xref: /freebsd/sys/vm/vm_page.h (revision 4cf49a43559ed9fdad601bdcccd2c55963008675)
1 /*
2  * Copyright (c) 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * The Mach Operating System project at Carnegie-Mellon University.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	from: @(#)vm_page.h	8.2 (Berkeley) 12/13/93
37  *
38  *
39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40  * All rights reserved.
41  *
42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43  *
44  * Permission to use, copy, modify and distribute this software and
45  * its documentation is hereby granted, provided that both the copyright
46  * notice and this permission notice appear in all copies of the
47  * software, derivative works or modified versions, and any portions
48  * thereof, and that both notices appear in supporting documentation.
49  *
50  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53  *
54  * Carnegie Mellon requests users of this software to return to
55  *
56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
57  *  School of Computer Science
58  *  Carnegie Mellon University
59  *  Pittsburgh PA 15213-3890
60  *
61  * any improvements or extensions that they make and grant Carnegie the
62  * rights to redistribute these changes.
63  *
64  * $FreeBSD$
65  */
66 
67 /*
68  *	Resident memory system definitions.
69  */
70 
71 #ifndef	_VM_PAGE_
72 #define	_VM_PAGE_
73 
74 #if !defined(KLD_MODULE)
75 #include "opt_vmpage.h"
76 #endif
77 
78 #include <vm/pmap.h>
79 #include <machine/atomic.h>
80 
81 /*
82  *	Management of resident (logical) pages.
83  *
84  *	A small structure is kept for each resident
85  *	page, indexed by page number.  Each structure
86  *	is an element of several lists:
87  *
88  *		A hash table bucket used to quickly
89  *		perform object/offset lookups
90  *
91  *		A list of all pages for a given object,
92  *		so they can be quickly deactivated at
93  *		time of deallocation.
94  *
95  *		An ordered list of pages due for pageout.
96  *
97  *	In addition, the structure contains the object
98  *	and offset to which this page belongs (for pageout),
99  *	and sundry status bits.
100  *
101  *	Fields in this structure are locked either by the lock on the
102  *	object that the page belongs to (O) or by the lock on the page
103  *	queues (P).
104  *
105  *	The 'valid' and 'dirty' fields are distinct.  A page may have dirty
106  *	bits set without having associated valid bits set.  This is used by
107  *	NFS to implement piecemeal writes.
108  */
109 
110 TAILQ_HEAD(pglist, vm_page);
111 
112 struct vm_page {
113 	TAILQ_ENTRY(vm_page) pageq;	/* queue info for FIFO queue or free list (P) */
114 	struct vm_page	*hnext;		/* hash table link (O,P)	*/
115 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) 	*/
116 
117 	vm_object_t object;		/* which object am I in (O,P)*/
118 	vm_pindex_t pindex;		/* offset into object (O,P) */
119 	vm_offset_t phys_addr;		/* physical address of page */
120 	u_short	queue;			/* page queue index */
121 	u_short	flags,			/* see below */
122 		pc;			/* page color */
123 	u_short wire_count;		/* wired down maps refs (P) */
124 	short hold_count;		/* page hold count */
125 	u_char	act_count;		/* page usage count */
126 	u_char	busy;			/* page busy count */
127 	/* NOTE that these must support one bit per DEV_BSIZE in a page!!! */
128 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
129 #if PAGE_SIZE == 4096
130 	u_char	valid;			/* map of valid DEV_BSIZE chunks */
131 	u_char	dirty;			/* map of dirty DEV_BSIZE chunks */
132 #elif PAGE_SIZE == 8192
133 	u_short	valid;			/* map of valid DEV_BSIZE chunks */
134 	u_short	dirty;			/* map of dirty DEV_BSIZE chunks */
135 #endif
136 };
137 
138 /*
139  * note: currently use SWAPBLK_NONE as an absolute value rather then
140  * a flag bit.
141  */
142 
143 #define SWAPBLK_MASK	((daddr_t)((u_daddr_t)-1 >> 1))		/* mask */
144 #define SWAPBLK_NONE	((daddr_t)((u_daddr_t)SWAPBLK_MASK + 1))/* flag */
145 
146 #if !defined(KLD_MODULE)
147 
148 /*
149  * Page coloring parameters
150  */
151 /* Each of PQ_FREE, and PQ_CACHE have PQ_HASH_SIZE entries */
152 
153 /* Define one of the following */
154 #if defined(PQ_HUGECACHE)
155 #define PQ_PRIME1 31	/* Prime number somewhat less than PQ_HASH_SIZE */
156 #define PQ_PRIME2 23	/* Prime number somewhat less than PQ_HASH_SIZE */
157 #define PQ_L2_SIZE 256	/* A number of colors opt for 1M cache */
158 #endif
159 
160 /* Define one of the following */
161 #if defined(PQ_LARGECACHE)
162 #define PQ_PRIME1 31	/* Prime number somewhat less than PQ_HASH_SIZE */
163 #define PQ_PRIME2 23	/* Prime number somewhat less than PQ_HASH_SIZE */
164 #define PQ_L2_SIZE 128	/* A number of colors opt for 512K cache */
165 #endif
166 
167 
168 /*
169  * Use 'options PQ_NOOPT' to disable page coloring
170  */
171 #if defined(PQ_NOOPT)
172 #define PQ_PRIME1 1
173 #define PQ_PRIME2 1
174 #define PQ_L2_SIZE 1
175 #endif
176 
177 #if defined(PQ_NORMALCACHE)
178 #define PQ_PRIME1 5	/* Prime number somewhat less than PQ_HASH_SIZE */
179 #define PQ_PRIME2 3	/* Prime number somewhat less than PQ_HASH_SIZE */
180 #define PQ_L2_SIZE 16	/* A reasonable number of colors (opt for 64K cache) */
181 #endif
182 
183 #if defined(PQ_MEDIUMCACHE)
184 #define PQ_PRIME1 13	/* Prime number somewhat less than PQ_HASH_SIZE */
185 #define PQ_PRIME2 7	/* Prime number somewhat less than PQ_HASH_SIZE */
186 #define PQ_L2_SIZE 64	/* A number of colors opt for 256K cache */
187 #endif
188 
189 #if !defined(PQ_L2_SIZE)
190 #define PQ_PRIME1 9	/* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */
191 #define PQ_PRIME2 5	/* Prime number somewhat less than PQ_HASH_SIZE */
192 #define PQ_L2_SIZE 32	/* 512KB or smaller, 4-way set-associative cache */
193 #endif
194 
195 #define PQ_L2_MASK (PQ_L2_SIZE - 1)
196 
197 #if 1
198 #define PQ_NONE 0
199 #define PQ_FREE	1
200 #define PQ_INACTIVE (1 + 1*PQ_L2_SIZE)
201 #define PQ_ACTIVE (2 + 1*PQ_L2_SIZE)
202 #define PQ_CACHE (3 + 1*PQ_L2_SIZE)
203 #define PQ_COUNT (3 + 2*PQ_L2_SIZE)
204 #else
205 #define PQ_NONE		PQ_COUNT
206 #define PQ_FREE		0
207 #define PQ_INACTIVE	PQ_L2_SIZE
208 #define PQ_ACTIVE	(1 +   PQ_L2_SIZE)
209 #define PQ_CACHE	(2 +   PQ_L2_SIZE)
210 #define PQ_COUNT	(2 + 2*PQ_L2_SIZE)
211 #endif
212 
213 extern struct vpgqueues {
214 	struct pglist *pl;
215 	int	*cnt;
216 	int	lcnt;
217 } vm_page_queues[PQ_COUNT];
218 
219 #endif
220 
221 /*
222  * These are the flags defined for vm_page.
223  *
224  * Note: PG_FILLED and PG_DIRTY are added for the filesystems.
225  */
226 #define	PG_BUSY		0x0001		/* page is in transit (O) */
227 #define	PG_WANTED	0x0002		/* someone is waiting for page (O) */
228 #define	PG_FICTITIOUS	0x0008		/* physical page doesn't exist (O) */
229 #define	PG_WRITEABLE	0x0010		/* page is mapped writeable */
230 #define PG_MAPPED	0x0020		/* page is mapped */
231 #define	PG_ZERO		0x0040		/* page is zeroed */
232 #define PG_REFERENCED	0x0080		/* page has been referenced */
233 #define PG_CLEANCHK	0x0100		/* page will be checked for cleaning */
234 #define PG_SWAPINPROG	0x0200		/* swap I/O in progress on page	     */
235 
236 /*
237  * Misc constants.
238  */
239 
240 #define ACT_DECLINE		1
241 #define ACT_ADVANCE		3
242 #define ACT_INIT		5
243 #define ACT_MAX			64
244 #define PFCLUSTER_BEHIND	3
245 #define PFCLUSTER_AHEAD		3
246 
247 #ifdef KERNEL
248 /*
249  * Each pageable resident page falls into one of four lists:
250  *
251  *	free
252  *		Available for allocation now.
253  *
254  * The following are all LRU sorted:
255  *
256  *	cache
257  *		Almost available for allocation. Still in an
258  *		object, but clean and immediately freeable at
259  *		non-interrupt times.
260  *
261  *	inactive
262  *		Low activity, candidates for reclamation.
263  *		This is the list of pages that should be
264  *		paged out next.
265  *
266  *	active
267  *		Pages that are "active" i.e. they have been
268  *		recently referenced.
269  *
270  *	zero
271  *		Pages that are really free and have been pre-zeroed
272  *
273  */
274 
275 #if !defined(KLD_MODULE)
276 
277 extern struct pglist vm_page_queue_free[PQ_L2_SIZE];/* memory free queue */
278 extern struct pglist vm_page_queue_active;	/* active memory queue */
279 extern struct pglist vm_page_queue_inactive;	/* inactive memory queue */
280 extern struct pglist vm_page_queue_cache[PQ_L2_SIZE];/* cache memory queue */
281 
282 #endif
283 
284 extern int vm_page_zero_count;
285 
286 extern vm_page_t vm_page_array;		/* First resident page in table */
287 extern long first_page;			/* first physical page number */
288 
289 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
290 
291 #define PHYS_TO_VM_PAGE(pa) \
292 		(&vm_page_array[atop(pa) - first_page ])
293 
294 /*
295  *	Functions implemented as macros
296  */
297 
298 static __inline void
299 vm_page_flag_set(vm_page_t m, unsigned int bits)
300 {
301 	atomic_set_short(&(m)->flags, bits);
302 }
303 
304 static __inline void
305 vm_page_flag_clear(vm_page_t m, unsigned int bits)
306 {
307 	atomic_clear_short(&(m)->flags, bits);
308 }
309 
310 #if 0
311 static __inline void
312 vm_page_assert_wait(vm_page_t m, int interruptible)
313 {
314 	vm_page_flag_set(m, PG_WANTED);
315 	assert_wait((int) m, interruptible);
316 }
317 #endif
318 
319 static __inline void
320 vm_page_busy(vm_page_t m)
321 {
322 	KASSERT((m->flags & PG_BUSY) == 0, ("vm_page_busy: page already busy!!!"));
323 	vm_page_flag_set(m, PG_BUSY);
324 }
325 
326 /*
327  *	vm_page_flash:
328  *
329  *	wakeup anyone waiting for the page.
330  */
331 
332 static __inline void
333 vm_page_flash(vm_page_t m)
334 {
335 	if (m->flags & PG_WANTED) {
336 		vm_page_flag_clear(m, PG_WANTED);
337 		wakeup(m);
338 	}
339 }
340 
341 /*
342  *	vm_page_wakeup:
343  *
344  *	clear the PG_BUSY flag and wakeup anyone waiting for the
345  *	page.
346  *
347  */
348 
349 static __inline void
350 vm_page_wakeup(vm_page_t m)
351 {
352 	KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
353 	vm_page_flag_clear(m, PG_BUSY);
354 	vm_page_flash(m);
355 }
356 
357 /*
358  *
359  *
360  */
361 
362 static __inline void
363 vm_page_io_start(vm_page_t m)
364 {
365 	atomic_add_char(&(m)->busy, 1);
366 }
367 
368 static __inline void
369 vm_page_io_finish(vm_page_t m)
370 {
371 	atomic_subtract_char(&m->busy, 1);
372 	if (m->busy == 0)
373 		vm_page_flash(m);
374 }
375 
376 
377 #if PAGE_SIZE == 4096
378 #define VM_PAGE_BITS_ALL 0xff
379 #endif
380 
381 #if PAGE_SIZE == 8192
382 #define VM_PAGE_BITS_ALL 0xffff
383 #endif
384 
385 #define VM_ALLOC_NORMAL		0
386 #define VM_ALLOC_INTERRUPT	1
387 #define VM_ALLOC_SYSTEM		2
388 #define	VM_ALLOC_ZERO		3
389 #define	VM_ALLOC_RETRY		0x80
390 
391 void vm_page_activate __P((vm_page_t));
392 vm_page_t vm_page_alloc __P((vm_object_t, vm_pindex_t, int));
393 vm_page_t vm_page_grab __P((vm_object_t, vm_pindex_t, int));
394 void vm_page_cache __P((register vm_page_t));
395 void vm_page_dontneed __P((register vm_page_t));
396 static __inline void vm_page_copy __P((vm_page_t, vm_page_t));
397 static __inline void vm_page_free __P((vm_page_t));
398 static __inline void vm_page_free_zero __P((vm_page_t));
399 void vm_page_deactivate __P((vm_page_t));
400 void vm_page_insert __P((vm_page_t, vm_object_t, vm_pindex_t));
401 vm_page_t vm_page_lookup __P((vm_object_t, vm_pindex_t));
402 void vm_page_remove __P((vm_page_t));
403 void vm_page_rename __P((vm_page_t, vm_object_t, vm_pindex_t));
404 vm_offset_t vm_page_startup __P((vm_offset_t, vm_offset_t, vm_offset_t));
405 void vm_page_unwire __P((vm_page_t, int));
406 void vm_page_wire __P((vm_page_t));
407 void vm_page_unqueue __P((vm_page_t));
408 void vm_page_unqueue_nowakeup __P((vm_page_t));
409 void vm_page_set_validclean __P((vm_page_t, int, int));
410 void vm_page_set_dirty __P((vm_page_t, int, int));
411 void vm_page_clear_dirty __P((vm_page_t, int, int));
412 void vm_page_set_invalid __P((vm_page_t, int, int));
413 static __inline boolean_t vm_page_zero_fill __P((vm_page_t));
414 int vm_page_is_valid __P((vm_page_t, int, int));
415 void vm_page_test_dirty __P((vm_page_t));
416 int vm_page_bits __P((int, int));
417 vm_page_t _vm_page_list_find __P((int, int));
418 #if 0
419 int vm_page_sleep(vm_page_t m, char *msg, char *busy);
420 int vm_page_asleep(vm_page_t m, char *msg, char *busy);
421 #endif
422 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
423 void vm_page_free_toq(vm_page_t m);
424 
425 /*
426  * Keep page from being freed by the page daemon
427  * much of the same effect as wiring, except much lower
428  * overhead and should be used only for *very* temporary
429  * holding ("wiring").
430  */
431 static __inline void
432 vm_page_hold(vm_page_t mem)
433 {
434 	mem->hold_count++;
435 }
436 
437 static __inline void
438 vm_page_unhold(vm_page_t mem)
439 {
440 	--mem->hold_count;
441 	KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!"));
442 }
443 
444 /*
445  * 	vm_page_protect:
446  *
447  *	Reduce the protection of a page.  This routine never
448  *	raises the protection and therefore can be safely
449  *	called if the page is already at VM_PROT_NONE ( it
450  *	will be a NOP effectively ).
451  */
452 
453 static __inline void
454 vm_page_protect(vm_page_t mem, int prot)
455 {
456 	if (prot == VM_PROT_NONE) {
457 		if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) {
458 			pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_NONE);
459 			vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED);
460 		}
461 	} else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) {
462 		pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_READ);
463 		vm_page_flag_clear(mem, PG_WRITEABLE);
464 	}
465 }
466 
467 /*
468  *	vm_page_zero_fill:
469  *
470  *	Zero-fill the specified page.
471  *	Written as a standard pagein routine, to
472  *	be used by the zero-fill object.
473  */
474 static __inline boolean_t
475 vm_page_zero_fill(m)
476 	vm_page_t m;
477 {
478 	pmap_zero_page(VM_PAGE_TO_PHYS(m));
479 	return (TRUE);
480 }
481 
482 /*
483  *	vm_page_copy:
484  *
485  *	Copy one page to another
486  */
487 static __inline void
488 vm_page_copy(src_m, dest_m)
489 	vm_page_t src_m;
490 	vm_page_t dest_m;
491 {
492 	pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m));
493 	dest_m->valid = VM_PAGE_BITS_ALL;
494 }
495 
496 /*
497  *	vm_page_free:
498  *
499  *	Free a page
500  *
501  *	The clearing of PG_ZERO is a temporary safety until the code can be
502  *	reviewed to determine that PG_ZERO is being properly cleared on
503  *	write faults or maps.  PG_ZERO was previously cleared in
504  *	vm_page_alloc().
505  */
506 static __inline void
507 vm_page_free(m)
508 	vm_page_t m;
509 {
510 	vm_page_flag_clear(m, PG_ZERO);
511 	vm_page_free_toq(m);
512 }
513 
514 /*
515  *	vm_page_free_zero:
516  *
517  *	Free a page to the zerod-pages queue
518  */
519 static __inline void
520 vm_page_free_zero(m)
521 	vm_page_t m;
522 {
523 	vm_page_flag_set(m, PG_ZERO);
524 	vm_page_free_toq(m);
525 }
526 
527 /*
528  *	vm_page_sleep_busy:
529  *
530  *	Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
531  *	m->busy is zero.  Returns TRUE if it had to sleep ( including if
532  *	it almost had to sleep and made temporary spl*() mods), FALSE
533  *	otherwise.
534  *
535  *	This routine assumes that interrupts can only remove the busy
536  *	status from a page, not set the busy status or change it from
537  *	PG_BUSY to m->busy or vise versa (which would create a timing
538  *	window).
539  *
540  *	Note that being an inline, this code will be well optimized.
541  */
542 
543 static __inline int
544 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
545 {
546 	if ((m->flags & PG_BUSY) || (also_m_busy && m->busy))  {
547 		int s = splvm();
548 		if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
549 			/*
550 			 * Page is busy. Wait and retry.
551 			 */
552 			vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
553 			tsleep(m, PVM, msg, 0);
554 		}
555 		splx(s);
556 		return(TRUE);
557 		/* not reached */
558 	}
559 	return(FALSE);
560 }
561 
562 /*
563  *	vm_page_dirty:
564  *
565  *	make page all dirty
566  */
567 
568 static __inline void
569 vm_page_dirty(vm_page_t m)
570 {
571 #if !defined(KLD_MODULE)
572 	KASSERT(m->queue - m->pc != PQ_CACHE, ("vm_page_dirty: page in cache!"));
573 #endif
574 	m->dirty = VM_PAGE_BITS_ALL;
575 }
576 
577 /*
578  *	vm_page_undirty:
579  *
580  *	Set page to not be dirty.  Note: does not clear pmap modify bits
581  */
582 
583 static __inline void
584 vm_page_undirty(vm_page_t m)
585 {
586 	m->dirty = 0;
587 }
588 
589 #if !defined(KLD_MODULE)
590 
591 static __inline vm_page_t
592 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
593 {
594 	vm_page_t m;
595 
596 #if PQ_L2_SIZE > 1
597 	if (prefer_zero) {
598 		m = TAILQ_LAST(vm_page_queues[basequeue+index].pl, pglist);
599 	} else {
600 		m = TAILQ_FIRST(vm_page_queues[basequeue+index].pl);
601 	}
602 	if (m == NULL)
603 		m = _vm_page_list_find(basequeue, index);
604 #else
605 	if (prefer_zero) {
606 		m = TAILQ_LAST(vm_page_queues[basequeue].pl, pglist);
607 	} else {
608 		m = TAILQ_FIRST(vm_page_queues[basequeue].pl);
609 	}
610 #endif
611 	return(m);
612 }
613 
614 #endif
615 
616 #endif				/* KERNEL */
617 #endif				/* !_VM_PAGE_ */
618