2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
36 * from: @(#)vm_page.h 8.2 (Berkeley) 12/13/93
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
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.
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.
54 * Carnegie Mellon requests users of this software to return to
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
64 * $FreeBSD: src/sys/vm/vm_page.h,v 1.75.2.8 2002/03/06 01:07:09 dillon Exp $
65 * $DragonFly: src/sys/vm/vm_page.h,v 1.8 2003/11/03 17:11:23 dillon Exp $
69 * Resident memory system definitions.
75 #if !defined(KLD_MODULE)
76 #include "opt_vmpage.h"
80 #include <machine/atomic.h>
83 * Management of resident (logical) pages.
85 * A small structure is kept for each resident
86 * page, indexed by page number. Each structure
87 * is an element of several lists:
89 * A hash table bucket used to quickly
90 * perform object/offset lookups
92 * A list of all pages for a given object,
93 * so they can be quickly deactivated at
94 * time of deallocation.
96 * An ordered list of pages due for pageout.
98 * In addition, the structure contains the object
99 * and offset to which this page belongs (for pageout),
100 * and sundry status bits.
102 * Fields in this structure are locked either by the lock on the
103 * object that the page belongs to (O) or by the lock on the page
106 * The 'valid' and 'dirty' fields are distinct. A page may have dirty
107 * bits set without having associated valid bits set. This is used by
108 * NFS to implement piecemeal writes.
111 TAILQ_HEAD(pglist, vm_page);
114 TAILQ_ENTRY(vm_page) pageq; /* queue info for FIFO queue or free list (P) */
115 struct vm_page *hnext; /* hash table link (O,P) */
116 TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */
118 vm_object_t object; /* which object am I in (O,P)*/
119 vm_pindex_t pindex; /* offset into object (O,P) */
120 vm_paddr_t phys_addr; /* physical address of page */
121 struct md_page md; /* machine dependant stuff */
122 u_short queue; /* page queue index */
123 u_short flags, /* see below */
125 u_short wire_count; /* wired down maps refs (P) */
126 short hold_count; /* page hold count */
127 u_char act_count; /* page usage count */
128 u_char busy; /* page busy count */
129 /* NOTE that these must support one bit per DEV_BSIZE in a page!!! */
130 /* so, on normal X86 kernels, they must be at least 8 bits wide */
131 #if PAGE_SIZE == 4096
132 u_char valid; /* map of valid DEV_BSIZE chunks */
133 u_char dirty; /* map of dirty DEV_BSIZE chunks */
134 #elif PAGE_SIZE == 8192
135 u_short valid; /* map of valid DEV_BSIZE chunks */
136 u_short dirty; /* map of dirty DEV_BSIZE chunks */
141 * note: currently use SWAPBLK_NONE as an absolute value rather then
145 #define SWAPBLK_MASK ((daddr_t)((u_daddr_t)-1 >> 1)) /* mask */
146 #define SWAPBLK_NONE ((daddr_t)((u_daddr_t)SWAPBLK_MASK + 1))/* flag */
148 #if !defined(KLD_MODULE)
151 * Page coloring parameters
153 /* Each of PQ_FREE, and PQ_CACHE have PQ_HASH_SIZE entries */
155 /* Backward compatibility for existing PQ_*CACHE config options. */
156 #if !defined(PQ_CACHESIZE)
157 #if defined(PQ_HUGECACHE)
158 #define PQ_CACHESIZE 1024
159 #elif defined(PQ_LARGECACHE)
160 #define PQ_CACHESIZE 512
161 #elif defined(PQ_MEDIUMCACHE)
162 #define PQ_CACHESIZE 256
163 #elif defined(PQ_NORMALCACHE)
164 #define PQ_CACHESIZE 64
165 #elif defined(PQ_NOOPT)
166 #define PQ_CACHESIZE 0
168 #define PQ_CACHESIZE 128
172 #if PQ_CACHESIZE >= 1024
173 #define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */
174 #define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */
175 #define PQ_L2_SIZE 256 /* A number of colors opt for 1M cache */
177 #elif PQ_CACHESIZE >= 512
178 #define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */
179 #define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */
180 #define PQ_L2_SIZE 128 /* A number of colors opt for 512K cache */
182 #elif PQ_CACHESIZE >= 256
183 #define PQ_PRIME1 13 /* Prime number somewhat less than PQ_HASH_SIZE */
184 #define PQ_PRIME2 7 /* Prime number somewhat less than PQ_HASH_SIZE */
185 #define PQ_L2_SIZE 64 /* A number of colors opt for 256K cache */
187 #elif PQ_CACHESIZE >= 128
188 #define PQ_PRIME1 9 /* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */
189 #define PQ_PRIME2 5 /* Prime number somewhat less than PQ_HASH_SIZE */
190 #define PQ_L2_SIZE 32 /* A number of colors opt for 128k cache */
192 #elif PQ_CACHESIZE >= 64
193 #define PQ_PRIME1 5 /* Prime number somewhat less than PQ_HASH_SIZE */
194 #define PQ_PRIME2 3 /* Prime number somewhat less than PQ_HASH_SIZE */
195 #define PQ_L2_SIZE 16 /* A reasonable number of colors (opt for 64K cache) */
198 #define PQ_PRIME1 1 /* Disable page coloring. */
204 #define PQ_L2_MASK (PQ_L2_SIZE - 1)
208 #define PQ_INACTIVE (1 + 1*PQ_L2_SIZE)
209 #define PQ_ACTIVE (2 + 1*PQ_L2_SIZE)
210 #define PQ_CACHE (3 + 1*PQ_L2_SIZE)
211 #define PQ_HOLD (3 + 2*PQ_L2_SIZE)
212 #define PQ_COUNT (4 + 2*PQ_L2_SIZE)
218 int flipflop; /* probably not the best place */
221 extern struct vpgqueues vm_page_queues[PQ_COUNT];
226 * These are the flags defined for vm_page.
228 * Note: PG_FILLED and PG_DIRTY are added for the filesystems.
230 * Note: PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
231 * not under PV management but otherwise should be treated as a
232 * normal page. Pages not under PV management cannot be paged out
233 * via the object/vm_page_t because there is no knowledge of their
234 * pte mappings, nor can they be removed from their objects via
235 * the object, and such pages are also not on any PQ queue.
237 #define PG_BUSY 0x0001 /* page is in transit (O) */
238 #define PG_WANTED 0x0002 /* someone is waiting for page (O) */
239 #define PG_WINATCFLS 0x0004 /* flush dirty page on inactive q */
240 #define PG_FICTITIOUS 0x0008 /* physical page doesn't exist (O) */
241 #define PG_WRITEABLE 0x0010 /* page is mapped writeable */
242 #define PG_MAPPED 0x0020 /* page is mapped */
243 #define PG_ZERO 0x0040 /* page is zeroed */
244 #define PG_REFERENCED 0x0080 /* page has been referenced */
245 #define PG_CLEANCHK 0x0100 /* page will be checked for cleaning */
246 #define PG_SWAPINPROG 0x0200 /* swap I/O in progress on page */
247 #define PG_NOSYNC 0x0400 /* do not collect for syncer */
248 #define PG_UNMANAGED 0x0800 /* No PV management for page */
249 #define PG_MARKER 0x1000 /* special queue marker page */
255 #define ACT_DECLINE 1
256 #define ACT_ADVANCE 3
259 #define PFCLUSTER_BEHIND 3
260 #define PFCLUSTER_AHEAD 3
264 * Each pageable resident page falls into one of four lists:
267 * Available for allocation now.
269 * The following are all LRU sorted:
272 * Almost available for allocation. Still in an
273 * object, but clean and immediately freeable at
274 * non-interrupt times.
277 * Low activity, candidates for reclamation.
278 * This is the list of pages that should be
282 * Pages that are "active" i.e. they have been
283 * recently referenced.
286 * Pages that are really free and have been pre-zeroed
290 extern int vm_page_zero_count;
292 extern vm_page_t vm_page_array; /* First resident page in table */
293 extern int vm_page_array_size; /* number of vm_page_t's */
294 extern long first_page; /* first physical page number */
296 #define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr)
298 #define PHYS_TO_VM_PAGE(pa) \
299 (&vm_page_array[atop(pa) - first_page ])
302 * Functions implemented as macros
306 vm_page_flag_set(vm_page_t m, unsigned int bits)
308 atomic_set_short(&(m)->flags, bits);
312 vm_page_flag_clear(vm_page_t m, unsigned int bits)
314 atomic_clear_short(&(m)->flags, bits);
319 vm_page_assert_wait(vm_page_t m, int interruptible)
321 vm_page_flag_set(m, PG_WANTED);
322 assert_wait((int) m, interruptible);
327 vm_page_busy(vm_page_t m)
329 KASSERT((m->flags & PG_BUSY) == 0, ("vm_page_busy: page already busy!!!"));
330 vm_page_flag_set(m, PG_BUSY);
336 * wakeup anyone waiting for the page.
340 vm_page_flash(vm_page_t m)
342 if (m->flags & PG_WANTED) {
343 vm_page_flag_clear(m, PG_WANTED);
351 * clear the PG_BUSY flag and wakeup anyone waiting for the
357 vm_page_wakeup(vm_page_t m)
359 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
360 vm_page_flag_clear(m, PG_BUSY);
370 vm_page_io_start(vm_page_t m)
372 atomic_add_char(&(m)->busy, 1);
376 vm_page_io_finish(vm_page_t m)
378 atomic_subtract_char(&m->busy, 1);
384 #if PAGE_SIZE == 4096
385 #define VM_PAGE_BITS_ALL 0xff
388 #if PAGE_SIZE == 8192
389 #define VM_PAGE_BITS_ALL 0xffff
392 #define VM_ALLOC_NORMAL 0
393 #define VM_ALLOC_INTERRUPT 1
394 #define VM_ALLOC_SYSTEM 2
395 #define VM_ALLOC_ZERO 3
396 #define VM_ALLOC_RETRY 0x80
398 void vm_page_unhold(vm_page_t mem);
400 void vm_page_activate (vm_page_t);
401 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
402 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
403 void vm_page_cache (vm_page_t);
404 int vm_page_try_to_cache (vm_page_t);
405 int vm_page_try_to_free (vm_page_t);
406 void vm_page_dontneed (vm_page_t);
407 static __inline void vm_page_copy (vm_page_t, vm_page_t);
408 static __inline void vm_page_free (vm_page_t);
409 static __inline void vm_page_free_zero (vm_page_t);
410 void vm_page_deactivate (vm_page_t);
411 void vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
412 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
413 void vm_page_remove (vm_page_t);
414 void vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
415 vm_offset_t vm_page_startup (vm_offset_t, vm_offset_t, vm_offset_t);
416 vm_page_t vm_add_new_page (vm_paddr_t pa);
417 void vm_page_unmanage (vm_page_t);
418 void vm_page_unwire (vm_page_t, int);
419 void vm_page_wire (vm_page_t);
420 void vm_page_unqueue (vm_page_t);
421 void vm_page_unqueue_nowakeup (vm_page_t);
422 void vm_page_set_validclean (vm_page_t, int, int);
423 void vm_page_set_dirty (vm_page_t, int, int);
424 void vm_page_clear_dirty (vm_page_t, int, int);
425 void vm_page_set_invalid (vm_page_t, int, int);
426 static __inline boolean_t vm_page_zero_fill (vm_page_t);
427 int vm_page_is_valid (vm_page_t, int, int);
428 void vm_page_test_dirty (vm_page_t);
429 int vm_page_bits (int, int);
430 vm_page_t _vm_page_list_find (int, int);
431 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
432 void vm_page_free_toq(vm_page_t m);
434 int vm_contig_pg_alloc(u_long, vm_paddr_t, vm_paddr_t, u_long, u_long);
435 vm_offset_t vm_contig_pg_kmap(int, u_long, vm_map_t);
436 void vm_contig_pg_free(int, u_long);
439 * Keep page from being freed by the page daemon
440 * much of the same effect as wiring, except much lower
441 * overhead and should be used only for *very* temporary
442 * holding ("wiring").
445 vm_page_hold(vm_page_t mem)
453 * Reduce the protection of a page. This routine never raises the
454 * protection and therefore can be safely called if the page is already
455 * at VM_PROT_NONE (it will be a NOP effectively ).
459 vm_page_protect(vm_page_t mem, int prot)
461 if (prot == VM_PROT_NONE) {
462 if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) {
463 pmap_page_protect(mem, VM_PROT_NONE);
464 vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED);
466 } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) {
467 pmap_page_protect(mem, VM_PROT_READ);
468 vm_page_flag_clear(mem, PG_WRITEABLE);
475 * Zero-fill the specified page.
476 * Written as a standard pagein routine, to
477 * be used by the zero-fill object.
479 static __inline boolean_t
483 pmap_zero_page(VM_PAGE_TO_PHYS(m));
490 * Copy one page to another
493 vm_page_copy(src_m, dest_m)
497 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m));
498 dest_m->valid = VM_PAGE_BITS_ALL;
506 * The clearing of PG_ZERO is a temporary safety until the code can be
507 * reviewed to determine that PG_ZERO is being properly cleared on
508 * write faults or maps. PG_ZERO was previously cleared in
515 vm_page_flag_clear(m, PG_ZERO);
522 * Free a page to the zerod-pages queue
528 vm_page_flag_set(m, PG_ZERO);
533 * vm_page_sleep_busy:
535 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
536 * m->busy is zero. Returns TRUE if it had to sleep ( including if
537 * it almost had to sleep and made temporary spl*() mods), FALSE
540 * This routine assumes that interrupts can only remove the busy
541 * status from a page, not set the busy status or change it from
542 * PG_BUSY to m->busy or vise versa (which would create a timing
545 * Note that being an inline, this code will be well optimized.
549 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
551 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
553 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
555 * Page is busy. Wait and retry.
557 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
558 tsleep(m, 0, msg, 0);
570 * make page all dirty
574 vm_page_dirty(vm_page_t m)
576 #if !defined(KLD_MODULE)
577 KASSERT(m->queue - m->pc != PQ_CACHE, ("vm_page_dirty: page in cache!"));
579 m->dirty = VM_PAGE_BITS_ALL;
585 * Set page to not be dirty. Note: does not clear pmap modify bits
589 vm_page_undirty(vm_page_t m)
594 #if !defined(KLD_MODULE)
596 static __inline vm_page_t
597 vm_page_list_find(int basequeue, int index, boolean_t prefer_zero)
603 m = TAILQ_LAST(&vm_page_queues[basequeue+index].pl, pglist);
605 m = TAILQ_FIRST(&vm_page_queues[basequeue+index].pl);
608 m = _vm_page_list_find(basequeue, index);
611 m = TAILQ_LAST(&vm_page_queues[basequeue].pl, pglist);
613 m = TAILQ_FIRST(&vm_page_queues[basequeue].pl);
622 #endif /* !_VM_PAGE_ */