sys/vm/vm_page.h

   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.
   5  *
   6  * This code is derived from software contributed to Berkeley by
   7  * The Mach Operating System project at Carnegie-Mellon University.
   8  *
   9  * This code is derived from software contributed to The DragonFly Project
  10  * by Matthew Dillon <dillon@backplane.com>
  11  *
  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.
  23  *
  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.
  35  *
  36  *      from: @(#)vm_page.h     8.2 (Berkeley) 12/13/93
  37  *
  38  *
  39  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
  40  * All rights reserved.
  41  *
  42  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
  43  *
  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.
  49  *
  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.
  53  *
  54  * Carnegie Mellon requests users of this software to return to
  55  *
  56  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  57  *  School of Computer Science
  58  *  Carnegie Mellon University
  59  *  Pittsburgh PA 15213-3890
  60  *
  61  * any improvements or extensions that they make and grant Carnegie the
  62  * rights to redistribute these changes.
  63  */
  64
  65 /*
  66  * Resident memory system definitions.
  67  */
  68
  69 #ifndef _VM_VM_PAGE_H_
  70 #define _VM_VM_PAGE_H_
  71
  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>
  85
  86 #ifdef _KERNEL
  87
  88 #ifndef _SYS_SYSTM_H_
  89 #include <sys/systm.h>
  90 #endif
  91 #ifndef _SYS_SPINLOCK_H_
  92 #include <sys/spinlock.h>
  93 #endif
  94
  95 #ifdef __x86_64__
  96 #include <machine/vmparam.h>
  97 #endif
  98
  99 #endif
 100
 101 /*
 102  * vm_page structure
 103  *
 104  * hard-busy: (PBUSY_LOCKED)
 105  *
 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.
 110  *
 111  * soft-busy: (PBUSY_MASK)
 112  *
 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).
 117  *
 118  * hold_count
 119  *
 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.
 124  *
 125  * wire_count
 126  *
 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.
 130  */
 131 TAILQ_HEAD(pglist, vm_page);
 132
 133 struct vm_object;
 134
 135 int rb_vm_page_compare(struct vm_page *, struct vm_page *);
 136
 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);
 141
 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
 165 };
 166
 167 #define PBUSY_LOCKED            0x80000000U
 168 #define PBUSY_WANTED            0x40000000U
 169 #define PBUSY_SWAPINPROG        0x20000000U
 170 #define PBUSY_MASK              0x1FFFFFFFU
 171
 172 #ifndef __VM_PAGE_T_DEFINED__
 173 #define __VM_PAGE_T_DEFINED__
 174 typedef struct vm_page *vm_page_t;
 175 #endif
 176
 177 /*
 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.
 183  *
 184  * Page coloring cannot be disabled.
 185  *
 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.
 189  * Using 2048 allows 8-way set-assoc with 256 logical cpus.
 190  */
 191 #define PQ_PRIME1 31    /* Prime number somewhat less than PQ_HASH_SIZE */
 192 #define PQ_PRIME2 23    /* Prime number somewhat less than PQ_HASH_SIZE */
 193 #define PQ_L2_SIZE 2048 /* Must be enough for maximal ncpus x hw set-assoc */
 194 #define PQ_L2_MASK      (PQ_L2_SIZE - 1)
 195
 196 #define PQ_NONE         0
 197 #define PQ_FREE         (1 + 0*PQ_L2_SIZE)
 198 #define PQ_INACTIVE     (1 + 1*PQ_L2_SIZE)
 199 #define PQ_ACTIVE       (1 + 2*PQ_L2_SIZE)
 200 #define PQ_CACHE        (1 + 3*PQ_L2_SIZE)
 201 #define PQ_HOLD         (1 + 4*PQ_L2_SIZE)
 202 #define PQ_COUNT        (1 + 5*PQ_L2_SIZE)
 203
 204 /*
 205  * Scan support
 206  */
 207 struct vm_map;
 208
 209 struct rb_vm_page_scan_info {
 210         vm_pindex_t     start_pindex;
 211         vm_pindex_t     end_pindex;
 212         int             limit;
 213         int             desired;
 214         int             error;
 215         int             pagerflags;
 216         int             count;
 217         int             unused01;
 218         vm_offset_t     addr;
 219         struct vm_map_entry *entry;
 220         struct vm_object *object;
 221         struct vm_object *dest_object;
 222         struct vm_page  *mpte;
 223         struct pmap     *pmap;
 224         struct vm_map   *map;
 225 };
 226
 227 int rb_vm_page_scancmp(struct vm_page *, void *);
 228
 229 struct vpgqueues {
 230         struct spinlock spin;
 231         struct pglist pl;
 232         long    lcnt;
 233         long    adds;           /* heuristic, add operations */
 234         int     cnt_offset;     /* offset into vmstats structure (int) */
 235 } __aligned(64);
 236
 237 extern struct vpgqueues vm_page_queues[PQ_COUNT];
 238
 239 /*
 240  * These are the flags defined for vm_page.
 241  *
 242  *  PG_UNMANAGED (used by OBJT_PHYS) indicates that the page is
 243  *  not under PV management but otherwise should be treated as a
 244  *  normal page.  Pages not under PV management cannot be paged out
 245  *  via the object/vm_page_t because there is no knowledge of their
 246  *  pte mappings, nor can they be removed from their objects via
 247  *  the object, and such pages are also not on any PQ queue.  The
 248  *  PG_MAPPED and PG_WRITEABLE flags are not applicable.
 249  *
 250  *  PG_MAPPED only applies to managed pages, indicating whether the page
 251  *  is mapped onto one or more pmaps.  A page might still be mapped to
 252  *  special pmaps in an unmanaged fashion, for example when mapped into a
 253  *  buffer cache buffer, without setting PG_MAPPED.
 254  *
 255  *  PG_WRITEABLE indicates that there may be a writeable managed pmap entry
 256  *  somewhere, and that the page can be dirtied by hardware at any time
 257  *  and may have to be tested for that.  The modified bit in unmanaged
 258  *  mappings or in the special clean map is not tested.
 259  *
 260  *  PG_SWAPPED indicates that the page is backed by a swap block.  Any
 261  *  VM object type other than OBJT_DEFAULT can have swap-backed pages now.
 262  */
 263 #define PG_UNUSED0001   0x00000001
 264 #define PG_UNUSED0002   0x00000002
 265 #define PG_WINATCFLS    0x00000004      /* flush dirty page on inactive q */
 266 #define PG_FICTITIOUS   0x00000008      /* physical page doesn't exist (O) */
 267 #define PG_WRITEABLE    0x00000010      /* page is writeable */
 268 #define PG_MAPPED       0x00000020      /* page is mapped (managed) */
 269 #define PG_UNUSED0040   0x00000040
 270 #define PG_REFERENCED   0x00000080      /* page has been referenced */
 271 #define PG_CLEANCHK     0x00000100      /* page will be checked for cleaning */
 272 #define PG_UNUSED0200   0x00000200
 273 #define PG_NOSYNC       0x00000400      /* do not collect for syncer */
 274 #define PG_UNMANAGED    0x00000800      /* No PV management for page */
 275 #define PG_MARKER       0x00001000      /* special queue marker page */
 276 #define PG_RAM          0x00002000      /* read ahead mark */
 277 #define PG_SWAPPED      0x00004000      /* backed by swap */
 278 #define PG_NOTMETA      0x00008000      /* do not back with swap */
 279 #define PG_UNUSED10000  0x00010000
 280 #define PG_UNUSED20000  0x00020000
 281 #define PG_NEED_COMMIT  0x00040000      /* clean page requires commit */
 282
 283 #define PG_KEEP_NEWPAGE_MASK    (0)
 284
 285 /*
 286  * Misc constants.
 287  */
 288
 289 #define ACT_DECLINE             1
 290 #define ACT_ADVANCE             3
 291 #define ACT_INIT                5
 292 #define ACT_MAX                 64
 293
 294 #ifdef VM_PAGE_DEBUG
 295 #define VM_PAGE_DEBUG_EXT(name) name ## _debug
 296 #define VM_PAGE_DEBUG_ARGS      , const char *func, int lineno
 297 #else
 298 #define VM_PAGE_DEBUG_EXT(name) name
 299 #define VM_PAGE_DEBUG_ARGS
 300 #endif
 301
 302 #ifdef _KERNEL
 303 /*
 304  * Each pageable resident page falls into one of four lists:
 305  *
 306  *      free
 307  *              Available for allocation now.
 308  *
 309  * The following are all LRU sorted:
 310  *
 311  *      cache
 312  *              Almost available for allocation. Still in an
 313  *              object, but clean and immediately freeable at
 314  *              non-interrupt times.
 315  *
 316  *      inactive
 317  *              Low activity, candidates for reclamation.
 318  *              This is the list of pages that should be
 319  *              paged out next.
 320  *
 321  *      active
 322  *              Pages that are "active" i.e. they have been
 323  *              recently referenced.
 324  *
 325  *      zero
 326  *              Pages that are really free and have been pre-zeroed
 327  *
 328  */
 329
 330 extern struct vm_page *vm_page_array;   /* First resident page in table */
 331 extern vm_pindex_t vm_page_array_size;  /* number of vm_page_t's */
 332 extern vm_pindex_t first_page;          /* first physical page number */
 333
 334 #define VM_PAGE_TO_PHYS(entry)  \
 335                 ((entry)->phys_addr)
 336
 337 #define PHYS_TO_VM_PAGE(pa)     \
 338                 (&vm_page_array[atop(pa) - first_page])
 339
 340
 341 #if PAGE_SIZE == 4096
 342 #define VM_PAGE_BITS_ALL 0xff
 343 #endif
 344
 345 /*
 346  * Note: the code will always use nominally free pages from the free list
 347  * before trying other flag-specified sources.
 348  *
 349  * At least one of VM_ALLOC_NORMAL|VM_ALLOC_SYSTEM|VM_ALLOC_INTERRUPT
 350  * must be specified.  VM_ALLOC_RETRY may only be specified if VM_ALLOC_NORMAL
 351  * is also specified.
 352  */
 353 #define VM_ALLOC_NORMAL         0x0001  /* ok to use cache pages */
 354 #define VM_ALLOC_SYSTEM         0x0002  /* ok to exhaust most of free list */
 355 #define VM_ALLOC_INTERRUPT      0x0004  /* ok to exhaust entire free list */
 356 #define VM_ALLOC_ZERO           0x0008  /* req pre-zero'd memory if avail */
 357 #define VM_ALLOC_QUICK          0x0010  /* like NORMAL but do not use cache */
 358 #define VM_ALLOC_FORCE_ZERO     0x0020  /* zero page even if already valid */
 359 #define VM_ALLOC_NULL_OK        0x0040  /* ok to return NULL on collision */
 360 #define VM_ALLOC_RETRY          0x0080  /* indefinite block (vm_page_grab()) */
 361 #define VM_ALLOC_USE_GD         0x0100  /* use per-gd cache */
 362 #define VM_ALLOC_CPU_SPEC       0x0200
 363
 364 #define VM_ALLOC_CPU_SHIFT      16
 365 #define VM_ALLOC_CPU(n)         (((n) << VM_ALLOC_CPU_SHIFT) | \
 366                                  VM_ALLOC_CPU_SPEC)
 367 #define VM_ALLOC_GETCPU(flags)  ((flags) >> VM_ALLOC_CPU_SHIFT)
 368
 369 void vm_page_queue_spin_lock(vm_page_t);
 370 void vm_page_queues_spin_lock(u_short);
 371 void vm_page_and_queue_spin_lock(vm_page_t);
 372
 373 void vm_page_queue_spin_unlock(vm_page_t);
 374 void vm_page_queues_spin_unlock(u_short);
 375 void vm_page_and_queue_spin_unlock(vm_page_t m);
 376
 377 void vm_page_init(vm_page_t m);
 378 void vm_page_io_finish(vm_page_t m);
 379 void vm_page_io_start(vm_page_t m);
 380 void vm_page_need_commit(vm_page_t m);
 381 void vm_page_clear_commit(vm_page_t m);
 382 void vm_page_wakeup(vm_page_t m);
 383 void vm_page_hold(vm_page_t);
 384 void vm_page_unhold(vm_page_t);
 385 void vm_page_activate (vm_page_t);
 386
 387 vm_size_t vm_contig_avail_pages(void);
 388 vm_page_t vm_page_alloc (struct vm_object *, vm_pindex_t, int);
 389 vm_page_t vm_page_alloc_contig(vm_paddr_t low, vm_paddr_t high,
 390                      unsigned long alignment, unsigned long boundary,
 391                      unsigned long size, vm_memattr_t memattr);
 392
 393 vm_page_t vm_page_grab (struct vm_object *, vm_pindex_t, int);
 394 void vm_page_cache (vm_page_t);
 395 int vm_page_try_to_cache (vm_page_t);
 396 int vm_page_try_to_free (vm_page_t);
 397 void vm_page_dontneed (vm_page_t);
 398 void vm_page_deactivate (vm_page_t);
 399 void vm_page_deactivate_locked (vm_page_t);
 400 void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
 401 int vm_page_insert (vm_page_t, struct vm_object *, vm_pindex_t);
 402
 403 vm_page_t vm_page_hash_get(vm_object_t object, vm_pindex_t pindex);
 404
 405 vm_page_t vm_page_lookup (struct vm_object *, vm_pindex_t);
 406 vm_page_t vm_page_lookup_sbusy_try(struct vm_object *object,
 407                 vm_pindex_t pindex, int pgoff, int pgbytes);
 408 vm_page_t VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_wait)(
 409                 struct vm_object *, vm_pindex_t, int, const char *
 410                 VM_PAGE_DEBUG_ARGS);
 411 vm_page_t VM_PAGE_DEBUG_EXT(vm_page_lookup_busy_try)(
 412                 struct vm_object *, vm_pindex_t, int, int *
 413                 VM_PAGE_DEBUG_ARGS);
 414 void vm_page_remove (vm_page_t);
 415 void vm_page_rename (vm_page_t, struct vm_object *, vm_pindex_t);
 416 void vm_page_startup (void);
 417 void vm_numa_organize(vm_paddr_t ran_beg, vm_paddr_t bytes, int physid);
 418 void vm_numa_organize_finalize(void);
 419 void vm_page_unmanage (vm_page_t);
 420 void vm_page_unwire (vm_page_t, int);
 421 void vm_page_wire (vm_page_t);
 422 void vm_page_unqueue (vm_page_t);
 423 void vm_page_unqueue_nowakeup (vm_page_t);
 424 vm_page_t vm_page_next (vm_page_t);
 425 void vm_page_set_validclean (vm_page_t, int, int);
 426 void vm_page_set_validdirty (vm_page_t, int, int);
 427 void vm_page_set_valid (vm_page_t, int, int);
 428 void vm_page_set_dirty (vm_page_t, int, int);
 429 void vm_page_clear_dirty (vm_page_t, int, int);
 430 void vm_page_set_invalid (vm_page_t, int, int);
 431 int vm_page_is_valid (vm_page_t, int, int);
 432 void vm_page_test_dirty (vm_page_t);
 433 int vm_page_bits (int, int);
 434 vm_page_t vm_page_list_find(int basequeue, int index);
 435 void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
 436 void vm_page_free_toq(vm_page_t m);
 437 void vm_page_free_contig(vm_page_t m, unsigned long size);
 438 vm_page_t vm_page_free_fromq_fast(void);
 439 void vm_page_dirty(vm_page_t m);
 440 void vm_page_sleep_busy(vm_page_t m, int also_m_busy, const char *msg);
 441 int vm_page_sbusy_try(vm_page_t m);
 442 void VM_PAGE_DEBUG_EXT(vm_page_busy_wait)(vm_page_t m,
 443                         int also_m_busy, const char *wmsg VM_PAGE_DEBUG_ARGS);
 444 int VM_PAGE_DEBUG_EXT(vm_page_busy_try)(vm_page_t m,
 445                         int also_m_busy VM_PAGE_DEBUG_ARGS);
 446 u_short vm_get_pg_color(int cpuid, vm_object_t object, vm_pindex_t pindex);
 447
 448 #ifdef VM_PAGE_DEBUG
 449
 450 #define vm_page_lookup_busy_wait(object, pindex, alsob, msg)            \
 451         vm_page_lookup_busy_wait_debug(object, pindex, alsob, msg,      \
 452                                         __func__, __LINE__)
 453
 454 #define vm_page_lookup_busy_try(object, pindex, alsob, errorp)          \
 455         vm_page_lookup_busy_try_debug(object, pindex, alsob, errorp,    \
 456                                         __func__, __LINE__)
 457
 458 #define vm_page_busy_wait(m, alsob, msg)                                \
 459         vm_page_busy_wait_debug(m, alsob, msg, __func__, __LINE__)
 460
 461 #define vm_page_busy_try(m, alsob)                                      \
 462         vm_page_busy_try_debug(m, alsob, __func__, __LINE__)
 463
 464 #endif
 465
 466 #endif                          /* _KERNEL */
 467 #endif                          /* !_VM_VM_PAGE_H_ */