xref: /freebsd/sys/vm/vm_page.h (revision 2357939bc239bd5334a169b62313806178dd8f30)
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  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	from: @(#)vm_page.h	8.2 (Berkeley) 12/13/93
33  *
34  *
35  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36  * All rights reserved.
37  *
38  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39  *
40  * Permission to use, copy, modify and distribute this software and
41  * its documentation is hereby granted, provided that both the copyright
42  * notice and this permission notice appear in all copies of the
43  * software, derivative works or modified versions, and any portions
44  * thereof, and that both notices appear in supporting documentation.
45  *
46  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49  *
50  * Carnegie Mellon requests users of this software to return to
51  *
52  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
53  *  School of Computer Science
54  *  Carnegie Mellon University
55  *  Pittsburgh PA 15213-3890
56  *
57  * any improvements or extensions that they make and grant Carnegie the
58  * rights to redistribute these changes.
59  *
60  * $FreeBSD$
61  */
62 
63 /*
64  *	Resident memory system definitions.
65  */
66 
67 #ifndef	_VM_PAGE_
68 #define	_VM_PAGE_
69 
70 #if !defined(KLD_MODULE)
71 #include "opt_vmpage.h"
72 #endif
73 
74 #include <vm/pmap.h>
75 
76 /*
77  *	Management of resident (logical) pages.
78  *
79  *	A small structure is kept for each resident
80  *	page, indexed by page number.  Each structure
81  *	is an element of several lists:
82  *
83  *		A hash table bucket used to quickly
84  *		perform object/offset lookups
85  *
86  *		A list of all pages for a given object,
87  *		so they can be quickly deactivated at
88  *		time of deallocation.
89  *
90  *		An ordered list of pages due for pageout.
91  *
92  *	In addition, the structure contains the object
93  *	and offset to which this page belongs (for pageout),
94  *	and sundry status bits.
95  *
96  *	Fields in this structure are locked either by the lock on the
97  *	object that the page belongs to (O) or by the lock on the page
98  *	queues (P).
99  *
100  *	The 'valid' and 'dirty' fields are distinct.  A page may have dirty
101  *	bits set without having associated valid bits set.  This is used by
102  *	NFS to implement piecemeal writes.
103  */
104 
105 TAILQ_HEAD(pglist, vm_page);
106 
107 struct vm_page {
108 	TAILQ_ENTRY(vm_page) pageq;	/* queue info for FIFO queue or free list (P) */
109 	TAILQ_ENTRY(vm_page) listq;	/* pages in same object (O) 	*/
110 	struct vm_page *left;		/* splay tree link (O)		*/
111 	struct vm_page *right;		/* splay tree link (O)		*/
112 
113 	vm_object_t object;		/* which object am I in (O,P)*/
114 	vm_pindex_t pindex;		/* offset into object (O,P) */
115 	vm_paddr_t phys_addr;		/* physical address of page */
116 	struct md_page md;		/* machine dependant stuff */
117 	u_short	queue;			/* page queue index */
118 	u_short	flags,			/* see below */
119 		pc;			/* page color */
120 	u_short wire_count;		/* wired down maps refs (P) */
121 	short hold_count;		/* page hold count */
122 	u_char	act_count;		/* page usage count */
123 	u_char	busy;			/* page busy count */
124 	/* NOTE that these must support one bit per DEV_BSIZE in a page!!! */
125 	/* so, on normal X86 kernels, they must be at least 8 bits wide */
126 #if PAGE_SIZE == 4096
127 	u_char	valid;			/* map of valid DEV_BSIZE chunks (O) */
128 	u_char	dirty;			/* map of dirty DEV_BSIZE chunks */
129 #elif PAGE_SIZE == 8192
130 	u_short	valid;			/* map of valid DEV_BSIZE chunks (O) */
131 	u_short	dirty;			/* map of dirty DEV_BSIZE chunks */
132 #elif PAGE_SIZE == 16384
133 	u_int valid;			/* map of valid DEV_BSIZE chunks (O) */
134 	u_int dirty;			/* map of dirty DEV_BSIZE chunks */
135 #elif PAGE_SIZE == 32768
136 	u_long valid;			/* map of valid DEV_BSIZE chunks (O) */
137 	u_long dirty;			/* map of dirty DEV_BSIZE chunks */
138 #endif
139 	u_int cow;			/* page cow mapping count */
140 };
141 
142 /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
143 #if PAGE_SIZE == 32768
144 #ifdef CTASSERT
145 CTASSERT(sizeof(u_long) >= 8);
146 #endif
147 #endif
148 
149 /*
150  * note: currently use SWAPBLK_NONE as an absolute value rather then
151  * a flag bit.
152  */
153 
154 #define SWAPBLK_MASK	((daddr_t)((u_daddr_t)-1 >> 1))		/* mask */
155 #define SWAPBLK_NONE	((daddr_t)((u_daddr_t)SWAPBLK_MASK + 1))/* flag */
156 
157 #if !defined(KLD_MODULE)
158 /*
159  * Page coloring parameters
160  */
161 /* Each of PQ_FREE, and PQ_CACHE have PQ_HASH_SIZE entries */
162 
163 /* Backward compatibility for existing PQ_*CACHE config options. */
164 #if !defined(PQ_CACHESIZE)
165 #if defined(PQ_HUGECACHE)
166 #define PQ_CACHESIZE 1024
167 #elif defined(PQ_LARGECACHE)
168 #define PQ_CACHESIZE 512
169 #elif defined(PQ_MEDIUMCACHE)
170 #define PQ_CACHESIZE 256
171 #elif defined(PQ_NORMALCACHE)
172 #define PQ_CACHESIZE 64
173 #elif defined(PQ_NOOPT)
174 #define PQ_CACHESIZE 0
175 #else
176 #define PQ_CACHESIZE 128
177 #endif
178 #endif			/* !defined(PQ_CACHESIZE) */
179 
180 #if PQ_CACHESIZE >= 1024
181 #define PQ_PRIME1 31	/* Prime number somewhat less than PQ_HASH_SIZE */
182 #define PQ_PRIME2 23	/* Prime number somewhat less than PQ_HASH_SIZE */
183 #define PQ_L2_SIZE 256	/* A number of colors opt for 1M cache */
184 
185 #elif PQ_CACHESIZE >= 512
186 #define PQ_PRIME1 31	/* Prime number somewhat less than PQ_HASH_SIZE */
187 #define PQ_PRIME2 23	/* Prime number somewhat less than PQ_HASH_SIZE */
188 #define PQ_L2_SIZE 128	/* A number of colors opt for 512K cache */
189 
190 #elif PQ_CACHESIZE >= 256
191 #define PQ_PRIME1 13	/* Prime number somewhat less than PQ_HASH_SIZE */
192 #define PQ_PRIME2 7	/* Prime number somewhat less than PQ_HASH_SIZE */
193 #define PQ_L2_SIZE 64	/* A number of colors opt for 256K cache */
194 
195 #elif PQ_CACHESIZE >= 128
196 #define PQ_PRIME1 9	/* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */
197 #define PQ_PRIME2 5	/* Prime number somewhat less than PQ_HASH_SIZE */
198 #define PQ_L2_SIZE 32	/* A number of colors opt for 128k cache */
199 
200 #elif PQ_CACHESIZE >= 64
201 #define PQ_PRIME1 5	/* Prime number somewhat less than PQ_HASH_SIZE */
202 #define PQ_PRIME2 3	/* Prime number somewhat less than PQ_HASH_SIZE */
203 #define PQ_L2_SIZE 16	/* A reasonable number of colors (opt for 64K cache) */
204 
205 #else
206 #define PQ_PRIME1 1	/* Disable page coloring. */
207 #define PQ_PRIME2 1
208 #define PQ_L2_SIZE 1
209 
210 #endif
211 
212 #define PQ_L2_MASK (PQ_L2_SIZE - 1)
213 
214 #define PQ_NONE 0
215 #define PQ_FREE	1
216 #define PQ_INACTIVE (1 + 1*PQ_L2_SIZE)
217 #define PQ_ACTIVE (2 + 1*PQ_L2_SIZE)
218 #define PQ_CACHE (3 + 1*PQ_L2_SIZE)
219 #define PQ_HOLD  (3 + 2*PQ_L2_SIZE)
220 #define PQ_COUNT (4 + 2*PQ_L2_SIZE)
221 
222 struct vpgqueues {
223 	struct pglist pl;
224 	int	*cnt;
225 	int	lcnt;
226 };
227 
228 extern struct vpgqueues vm_page_queues[PQ_COUNT];
229 extern struct mtx vm_page_queue_free_mtx;
230 
231 #endif			/* !defined(KLD_MODULE) */
232 
233 /*
234  * These are the flags defined for vm_page.
235  *
236  * Note: PG_FILLED and PG_DIRTY are added for the filesystems.
237  *
238  * Note: PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
239  * 	 not under PV management but otherwise should be treated as a
240  *	 normal page.  Pages not under PV management cannot be paged out
241  *	 via the object/vm_page_t because there is no knowledge of their
242  *	 pte mappings, nor can they be removed from their objects via
243  *	 the object, and such pages are also not on any PQ queue.
244  */
245 #define	PG_BUSY		0x0001		/* page is in transit (O) */
246 #define	PG_WANTED	0x0002		/* someone is waiting for page (O) */
247 #define PG_WINATCFLS	0x0004		/* flush dirty page on inactive q */
248 #define	PG_FICTITIOUS	0x0008		/* physical page doesn't exist (O) */
249 #define	PG_WRITEABLE	0x0010		/* page is mapped writeable */
250 #define	PG_ZERO		0x0040		/* page is zeroed */
251 #define PG_REFERENCED	0x0080		/* page has been referenced */
252 #define PG_CLEANCHK	0x0100		/* page will be checked for cleaning */
253 #define PG_SWAPINPROG	0x0200		/* swap I/O in progress on page	     */
254 #define PG_NOSYNC	0x0400		/* do not collect for syncer */
255 #define PG_UNMANAGED	0x0800		/* No PV management for page */
256 #define PG_MARKER	0x1000		/* special queue marker page */
257 #define	PG_SLAB		0x2000		/* object pointer is actually a slab */
258 
259 /*
260  * Misc constants.
261  */
262 #define ACT_DECLINE		1
263 #define ACT_ADVANCE		3
264 #define ACT_INIT		5
265 #define ACT_MAX			64
266 
267 #ifdef _KERNEL
268 /*
269  * Each pageable resident page falls into one of four lists:
270  *
271  *	free
272  *		Available for allocation now.
273  *
274  * The following are all LRU sorted:
275  *
276  *	cache
277  *		Almost available for allocation. Still in an
278  *		object, but clean and immediately freeable at
279  *		non-interrupt times.
280  *
281  *	inactive
282  *		Low activity, candidates for reclamation.
283  *		This is the list of pages that should be
284  *		paged out next.
285  *
286  *	active
287  *		Pages that are "active" i.e. they have been
288  *		recently referenced.
289  *
290  *	zero
291  *		Pages that are really free and have been pre-zeroed
292  *
293  */
294 
295 extern int vm_page_zero_count;
296 
297 extern vm_page_t vm_page_array;		/* First resident page in table */
298 extern int vm_page_array_size;		/* number of vm_page_t's */
299 extern long first_page;			/* first physical page number */
300 
301 #define VM_PAGE_TO_PHYS(entry)	((entry)->phys_addr)
302 
303 #define PHYS_TO_VM_PAGE(pa) \
304 		(&vm_page_array[atop(pa) - first_page ])
305 
306 extern struct mtx vm_page_queue_mtx;
307 #define vm_page_lock_queues()   mtx_lock(&vm_page_queue_mtx)
308 #define vm_page_unlock_queues() mtx_unlock(&vm_page_queue_mtx)
309 
310 #if PAGE_SIZE == 4096
311 #define VM_PAGE_BITS_ALL 0xffu
312 #elif PAGE_SIZE == 8192
313 #define VM_PAGE_BITS_ALL 0xffffu
314 #elif PAGE_SIZE == 16384
315 #define VM_PAGE_BITS_ALL 0xffffffffu
316 #elif PAGE_SIZE == 32768
317 #define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
318 #endif
319 
320 /* page allocation classes: */
321 #define VM_ALLOC_NORMAL		0
322 #define VM_ALLOC_INTERRUPT	1
323 #define VM_ALLOC_SYSTEM		2
324 #define	VM_ALLOC_CLASS_MASK	3
325 /* page allocation flags: */
326 #define	VM_ALLOC_WIRED		0x0020	/* non pageable */
327 #define	VM_ALLOC_ZERO		0x0040	/* Try to obtain a zeroed page */
328 #define	VM_ALLOC_RETRY		0x0080	/* vm_page_grab() only */
329 #define	VM_ALLOC_NOOBJ		0x0100	/* No associated object */
330 
331 void vm_page_flag_set(vm_page_t m, unsigned short bits);
332 void vm_page_flag_clear(vm_page_t m, unsigned short bits);
333 void vm_page_busy(vm_page_t m);
334 void vm_page_flash(vm_page_t m);
335 void vm_page_io_start(vm_page_t m);
336 void vm_page_io_finish(vm_page_t m);
337 void vm_page_hold(vm_page_t mem);
338 void vm_page_unhold(vm_page_t mem);
339 void vm_page_free(vm_page_t m);
340 void vm_page_free_zero(vm_page_t m);
341 int vm_page_sleep_if_busy(vm_page_t m, int also_m_busy, const char *msg);
342 void vm_page_dirty(vm_page_t m);
343 void vm_page_wakeup(vm_page_t m);
344 
345 void vm_pageq_init(void);
346 vm_page_t vm_pageq_add_new_page(vm_paddr_t pa);
347 void vm_pageq_enqueue(int queue, vm_page_t m);
348 void vm_pageq_remove_nowakeup(vm_page_t m);
349 void vm_pageq_remove(vm_page_t m);
350 vm_page_t vm_pageq_find(int basequeue, int index, boolean_t prefer_zero);
351 void vm_pageq_requeue(vm_page_t m);
352 
353 void vm_page_activate (vm_page_t);
354 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
355 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
356 void vm_page_cache (register vm_page_t);
357 int vm_page_try_to_cache (vm_page_t);
358 int vm_page_try_to_free (vm_page_t);
359 void vm_page_dontneed (register vm_page_t);
360 void vm_page_deactivate (vm_page_t);
361 void vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
362 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
363 void vm_page_remove (vm_page_t);
364 void vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
365 vm_page_t vm_page_select_cache(int);
366 vm_page_t vm_page_splay(vm_pindex_t, vm_page_t);
367 vm_offset_t vm_page_startup(vm_offset_t vaddr);
368 void vm_page_unmanage (vm_page_t);
369 void vm_page_unwire (vm_page_t, int);
370 void vm_page_wire (vm_page_t);
371 void vm_page_set_validclean (vm_page_t, int, int);
372 void vm_page_clear_dirty (vm_page_t, int, int);
373 void vm_page_set_invalid (vm_page_t, int, int);
374 int vm_page_is_valid (vm_page_t, int, int);
375 void vm_page_test_dirty (vm_page_t);
376 int vm_page_bits (int, int);
377 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
378 void vm_page_free_toq(vm_page_t m);
379 void vm_page_zero_idle_wakeup(void);
380 void vm_page_cowfault (vm_page_t);
381 void vm_page_cowsetup (vm_page_t);
382 void vm_page_cowclear (vm_page_t);
383 
384 /*
385  *	vm_page_undirty:
386  *
387  *	Set page to not be dirty.  Note: does not clear pmap modify bits
388  */
389 static __inline void
390 vm_page_undirty(vm_page_t m)
391 {
392 	m->dirty = 0;
393 }
394 
395 #endif				/* _KERNEL */
396 #endif				/* !_VM_PAGE_ */
397