kernel - VM rework part 16 - Optimization & cleanup pass
[dragonfly.git] / sys / vm / vm_page.h
blob8634dc88c79ede3ab036c80660de3e22e7b5b5e7
1 /*
2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 2003-2017 The DragonFly Project. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * The Mach Operating System project at Carnegie-Mellon University.
9 * This code is derived from software contributed to The DragonFly Project
10 * by Matthew Dillon <dillon@backplane.com>
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. 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
34 * SUCH DAMAGE.
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.
66 * Resident memory system definitions.
69 #ifndef _VM_VM_PAGE_H_
70 #define _VM_VM_PAGE_H_
72 #ifndef _SYS_TYPES_H_
73 #include <sys/types.h>
74 #endif
75 #ifndef _SYS_TREE_H_
76 #include <sys/tree.h>
77 #endif
78 #ifndef _MACHINE_PMAP_H_
79 #include <machine/pmap.h>
80 #endif
81 #ifndef _VM_PMAP_H_
82 #include <vm/pmap.h>
83 #endif
84 #include <machine/atomic.h>
86 #ifdef _KERNEL
88 #ifndef _SYS_SYSTM_H_
89 #include <sys/systm.h>
90 #endif
91 #ifndef _SYS_SPINLOCK_H_
92 #include <sys/spinlock.h>
93 #endif
95 #ifdef __x86_64__
96 #include <machine/vmparam.h>
97 #endif
99 #endif
102 * vm_page structure
104 * hard-busy: (PBUSY_LOCKED)
106 * Hard-busying a page allows major manipulation of the page structure.
107 * No new soft-busies can accumulate while a page is hard-busied. The
108 * page busying code typically waits for all soft-busies to drop before
109 * allowing the hard-busy.
111 * soft-busy: (PBUSY_MASK)
113 * Soft-busying a page typically indicates I/O or read-only use of
114 * the content. A page can have multiple soft-busies on it. New
115 * soft-busies block on any hard-busied page (wait for the hard-busy
116 * to go away).
118 * hold_count
120 * This prevents a page from being freed. This does not prevent any
121 * other operation. The page may still be disassociated from its
122 * object and essentially scrapped. It just won't be reused while
123 * a non-zero hold_count is present.
125 * wire_count
127 * This indicates that the page has been wired into memory somewhere
128 * (typically a buffer cache buffer, or a user wire). The pageout
129 * daemon will skip wired pages.
131 TAILQ_HEAD(pglist, vm_page);
133 struct vm_object;
135 int rb_vm_page_compare(struct vm_page *, struct vm_page *);
137 struct vm_page_rb_tree;
138 RB_PROTOTYPE2(vm_page_rb_tree, vm_page, rb_entry,
139 rb_vm_page_compare, vm_pindex_t);
140 RB_HEAD(vm_page_rb_tree, vm_page);
142 struct vm_page {
143 TAILQ_ENTRY(vm_page) pageq; /* vm_page_queues[] list (P) */
144 RB_ENTRY(vm_page) rb_entry; /* Red-Black tree based at object */
145 struct spinlock spin;
146 struct vm_object *object; /* which object am I in (O,P)*/
147 vm_pindex_t pindex; /* offset into object (O,P) */
148 vm_paddr_t phys_addr; /* physical address of page */
149 struct md_page md; /* machine dependant stuff */
150 uint16_t queue; /* page queue index */
151 uint16_t pc; /* page color */
152 uint8_t act_count; /* page usage count */
153 uint8_t pat_mode; /* hardware page attribute */
154 uint8_t valid; /* map of valid DEV_BSIZE chunks */
155 uint8_t dirty; /* map of dirty DEV_BSIZE chunks */
156 uint32_t flags; /* see below */
157 uint32_t wire_count; /* wired down maps refs (P) */
158 uint32_t busy_count; /* soft-busy and hard-busy */
159 int hold_count; /* page hold count */
160 int ku_pagecnt; /* kmalloc helper */
161 #ifdef VM_PAGE_DEBUG
162 const char *busy_func;
163 int busy_line;
164 #endif
167 #define PBUSY_LOCKED 0x80000000U
168 #define PBUSY_WANTED 0x40000000U
169 #define PBUSY_SWAPINPROG 0x20000000U
170 #define PBUSY_MASK 0x1FFFFFFFU
172 #ifndef __VM_PAGE_T_DEFINED__
173 #define __VM_PAGE_T_DEFINED__
174 typedef struct vm_page *vm_page_t;
175 #endif
178 * Page coloring parameters. We use generous parameters designed to
179 * statistically spread pages over available cpu cache space. This has
180 * become less important over time as cache associativity is higher
181 * in modern times but we still use the core algorithm to help reduce
182 * lock contention between cpus.
184 * Page coloring cannot be disabled.
186 * In today's world of many-core systems, we must be able to provide enough VM
187 * page queues for each logical cpu thread to cover the L1/L2/L3 cache set
188 * associativity. If we don't, the cpu caches will not be properly utilized.
190 * Using 2048 allows 8-way set-assoc with 256 logical cpus, but seems to
191 * have a number of downsides when queues are assymetrically starved.
193 * Using 1024 allows 4-way set-assoc with 256 logical cpus, and more with
194 * fewer cpus.
196 #define PQ_PRIME1 31 /* Prime number somewhat less than PQ_HASH_SIZE */
197 #define PQ_PRIME2 23 /* Prime number somewhat less than PQ_HASH_SIZE */
198 #define PQ_L2_SIZE 1024 /* Must be enough for maximal ncpus x hw set-assoc */
199 #define PQ_L2_MASK (PQ_L2_SIZE - 1)
201 #define PQ_NONE 0
202 #define PQ_FREE (1 + 0*PQ_L2_SIZE)
203 #define PQ_INACTIVE (1 + 1*PQ_L2_SIZE)
204 #define PQ_ACTIVE (1 + 2*PQ_L2_SIZE)
205 #define PQ_CACHE (1 + 3*PQ_L2_SIZE)
206 #define PQ_HOLD (1 + 4*PQ_L2_SIZE)
207 #define PQ_COUNT (1 + 5*PQ_L2_SIZE)
210 * Scan support
212 struct vm_map;
214 struct rb_vm_page_scan_info {
215 vm_pindex_t start_pindex;
216 vm_pindex_t end_pindex;
217 int limit;
218 int desired;
219 int error;
220 int pagerflags;
221 int count;
222 int unused01;
223 vm_offset_t addr;
224 struct vm_map_entry *entry;
225 struct vm_object *object;
226 struct vm_object *dest_object;
227 struct vm_page *mpte;
228 struct pmap *pmap;
229 struct vm_map *map;
232 int rb_vm_page_scancmp(struct vm_page *, void *);
234 struct vpgqueues {
235 struct spinlock spin;
236 struct pglist pl;
237 long lcnt;
238 long adds; /* heuristic, add operations */
239 int cnt_offset; /* offset into vmstats structure (int) */
240 int lastq; /* heuristic, skip empty queues */
241 } __aligned(64);
243 extern struct vpgqueues vm_page_queues[PQ_COUNT];
244 extern long vmmeter_neg_slop_cnt;
247 * These are the flags defined for vm_page.
249 * PG_FICTITIOUS It is not possible to translate the pte's physical
250 * address back to a vm_page_t. The vm_page_t is fake
251 * or there isn't one at all.
253 * Fictitious vm_page_t's can be placed in objects and
254 * it is possible to perform pmap functions on them
255 * by virtual address range and by their vm_page_t.
256 * However, pmap_count and writeable_count cannot be
257 * tracked since there is no way to reverse-map the
258 * pte back to the vm_page.
260 * (pmap operations by-vm_page can still be used to
261 * adjust protections or remove the page from the pmap,
262 * and will go only by the PG_MAPPED flag).
264 * NOTE: The contiguous memory management will flag
265 * PG_FICTITIOUS on pages in the vm_page_array,
266 * even though the physical addrses can be
267 * translated back to a vm_page_t.
269 * NOTE: Implies PG_UNQUEUED. PG_UNQUEUED must also
270 * be set. No queue management may be performed
271 * on fictitious pages.
273 * PG_UNQUEUED The page is not to participate in any VM page queue
274 * manipulation (even if it is otherwise a normal page).
276 * PG_MAPPED Only applies to non-fictitious regular pages, this
277 * flag indicates that the page MIGHT be mapped into
278 * zero or more pmaps via normal managed operations..
280 * The page might still be mapped in a specialized manner
281 * (i.e. pmap_kenter(), or mapped into the buffer cache,
282 * and so forth) without setting this flag.
284 * If this flag is clear it indicates that the page is
285 * absolutely not mapped into a regular pmap by normal
286 * means. If set, the status is unknown.
288 * PG_WRITEABLE Similar to PG_MAPPED, indicates that the page might
289 * be mapped RW into zero or more pmaps via normal
290 * managed operations.
292 * If this flag is clear it indicates that the page is
293 * absolutely not mapped RW into a regular pmap by normal
294 * means. If set, the status is unknown.
296 * PG_SWAPPED Indicates that the page is backed by a swap block.
297 * Any VM object type other than OBJT_DEFAULT can contain
298 * swap-backed pages now.
300 #define PG_UNUSED0001 0x00000001
301 #define PG_UNUSED0002 0x00000002
302 #define PG_WINATCFLS 0x00000004 /* flush dirty page on inactive q */
303 #define PG_FICTITIOUS 0x00000008 /* No reverse-map or tracking */
304 #define PG_WRITEABLE 0x00000010 /* page may be writeable */
305 #define PG_MAPPED 0x00000020 /* page may be mapped (managed) */
306 #define PG_UNUSED0040 0x00000040
307 #define PG_REFERENCED 0x00000080 /* page has been referenced */
308 #define PG_CLEANCHK 0x00000100 /* page will be checked for cleaning */
309 #define PG_UNUSED0200 0x00000200
310 #define PG_NOSYNC 0x00000400 /* do not collect for syncer */
311 #define PG_UNQUEUED 0x00000800 /* No queue management for page */
312 #define PG_MARKER 0x00001000 /* special queue marker page */
313 #define PG_RAM 0x00002000 /* read ahead mark */
314 #define PG_SWAPPED 0x00004000 /* backed by swap */
315 #define PG_NOTMETA 0x00008000 /* do not back with swap */
316 #define PG_UNUSED10000 0x00010000
317 #define PG_UNUSED20000 0x00020000
318 #define PG_NEED_COMMIT 0x00040000 /* clean page requires commit */
320 #define PG_KEEP_NEWPAGE_MASK (0)
323 * Misc constants.
326 #define ACT_DECLINE 1
327 #define ACT_ADVANCE 3
328 #define ACT_INIT 5
329 #define ACT_MAX 64
331 #ifdef VM_PAGE_DEBUG
332 #define VM_PAGE_DEBUG_EXT(name) name ## _debug
333 #define VM_PAGE_DEBUG_ARGS , const char *func, int lineno
334 #else
335 #define VM_PAGE_DEBUG_EXT(name) name
336 #define VM_PAGE_DEBUG_ARGS
337 #endif
339 #ifdef _KERNEL
341 * Each pageable resident page falls into one of four lists:
343 * free
344 * Available for allocation now.
346 * The following are all LRU sorted:
348 * cache
349 * Almost available for allocation. Still in an
350 * object, but clean and immediately freeable at
351 * non-interrupt times.
353 * inactive
354 * Low activity, candidates for reclamation.
355 * This is the list of pages that should be
356 * paged out next.
358 * active
359 * Pages that are "active" i.e. they have been
360 * recently referenced.
362 * zero
363 * Pages that are really free and have been pre-zeroed
367 extern struct vm_page *vm_page_array; /* First resident page in table */
368 extern vm_pindex_t vm_page_array_size; /* number of vm_page_t's */
369 extern vm_pindex_t first_page; /* first physical page number */
371 #define VM_PAGE_TO_PHYS(entry) \
372 ((entry)->phys_addr)
374 #define PHYS_TO_VM_PAGE(pa) \
375 (&vm_page_array[atop(pa) - first_page])
378 #if PAGE_SIZE == 4096
379 #define VM_PAGE_BITS_ALL 0xff
380 #endif
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
388 * is also specified.
390 #define VM_ALLOC_NORMAL 0x0001 /* ok to use cache pages */
391 #define VM_ALLOC_SYSTEM 0x0002 /* ok to exhaust most of free list */
392 #define VM_ALLOC_INTERRUPT 0x0004 /* ok to exhaust entire free list */
393 #define VM_ALLOC_ZERO 0x0008 /* req pre-zero'd memory if avail */
394 #define VM_ALLOC_QUICK 0x0010 /* like NORMAL but do not use cache */
395 #define VM_ALLOC_FORCE_ZERO 0x0020 /* zero page even if already valid */
396 #define VM_ALLOC_NULL_OK 0x0040 /* ok to return NULL on collision */
397 #define VM_ALLOC_RETRY 0x0080 /* indefinite block (vm_page_grab()) */
398 #define VM_ALLOC_USE_GD 0x0100 /* use per-gd cache */
399 #define VM_ALLOC_CPU_SPEC 0x0200
401 #define VM_ALLOC_CPU_SHIFT 16
402 #define VM_ALLOC_CPU(n) (((n) << VM_ALLOC_CPU_SHIFT) | \
403 VM_ALLOC_CPU_SPEC)
404 #define VM_ALLOC_GETCPU(flags) ((flags) >> VM_ALLOC_CPU_SHIFT)
406 void vm_page_queue_spin_lock(vm_page_t);
407 void vm_page_queues_spin_lock(u_short);
408 void vm_page_and_queue_spin_lock(vm_page_t);
410 void vm_page_queue_spin_unlock(vm_page_t);
411 void vm_page_queues_spin_unlock(u_short);
412 void vm_page_and_queue_spin_unlock(vm_page_t m);
414 void vm_page_init(vm_page_t m);
415 void vm_page_io_finish(vm_page_t m);
416 void vm_page_io_start(vm_page_t m);
417 void vm_page_need_commit(vm_page_t m);
418 void vm_page_clear_commit(vm_page_t m);
419 void vm_page_wakeup(vm_page_t m);
420 void vm_page_hold(vm_page_t);
421 void vm_page_unhold(vm_page_t);
422 void vm_page_activate (vm_page_t);
423 void vm_page_soft_activate (vm_page_t);
425 vm_size_t vm_contig_avail_pages(void);
426 vm_page_t vm_page_alloc (struct vm_object *, vm_pindex_t, int);
427 vm_page_t vm_page_alloc_contig(vm_paddr_t low, vm_paddr_t high,
428 unsigned long alignment, unsigned long boundary,
429 unsigned long size, vm_memattr_t memattr);
431 vm_page_t vm_page_grab (struct vm_object *, vm_pindex_t, int);
432 void vm_page_cache (vm_page_t);
433 int vm_page_try_to_cache (vm_page_t);
434 int vm_page_try_to_free (vm_page_t);
435 void vm_page_dontneed (vm_page_t);
436 void vm_page_deactivate (vm_page_t);
437 void vm_page_deactivate_locked (vm_page_t);
438 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
439 int vm_page_insert (vm_page_t, struct vm_object *, vm_pindex_t);
441 vm_page_t vm_page_hash_get(vm_object_t object, vm_pindex_t pindex);
443 vm_page_t vm_page_lookup (struct vm_object *, vm_pindex_t);
444 vm_page_t vm_page_lookup_sbusy_try(struct vm_object *object,
445 vm_pindex_t pindex, int pgoff, int pgbytes);
446 vm_page_t VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_wait)(
447 struct vm_object *, vm_pindex_t, int, const char *
448 VM_PAGE_DEBUG_ARGS);
449 vm_page_t VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_try)(
450 struct vm_object *, vm_pindex_t, int, int *
451 VM_PAGE_DEBUG_ARGS);
452 void vm_page_remove (vm_page_t);
453 void vm_page_rename (vm_page_t, struct vm_object *, vm_pindex_t);
454 void vm_page_startup (void);
455 void vm_numa_organize(vm_paddr_t ran_beg, vm_paddr_t bytes, int physid);
456 void vm_numa_organize_finalize(void);
457 void vm_page_unwire (vm_page_t, int);
458 void vm_page_wire (vm_page_t);
459 void vm_page_unqueue (vm_page_t);
460 void vm_page_unqueue_nowakeup (vm_page_t);
461 vm_page_t vm_page_next (vm_page_t);
462 void vm_page_set_validclean (vm_page_t, int, int);
463 void vm_page_set_validdirty (vm_page_t, int, int);
464 void vm_page_set_valid (vm_page_t, int, int);
465 void vm_page_set_dirty (vm_page_t, int, int);
466 void vm_page_clear_dirty (vm_page_t, int, int);
467 void vm_page_set_invalid (vm_page_t, int, int);
468 int vm_page_is_valid (vm_page_t, int, int);
469 void vm_page_test_dirty (vm_page_t);
470 int vm_page_bits (int, int);
471 vm_page_t vm_page_list_find(int basequeue, int index);
472 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
473 void vm_page_free_toq(vm_page_t m);
474 void vm_page_free_contig(vm_page_t m, unsigned long size);
475 vm_page_t vm_page_free_fromq_fast(void);
476 void vm_page_dirty(vm_page_t m);
477 void vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg);
478 int vm_page_sbusy_try(vm_page_t m);
479 void VM_PAGE_DEBUG_EXT(vm_page_busy_wait)(vm_page_t m,
480 int also_m_busy, const char *wmsg VM_PAGE_DEBUG_ARGS);
481 int VM_PAGE_DEBUG_EXT(vm_page_busy_try)(vm_page_t m,
482 int also_m_busy VM_PAGE_DEBUG_ARGS);
483 u_short vm_get_pg_color(int cpuid, vm_object_t object, vm_pindex_t pindex);
485 #ifdef VM_PAGE_DEBUG
487 #define vm_page_lookup_busy_wait(object, pindex, alsob, msg) \
488 vm_page_lookup_busy_wait_debug(object, pindex, alsob, msg, \
489 __func__, __LINE__)
491 #define vm_page_lookup_busy_try(object, pindex, alsob, errorp) \
492 vm_page_lookup_busy_try_debug(object, pindex, alsob, errorp, \
493 __func__, __LINE__)
495 #define vm_page_busy_wait(m, alsob, msg) \
496 vm_page_busy_wait_debug(m, alsob, msg, __func__, __LINE__)
498 #define vm_page_busy_try(m, alsob) \
499 vm_page_busy_try_debug(m, alsob, __func__, __LINE__)
501 #endif
503 #endif /* _KERNEL */
504 #endif /* !_VM_VM_PAGE_H_ */