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 1.20 2005/06/14 17:16:00 dillon Exp $
69 * Resident memory system definitions.
75 #if !defined(KLD_MODULE) && defined(_KERNEL)
76 #include "opt_vmpage.h"
80 #include <machine/atomic.h>
81 #include <sys/thread2.h>
84 * Management of resident (logical) pages.
86 * A small structure is kept for each resident
87 * page, indexed by page number. Each structure
88 * is an element of several lists:
90 * A hash table bucket used to quickly
91 * perform object/offset lookups
93 * A list of all pages for a given object,
94 * so they can be quickly deactivated at
95 * time of deallocation.
97 * An ordered list of pages due for pageout.
99 * In addition, the structure contains the object
100 * and offset to which this page belongs (for pageout),
101 * and sundry status bits.
103 * Fields in this structure are locked either by the lock on the
104 * object that the page belongs to (O) or by the lock on the page
107 * The 'valid' and 'dirty' fields are distinct. A page may have dirty
108 * bits set without having associated valid bits set. This is used by
109 * NFS to implement piecemeal writes.
112 TAILQ_HEAD(pglist, vm_page);
116 TAILQ_ENTRY(vm_page) pageq; /* vm_page_queues[] list (P) */
117 struct vm_page *hnext; /* hash table link (O,P) */
118 TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */
120 vm_object_t object; /* which object am I in (O,P)*/
121 vm_pindex_t pindex; /* offset into object (O,P) */
122 vm_paddr_t phys_addr; /* physical address of page */
123 struct md_page md; /* machine dependant stuff */
124 u_short queue; /* page queue index */
125 u_short flags; /* see below */
126 u_short pc; /* page color */
127 u_short wire_count; /* wired down maps refs (P) */
128 short hold_count; /* page hold count */
129 u_char act_count; /* page usage count */
130 u_char busy; /* page busy count */
133 * NOTE that these must support one bit per DEV_BSIZE in a page!!!
134 * so, on normal X86 kernels, they must be at least 8 bits wide.
136 #if PAGE_SIZE == 4096
137 u_char valid; /* map of valid DEV_BSIZE chunks */
138 u_char dirty; /* map of dirty DEV_BSIZE chunks */
141 #elif PAGE_SIZE == 8192
142 u_short valid; /* map of valid DEV_BSIZE chunks */
143 u_short dirty; /* map of dirty DEV_BSIZE chunks */
145 struct msf_buf *msf_hint; /* first page of an msfbuf map */
149 * note: currently use SWAPBLK_NONE as an absolute value rather then
152 #define SWAPBLK_MASK ((daddr_t)((u_daddr_t)-1 >> 1)) /* mask */
153 #define SWAPBLK_NONE ((daddr_t)((u_daddr_t)SWAPBLK_MASK + 1))/* flag */
156 * Page coloring parameters. We default to a middle of the road optimization.
157 * Larger selections would not really hurt us but if a machine does not have
158 * a lot of memory it could cause vm_page_alloc() to eat more cpu cycles
159 * looking for free pages.
161 * Page coloring cannot be disabled. Modules do not have access to most PQ
162 * constants because they can change between builds.
164 #if defined(_KERNEL) && !defined(KLD_MODULE)
166 #if !defined(PQ_CACHESIZE)
167 #define PQ_CACHESIZE 256 /* max is 1024 (MB) */
170 #if PQ_CACHESIZE >= 1024
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 256 /* A number of colors opt for 1M cache */
175 #elif PQ_CACHESIZE >= 512
176 #define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */
177 #define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */
178 #define PQ_L2_SIZE 128 /* A number of colors opt for 512K cache */
180 #elif PQ_CACHESIZE >= 256
181 #define PQ_PRIME1 13 /* Prime number somewhat less than PQ_HASH_SIZE */
182 #define PQ_PRIME2 7 /* Prime number somewhat less than PQ_HASH_SIZE */
183 #define PQ_L2_SIZE 64 /* A number of colors opt for 256K cache */
185 #elif PQ_CACHESIZE >= 128
186 #define PQ_PRIME1 9 /* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */
187 #define PQ_PRIME2 5 /* Prime number somewhat less than PQ_HASH_SIZE */
188 #define PQ_L2_SIZE 32 /* A number of colors opt for 128k cache */
191 #define PQ_PRIME1 5 /* Prime number somewhat less than PQ_HASH_SIZE */
192 #define PQ_PRIME2 3 /* Prime number somewhat less than PQ_HASH_SIZE */
193 #define PQ_L2_SIZE 16 /* A reasonable number of colors (opt for 64K cache) */
197 #define PQ_L2_MASK (PQ_L2_SIZE - 1)
199 #endif /* KERNEL && !KLD_MODULE */
203 * The queue array is always based on PQ_MAXL2_SIZE regardless of the actual
204 * cache size chosen in order to present a uniform interface for modules.
206 #define PQ_MAXL2_SIZE 256 /* fixed maximum (in pages) / module compat */
208 #if PQ_L2_SIZE > PQ_MAXL2_SIZE
209 #error "Illegal PQ_L2_SIZE"
214 #define PQ_INACTIVE (1 + 1*PQ_MAXL2_SIZE)
215 #define PQ_ACTIVE (2 + 1*PQ_MAXL2_SIZE)
216 #define PQ_CACHE (3 + 1*PQ_MAXL2_SIZE)
217 #define PQ_HOLD (3 + 2*PQ_MAXL2_SIZE)
218 #define PQ_COUNT (4 + 2*PQ_MAXL2_SIZE)
224 int flipflop; /* probably not the best place */
227 extern struct vpgqueues vm_page_queues[PQ_COUNT];
230 * These are the flags defined for vm_page.
232 * Note: PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
233 * not under PV management but otherwise should be treated as a
234 * normal page. Pages not under PV management cannot be paged out
235 * via the object/vm_page_t because there is no knowledge of their
236 * pte mappings, nor can they be removed from their objects via
237 * the object, and such pages are also not on any PQ queue.
239 #define PG_BUSY 0x0001 /* page is in transit (O) */
240 #define PG_WANTED 0x0002 /* someone is waiting for page (O) */
241 #define PG_WINATCFLS 0x0004 /* flush dirty page on inactive q */
242 #define PG_FICTITIOUS 0x0008 /* physical page doesn't exist (O) */
243 #define PG_WRITEABLE 0x0010 /* page is mapped writeable */
244 #define PG_MAPPED 0x0020 /* page is mapped */
245 #define PG_ZERO 0x0040 /* page is zeroed */
246 #define PG_REFERENCED 0x0080 /* page has been referenced */
247 #define PG_CLEANCHK 0x0100 /* page will be checked for cleaning */
248 #define PG_SWAPINPROG 0x0200 /* swap I/O in progress on page */
249 #define PG_NOSYNC 0x0400 /* do not collect for syncer */
250 #define PG_UNMANAGED 0x0800 /* No PV management for page */
251 #define PG_MARKER 0x1000 /* special queue marker page */
257 #define ACT_DECLINE 1
258 #define ACT_ADVANCE 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;
291 extern vm_page_t vm_page_array; /* First resident page in table */
292 extern int vm_page_array_size; /* number of vm_page_t's */
293 extern long first_page; /* first physical page number */
295 #define VM_PAGE_TO_PHYS(entry) \
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);
318 vm_page_busy(vm_page_t m)
320 KASSERT((m->flags & PG_BUSY) == 0,
321 ("vm_page_busy: page already busy!!!"));
322 vm_page_flag_set(m, PG_BUSY);
328 * wakeup anyone waiting for the page.
332 vm_page_flash(vm_page_t m)
334 if (m->flags & PG_WANTED) {
335 vm_page_flag_clear(m, PG_WANTED);
341 * Clear the PG_BUSY flag and wakeup anyone waiting for the page. This
342 * is typically the last call you make on a page before moving onto
346 vm_page_wakeup(vm_page_t m)
348 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
349 vm_page_flag_clear(m, PG_BUSY);
354 * These routines manipulate the 'soft busy' count for a page. A soft busy
355 * is almost like PG_BUSY except that it allows certain compatible operations
356 * to occur on the page while it is busy. For example, a page undergoing a
357 * write can still be mapped read-only.
360 vm_page_io_start(vm_page_t m)
362 atomic_add_char(&(m)->busy, 1);
366 vm_page_io_finish(vm_page_t m)
368 atomic_subtract_char(&m->busy, 1);
374 #if PAGE_SIZE == 4096
375 #define VM_PAGE_BITS_ALL 0xff
378 #if PAGE_SIZE == 8192
379 #define VM_PAGE_BITS_ALL 0xffff
383 * Note: the code will always use nominally free pages from the free list
384 * before trying other flag-specified sources.
386 * At least one of VM_ALLOC_NORMAL|VM_ALLOC_SYSTEM|VM_ALLOC_INTERRUPT
387 * must be specified. VM_ALLOC_RETRY may only be specified if VM_ALLOC_NORMAL
390 #define VM_ALLOC_NORMAL 0x01 /* ok to use cache pages */
391 #define VM_ALLOC_SYSTEM 0x02 /* ok to exhaust most of free list */
392 #define VM_ALLOC_INTERRUPT 0x04 /* ok to exhaust entire free list */
393 #define VM_ALLOC_ZERO 0x08 /* req pre-zero'd memory if avail */
394 #define VM_ALLOC_RETRY 0x80 /* indefinite block (vm_page_grab()) */
396 void vm_page_unhold(vm_page_t mem);
397 void vm_page_activate (vm_page_t);
398 vm_page_t vm_page_alloc (vm_object_t, vm_pindex_t, int);
399 vm_page_t vm_page_grab (vm_object_t, vm_pindex_t, int);
400 void vm_page_cache (vm_page_t);
401 int vm_page_try_to_cache (vm_page_t);
402 int vm_page_try_to_free (vm_page_t);
403 void vm_page_dontneed (vm_page_t);
404 void vm_page_deactivate (vm_page_t);
405 void vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
406 vm_page_t vm_page_lookup (vm_object_t, vm_pindex_t);
407 void vm_page_remove (vm_page_t);
408 void vm_page_rename (vm_page_t, vm_object_t, vm_pindex_t);
409 vm_offset_t vm_page_startup (vm_offset_t);
410 vm_page_t vm_add_new_page (vm_paddr_t pa);
411 void vm_page_unmanage (vm_page_t);
412 void vm_page_unwire (vm_page_t, int);
413 void vm_page_wire (vm_page_t);
414 void vm_page_unqueue (vm_page_t);
415 void vm_page_unqueue_nowakeup (vm_page_t);
416 void vm_page_set_validclean (vm_page_t, int, int);
417 void vm_page_set_dirty (vm_page_t, int, int);
418 void vm_page_clear_dirty (vm_page_t, int, int);
419 void vm_page_set_invalid (vm_page_t, int, int);
420 int vm_page_is_valid (vm_page_t, int, int);
421 void vm_page_test_dirty (vm_page_t);
422 int vm_page_bits (int, int);
423 vm_page_t vm_page_list_find(int basequeue, int index, boolean_t prefer_zero);
424 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
425 void vm_page_free_toq(vm_page_t m);
426 vm_offset_t vm_contig_pg_kmap(int, u_long, vm_map_t, int);
427 void vm_contig_pg_free(int, u_long);
430 * Holding a page keeps it from being reused. Other parts of the system
431 * can still disassociate the page from its current object and free it, or
432 * perform read or write I/O on it and/or otherwise manipulate the page,
433 * but if the page is held the VM system will leave the page and its data
434 * intact and not reuse the page for other purposes until the last hold
435 * reference is released. (see vm_page_wire() if you want to prevent the
436 * page from being disassociated from its object too).
438 * This routine must be called while at splvm() or better.
440 * The caller must still validate the contents of the page and, if necessary,
441 * wait for any pending I/O (e.g. vm_page_sleep_busy() loop) to complete
442 * before manipulating the page.
445 vm_page_hold(vm_page_t mem)
451 * Reduce the protection of a page. This routine never raises the
452 * protection and therefore can be safely called if the page is already
453 * at VM_PROT_NONE (it will be a NOP effectively ).
455 * VM_PROT_NONE will remove all user mappings of a page. This is often
456 * necessary when a page changes state (for example, turns into a copy-on-write
457 * page or needs to be frozen for write I/O) in order to force a fault, or
458 * to force a page's dirty bits to be synchronized and avoid hardware
459 * (modified/accessed) bit update races with pmap changes.
461 * Since 'prot' is usually a constant, this inline usually winds up optimizing
462 * out the primary conditional.
465 vm_page_protect(vm_page_t mem, int prot)
467 if (prot == VM_PROT_NONE) {
468 if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) {
469 pmap_page_protect(mem, VM_PROT_NONE);
470 vm_page_flag_clear(mem, PG_WRITEABLE|PG_MAPPED);
472 } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) {
473 pmap_page_protect(mem, VM_PROT_READ);
474 vm_page_flag_clear(mem, PG_WRITEABLE);
479 * Zero-fill the specified page. The entire contents of the page will be
482 static __inline boolean_t
483 vm_page_zero_fill(vm_page_t m)
485 pmap_zero_page(VM_PAGE_TO_PHYS(m));
490 * Copy the contents of src_m to dest_m. The pages must be stable but spl
491 * and other protections depend on context.
494 vm_page_copy(vm_page_t src_m, vm_page_t dest_m)
496 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m));
497 dest_m->valid = VM_PAGE_BITS_ALL;
501 * Free a page. The page must be marked BUSY.
503 * The clearing of PG_ZERO is a temporary safety until the code can be
504 * reviewed to determine that PG_ZERO is being properly cleared on
505 * write faults or maps. PG_ZERO was previously cleared in
509 vm_page_free(vm_page_t m)
511 vm_page_flag_clear(m, PG_ZERO);
516 * Free a page to the zerod-pages queue
519 vm_page_free_zero(vm_page_t m)
521 vm_page_flag_set(m, PG_ZERO);
526 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
527 * m->busy is zero. Returns TRUE if it had to sleep ( including if
528 * it almost had to sleep and made temporary spl*() mods), FALSE
531 * This routine assumes that interrupts can only remove the busy
532 * status from a page, not set the busy status or change it from
533 * PG_BUSY to m->busy or vise versa (which would create a timing
536 * Note: as an inline, 'also_m_busy' is usually a constant and well
540 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
542 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
544 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
546 * Page is busy. Wait and retry.
548 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
549 tsleep(m, 0, msg, 0);
559 * Make page all dirty
562 _vm_page_dirty(vm_page_t m, const char *info)
564 int pqtype = m->queue - m->pc;
565 KASSERT(pqtype != PQ_CACHE && pqtype != PQ_FREE,
566 ("vm_page_dirty: page in free/cache queue!"));
567 m->dirty = VM_PAGE_BITS_ALL;
570 #define vm_page_dirty(m) _vm_page_dirty(m, __FUNCTION__)
573 * Set page to not be dirty. Note: does not clear pmap modify bits .
576 vm_page_undirty(vm_page_t m)
582 #endif /* !_VM_PAGE_ */