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 1.14 2004/05/31 01:51:23 hmp 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; /* vm_page_queues[] 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 */
124 u_short pc; /* page color */
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 */
131 * NOTE that these must support one bit per DEV_BSIZE in a page!!!
132 * so, on normal X86 kernels, they must be at least 8 bits wide.
134 #if PAGE_SIZE == 4096
135 u_char valid; /* map of valid DEV_BSIZE chunks */
136 u_char dirty; /* map of dirty DEV_BSIZE chunks */
137 #elif PAGE_SIZE == 8192
138 u_short valid; /* map of valid DEV_BSIZE chunks */
139 u_short dirty; /* map of dirty DEV_BSIZE chunks */
144 * note: currently use SWAPBLK_NONE as an absolute value rather then
147 #define SWAPBLK_MASK ((daddr_t)((u_daddr_t)-1 >> 1)) /* mask */
148 #define SWAPBLK_NONE ((daddr_t)((u_daddr_t)SWAPBLK_MASK + 1))/* flag */
151 * Page coloring parameters. We default to a middle of the road optimization.
152 * Larger selections would not really hurt us but if a machine does not have
153 * a lot of memory it could cause vm_page_alloc() to eat more cpu cycles
154 * looking for free pages.
156 * Page coloring cannot be disabled. Modules do not have access to most PQ
157 * constants because they can change between builds.
159 #if defined(_KERNEL) && !defined(KLD_MODULE)
161 #if !defined(PQ_CACHESIZE)
162 #define PQ_CACHESIZE 256 /* max is 1024 (MB) */
165 #if PQ_CACHESIZE >= 1024
166 #define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */
167 #define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */
168 #define PQ_L2_SIZE 256 /* A number of colors opt for 1M cache */
170 #elif PQ_CACHESIZE >= 512
171 #define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */
172 #define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */
173 #define PQ_L2_SIZE 128 /* A number of colors opt for 512K cache */
175 #elif PQ_CACHESIZE >= 256
176 #define PQ_PRIME1 13 /* Prime number somewhat less than PQ_HASH_SIZE */
177 #define PQ_PRIME2 7 /* Prime number somewhat less than PQ_HASH_SIZE */
178 #define PQ_L2_SIZE 64 /* A number of colors opt for 256K cache */
180 #elif PQ_CACHESIZE >= 128
181 #define PQ_PRIME1 9 /* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */
182 #define PQ_PRIME2 5 /* Prime number somewhat less than PQ_HASH_SIZE */
183 #define PQ_L2_SIZE 32 /* A number of colors opt for 128k cache */
186 #define PQ_PRIME1 5 /* Prime number somewhat less than PQ_HASH_SIZE */
187 #define PQ_PRIME2 3 /* Prime number somewhat less than PQ_HASH_SIZE */
188 #define PQ_L2_SIZE 16 /* A reasonable number of colors (opt for 64K cache) */
192 #define PQ_L2_MASK (PQ_L2_SIZE - 1)
194 #endif /* KERNEL && !KLD_MODULE */
198 * The queue array is always based on PQ_MAXL2_SIZE regardless of the actual
199 * cache size chosen in order to present a uniform interface for modules.
201 #define PQ_MAXL2_SIZE 256 /* fixed maximum (in pages) / module compat */
203 #if PQ_L2_SIZE > PQ_MAXL2_SIZE
204 #error "Illegal PQ_L2_SIZE"
209 #define PQ_INACTIVE (1 + 1*PQ_MAXL2_SIZE)
210 #define PQ_ACTIVE (2 + 1*PQ_MAXL2_SIZE)
211 #define PQ_CACHE (3 + 1*PQ_MAXL2_SIZE)
212 #define PQ_HOLD (3 + 2*PQ_MAXL2_SIZE)
213 #define PQ_COUNT (4 + 2*PQ_MAXL2_SIZE)
219 int flipflop; /* probably not the best place */
222 extern struct vpgqueues vm_page_queues[PQ_COUNT];
225 * These are the flags defined for vm_page.
227 * Note: PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
228 * not under PV management but otherwise should be treated as a
229 * normal page. Pages not under PV management cannot be paged out
230 * via the object/vm_page_t because there is no knowledge of their
231 * pte mappings, nor can they be removed from their objects via
232 * the object, and such pages are also not on any PQ queue.
234 #define PG_BUSY 0x0001 /* page is in transit (O) */
235 #define PG_WANTED 0x0002 /* someone is waiting for page (O) */
236 #define PG_WINATCFLS 0x0004 /* flush dirty page on inactive q */
237 #define PG_FICTITIOUS 0x0008 /* physical page doesn't exist (O) */
238 #define PG_WRITEABLE 0x0010 /* page is mapped writeable */
239 #define PG_MAPPED 0x0020 /* page is mapped */
240 #define PG_ZERO 0x0040 /* page is zeroed */
241 #define PG_REFERENCED 0x0080 /* page has been referenced */
242 #define PG_CLEANCHK 0x0100 /* page will be checked for cleaning */
243 #define PG_SWAPINPROG 0x0200 /* swap I/O in progress on page */
244 #define PG_NOSYNC 0x0400 /* do not collect for syncer */
245 #define PG_UNMANAGED 0x0800 /* No PV management for page */
246 #define PG_MARKER 0x1000 /* special queue marker page */
252 #define ACT_DECLINE 1
253 #define ACT_ADVANCE 3
256 #define PFCLUSTER_BEHIND 3
257 #define PFCLUSTER_AHEAD 3
261 * Each pageable resident page falls into one of four lists:
264 * Available for allocation now.
266 * The following are all LRU sorted:
269 * Almost available for allocation. Still in an
270 * object, but clean and immediately freeable at
271 * non-interrupt times.
274 * Low activity, candidates for reclamation.
275 * This is the list of pages that should be
279 * Pages that are "active" i.e. they have been
280 * recently referenced.
283 * Pages that are really free and have been pre-zeroed
287 extern int vm_page_zero_count;
288 extern vm_page_t vm_page_array; /* First resident page in table */
289 extern int vm_page_array_size; /* number of vm_page_t's */
290 extern long first_page; /* first physical page number */
292 #define VM_PAGE_TO_PHYS(entry) \
295 #define PHYS_TO_VM_PAGE(pa) \
296 (&vm_page_array[atop(pa) - first_page])
299 * Functions implemented as macros
303 vm_page_flag_set(vm_page_t m, unsigned int bits)
305 atomic_set_short(&(m)->flags, bits);
309 vm_page_flag_clear(vm_page_t m, unsigned int bits)
311 atomic_clear_short(&(m)->flags, bits);
315 vm_page_busy(vm_page_t m)
317 KASSERT((m->flags & PG_BUSY) == 0,
318 ("vm_page_busy: page already busy!!!"));
319 vm_page_flag_set(m, PG_BUSY);
325 * wakeup anyone waiting for the page.
329 vm_page_flash(vm_page_t m)
331 if (m->flags & PG_WANTED) {
332 vm_page_flag_clear(m, PG_WANTED);
338 * Clear the PG_BUSY flag and wakeup anyone waiting for the page. This
339 * is typically the last call you make on a page before moving onto
343 vm_page_wakeup(vm_page_t m)
345 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
346 vm_page_flag_clear(m, PG_BUSY);
351 * These routines manipulate the 'soft busy' count for a page. A soft busy
352 * is almost like PG_BUSY except that it allows certain compatible operations
353 * to occur on the page while it is busy. For example, a page undergoing a
354 * write can still be mapped read-only.
357 vm_page_io_start(vm_page_t m)
359 atomic_add_char(&(m)->busy, 1);
363 vm_page_io_finish(vm_page_t m)
365 atomic_subtract_char(&m->busy, 1);
371 #if PAGE_SIZE == 4096
372 #define VM_PAGE_BITS_ALL 0xff
375 #if PAGE_SIZE == 8192
376 #define VM_PAGE_BITS_ALL 0xffff
380 * Note: the code will always use nominally free pages from the free list
381 * before trying other flag-specified sources.
383 * At least one of VM_ALLOC_NORMAL|VM_ALLOC_SYSTEM|VM_ALLOC_INTERRUPT
384 * must be specified. VM_ALLOC_RETRY may only be specified if VM_ALLOC_NORMAL
387 #define VM_ALLOC_NORMAL 0x01 /* ok to use cache pages */
388 #define VM_ALLOC_SYSTEM 0x02 /* ok to exhaust most of free list */
389 #define VM_ALLOC_INTERRUPT 0x04 /* ok to exhaust entire free list */
390 #define VM_ALLOC_ZERO 0x08 /* req pre-zero'd memory if avail */
391 #define VM_ALLOC_RETRY 0x80 /* indefinite block (vm_page_grab()) */
393 void vm_page_unhold(vm_page_t mem);
394 void vm_page_activate (vm_page_t);
395 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
396 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
397 void vm_page_cache (vm_page_t);
398 int vm_page_try_to_cache (vm_page_t);
399 int vm_page_try_to_free (vm_page_t);
400 void vm_page_dontneed (vm_page_t);
401 void vm_page_deactivate (vm_page_t);
402 void vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
403 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
404 void vm_page_remove (vm_page_t);
405 void vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
406 vm_offset_t vm_page_startup (vm_offset_t, vm_offset_t, vm_offset_t);
407 vm_page_t vm_add_new_page (vm_paddr_t pa);
408 void vm_page_unmanage (vm_page_t);
409 void vm_page_unwire (vm_page_t, int);
410 void vm_page_wire (vm_page_t);
411 void vm_page_unqueue (vm_page_t);
412 void vm_page_unqueue_nowakeup (vm_page_t);
413 void vm_page_set_validclean (vm_page_t, int, int);
414 void vm_page_set_dirty (vm_page_t, int, int);
415 void vm_page_clear_dirty (vm_page_t, int, int);
416 void vm_page_set_invalid (vm_page_t, int, int);
417 int vm_page_is_valid (vm_page_t, int, int);
418 void vm_page_test_dirty (vm_page_t);
419 int vm_page_bits (int, int);
420 vm_page_t vm_page_list_find(int basequeue, int index, boolean_t prefer_zero);
421 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
422 void vm_page_free_toq(vm_page_t m);
423 int vm_contig_pg_alloc(u_long, vm_paddr_t, vm_paddr_t, u_long, u_long);
424 vm_offset_t vm_contig_pg_kmap(int, u_long, vm_map_t, int);
425 void vm_contig_pg_free(int, u_long);
428 * Holding a page keeps it from being reused. Other parts of the system
429 * can still disassociate the page from its current object and free it, or
430 * perform read or write I/O on it and/or otherwise manipulate the page,
431 * but if the page is held the VM system will leave the page and its data
432 * intact and not reuse the page for other purposes until the last hold
433 * reference is released. (see vm_page_wire() if you want to prevent the
434 * page from being disassociated from its object too).
436 * This routine must be called while at splvm() or better.
438 * The caller must still validate the contents of the page and, if necessary,
439 * wait for any pending I/O (e.g. vm_page_sleep_busy() loop) to complete
440 * before manipulating the page.
443 vm_page_hold(vm_page_t mem)
449 * Reduce the protection of a page. This routine never raises the
450 * protection and therefore can be safely called if the page is already
451 * at VM_PROT_NONE (it will be a NOP effectively ).
453 * VM_PROT_NONE will remove all user mappings of a page. This is often
454 * necessary when a page changes state (for example, turns into a copy-on-write
455 * page or needs to be frozen for write I/O) in order to force a fault, or
456 * to force a page's dirty bits to be synchronized and avoid hardware
457 * (modified/accessed) bit update races with pmap changes.
459 * Since 'prot' is usually a constant, this inline usually winds up optimizing
460 * out the primary conditional.
463 vm_page_protect(vm_page_t mem, int prot)
465 if (prot == VM_PROT_NONE) {
466 if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) {
467 pmap_page_protect(mem, VM_PROT_NONE);
468 vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED);
470 } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) {
471 pmap_page_protect(mem, VM_PROT_READ);
472 vm_page_flag_clear(mem, PG_WRITEABLE);
477 * Zero-fill the specified page. The entire contents of the page will be
480 static __inline boolean_t
481 vm_page_zero_fill(vm_page_t m)
483 pmap_zero_page(VM_PAGE_TO_PHYS(m));
488 * Copy the contents of src_m to dest_m. The pages must be stable but spl
489 * and other protections depend on context.
492 vm_page_copy(vm_page_t src_m, vm_page_t dest_m)
494 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m));
495 dest_m->valid = VM_PAGE_BITS_ALL;
499 * Free a page. The page must be marked BUSY.
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
507 vm_page_free(vm_page_t m)
509 vm_page_flag_clear(m, PG_ZERO);
514 * Free a page to the zerod-pages queue
517 vm_page_free_zero(vm_page_t m)
519 vm_page_flag_set(m, PG_ZERO);
524 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
525 * m->busy is zero. Returns TRUE if it had to sleep ( including if
526 * it almost had to sleep and made temporary spl*() mods), FALSE
529 * This routine assumes that interrupts can only remove the busy
530 * status from a page, not set the busy status or change it from
531 * PG_BUSY to m->busy or vise versa (which would create a timing
534 * Note: as an inline, 'also_m_busy' is usually a constant and well
538 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
540 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
542 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
544 * Page is busy. Wait and retry.
546 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
547 tsleep(m, 0, msg, 0);
557 * Make page all dirty
560 vm_page_dirty(vm_page_t m)
562 KASSERT(m->queue - m->pc != PQ_CACHE,
563 ("vm_page_dirty: page in cache!"));
564 m->dirty = VM_PAGE_BITS_ALL;
568 * Set page to not be dirty. Note: does not clear pmap modify bits .
571 vm_page_undirty(vm_page_t m)
577 #endif /* !_VM_PAGE_ */