kernel - Implement CPU localization hinting for low level page allocations
[dragonfly.git] / sys / vm / vm_page.h
CommitLineData
984263bc
MD
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.
2c64e990 16 * 3. Neither the name of the University nor the names of its contributors
984263bc
MD
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: src/sys/vm/vm_page.h,v 1.75.2.8 2002/03/06 01:07:09 dillon Exp $
61 */
62
63/*
64 * Resident memory system definitions.
65 */
66
1bd40720
MD
67#ifndef _VM_VM_PAGE_H_
68#define _VM_VM_PAGE_H_
984263bc 69
1bd40720
MD
70#ifndef _SYS_TYPES_H_
71#include <sys/types.h>
72#endif
1f804340
MD
73#ifndef _SYS_TREE_H_
74#include <sys/tree.h>
75#endif
1bd40720
MD
76#ifndef _MACHINE_PMAP_H_
77#include <machine/pmap.h>
78#endif
79#ifndef _VM_PMAP_H_
984263bc 80#include <vm/pmap.h>
1bd40720 81#endif
984263bc 82#include <machine/atomic.h>
1bd40720 83
668b1228 84#ifdef _KERNEL
1bd40720 85
03d6a592
MD
86#ifndef _SYS_SYSTM_H_
87#include <sys/systm.h>
88#endif
1bd40720 89#ifndef _SYS_THREAD2_H_
cdd46d2e 90#include <sys/thread2.h>
668b1228 91#endif
984263bc 92
b2b3ffcd 93#ifdef __x86_64__
973c11b9
MD
94#include <machine/vmparam.h>
95#endif
96
1bd40720
MD
97#endif
98
10192bae
MD
99typedef enum vm_page_event { VMEVENT_NONE, VMEVENT_COW } vm_page_event_t;
100
101struct vm_page_action {
102 LIST_ENTRY(vm_page_action) entry;
906c754c 103 struct vm_page *m;
10192bae
MD
104 vm_page_event_t event;
105 void (*func)(struct vm_page *,
106 struct vm_page_action *);
107 void *data;
108};
109
110typedef struct vm_page_action *vm_page_action_t;
111
984263bc
MD
112/*
113 * Management of resident (logical) pages.
114 *
115 * A small structure is kept for each resident
116 * page, indexed by page number. Each structure
117 * is an element of several lists:
118 *
119 * A hash table bucket used to quickly
120 * perform object/offset lookups
121 *
122 * A list of all pages for a given object,
123 * so they can be quickly deactivated at
124 * time of deallocation.
125 *
126 * An ordered list of pages due for pageout.
127 *
128 * In addition, the structure contains the object
129 * and offset to which this page belongs (for pageout),
130 * and sundry status bits.
131 *
132 * Fields in this structure are locked either by the lock on the
133 * object that the page belongs to (O) or by the lock on the page
134 * queues (P).
135 *
136 * The 'valid' and 'dirty' fields are distinct. A page may have dirty
137 * bits set without having associated valid bits set. This is used by
138 * NFS to implement piecemeal writes.
139 */
140
141TAILQ_HEAD(pglist, vm_page);
142
03d6a592
MD
143struct vm_object;
144
1f804340
MD
145int rb_vm_page_compare(struct vm_page *, struct vm_page *);
146
147struct vm_page_rb_tree;
148RB_PROTOTYPE2(vm_page_rb_tree, vm_page, rb_entry, rb_vm_page_compare, vm_pindex_t);
149
984263bc 150struct vm_page {
a441ad78 151 TAILQ_ENTRY(vm_page) pageq; /* vm_page_queues[] list (P) */
1f804340 152 RB_ENTRY(vm_page) rb_entry; /* Red-Black tree based at object */
984263bc 153
03d6a592 154 struct vm_object *object; /* which object am I in (O,P)*/
984263bc 155 vm_pindex_t pindex; /* offset into object (O,P) */
6ef943a3 156 vm_paddr_t phys_addr; /* physical address of page */
984263bc
MD
157 struct md_page md; /* machine dependant stuff */
158 u_short queue; /* page queue index */
a441ad78 159 u_short pc; /* page color */
984263bc
MD
160 u_char act_count; /* page usage count */
161 u_char busy; /* page busy count */
b524ca76 162 u_char pat_mode; /* hardware page attribute */
906c754c
MD
163 u_char unused02;
164 u_int32_t flags; /* see below */
10192bae
MD
165 u_int wire_count; /* wired down maps refs (P) */
166 int hold_count; /* page hold count */
a441ad78
MD
167
168 /*
169 * NOTE that these must support one bit per DEV_BSIZE in a page!!!
170 * so, on normal X86 kernels, they must be at least 8 bits wide.
171 */
984263bc
MD
172 u_char valid; /* map of valid DEV_BSIZE chunks */
173 u_char dirty; /* map of dirty DEV_BSIZE chunks */
d797ef1c 174
722871d3 175 int ku_pagecnt; /* kmalloc helper */
d0aa00e8
MD
176#ifdef VM_PAGE_DEBUG
177 const char *busy_func;
178 int busy_line;
179#endif
984263bc
MD
180};
181
b12defdc
MD
182#ifdef VM_PAGE_DEBUG
183#define VM_PAGE_DEBUG_EXT(name) name ## _debug
184#define VM_PAGE_DEBUG_ARGS , const char *func, int lineno
185#else
186#define VM_PAGE_DEBUG_EXT(name) name
187#define VM_PAGE_DEBUG_ARGS
188#endif
189
03d6a592
MD
190#ifndef __VM_PAGE_T_DEFINED__
191#define __VM_PAGE_T_DEFINED__
192typedef struct vm_page *vm_page_t;
193#endif
194
984263bc 195/*
51c99c61
MD
196 * Page coloring parameters. We use generous parameters designed to
197 * statistically spread pages over available cpu cache space. This has
198 * become less important over time as cache associativity is higher
199 * in modern times but we still use the core algorithm to help reduce
200 * lock contention between cpus.
74232d8e 201 *
51c99c61 202 * Page coloring cannot be disabled.
984263bc 203 */
984263bc 204
984263bc
MD
205#define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */
206#define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */
207#define PQ_L2_SIZE 256 /* A number of colors opt for 1M cache */
74232d8e 208#define PQ_L2_MASK (PQ_L2_SIZE - 1)
984263bc 209
74232d8e 210#define PQ_NONE 0
51c99c61
MD
211#define PQ_FREE (1 + 0*PQ_L2_SIZE)
212#define PQ_INACTIVE (1 + 1*PQ_L2_SIZE)
213#define PQ_ACTIVE (1 + 2*PQ_L2_SIZE)
214#define PQ_CACHE (1 + 3*PQ_L2_SIZE)
215#define PQ_HOLD (1 + 4*PQ_L2_SIZE)
216#define PQ_COUNT (1 + 5*PQ_L2_SIZE)
984263bc 217
1f804340
MD
218/*
219 * Scan support
220 */
221struct vm_map;
222
223struct rb_vm_page_scan_info {
224 vm_pindex_t start_pindex;
225 vm_pindex_t end_pindex;
226 int limit;
227 int desired;
228 int error;
229 int pagerflags;
e674353b
MD
230 int count;
231 int unused01;
1f804340
MD
232 vm_offset_t addr;
233 vm_pindex_t backing_offset_index;
234 struct vm_object *object;
235 struct vm_object *backing_object;
236 struct vm_page *mpte;
237 struct pmap *pmap;
238 struct vm_map *map;
239};
240
241int rb_vm_page_scancmp(struct vm_page *, void *);
242
984263bc
MD
243struct vpgqueues {
244 struct pglist pl;
245 int *cnt;
246 int lcnt;
161399b3 247 int flipflop; /* probably not the best place */
b12defdc 248 struct spinlock spin;
bca42d4f
MD
249 int zero_count; /* only applies to PQ_FREE queues */
250 int unused00;
b12defdc 251 char unused[64 - sizeof(struct pglist) -
bca42d4f 252 sizeof(int *) - sizeof(int) * 4];
984263bc
MD
253};
254
255extern struct vpgqueues vm_page_queues[PQ_COUNT];
256
984263bc
MD
257/*
258 * These are the flags defined for vm_page.
259 *
17cde63e
MD
260 * PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
261 * not under PV management but otherwise should be treated as a
262 * normal page. Pages not under PV management cannot be paged out
263 * via the object/vm_page_t because there is no knowledge of their
264 * pte mappings, nor can they be removed from their objects via
265 * the object, and such pages are also not on any PQ queue. The
266 * PG_MAPPED and PG_WRITEABLE flags are not applicable.
267 *
268 * PG_MAPPED only applies to managed pages, indicating whether the page
269 * is mapped onto one or more pmaps. A page might still be mapped to
270 * special pmaps in an unmanaged fashion, for example when mapped into a
271 * buffer cache buffer, without setting PG_MAPPED.
272 *
273 * PG_WRITEABLE indicates that there may be a writeable managed pmap entry
274 * somewhere, and that the page can be dirtied by hardware at any time
275 * and may have to be tested for that. The modified bit in unmanaged
276 * mappings or in the special clean map is not tested.
67803f3e
MD
277 *
278 * PG_SWAPPED indicates that the page is backed by a swap block. Any
279 * VM object type other than OBJT_DEFAULT can have swap-backed pages now.
b12defdc
MD
280 *
281 * PG_SBUSY is set when m->busy != 0. PG_SBUSY and m->busy are only modified
282 * when the page is PG_BUSY.
984263bc 283 */
906c754c
MD
284#define PG_BUSY 0x00000001 /* page is in transit (O) */
285#define PG_WANTED 0x00000002 /* someone is waiting for page (O) */
286#define PG_WINATCFLS 0x00000004 /* flush dirty page on inactive q */
287#define PG_FICTITIOUS 0x00000008 /* physical page doesn't exist (O) */
288#define PG_WRITEABLE 0x00000010 /* page is writeable */
289#define PG_MAPPED 0x00000020 /* page is mapped (managed) */
afd2da4d 290#define PG_UNUSED0040 0x00000040
906c754c
MD
291#define PG_REFERENCED 0x00000080 /* page has been referenced */
292#define PG_CLEANCHK 0x00000100 /* page will be checked for cleaning */
293#define PG_SWAPINPROG 0x00000200 /* swap I/O in progress on page */
294#define PG_NOSYNC 0x00000400 /* do not collect for syncer */
295#define PG_UNMANAGED 0x00000800 /* No PV management for page */
296#define PG_MARKER 0x00001000 /* special queue marker page */
297#define PG_RAM 0x00002000 /* read ahead mark */
298#define PG_SWAPPED 0x00004000 /* backed by swap */
299#define PG_NOTMETA 0x00008000 /* do not back with swap */
300#define PG_ACTIONLIST 0x00010000 /* lookaside action list present */
b12defdc 301#define PG_SBUSY 0x00020000 /* soft-busy also set */
9bf025db 302#define PG_NEED_COMMIT 0x00040000 /* clean page requires commit */
984263bc
MD
303
304/*
305 * Misc constants.
306 */
307
308#define ACT_DECLINE 1
309#define ACT_ADVANCE 3
310#define ACT_INIT 5
311#define ACT_MAX 64
984263bc
MD
312
313#ifdef _KERNEL
314/*
315 * Each pageable resident page falls into one of four lists:
316 *
317 * free
318 * Available for allocation now.
319 *
320 * The following are all LRU sorted:
321 *
322 * cache
323 * Almost available for allocation. Still in an
324 * object, but clean and immediately freeable at
325 * non-interrupt times.
326 *
327 * inactive
328 * Low activity, candidates for reclamation.
329 * This is the list of pages that should be
330 * paged out next.
331 *
332 * active
333 * Pages that are "active" i.e. they have been
334 * recently referenced.
335 *
336 * zero
337 * Pages that are really free and have been pre-zeroed
338 *
339 */
340
03d6a592 341extern struct vm_page *vm_page_array; /* First resident page in table */
984263bc
MD
342extern int vm_page_array_size; /* number of vm_page_t's */
343extern long first_page; /* first physical page number */
344
a441ad78
MD
345#define VM_PAGE_TO_PHYS(entry) \
346 ((entry)->phys_addr)
984263bc 347
a441ad78
MD
348#define PHYS_TO_VM_PAGE(pa) \
349 (&vm_page_array[atop(pa) - first_page])
984263bc 350
984263bc 351
984263bc
MD
352#if PAGE_SIZE == 4096
353#define VM_PAGE_BITS_ALL 0xff
354#endif
355
dc1fd4b3
MD
356/*
357 * Note: the code will always use nominally free pages from the free list
358 * before trying other flag-specified sources.
359 *
360 * At least one of VM_ALLOC_NORMAL|VM_ALLOC_SYSTEM|VM_ALLOC_INTERRUPT
361 * must be specified. VM_ALLOC_RETRY may only be specified if VM_ALLOC_NORMAL
362 * is also specified.
363 */
54341a3b
MD
364#define VM_ALLOC_NORMAL 0x0001 /* ok to use cache pages */
365#define VM_ALLOC_SYSTEM 0x0002 /* ok to exhaust most of free list */
366#define VM_ALLOC_INTERRUPT 0x0004 /* ok to exhaust entire free list */
367#define VM_ALLOC_ZERO 0x0008 /* req pre-zero'd memory if avail */
368#define VM_ALLOC_QUICK 0x0010 /* like NORMAL but do not use cache */
369#define VM_ALLOC_FORCE_ZERO 0x0020 /* zero page even if already valid */
370#define VM_ALLOC_NULL_OK 0x0040 /* ok to return NULL on collision */
371#define VM_ALLOC_RETRY 0x0080 /* indefinite block (vm_page_grab()) */
372#define VM_ALLOC_USE_GD 0x0100 /* use per-gd cache */
070a58b3
MD
373#define VM_ALLOC_CPU_SPEC 0x0200
374
375#define VM_ALLOC_CPU_SHIFT 16
376#define VM_ALLOC_CPU(n) (((n) << VM_ALLOC_CPU_SHIFT) | \
377 VM_ALLOC_CPU_SPEC)
378#define VM_ALLOC_GETCPU(flags) ((flags) >> VM_ALLOC_CPU_SHIFT)
984263bc 379
b12defdc
MD
380void vm_page_queue_spin_lock(vm_page_t);
381void vm_page_queues_spin_lock(u_short);
382void vm_page_and_queue_spin_lock(vm_page_t);
383
384void vm_page_queue_spin_unlock(vm_page_t);
385void vm_page_queues_spin_unlock(u_short);
386void vm_page_and_queue_spin_unlock(vm_page_t m);
387
9e5e1578 388void vm_page_init(vm_page_t m);
b12defdc
MD
389void vm_page_io_finish(vm_page_t m);
390void vm_page_io_start(vm_page_t m);
9bf025db
MD
391void vm_page_need_commit(vm_page_t m);
392void vm_page_clear_commit(vm_page_t m);
b12defdc 393void vm_page_wakeup(vm_page_t m);
573fb415
MD
394void vm_page_hold(vm_page_t);
395void vm_page_unhold(vm_page_t);
984263bc 396void vm_page_activate (vm_page_t);
888a40a7
MD
397
398vm_size_t vm_contig_avail_pages(void);
03d6a592 399vm_page_t vm_page_alloc (struct vm_object *, vm_pindex_t, int);
79d182b0
MD
400vm_page_t vm_page_alloc_contig(vm_paddr_t low, vm_paddr_t high,
401 unsigned long alignment, unsigned long boundary,
8b9ed12e 402 unsigned long size, vm_memattr_t memattr);
888a40a7 403
03d6a592 404vm_page_t vm_page_grab (struct vm_object *, vm_pindex_t, int);
5f910b2f 405void vm_page_cache (vm_page_t);
984263bc
MD
406int vm_page_try_to_cache (vm_page_t);
407int vm_page_try_to_free (vm_page_t);
5f910b2f 408void vm_page_dontneed (vm_page_t);
984263bc 409void vm_page_deactivate (vm_page_t);
b12defdc 410void vm_page_deactivate_locked (vm_page_t);
3023924a 411void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
d2d8515b 412int vm_page_insert (vm_page_t, struct vm_object *, vm_pindex_t);
03d6a592 413vm_page_t vm_page_lookup (struct vm_object *, vm_pindex_t);
79d182b0
MD
414vm_page_t VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_wait)(
415 struct vm_object *, vm_pindex_t, int, const char *
416 VM_PAGE_DEBUG_ARGS);
417vm_page_t VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_try)(
418 struct vm_object *, vm_pindex_t, int, int *
419 VM_PAGE_DEBUG_ARGS);
dc6a6bd2 420vm_page_t vm_page_repurpose(struct vm_object *, vm_pindex_t, int, int *,
962f16a7 421 vm_page_t, int *, int *);
984263bc 422void vm_page_remove (vm_page_t);
03d6a592 423void vm_page_rename (vm_page_t, struct vm_object *, vm_pindex_t);
da23a592 424void vm_page_startup (void);
c7f9edd8 425void vm_numa_organize(vm_paddr_t ran_beg, vm_paddr_t bytes, int physid);
984263bc
MD
426void vm_page_unmanage (vm_page_t);
427void vm_page_unwire (vm_page_t, int);
428void vm_page_wire (vm_page_t);
429void vm_page_unqueue (vm_page_t);
430void vm_page_unqueue_nowakeup (vm_page_t);
b12defdc 431vm_page_t vm_page_next (vm_page_t);
984263bc 432void vm_page_set_validclean (vm_page_t, int, int);
0a8aee15 433void vm_page_set_validdirty (vm_page_t, int, int);
1a54183b 434void vm_page_set_valid (vm_page_t, int, int);
984263bc
MD
435void vm_page_set_dirty (vm_page_t, int, int);
436void vm_page_clear_dirty (vm_page_t, int, int);
437void vm_page_set_invalid (vm_page_t, int, int);
984263bc
MD
438int vm_page_is_valid (vm_page_t, int, int);
439void vm_page_test_dirty (vm_page_t);
440int vm_page_bits (int, int);
74232d8e 441vm_page_t vm_page_list_find(int basequeue, int index, boolean_t prefer_zero);
984263bc
MD
442void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
443void vm_page_free_toq(vm_page_t m);
79d182b0 444void vm_page_free_contig(vm_page_t m, unsigned long size);
bb6811be 445vm_page_t vm_page_free_fromq_fast(void);
10192bae 446void vm_page_event_internal(vm_page_t, vm_page_event_t);
17cde63e 447void vm_page_dirty(vm_page_t m);
906c754c
MD
448void vm_page_register_action(vm_page_action_t action, vm_page_event_t event);
449void vm_page_unregister_action(vm_page_action_t action);
b12defdc 450void vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg);
9002b0d5
MD
451void VM_PAGE_DEBUG_EXT(vm_page_busy_wait)(vm_page_t m,
452 int also_m_busy, const char *wmsg VM_PAGE_DEBUG_ARGS);
453int VM_PAGE_DEBUG_EXT(vm_page_busy_try)(vm_page_t m,
454 int also_m_busy VM_PAGE_DEBUG_ARGS);
070a58b3 455u_short vm_get_pg_color(int cpuid, vm_object_t object, vm_pindex_t pindex);
b12defdc
MD
456
457#ifdef VM_PAGE_DEBUG
458
459#define vm_page_lookup_busy_wait(object, pindex, alsob, msg) \
460 vm_page_lookup_busy_wait_debug(object, pindex, alsob, msg, \
461 __func__, __LINE__)
462
463#define vm_page_lookup_busy_try(object, pindex, alsob, errorp) \
464 vm_page_lookup_busy_try_debug(object, pindex, alsob, errorp, \
465 __func__, __LINE__)
466
467#define vm_page_busy_wait(m, alsob, msg) \
468 vm_page_busy_wait_debug(m, alsob, msg, __func__, __LINE__)
469
470#define vm_page_busy_try(m, alsob) \
471 vm_page_busy_try_debug(m, alsob, __func__, __LINE__)
472
473#endif
d1fcdd16 474
984263bc 475#endif /* _KERNEL */
1bd40720 476#endif /* !_VM_VM_PAGE_H_ */