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 $
68 * Resident memory system definitions.
71 #ifndef _VM_VM_PAGE_H_
72 #define _VM_VM_PAGE_H_
74 #if !defined(KLD_MODULE) && defined(_KERNEL)
75 #include "opt_vmpage.h"
79 #include <sys/types.h>
84 #ifndef _MACHINE_PMAP_H_
85 #include <machine/pmap.h>
90 #include <machine/atomic.h>
95 #include <sys/systm.h>
97 #ifndef _SYS_THREAD2_H_
98 #include <sys/thread2.h>
102 #include <machine/vmparam.h>
107 typedef enum vm_page_event { VMEVENT_NONE, VMEVENT_COW } vm_page_event_t;
109 struct vm_page_action {
110 LIST_ENTRY(vm_page_action) entry;
112 vm_page_event_t event;
113 void (*func)(struct vm_page *,
114 struct vm_page_action *);
118 typedef struct vm_page_action *vm_page_action_t;
121 * Management of resident (logical) pages.
123 * A small structure is kept for each resident
124 * page, indexed by page number. Each structure
125 * is an element of several lists:
127 * A hash table bucket used to quickly
128 * perform object/offset lookups
130 * A list of all pages for a given object,
131 * so they can be quickly deactivated at
132 * time of deallocation.
134 * An ordered list of pages due for pageout.
136 * In addition, the structure contains the object
137 * and offset to which this page belongs (for pageout),
138 * and sundry status bits.
140 * Fields in this structure are locked either by the lock on the
141 * object that the page belongs to (O) or by the lock on the page
144 * The 'valid' and 'dirty' fields are distinct. A page may have dirty
145 * bits set without having associated valid bits set. This is used by
146 * NFS to implement piecemeal writes.
149 TAILQ_HEAD(pglist, vm_page);
153 int rb_vm_page_compare(struct vm_page *, struct vm_page *);
155 struct vm_page_rb_tree;
156 RB_PROTOTYPE2(vm_page_rb_tree, vm_page, rb_entry, rb_vm_page_compare, vm_pindex_t);
159 TAILQ_ENTRY(vm_page) pageq; /* vm_page_queues[] list (P) */
160 RB_ENTRY(vm_page) rb_entry; /* Red-Black tree based at object */
162 struct vm_object *object; /* which object am I in (O,P)*/
163 vm_pindex_t pindex; /* offset into object (O,P) */
164 vm_paddr_t phys_addr; /* physical address of page */
165 struct md_page md; /* machine dependant stuff */
166 u_short queue; /* page queue index */
167 u_short pc; /* page color */
168 u_char act_count; /* page usage count */
169 u_char busy; /* page busy count */
172 u_int32_t flags; /* see below */
173 u_int wire_count; /* wired down maps refs (P) */
174 int hold_count; /* page hold count */
177 * NOTE that these must support one bit per DEV_BSIZE in a page!!!
178 * so, on normal X86 kernels, they must be at least 8 bits wide.
180 u_char valid; /* map of valid DEV_BSIZE chunks */
181 u_char dirty; /* map of dirty DEV_BSIZE chunks */
183 int ku_pagecnt; /* kmalloc helper */
185 const char *busy_func;
190 #ifndef __VM_PAGE_T_DEFINED__
191 #define __VM_PAGE_T_DEFINED__
192 typedef struct vm_page *vm_page_t;
196 * Page coloring parameters. We default to a middle of the road optimization.
197 * Larger selections would not really hurt us but if a machine does not have
198 * a lot of memory it could cause vm_page_alloc() to eat more cpu cycles
199 * looking for free pages.
201 * Page coloring cannot be disabled. Modules do not have access to most PQ
202 * constants because they can change between builds.
204 #if defined(_KERNEL) && !defined(KLD_MODULE)
206 #if !defined(PQ_CACHESIZE)
207 #define PQ_CACHESIZE 256 /* max is 1024 (MB) */
210 #if PQ_CACHESIZE >= 1024
211 #define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */
212 #define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */
213 #define PQ_L2_SIZE 256 /* A number of colors opt for 1M cache */
215 #elif PQ_CACHESIZE >= 512
216 #define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */
217 #define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */
218 #define PQ_L2_SIZE 128 /* A number of colors opt for 512K cache */
220 #elif PQ_CACHESIZE >= 256
221 #define PQ_PRIME1 13 /* Prime number somewhat less than PQ_HASH_SIZE */
222 #define PQ_PRIME2 7 /* Prime number somewhat less than PQ_HASH_SIZE */
223 #define PQ_L2_SIZE 64 /* A number of colors opt for 256K cache */
225 #elif PQ_CACHESIZE >= 128
226 #define PQ_PRIME1 9 /* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */
227 #define PQ_PRIME2 5 /* Prime number somewhat less than PQ_HASH_SIZE */
228 #define PQ_L2_SIZE 32 /* A number of colors opt for 128k cache */
231 #define PQ_PRIME1 5 /* Prime number somewhat less than PQ_HASH_SIZE */
232 #define PQ_PRIME2 3 /* Prime number somewhat less than PQ_HASH_SIZE */
233 #define PQ_L2_SIZE 16 /* A reasonable number of colors (opt for 64K cache) */
237 #define PQ_L2_MASK (PQ_L2_SIZE - 1)
239 #endif /* KERNEL && !KLD_MODULE */
243 * The queue array is always based on PQ_MAXL2_SIZE regardless of the actual
244 * cache size chosen in order to present a uniform interface for modules.
246 #define PQ_MAXL2_SIZE 256 /* fixed maximum (in pages) / module compat */
248 #if PQ_L2_SIZE > PQ_MAXL2_SIZE
249 #error "Illegal PQ_L2_SIZE"
254 #define PQ_INACTIVE (1 + 1*PQ_MAXL2_SIZE)
255 #define PQ_ACTIVE (2 + 1*PQ_MAXL2_SIZE)
256 #define PQ_CACHE (3 + 1*PQ_MAXL2_SIZE)
257 #define PQ_HOLD (3 + 2*PQ_MAXL2_SIZE)
258 #define PQ_COUNT (4 + 2*PQ_MAXL2_SIZE)
265 struct rb_vm_page_scan_info {
266 vm_pindex_t start_pindex;
267 vm_pindex_t end_pindex;
273 vm_pindex_t backing_offset_index;
274 struct vm_object *object;
275 struct vm_object *backing_object;
276 struct vm_page *mpte;
281 int rb_vm_page_scancmp(struct vm_page *, void *);
287 int flipflop; /* probably not the best place */
290 extern struct vpgqueues vm_page_queues[PQ_COUNT];
293 * These are the flags defined for vm_page.
295 * PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
296 * not under PV management but otherwise should be treated as a
297 * normal page. Pages not under PV management cannot be paged out
298 * via the object/vm_page_t because there is no knowledge of their
299 * pte mappings, nor can they be removed from their objects via
300 * the object, and such pages are also not on any PQ queue. The
301 * PG_MAPPED and PG_WRITEABLE flags are not applicable.
303 * PG_MAPPED only applies to managed pages, indicating whether the page
304 * is mapped onto one or more pmaps. A page might still be mapped to
305 * special pmaps in an unmanaged fashion, for example when mapped into a
306 * buffer cache buffer, without setting PG_MAPPED.
308 * PG_WRITEABLE indicates that there may be a writeable managed pmap entry
309 * somewhere, and that the page can be dirtied by hardware at any time
310 * and may have to be tested for that. The modified bit in unmanaged
311 * mappings or in the special clean map is not tested.
313 * PG_SWAPPED indicates that the page is backed by a swap block. Any
314 * VM object type other than OBJT_DEFAULT can have swap-backed pages now.
316 #define PG_BUSY 0x00000001 /* page is in transit (O) */
317 #define PG_WANTED 0x00000002 /* someone is waiting for page (O) */
318 #define PG_WINATCFLS 0x00000004 /* flush dirty page on inactive q */
319 #define PG_FICTITIOUS 0x00000008 /* physical page doesn't exist (O) */
320 #define PG_WRITEABLE 0x00000010 /* page is writeable */
321 #define PG_MAPPED 0x00000020 /* page is mapped (managed) */
322 #define PG_ZERO 0x00000040 /* page is zeroed */
323 #define PG_REFERENCED 0x00000080 /* page has been referenced */
324 #define PG_CLEANCHK 0x00000100 /* page will be checked for cleaning */
325 #define PG_SWAPINPROG 0x00000200 /* swap I/O in progress on page */
326 #define PG_NOSYNC 0x00000400 /* do not collect for syncer */
327 #define PG_UNMANAGED 0x00000800 /* No PV management for page */
328 #define PG_MARKER 0x00001000 /* special queue marker page */
329 #define PG_RAM 0x00002000 /* read ahead mark */
330 #define PG_SWAPPED 0x00004000 /* backed by swap */
331 #define PG_NOTMETA 0x00008000 /* do not back with swap */
332 #define PG_ACTIONLIST 0x00010000 /* lookaside action list present */
333 /* u_short, only 16 flag bits */
339 #define ACT_DECLINE 1
340 #define ACT_ADVANCE 3
346 * Each pageable resident page falls into one of four lists:
349 * Available for allocation now.
351 * The following are all LRU sorted:
354 * Almost available for allocation. Still in an
355 * object, but clean and immediately freeable at
356 * non-interrupt times.
359 * Low activity, candidates for reclamation.
360 * This is the list of pages that should be
364 * Pages that are "active" i.e. they have been
365 * recently referenced.
368 * Pages that are really free and have been pre-zeroed
372 extern int vm_page_zero_count;
373 extern struct vm_page *vm_page_array; /* First resident page in table */
374 extern int vm_page_array_size; /* number of vm_page_t's */
375 extern long first_page; /* first physical page number */
377 #define VM_PAGE_TO_PHYS(entry) \
380 #define PHYS_TO_VM_PAGE(pa) \
381 (&vm_page_array[atop(pa) - first_page])
384 * Functions implemented as macros
388 vm_page_flag_set(vm_page_t m, unsigned int bits)
390 atomic_set_int(&(m)->flags, bits);
394 vm_page_flag_clear(vm_page_t m, unsigned int bits)
396 atomic_clear_int(&(m)->flags, bits);
402 _vm_page_busy(vm_page_t m, const char *func, int lineno)
404 ASSERT_LWKT_TOKEN_HELD(&vm_token);
405 KASSERT((m->flags & PG_BUSY) == 0,
406 ("vm_page_busy: page already busy!!!"));
407 vm_page_flag_set(m, PG_BUSY);
409 m->busy_line = lineno;
412 #define vm_page_busy(m) _vm_page_busy(m, __func__, __LINE__)
417 vm_page_busy(vm_page_t m)
419 ASSERT_LWKT_TOKEN_HELD(&vm_token);
420 KASSERT((m->flags & PG_BUSY) == 0,
421 ("vm_page_busy: page already busy!!!"));
422 vm_page_flag_set(m, PG_BUSY);
430 * wakeup anyone waiting for the page.
434 vm_page_flash(vm_page_t m)
436 lwkt_gettoken(&vm_token);
437 if (m->flags & PG_WANTED) {
438 vm_page_flag_clear(m, PG_WANTED);
441 lwkt_reltoken(&vm_token);
445 * Clear the PG_BUSY flag and wakeup anyone waiting for the page. This
446 * is typically the last call you make on a page before moving onto
450 vm_page_wakeup(vm_page_t m)
452 KASSERT(m->flags & PG_BUSY, ("vm_page_wakeup: page not busy!!!"));
453 vm_page_flag_clear(m, PG_BUSY);
458 * These routines manipulate the 'soft busy' count for a page. A soft busy
459 * is almost like PG_BUSY except that it allows certain compatible operations
460 * to occur on the page while it is busy. For example, a page undergoing a
461 * write can still be mapped read-only.
464 vm_page_io_start(vm_page_t m)
466 atomic_add_char(&(m)->busy, 1);
470 vm_page_io_finish(vm_page_t m)
472 atomic_subtract_char(&m->busy, 1);
478 #if PAGE_SIZE == 4096
479 #define VM_PAGE_BITS_ALL 0xff
483 * Note: the code will always use nominally free pages from the free list
484 * before trying other flag-specified sources.
486 * At least one of VM_ALLOC_NORMAL|VM_ALLOC_SYSTEM|VM_ALLOC_INTERRUPT
487 * must be specified. VM_ALLOC_RETRY may only be specified if VM_ALLOC_NORMAL
490 #define VM_ALLOC_NORMAL 0x01 /* ok to use cache pages */
491 #define VM_ALLOC_SYSTEM 0x02 /* ok to exhaust most of free list */
492 #define VM_ALLOC_INTERRUPT 0x04 /* ok to exhaust entire free list */
493 #define VM_ALLOC_ZERO 0x08 /* req pre-zero'd memory if avail */
494 #define VM_ALLOC_QUICK 0x10 /* like NORMAL but do not use cache */
495 #define VM_ALLOC_RETRY 0x80 /* indefinite block (vm_page_grab()) */
497 void vm_page_hold(vm_page_t);
498 void vm_page_unhold(vm_page_t);
499 void vm_page_activate (vm_page_t);
500 vm_page_t vm_page_alloc (struct vm_object *, vm_pindex_t, int);
501 vm_page_t vm_page_grab (struct vm_object *, vm_pindex_t, int);
502 void vm_page_cache (vm_page_t);
503 int vm_page_try_to_cache (vm_page_t);
504 int vm_page_try_to_free (vm_page_t);
505 void vm_page_dontneed (vm_page_t);
506 void vm_page_deactivate (vm_page_t);
507 void vm_page_insert (vm_page_t, struct vm_object *, vm_pindex_t);
508 vm_page_t vm_page_lookup (struct vm_object *, vm_pindex_t);
509 void vm_page_remove (vm_page_t);
510 void vm_page_rename (vm_page_t, struct vm_object *, vm_pindex_t);
511 void vm_page_startup (void);
512 void vm_page_unmanage (vm_page_t);
513 void vm_page_unwire (vm_page_t, int);
514 void vm_page_wire (vm_page_t);
515 void vm_page_unqueue (vm_page_t);
516 void vm_page_unqueue_nowakeup (vm_page_t);
517 void vm_page_set_validclean (vm_page_t, int, int);
518 void vm_page_set_validdirty (vm_page_t, int, int);
519 void vm_page_set_valid (vm_page_t, int, int);
520 void vm_page_set_dirty (vm_page_t, int, int);
521 void vm_page_clear_dirty (vm_page_t, int, int);
522 void vm_page_set_invalid (vm_page_t, int, int);
523 int vm_page_is_valid (vm_page_t, int, int);
524 void vm_page_test_dirty (vm_page_t);
525 int vm_page_bits (int, int);
526 vm_page_t vm_page_list_find(int basequeue, int index, boolean_t prefer_zero);
527 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
528 void vm_page_free_toq(vm_page_t m);
529 vm_page_t vm_page_free_fromq_fast(void);
530 void vm_page_event_internal(vm_page_t, vm_page_event_t);
531 void vm_page_dirty(vm_page_t m);
532 void vm_page_register_action(vm_page_action_t action, vm_page_event_t event);
533 void vm_page_unregister_action(vm_page_action_t action);
536 * Reduce the protection of a page. This routine never raises the
537 * protection and therefore can be safely called if the page is already
538 * at VM_PROT_NONE (it will be a NOP effectively ).
540 * VM_PROT_NONE will remove all user mappings of a page. This is often
541 * necessary when a page changes state (for example, turns into a copy-on-write
542 * page or needs to be frozen for write I/O) in order to force a fault, or
543 * to force a page's dirty bits to be synchronized and avoid hardware
544 * (modified/accessed) bit update races with pmap changes.
546 * Since 'prot' is usually a constant, this inline usually winds up optimizing
547 * out the primary conditional.
549 * WARNING: VM_PROT_NONE can block, but will loop until all mappings have
550 * been cleared. Callers should be aware that other page related elements
551 * might have changed, however.
554 vm_page_protect(vm_page_t mem, int prot)
556 if (prot == VM_PROT_NONE) {
557 if (mem->flags & (PG_WRITEABLE|PG_MAPPED)) {
558 pmap_page_protect(mem, VM_PROT_NONE);
559 /* PG_WRITEABLE & PG_MAPPED cleared by call */
561 } else if ((prot == VM_PROT_READ) && (mem->flags & PG_WRITEABLE)) {
562 pmap_page_protect(mem, VM_PROT_READ);
563 /* PG_WRITEABLE cleared by call */
568 * Zero-fill the specified page. The entire contents of the page will be
571 static __inline boolean_t
572 vm_page_zero_fill(vm_page_t m)
574 pmap_zero_page(VM_PAGE_TO_PHYS(m));
579 * Copy the contents of src_m to dest_m. The pages must be stable but spl
580 * and other protections depend on context.
583 vm_page_copy(vm_page_t src_m, vm_page_t dest_m)
585 pmap_copy_page(VM_PAGE_TO_PHYS(src_m), VM_PAGE_TO_PHYS(dest_m));
586 dest_m->valid = VM_PAGE_BITS_ALL;
587 dest_m->dirty = VM_PAGE_BITS_ALL;
591 * Free a page. The page must be marked BUSY.
593 * The clearing of PG_ZERO is a temporary safety until the code can be
594 * reviewed to determine that PG_ZERO is being properly cleared on
595 * write faults or maps. PG_ZERO was previously cleared in
599 vm_page_free(vm_page_t m)
601 vm_page_flag_clear(m, PG_ZERO);
606 * Free a page to the zerod-pages queue
609 vm_page_free_zero(vm_page_t m)
612 /* JG DEBUG64 We check if the page is really zeroed. */
613 char *p = (char *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
616 for (i = 0; i < PAGE_SIZE; i++) {
618 panic("non-zero page in vm_page_free_zero()");
623 vm_page_flag_set(m, PG_ZERO);
628 * Wait until page is no longer PG_BUSY or (if also_m_busy is TRUE)
629 * m->busy is zero. Returns TRUE if it had to sleep ( including if
630 * it almost had to sleep and made temporary spl*() mods), FALSE
633 * This routine assumes that interrupts can only remove the busy
634 * status from a page, not set the busy status or change it from
635 * PG_BUSY to m->busy or vise versa (which would create a timing
638 * Note: as an inline, 'also_m_busy' is usually a constant and well
642 vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg)
644 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
645 lwkt_gettoken(&vm_token);
646 if ((m->flags & PG_BUSY) || (also_m_busy && m->busy)) {
648 * Page is busy. Wait and retry.
650 vm_page_flag_set(m, PG_WANTED | PG_REFERENCED);
651 tsleep(m, 0, msg, 0);
653 lwkt_reltoken(&vm_token);
661 * Set page to not be dirty. Note: does not clear pmap modify bits .
664 vm_page_undirty(vm_page_t m)
670 #endif /* !_VM_VM_PAGE_H_ */