
895
896
897
898
	}
	if (msucc != NULL) {
		if ((rv = vm_reserv_from_page(msucc))->object != object)
			rightcap = msucc->pindex;
		else
			rightcap = rv->pindex;
899
900
		if (first + VM_LEVEL_0_NPAGES > rightcap) {
			vm_reserv_object_unlock(object);
901
			return (NULL);
902
		}
903
	}
904
	vm_reserv_object_unlock(object);
905
906
907
908
909
910
911
912
913
914
915
916
917
918

	/*
	 * Would a new reservation extend past the end of the object? 
	 */
	if (first + VM_LEVEL_0_NPAGES > object->size) {
		/*
		 * Don't allocate a new reservation if the object is a vnode or
		 * backed by another object that is a vnode. 
		 */
		if (object->type == OBJT_VNODE ||
		    (object->backing_object != NULL &&
		    object->backing_object->type == OBJT_VNODE))
			return (NULL);
		/* Speculate that the object may grow. */
919
	}
920
921
922
923

	/*
	 * Allocate and populate the new reservation.
	 */
924
	m = vm_phys_alloc_pages(domain, VM_FREEPOOL_DEFAULT, VM_LEVEL_0_ORDER);
925
926
927
928
929
	if (m == NULL)
		return (NULL);
	rv = vm_reserv_from_page(m);
	KASSERT(rv->pages == m,
	    ("vm_reserv_alloc_page: reserv %p's pages is corrupted", rv));
930
	vm_reserv_insert(rv, object, first);
Alan Cox's avatar
Alan Cox committed
931
932
933
	index = VM_RESERV_INDEX(object, pindex);
	vm_reserv_populate(rv, index);
	return (&rv->pages[index]);
934
935
}

Alan Cox's avatar
Alan Cox committed
936
/*
937
938
939
940
 * Breaks the given reservation.  Except for the specified free page, all free
 * pages in the reservation are returned to the physical memory allocator.
 * The reservation's population count and map are reset to their initial
 * state.
Alan Cox's avatar
Alan Cox committed
941
 *
942
 * The given reservation must not be in the partially populated reservation
Alan Cox's avatar
Alan Cox committed
943
944
945
946
947
948
949
 * queue.  The free page queue lock must be held.
 */
static void
vm_reserv_break(vm_reserv_t rv, vm_page_t m)
{
	int begin_zeroes, hi, i, lo;

950
951
	vm_domain_free_assert_locked(VM_DOMAIN(rv->domain));
	vm_reserv_remove(rv);
Alan Cox's avatar
Alan Cox committed
952
953
954
955
956
957
958
	if (m != NULL) {
		/*
		 * Since the reservation is being broken, there is no harm in
		 * abusing the population map to stop "m" from being returned
		 * to the physical memory allocator.
		 */
		i = m - rv->pages;
959
		KASSERT(popmap_is_clear(rv->popmap, i),
Alan Cox's avatar
Alan Cox committed
960
		    ("vm_reserv_break: reserv %p's popmap is corrupted", rv));
961
		popmap_set(rv->popmap, i);
Alan Cox's avatar
Alan Cox committed
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
		rv->popcnt++;
	}
	i = hi = 0;
	do {
		/* Find the next 0 bit.  Any previous 0 bits are < "hi". */
		lo = ffsl(~(((1UL << hi) - 1) | rv->popmap[i]));
		if (lo == 0) {
			/* Redundantly clears bits < "hi". */
			rv->popmap[i] = 0;
			rv->popcnt -= NBPOPMAP - hi;
			while (++i < NPOPMAP) {
				lo = ffsl(~rv->popmap[i]);
				if (lo == 0) {
					rv->popmap[i] = 0;
					rv->popcnt -= NBPOPMAP;
				} else
					break;
			}
			if (i == NPOPMAP)
				break;
			hi = 0;
		}
		KASSERT(lo > 0, ("vm_reserv_break: lo is %d", lo));
		/* Convert from ffsl() to ordinary bit numbering. */
		lo--;
		if (lo > 0) {
			/* Redundantly clears bits < "hi". */
			rv->popmap[i] &= ~((1UL << lo) - 1);
			rv->popcnt -= lo - hi;
		}
		begin_zeroes = NBPOPMAP * i + lo;
		/* Find the next 1 bit. */
		do
			hi = ffsl(rv->popmap[i]);
		while (hi == 0 && ++i < NPOPMAP);
		if (i != NPOPMAP)
			/* Convert from ffsl() to ordinary bit numbering. */
			hi--;
		vm_phys_free_contig(&rv->pages[begin_zeroes], NBPOPMAP * i +
For faster browsing, not all history is shown. View entire blame