xref: /freebsd/sys/vm/vm_page.h (revision daf1cffce2e07931f27c6c6998652e90df6ba87e)
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 struct vpgqueues {
214 	struct pglist pl;
215 	int	*cnt;
216 	int	lcnt;
217 };
218 
219 extern struct vpgqueues vm_page_queues[PQ_COUNT];
220 
221 #endif
222 
223 /*
224  * These are the flags defined for vm_page.
225  *
226  * Note: PG_FILLED and PG_DIRTY are added for the filesystems.
227  */
228 #define	PG_BUSY		0x0001		/* page is in transit (O) */
229 #define	PG_WANTED	0x0002		/* someone is waiting for page (O) */
230 #define	PG_FICTITIOUS	0x0008		/* physical page doesn't exist (O) */
231 #define	PG_WRITEABLE	0x0010		/* page is mapped writeable */
232 #define PG_MAPPED	0x0020		/* page is mapped */
233 #define	PG_ZERO		0x0040		/* page is zeroed */
234 #define PG_REFERENCED	0x0080		/* page has been referenced */
235 #define PG_CLEANCHK	0x0100		/* page will be checked for cleaning */
236 #define PG_SWAPINPROG	0x0200		/* swap I/O in progress on page	     */
237 #define PG_NOSYNC	0x0400		/* do not collect for syncer */
238 
239 /*
240  * Misc constants.
241  */
242 
243 #define ACT_DECLINE		1
244 #define ACT_ADVANCE		3
245 #define ACT_INIT		5
246 #define ACT_MAX			64
247 #define PFCLUSTER_BEHIND	3
248 #define PFCLUSTER_AHEAD		3
249 
250 #ifdef _KERNEL
251 /*
252  * Each pageable resident page falls into one of four lists:
253  *
254  *	free
255  *		Available for allocation now.
256  *
257  * The following are all LRU sorted:
258  *
259  *	cache
260  *		Almost available for allocation. Still in an
261  *		object, but clean and immediately freeable at
262  *		non-interrupt times.
263  *
264  *	inactive
265  *		Low activity, candidates for reclamation.
266  *		This is the list of pages that should be
267  *		paged out next.
268  *
269  *	active
270  *		Pages that are "active" i.e. they have been
271  *		recently referenced.
272  *
273  *	zero
274  *		Pages that are really free and have been pre-zeroed
275  *
276  */
277 
278 extern int vm_page_zero_count;
279 
280 extern vm_page_t vm_page_array;		/* First resident page in table */
281 extern long first_page;			/* first physical page number */
282 
283 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
284 
285 #define PHYS_TO_VM_PAGE(pa) \
286 		(&vm_page_array[atop(pa) - first_page ])
287 
288 /*
289  *	Functions implemented as macros
290  */
291 
292 static __inline void
293 vm_page_flag_set(vm_page_t m, unsigned int bits)
294 {
295 	atomic_set_short(&(m)->flags, bits);
296 }
297 
298 static __inline void
299 vm_page_flag_clear(vm_page_t m, unsigned int bits)
300 {
301 	atomic_clear_short(&(m)->flags, bits);
302 }
303 
304 #if 0
305 static __inline void
306 vm_page_assert_wait(vm_page_t m, int interruptible)
307 {
308 	vm_page_flag_set(m, PG_WANTED);
309 	assert_wait((int) m, interruptible);
310 }
311 #endif
312 
313 static __inline void
314 vm_page_busy(vm_page_t m)
315 {
316 	KASSERT((m->flags & PG_BUSY) == 0, ("vm_page_busy: page already busy!!!"));
317 	vm_page_flag_set(m, PG_BUSY);
318 }
319 
320 /*
321  *	vm_page_flash:
322  *
323  *	wakeup anyone waiting for the page.
324  */
325 
326 static __inline void
327 vm_page_flash(vm_page_t m)
328 {
329 	if (m->flags & PG_WANTED) {
330 		vm_page_flag_clear(m, PG_WANTED);
331 		wakeup(m);
332 	}
333 }
334 
335 /*
336  *	vm_page_wakeup:
337  *
338  *	clear the PG_BUSY flag and wakeup anyone waiting for the
339  *	page.
340  *
341  */
342 
343 static __inline void
344 vm_page_wakeup(vm_page_t m)
345 {
346 	KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
347 	vm_page_flag_clear(m, PG_BUSY);
348 	vm_page_flash(m);
349 }
350 
351 /*
352  *
353  *
354  */
355 
356 static __inline void
357 vm_page_io_start(vm_page_t m)
358 {
359 	atomic_add_char(&(m)->busy, 1);
360 }
361 
362 static __inline void
363 vm_page_io_finish(vm_page_t m)
364 {
365 	atomic_subtract_char(&m->busy, 1);
366 	if (m->busy == 0)
367 		vm_page_flash(m);
368 }
369 
370 
371 #if PAGE_SIZE == 4096
372 #define VM_PAGE_BITS_ALL 0xff
373 #endif
374 
375 #if PAGE_SIZE == 8192
376 #define VM_PAGE_BITS_ALL 0xffff
377 #endif
378 
379 #define VM_ALLOC_NORMAL		0
380 #define VM_ALLOC_INTERRUPT	1
381 #define VM_ALLOC_SYSTEM		2
382 #define	VM_ALLOC_ZERO		3
383 #define	VM_ALLOC_RETRY		0x80
384 
385 void vm_page_activate __P((vm_page_t));
386 vm_page_t vm_page_alloc __P((vm_object_t, vm_pindex_t, int));
387 vm_page_t vm_page_grab __P((vm_object_t, vm_pindex_t, int));
388 void vm_page_cache __P((register vm_page_t));
389 void vm_page_dontneed __P((register vm_page_t));
390 static __inline void vm_page_copy __P((vm_page_t, vm_page_t));
391 static __inline void vm_page_free __P((vm_page_t));
392 static __inline void vm_page_free_zero __P((vm_page_t));
393 void vm_page_deactivate __P((vm_page_t));
394 void vm_page_insert __P((vm_page_t, vm_object_t, vm_pindex_t));
395 vm_page_t vm_page_lookup __P((vm_object_t, vm_pindex_t));
396 void vm_page_remove __P((vm_page_t));
397 void vm_page_rename __P((vm_page_t, vm_object_t, vm_pindex_t));
398 vm_offset_t vm_page_startup __P((vm_offset_t, vm_offset_t, vm_offset_t));
399 void vm_page_unwire __P((vm_page_t, int));
400 void vm_page_wire __P((vm_page_t));
401 void vm_page_unqueue __P((vm_page_t));
402 void vm_page_unqueue_nowakeup __P((vm_page_t));
403 void vm_page_set_validclean __P((vm_page_t, int, int));
404 void vm_page_set_dirty __P((vm_page_t, int, int));
405 void vm_page_clear_dirty __P((vm_page_t, int, int));
406 void vm_page_set_invalid __P((vm_page_t, int, int));
407 static __inline boolean_t vm_page_zero_fill __P((vm_page_t));
408 int vm_page_is_valid __P((vm_page_t, int, int));
409 void vm_page_test_dirty __P((vm_page_t));
410 int vm_page_bits __P((int, int));
411 vm_page_t _vm_page_list_find __P((int, int));
412 #if 0
413 int vm_page_sleep(vm_page_t m, char *msg, char *busy);
414 int vm_page_asleep(vm_page_t m, char *msg, char *busy);
415 #endif
416 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
417 void vm_page_free_toq(vm_page_t m);
418 
419 /*
420  * Keep page from being freed by the page daemon
421  * much of the same effect as wiring, except much lower
422  * overhead and should be used only for *very* temporary
423  * holding ("wiring").
424  */
425 static __inline void
426 vm_page_hold(vm_page_t mem)
427 {
428 	mem->hold_count++;
429 }
430 
431 static __inline void
432 vm_page_unhold(vm_page_t mem)
433 {
434 	--mem->hold_count;
435 	KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!"));
436 }
437 
438 /*
439  * 	vm_page_protect:
440  *
441  *	Reduce the protection of a page.  This routine never raises the
442  *	protection and therefore can be safely called if the page is already
443  *	at VM_PROT_NONE (it will be a NOP effectively ).
444  */
445 
446 static __inline void
447 vm_page_protect(vm_page_t mem, int prot)
448 {
449 	if (prot == VM_PROT_NONE) {
450 		if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) {
451 			pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_NONE);
452 			vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED);
453 		}
454 	} else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) {
455 		pmap_page_protect(VM_PAGE_TO_PHYS(mem), VM_PROT_READ);
456 		vm_page_flag_clear(mem, PG_WRITEABLE);
457 	}
458 }
459 
460 /*
461  *	vm_page_zero_fill:
462  *
463  *	Zero-fill the specified page.
464  *	Written as a standard pagein routine, to
465  *	be used by the zero-fill object.
466  */
467 static __inline boolean_t
468 vm_page_zero_fill(m)
469 	vm_page_t m;
470 {
471 	pmap_zero_page(VM_PAGE_TO_PHYS(m));
472 	return (TRUE);
473 }
474 
475 /*
476  *	vm_page_copy:
477  *
478  *	Copy one page to another
479  */
480 static __inline void
481 vm_page_copy(src_m, dest_m)
482 	vm_page_t src_m;
483 	vm_page_t dest_m;
484 {
485 	pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m));
486 	dest_m->valid = VM_PAGE_BITS_ALL;
487 }
488 
489 /*
490  *	vm_page_free:
491  *
492  *	Free a page
493  *
494  *	The clearing of PG_ZERO is a temporary safety until the code can be
495  *	reviewed to determine that PG_ZERO is being properly cleared on
496  *	write faults or maps.  PG_ZERO was previously cleared in
497  *	vm_page_alloc().
498  */
499 static __inline void
500 vm_page_free(m)
501 	vm_page_t m;
502 {
503 	vm_page_flag_clear(m, PG_ZERO);
504 	vm_page_free_toq(m);
505 }
506 
507 /*
508  *	vm_page_free_zero:
509  *
510  *	Free a page to the zerod-pages queue
511  */
512 static __inline void
513 vm_page_free_zero(m)
514 	vm_page_t m;
515 {
516 	vm_page_flag_set(m, PG_ZERO);
517 	vm_page_free_toq(m);
518 }
519 
520 /*
521  *	vm_page_sleep_busy:
522  *
523  *	Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
524  *	m->busy is zero.  Returns TRUE if it had to sleep ( including if
525  *	it almost had to sleep and made temporary spl*() mods), FALSE
526  *	otherwise.
527  *
528  *	This routine assumes that interrupts can only remove the busy
529  *	status from a page, not set the busy status or change it from
530  *	PG_BUSY to m->busy or vise versa (which would create a timing
531  *	window).
532  *
533  *	Note that being an inline, this code will be well optimized.
534  */
535 
536 static __inline int
537 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
538 {
539 	if ((m->flags & PG_BUSY) || (also_m_busy && m->busy))  {
540 		int s = splvm();
541 		if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
542 			/*
543 			 * Page is busy. Wait and retry.
544 			 */
545 			vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
546 			tsleep(m, PVM, msg, 0);
547 		}
548 		splx(s);
549 		return(TRUE);
550 		/* not reached */
551 	}
552 	return(FALSE);
553 }
554 
555 /*
556  *	vm_page_dirty:
557  *
558  *	make page all dirty
559  */
560 
561 static __inline void
562 vm_page_dirty(vm_page_t m)
563 {
564 #if !defined(KLD_MODULE)
565 	KASSERT(m->queue - m->pc != PQ_CACHE, ("vm_page_dirty: page in cache!"));
566 #endif
567 	m->dirty = VM_PAGE_BITS_ALL;
568 }
569 
570 /*
571  *	vm_page_undirty:
572  *
573  *	Set page to not be dirty.  Note: does not clear pmap modify bits
574  */
575 
576 static __inline void
577 vm_page_undirty(vm_page_t m)
578 {
579 	m->dirty = 0;
580 }
581 
582 #if !defined(KLD_MODULE)
583 
584 static __inline vm_page_t
585 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
586 {
587 	vm_page_t m;
588 
589 #if PQ_L2_SIZE > 1
590 	if (prefer_zero) {
591 		m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist);
592 	} else {
593 		m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl);
594 	}
595 	if (m == NULL)
596 		m = _vm_page_list_find(basequeue, index);
597 #else
598 	if (prefer_zero) {
599 		m = TAILQ_LAST(&vm_page_queues[basequeue].pl, pglist);
600 	} else {
601 		m = TAILQ_FIRST(&vm_page_queues[basequeue].pl);
602 	}
603 #endif
604 	return(m);
605 }
606 
607 #endif
608 
609 #endif				/* _KERNEL */
610 #endif				/* !_VM_PAGE_ */
611