sys/vm/vm_page.h

   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.
  16  * 3. Neither the name of the University nor the names of its contributors
  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
  67 #ifndef _VM_VM_PAGE_H_
  68 #define _VM_VM_PAGE_H_
  69
  70 #ifndef _SYS_TYPES_H_
  71 #include <sys/types.h>
  72 #endif
  73 #ifndef _SYS_TREE_H_
  74 #include <sys/tree.h>
  75 #endif
  76 #ifndef _MACHINE_PMAP_H_
  77 #include <machine/pmap.h>
  78 #endif
  79 #ifndef _VM_PMAP_H_
  80 #include <vm/pmap.h>
  81 #endif
  82 #include <machine/atomic.h>
  83
  84 #ifdef _KERNEL
  85
  86 #ifndef _SYS_SYSTM_H_
  87 #include <sys/systm.h>
  88 #endif
  89 #ifndef _SYS_THREAD2_H_
  90 #include <sys/thread2.h>
  91 #endif
  92
  93 #ifdef __x86_64__
  94 #include <machine/vmparam.h>
  95 #endif
  96
  97 #endif
  98
  99 typedef enum vm_page_event { VMEVENT_NONE, VMEVENT_COW } vm_page_event_t;
 100
 101 struct vm_page_action {
 102         LIST_ENTRY(vm_page_action) entry;
 103         struct vm_page          *m;
 104         vm_page_event_t         event;
 105         void                    (*func)(struct vm_page *,
 106                                         struct vm_page_action *);
 107         void                    *data;
 108 };
 109
 110 typedef struct vm_page_action *vm_page_action_t;
 111
 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
 141 TAILQ_HEAD(pglist, vm_page);
 142
 143 struct vm_object;
 144
 145 int rb_vm_page_compare(struct vm_page *, struct vm_page *);
 146
 147 struct vm_page_rb_tree;
 148 RB_PROTOTYPE2(vm_page_rb_tree, vm_page, rb_entry, rb_vm_page_compare, vm_pindex_t);
 149
 150 struct vm_page {
 151         TAILQ_ENTRY(vm_page) pageq;     /* vm_page_queues[] list (P)    */
 152         RB_ENTRY(vm_page) rb_entry;     /* Red-Black tree based at object */
 153
 154         struct vm_object *object;       /* which object am I in (O,P)*/
 155         vm_pindex_t pindex;             /* offset into object (O,P) */
 156         vm_paddr_t phys_addr;           /* physical address of page */
 157         struct md_page md;              /* machine dependant stuff */
 158         u_short queue;                  /* page queue index */
 159         u_short pc;                     /* page color */
 160         u_char  act_count;              /* page usage count */
 161         u_char  busy;                   /* page busy count */
 162         u_char  pat_mode;               /* hardware page attribute */
 163         u_char  unused02;
 164         u_int32_t flags;                /* see below */
 165         u_int   wire_count;             /* wired down maps refs (P) */
 166         int     hold_count;             /* page hold count */
 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          */
 172         u_char  valid;                  /* map of valid DEV_BSIZE chunks */
 173         u_char  dirty;                  /* map of dirty DEV_BSIZE chunks */
 174
 175         int     ku_pagecnt;             /* kmalloc helper */
 176 #ifdef VM_PAGE_DEBUG
 177         const char *busy_func;
 178         int     busy_line;
 179 #endif
 180 };
 181
 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
 190 #ifndef __VM_PAGE_T_DEFINED__
 191 #define __VM_PAGE_T_DEFINED__
 192 typedef struct vm_page *vm_page_t;
 193 #endif
 194
 195 /*
 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.
 201  *
 202  * Page coloring cannot be disabled.
 203  */
 204
 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 */
 208
 209 #if 0
 210 #define PQ_PRIME1 31    /* Prime number somewhat less than PQ_HASH_SIZE */
 211 #define PQ_PRIME2 23    /* Prime number somewhat less than PQ_HASH_SIZE */
 212 #define PQ_L2_SIZE 128  /* A number of colors opt for 512K cache */
 213
 214 #define PQ_PRIME1 13    /* Prime number somewhat less than PQ_HASH_SIZE */
 215 #define PQ_PRIME2 7     /* Prime number somewhat less than PQ_HASH_SIZE */
 216 #define PQ_L2_SIZE 64   /* A number of colors opt for 256K cache */
 217
 218 #define PQ_PRIME1 9     /* Produces a good PQ_L2_SIZE/3 + PQ_PRIME1 */
 219 #define PQ_PRIME2 5     /* Prime number somewhat less than PQ_HASH_SIZE */
 220 #define PQ_L2_SIZE 32   /* A number of colors opt for 128k cache */
 221
 222 #define PQ_PRIME1 5     /* Prime number somewhat less than PQ_HASH_SIZE */
 223 #define PQ_PRIME2 3     /* Prime number somewhat less than PQ_HASH_SIZE */
 224 #define PQ_L2_SIZE 16   /* A reasonable number of colors (opt for 64K cache) */
 225 #endif
 226
 227 #define PQ_L2_MASK      (PQ_L2_SIZE - 1)
 228
 229 #define PQ_NONE         0
 230 #define PQ_FREE         (1 + 0*PQ_L2_SIZE)
 231 #define PQ_INACTIVE     (1 + 1*PQ_L2_SIZE)
 232 #define PQ_ACTIVE       (1 + 2*PQ_L2_SIZE)
 233 #define PQ_CACHE        (1 + 3*PQ_L2_SIZE)
 234 #define PQ_HOLD         (1 + 4*PQ_L2_SIZE)
 235 #define PQ_COUNT        (1 + 5*PQ_L2_SIZE)
 236
 237 /*
 238  * Scan support
 239  */
 240 struct vm_map;
 241
 242 struct rb_vm_page_scan_info {
 243         vm_pindex_t     start_pindex;
 244         vm_pindex_t     end_pindex;
 245         int             limit;
 246         int             desired;
 247         int             error;
 248         int             pagerflags;
 249         int             count;
 250         int             unused01;
 251         vm_offset_t     addr;
 252         vm_pindex_t     backing_offset_index;
 253         struct vm_object *object;
 254         struct vm_object *backing_object;
 255         struct vm_page  *mpte;
 256         struct pmap     *pmap;
 257         struct vm_map   *map;
 258 };
 259
 260 int rb_vm_page_scancmp(struct vm_page *, void *);
 261
 262 struct vpgqueues {
 263         struct pglist pl;
 264         int     *cnt;
 265         int     lcnt;
 266         int     flipflop;       /* probably not the best place */
 267         struct spinlock spin;
 268         char    unused[64 - sizeof(struct pglist) -
 269                         sizeof(int *) - sizeof(int) * 2];
 270 };
 271
 272 extern struct vpgqueues vm_page_queues[PQ_COUNT];
 273
 274 /*
 275  * These are the flags defined for vm_page.
 276  *
 277  *  PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
 278  *  not under PV management but otherwise should be treated as a
 279  *  normal page.  Pages not under PV management cannot be paged out
 280  *  via the object/vm_page_t because there is no knowledge of their
 281  *  pte mappings, nor can they be removed from their objects via
 282  *  the object, and such pages are also not on any PQ queue.  The
 283  *  PG_MAPPED and PG_WRITEABLE flags are not applicable.
 284  *
 285  *  PG_MAPPED only applies to managed pages, indicating whether the page
 286  *  is mapped onto one or more pmaps.  A page might still be mapped to
 287  *  special pmaps in an unmanaged fashion, for example when mapped into a
 288  *  buffer cache buffer, without setting PG_MAPPED.
 289  *
 290  *  PG_WRITEABLE indicates that there may be a writeable managed pmap entry
 291  *  somewhere, and that the page can be dirtied by hardware at any time
 292  *  and may have to be tested for that.  The modified bit in unmanaged
 293  *  mappings or in the special clean map is not tested.
 294  *
 295  *  PG_SWAPPED indicates that the page is backed by a swap block.  Any
 296  *  VM object type other than OBJT_DEFAULT can have swap-backed pages now.
 297  *
 298  *  PG_SBUSY is set when m->busy != 0.  PG_SBUSY and m->busy are only modified
 299  *  when the page is PG_BUSY.
 300  */
 301 #define PG_BUSY         0x00000001      /* page is in transit (O) */
 302 #define PG_WANTED       0x00000002      /* someone is waiting for page (O) */
 303 #define PG_WINATCFLS    0x00000004      /* flush dirty page on inactive q */
 304 #define PG_FICTITIOUS   0x00000008      /* physical page doesn't exist (O) */
 305 #define PG_WRITEABLE    0x00000010      /* page is writeable */
 306 #define PG_MAPPED       0x00000020      /* page is mapped (managed) */
 307 #define PG_ZERO         0x00000040      /* page is zeroed */
 308 #define PG_REFERENCED   0x00000080      /* page has been referenced */
 309 #define PG_CLEANCHK     0x00000100      /* page will be checked for cleaning */
 310 #define PG_SWAPINPROG   0x00000200      /* swap I/O in progress on page      */
 311 #define PG_NOSYNC       0x00000400      /* do not collect for syncer */
 312 #define PG_UNMANAGED    0x00000800      /* No PV management for page */
 313 #define PG_MARKER       0x00001000      /* special queue marker page */
 314 #define PG_RAM          0x00002000      /* read ahead mark */
 315 #define PG_SWAPPED      0x00004000      /* backed by swap */
 316 #define PG_NOTMETA      0x00008000      /* do not back with swap */
 317 #define PG_ACTIONLIST   0x00010000      /* lookaside action list present */
 318 #define PG_SBUSY        0x00020000      /* soft-busy also set */
 319 #define PG_NEED_COMMIT  0x00040000      /* clean page requires commit */
 320
 321 /*
 322  * Misc constants.
 323  */
 324
 325 #define ACT_DECLINE             1
 326 #define ACT_ADVANCE             3
 327 #define ACT_INIT                5
 328 #define ACT_MAX                 64
 329
 330 #ifdef _KERNEL
 331 /*
 332  * Each pageable resident page falls into one of four lists:
 333  *
 334  *      free
 335  *              Available for allocation now.
 336  *
 337  * The following are all LRU sorted:
 338  *
 339  *      cache
 340  *              Almost available for allocation. Still in an
 341  *              object, but clean and immediately freeable at
 342  *              non-interrupt times.
 343  *
 344  *      inactive
 345  *              Low activity, candidates for reclamation.
 346  *              This is the list of pages that should be
 347  *              paged out next.
 348  *
 349  *      active
 350  *              Pages that are "active" i.e. they have been
 351  *              recently referenced.
 352  *
 353  *      zero
 354  *              Pages that are really free and have been pre-zeroed
 355  *
 356  */
 357
 358 extern int vm_page_zero_count;
 359 extern struct vm_page *vm_page_array;   /* First resident page in table */
 360 extern int vm_page_array_size;          /* number of vm_page_t's */
 361 extern long first_page;                 /* first physical page number */
 362
 363 #define VM_PAGE_TO_PHYS(entry)  \
 364                 ((entry)->phys_addr)
 365
 366 #define PHYS_TO_VM_PAGE(pa)     \
 367                 (&vm_page_array[atop(pa) - first_page])
 368
 369
 370 #if PAGE_SIZE == 4096
 371 #define VM_PAGE_BITS_ALL 0xff
 372 #endif
 373
 374 /*
 375  * Note: the code will always use nominally free pages from the free list
 376  * before trying other flag-specified sources.
 377  *
 378  * At least one of VM_ALLOC_NORMAL|VM_ALLOC_SYSTEM|VM_ALLOC_INTERRUPT
 379  * must be specified.  VM_ALLOC_RETRY may only be specified if VM_ALLOC_NORMAL
 380  * is also specified.
 381  */
 382 #define VM_ALLOC_NORMAL         0x0001  /* ok to use cache pages */
 383 #define VM_ALLOC_SYSTEM         0x0002  /* ok to exhaust most of free list */
 384 #define VM_ALLOC_INTERRUPT      0x0004  /* ok to exhaust entire free list */
 385 #define VM_ALLOC_ZERO           0x0008  /* req pre-zero'd memory if avail */
 386 #define VM_ALLOC_QUICK          0x0010  /* like NORMAL but do not use cache */
 387 #define VM_ALLOC_FORCE_ZERO     0x0020  /* zero page even if already valid */
 388 #define VM_ALLOC_NULL_OK        0x0040  /* ok to return NULL on collision */
 389 #define VM_ALLOC_RETRY          0x0080  /* indefinite block (vm_page_grab()) */
 390 #define VM_ALLOC_USE_GD         0x0100  /* use per-gd cache */
 391
 392 void vm_page_queue_spin_lock(vm_page_t);
 393 void vm_page_queues_spin_lock(u_short);
 394 void vm_page_and_queue_spin_lock(vm_page_t);
 395
 396 void vm_page_queue_spin_unlock(vm_page_t);
 397 void vm_page_queues_spin_unlock(u_short);
 398 void vm_page_and_queue_spin_unlock(vm_page_t m);
 399
 400 void vm_page_init(vm_page_t m);
 401 void vm_page_io_finish(vm_page_t m);
 402 void vm_page_io_start(vm_page_t m);
 403 void vm_page_need_commit(vm_page_t m);
 404 void vm_page_clear_commit(vm_page_t m);
 405 void vm_page_wakeup(vm_page_t m);
 406 void vm_page_hold(vm_page_t);
 407 void vm_page_unhold(vm_page_t);
 408 void vm_page_activate (vm_page_t);
 409 void vm_page_pcpu_cache(void);
 410
 411 vm_size_t vm_contig_avail_pages(void);
 412 vm_page_t vm_page_alloc (struct vm_object *, vm_pindex_t, int);
 413 vm_page_t vm_page_alloc_contig(vm_paddr_t low, vm_paddr_t high,
 414                      unsigned long alignment, unsigned long boundary,
 415                      unsigned long size, vm_memattr_t memattr);
 416
 417 vm_page_t vm_page_grab (struct vm_object *, vm_pindex_t, int);
 418 void vm_page_cache (vm_page_t);
 419 int vm_page_try_to_cache (vm_page_t);
 420 int vm_page_try_to_free (vm_page_t);
 421 void vm_page_dontneed (vm_page_t);
 422 void vm_page_deactivate (vm_page_t);
 423 void vm_page_deactivate_locked (vm_page_t);
 424 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
 425 int vm_page_insert (vm_page_t, struct vm_object *, vm_pindex_t);
 426 vm_page_t vm_page_lookup (struct vm_object *, vm_pindex_t);
 427 vm_page_t VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_wait)(
 428                 struct vm_object *, vm_pindex_t, int, const char *
 429                 VM_PAGE_DEBUG_ARGS);
 430 vm_page_t VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_try)(
 431                 struct vm_object *, vm_pindex_t, int, int *
 432                 VM_PAGE_DEBUG_ARGS);
 433 vm_page_t vm_page_repurpose(struct vm_object *, vm_pindex_t, int, int *,
 434                 vm_page_t, int *);
 435 void vm_page_remove (vm_page_t);
 436 void vm_page_rename (vm_page_t, struct vm_object *, vm_pindex_t);
 437 void vm_page_startup (void);
 438 void vm_page_unmanage (vm_page_t);
 439 void vm_page_unwire (vm_page_t, int);
 440 void vm_page_wire (vm_page_t);
 441 void vm_page_unqueue (vm_page_t);
 442 void vm_page_unqueue_nowakeup (vm_page_t);
 443 vm_page_t vm_page_next (vm_page_t);
 444 void vm_page_set_validclean (vm_page_t, int, int);
 445 void vm_page_set_validdirty (vm_page_t, int, int);
 446 void vm_page_set_valid (vm_page_t, int, int);
 447 void vm_page_set_dirty (vm_page_t, int, int);
 448 void vm_page_clear_dirty (vm_page_t, int, int);
 449 void vm_page_set_invalid (vm_page_t, int, int);
 450 int vm_page_is_valid (vm_page_t, int, int);
 451 void vm_page_test_dirty (vm_page_t);
 452 int vm_page_bits (int, int);
 453 vm_page_t vm_page_list_find(int basequeue, int index, boolean_t prefer_zero);
 454 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
 455 void vm_page_free_toq(vm_page_t m);
 456 void vm_page_free_contig(vm_page_t m, unsigned long size);
 457 vm_page_t vm_page_free_fromq_fast(void);
 458 void vm_page_event_internal(vm_page_t, vm_page_event_t);
 459 void vm_page_dirty(vm_page_t m);
 460 void vm_page_register_action(vm_page_action_t action, vm_page_event_t event);
 461 void vm_page_unregister_action(vm_page_action_t action);
 462 void vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg);
 463 void VM_PAGE_DEBUG_EXT(vm_page_busy_wait)(vm_page_t m, int also_m_busy, const char *wmsg VM_PAGE_DEBUG_ARGS);
 464 int VM_PAGE_DEBUG_EXT(vm_page_busy_try)(vm_page_t m, int also_m_busy VM_PAGE_DEBUG_ARGS);
 465
 466 #ifdef VM_PAGE_DEBUG
 467
 468 #define vm_page_lookup_busy_wait(object, pindex, alsob, msg)            \
 469         vm_page_lookup_busy_wait_debug(object, pindex, alsob, msg,      \
 470                                         __func__, __LINE__)
 471
 472 #define vm_page_lookup_busy_try(object, pindex, alsob, errorp)          \
 473         vm_page_lookup_busy_try_debug(object, pindex, alsob, errorp,    \
 474                                         __func__, __LINE__)
 475
 476 #define vm_page_busy_wait(m, alsob, msg)                                \
 477         vm_page_busy_wait_debug(m, alsob, msg, __func__, __LINE__)
 478
 479 #define vm_page_busy_try(m, alsob)                                      \
 480         vm_page_busy_try_debug(m, alsob, __func__, __LINE__)
 481
 482 #endif
 483
 484 #endif                          /* _KERNEL */
 485 #endif                          /* !_VM_VM_PAGE_H_ */