man2/madvise.2

   1 .\" Copyright (C) 2001 David Gómez <davidge@jazzfree.com>
   2 .\"
   3 .\" %%%LICENSE_START(VERBATIM)
   4 .\" Permission is granted to make and distribute verbatim copies of this
   5 .\" manual provided the copyright notice and this permission notice are
   6 .\" preserved on all copies.
   7 .\"
   8 .\" Permission is granted to copy and distribute modified versions of this
   9 .\" manual under the conditions for verbatim copying, provided that the
  10 .\" entire resulting derived work is distributed under the terms of a
  11 .\" permission notice identical to this one.
  12 .\"
  13 .\" Since the Linux kernel and libraries are constantly changing, this
  14 .\" manual page may be incorrect or out-of-date.  The author(s) assume no
  15 .\" responsibility for errors or omissions, or for damages resulting from
  16 .\" the use of the information contained herein.  The author(s) may not
  17 .\" have taken the same level of care in the production of this manual,
  18 .\" which is licensed free of charge, as they might when working
  19 .\" professionally.
  20 .\"
  21 .\" Formatted or processed versions of this manual, if unaccompanied by
  22 .\" the source, must acknowledge the copyright and authors of this work.
  23 .\" %%%LICENSE_END
  24 .\"
  25 .\" Based on comments from mm/filemap.c. Last modified on 10-06-2001
  26 .\" Modified, 25 Feb 2002, Michael Kerrisk, <mtk.manpages@gmail.com>
  27 .\"     Added notes on MADV_DONTNEED
  28 .\" 2010-06-19, mtk, Added documentation of MADV_MERGEABLE and
  29 .\"     MADV_UNMERGEABLE
  30 .\" 2010-06-15, Andi Kleen, Add documentation of MADV_HWPOISON.
  31 .\" 2010-06-19, Andi Kleen, Add documentation of MADV_SOFT_OFFLINE.
  32 .\" 2011-09-18, Doug Goldstein <cardoe@cardoe.com>
  33 .\"     Document MADV_HUGEPAGE and MADV_NOHUGEPAGE
  34 .\"
  35 .TH MADVISE 2 2015-02-21 "Linux" "Linux Programmer's Manual"
  36 .SH NAME
  37 madvise \- give advice about use of memory
  38 .SH SYNOPSIS
  39 .B #include <sys/mman.h>
  40 .sp
  41 .BI "int madvise(void *" addr ", size_t " length ", int " advice );
  42 .sp
  43 .in -4n
  44 Feature Test Macro Requirements for glibc (see
  45 .BR feature_test_macros (7)):
  46 .in
  47 .sp
  48 .BR madvise ():
  49 _BSD_SOURCE
  50 .SH DESCRIPTION
  51 The
  52 .BR madvise ()
  53 system call is used to give advice or directions to the kernel
  54 about the address range beginning at address
  55 .I addr
  56 and with size
  57 .I length
  58 bytes.
  59 Initially, the system call supported a set of "conventional"
  60 .I advice
  61 values, which are also available on several other implementations.
  62 (Note, though, that
  63 .BR madvise ()
  64 is not specified in POSIX.)
  65 Subsequently, a number of Linux-specific
  66 .IR advice
  67 values have been added.
  68 .\"
  69 .\" ======================================================================
  70 .\"
  71 .SS Conventional advice values
  72 The
  73 .I advice
  74 values listed below
  75 allow an application to tell the kernel how it expects to use
  76 some mapped or shared memory areas, so that the kernel can choose
  77 appropriate read-ahead and caching techniques.
  78 These
  79 .I advice
  80 values do not influence the semantics of the application
  81 (except in the case of
  82 .BR MADV_DONTNEED ),
  83 but may influence its performance.
  84 All of the
  85 .I advice
  86 values listed here have analogs in the POSIX-specified
  87 .BR posix_madvise (3)
  88 function, and the values have the same meanings, with the exception of
  89 .BR MADV_DONTNEED .
  90 .LP
  91 The advice is indicated in the
  92 .I advice
  93 argument, which is one of the following:
  94 .TP
  95 .B MADV_NORMAL
  96 No special treatment.
  97 This is the default.
  98 .TP
  99 .B MADV_RANDOM
 100 Expect page references in random order.
 101 (Hence, read ahead may be less useful than normally.)
 102 .TP
 103 .B MADV_SEQUENTIAL
 104 Expect page references in sequential order.
 105 (Hence, pages in the given range can be aggressively read ahead,
 106 and may be freed soon after they are accessed.)
 107 .TP
 108 .B MADV_WILLNEED
 109 Expect access in the near future.
 110 (Hence, it might be a good idea to read some pages ahead.)
 111 .TP
 112 .B MADV_DONTNEED
 113 Do not expect access in the near future.
 114 (For the time being, the application is finished with the given range,
 115 so the kernel can free resources associated with it.)
 116
 117 After a successful
 118 .B MADV_DONTNEED
 119 operation,
 120 the semantics of memory access in the specified region are changed:
 121 subsequent accesses of pages in the range will succeed, but will result
 122 in either repopulating the memory contents from the
 123 up-to-date contents of the underlying mapped file
 124 (for shared file mappings, shared anonymous mappings,
 125 and shmem-based techniques such as System V shared memory segments)
 126 or zero-fill-on-demand pages for anonymous private mappings.
 127
 128 Note that, when applied to shared mappings,
 129 .BR MADV_DONTNEED
 130 might not lead to immediate freeing of the pages in the range.
 131 The kernel is free to delay freeing the pages until an appropriate moment.
 132 The resident set size (RSS) of the calling process will be immediately
 133 reduced however.
 134
 135 .B MADV_DONTNEED
 136 cannot be applied to locked pages, Huge TLB pages, or
 137 .BR VM_PFNMAP
 138 pages.
 139 (Pages marked with the kernel-internal
 140 .B VM_PFNMAP
 141 .\" http://lwn.net/Articles/162860/
 142 flag are special memory areas that are not managed
 143 by the virtual memory subsystem.
 144 Such pages are typically created by device drivers that
 145 map the pages into user space.)
 146 .\"
 147 .\" ======================================================================
 148 .\"
 149 .SS Linux-specific advice values
 150 The following Linux-specific
 151 .I advice
 152 values have no counterparts in the POSIX-specified
 153 .BR posix_madvise (3),
 154 and may or may not have counterparts in the
 155 .BR madvise ()
 156 interface available on other implementations.
 157 Note that some of these operations change the semantics of memory accesses.
 158 .TP
 159 .BR MADV_REMOVE " (since Linux 2.6.16)"
 160 .\" commit f6b3ec238d12c8cc6cc71490c6e3127988460349
 161 Free up a given range of pages
 162 and its associated backing store.
 163 This is equivalent to punching a hole in the corresponding byte
 164 range of the backing store (see
 165 .BR fallocate (2)).
 166 Subsequent accesses in the specified address range will see
 167 bytes containing zero.
 168 .\" Databases want to use this feature to drop a section of their
 169 .\" bufferpool (shared memory segments) - without writing back to
 170 .\" disk/swap space.  This feature is also useful for supporting
 171 .\" hot-plug memory on UML.
 172
 173 The specified address range must be mapped shared and writable.
 174 This flag cannot be applied to locked pages, Huge TLB pages, or
 175 .BR VM_PFNMAP
 176 pages.
 177
 178 In the initial implementation, only shmfs/tmpfs supported
 179 .BR MADV_REMOVE ;
 180 but since Linux 3.5,
 181 .\" commit 3f31d07571eeea18a7d34db9af21d2285b807a17
 182 any filesystem which supports the
 183 .BR fallocate (2)
 184 .BR FALLOC_FL_PUNCH_HOLE
 185 mode also supports
 186 .BR MADV_REMOVE .
 187 Other filesystems fail with the error
 188 .BR EOPNOTSUPP .
 189 .TP
 190 .BR MADV_DONTFORK " (since Linux 2.6.16)"
 191 .\" commit f822566165dd46ff5de9bf895cfa6c51f53bb0c4
 192 .\" See http://lwn.net/Articles/171941/
 193 Do not make the pages in this range available to the child after a
 194 .BR fork (2).
 195 This is useful to prevent copy-on-write semantics from changing
 196 the physical location of a page if the parent writes to it after a
 197 .BR fork (2).
 198 (Such page relocations cause problems for hardware that
 199 DMAs into the page.)
 200 .\" [PATCH] madvise MADV_DONTFORK/MADV_DOFORK
 201 .\" Currently, copy-on-write may change the physical address of
 202 .\" a page even if the user requested that the page is pinned in
 203 .\" memory (either by mlock or by get_user_pages).  This happens
 204 .\" if the process forks meanwhile, and the parent writes to that
 205 .\" page.  As a result, the page is orphaned: in case of
 206 .\" get_user_pages, the application will never see any data hardware
 207 .\" DMA's into this page after the COW.  In case of mlock'd memory,
 208 .\" the parent is not getting the realtime/security benefits of mlock.
 209 .\"
 210 .\" In particular, this affects the Infiniband modules which do DMA from
 211 .\" and into user pages all the time.
 212 .\"
 213 .\" This patch adds madvise options to control whether memory range is
 214 .\" inherited across fork. Useful e.g. for when hardware is doing DMA
 215 .\" from/into these pages.  Could also be useful to an application
 216 .\" wanting to speed up its forks by cutting large areas out of
 217 .\" consideration.
 218 .\"
 219 .\" SEE ALSO: http://lwn.net/Articles/171941/
 220 .\" "Tweaks to madvise() and posix_fadvise()", 14 Feb 2006
 221 .TP
 222 .BR MADV_DOFORK " (since Linux 2.6.16)"
 223 Undo the effect of
 224 .BR MADV_DONTFORK ,
 225 restoring the default behavior, whereby a mapping is inherited across
 226 .BR fork (2).
 227 .TP
 228 .BR MADV_HWPOISON " (since Linux 2.6.32)
 229 .\" commit 9893e49d64a4874ea67849ee2cfbf3f3d6817573
 230 Poison a page and handle it like a hardware memory corruption.
 231 This operation is available only for privileged
 232 .RB ( CAP_SYS_ADMIN )
 233 processes.
 234 This operation may result in the calling process receiving a
 235 .B SIGBUS
 236 and the page being unmapped.
 237
 238 This feature is intended for testing of memory error-handling code;
 239 it is available only if the kernel was configured with
 240 .BR CONFIG_MEMORY_FAILURE .
 241 .TP
 242 .BR MADV_SOFT_OFFLINE " (since Linux 2.6.33)
 243 .\" commit afcf938ee0aac4ef95b1a23bac704c6fbeb26de6
 244 Soft offline the pages in the range specified by
 245 .I addr
 246 and
 247 .IR length .
 248 The memory of each page in the specified range is preserved
 249 (i.e., when next accessed, the same content will be visible,
 250 but in a new physical page frame),
 251 and the original page is offlined
 252 (i.e., no longer used, and taken out of normal memory management).
 253 The effect of the
 254 .B MADV_SOFT_OFFLINE
 255 operation is invisible to (i.e., does not change the semantics of)
 256 the calling process.
 257
 258 This feature is intended for testing of memory error-handling code;
 259 it is available only if the kernel was configured with
 260 .BR CONFIG_MEMORY_FAILURE .
 261 .TP
 262 .BR MADV_MERGEABLE " (since Linux 2.6.32)"
 263 .\" commit f8af4da3b4c14e7267c4ffb952079af3912c51c5
 264 Enable Kernel Samepage Merging (KSM) for the pages in the range specified by
 265 .I addr
 266 and
 267 .IR length .
 268 The kernel regularly scans those areas of user memory that have
 269 been marked as mergeable,
 270 looking for pages with identical content.
 271 These are replaced by a single write-protected page (which is automatically
 272 copied if a process later wants to update the content of the page).
 273 KSM merges only private anonymous pages (see
 274 .BR mmap (2)).
 275
 276 The KSM feature is intended for applications that generate many
 277 instances of the same data (e.g., virtualization systems such as KVM).
 278 It can consume a lot of processing power; use with care.
 279 See the Linux kernel source file
 280 .I Documentation/vm/ksm.txt
 281 for more details.
 282
 283 The
 284 .BR MADV_MERGEABLE
 285 and
 286 .BR MADV_UNMERGEABLE
 287 operations are available only if the kernel was configured with
 288 .BR CONFIG_KSM .
 289 .TP
 290 .BR MADV_UNMERGEABLE " (since Linux 2.6.32)"
 291 Undo the effect of an earlier
 292 .BR MADV_MERGEABLE
 293 operation on the specified address range;
 294 KSM unmerges whatever pages it had merged in the address range specified by
 295 .IR addr
 296 and
 297 .IR length .
 298 .TP
 299 .BR MADV_HUGEPAGE " (since Linux 2.6.38)"
 300 .\" commit 0af4e98b6b095c74588af04872f83d333c958c32
 301 .\" http://lwn.net/Articles/358904/
 302 .\" https://lwn.net/Articles/423584/
 303 Enable Transparent Huge Pages (THP) for pages in the range specified by
 304 .I addr
 305 and
 306 .IR length .
 307 Currently, Transparent Huge Pages work only with private anonymous pages (see
 308 .BR mmap (2)).
 309 The kernel will regularly scan the areas marked as huge page candidates
 310 to replace them with huge pages.
 311 The kernel will also allocate huge pages directly when the region is
 312 naturally aligned to the huge page size (see
 313 .BR posix_memalign (2)).
 314
 315 This feature is primarily aimed at applications that use large mappings of
 316 data and access large regions of that memory at a time (e.g., virtualization
 317 systems such as QEMU).
 318 It can very easily waste memory (e.g., a 2MB mapping that only ever accesses
 319 1 byte will result in 2MB of wired memory instead of one 4KB page).
 320 See the Linux kernel source file
 321 .I Documentation/vm/transhuge.txt
 322 for more details.
 323
 324 The
 325 .BR MADV_HUGEPAGE
 326 and
 327 .BR MADV_NOHUGEPAGE
 328 operations are available only if the kernel was configured with
 329 .BR CONFIG_TRANSPARENT_HUGEPAGE .
 330 .TP
 331 .BR MADV_NOHUGEPAGE " (since Linux 2.6.38)"
 332 Ensures that memory in the address range specified by
 333 .IR addr
 334 and
 335 .IR length
 336 will not be collapsed into huge pages.
 337 .TP
 338 .BR MADV_DONTDUMP " (since Linux 3.4)"
 339 .\" commit 909af768e88867016f427264ae39d27a57b6a8ed
 340 .\" commit accb61fe7bb0f5c2a4102239e4981650f9048519
 341 Exclude from a core dump those pages in the range specified by
 342 .I addr
 343 and
 344 .IR length .
 345 This is useful in applications that have large areas of memory
 346 that are known not to be useful in a core dump.
 347 The effect of
 348 .BR MADV_DONTDUMP
 349 takes precedence over the bit mask that is set via the
 350 .I /proc/PID/coredump_filter
 351 file (see
 352 .BR core (5)).
 353 .TP
 354 .BR MADV_DODUMP " (since Linux 3.4)"
 355 Undo the effect of an earlier
 356 .BR MADV_DONTDUMP .
 357 .SH RETURN VALUE
 358 On success,
 359 .BR madvise ()
 360 returns zero.
 361 On error, it returns \-1 and
 362 .I errno
 363 is set appropriately.
 364 .SH ERRORS
 365 .TP
 366 .B EACCES
 367 .I advice
 368 is
 369 .BR MADV_REMOVE ,
 370 but the specified address range is not a shared writable mapping.
 371 .TP
 372 .B EAGAIN
 373 A kernel resource was temporarily unavailable.
 374 .TP
 375 .B EBADF
 376 The map exists, but the area maps something that isn't a file.
 377 .TP
 378 .B EINVAL
 379 .I addr
 380 is not page-aligned or
 381 .I length
 382 is negative.
 383 .\" .I length
 384 .\" is zero,
 385 .TP
 386 .B EINVAL
 387 .I advice
 388 is not a valid.
 389 .TP
 390 .B EINVAL
 391 .I advice
 392 is
 393 .B MADV_DONTNEED
 394 or
 395 .BR MADV_REMOVE
 396 and the specified address range includes locked, Huge TLB pages, or
 397 .B VM_PFNMAP
 398 pages.
 399 .TP
 400 .B EINVAL
 401 .I advice
 402 is
 403 .BR MADV_MERGEABLE
 404 or
 405 .BR MADV_UNMERGEABLE ,
 406 but the kernel was not configured with
 407 .BR CONFIG_KSM .
 408 .TP
 409 .B EIO
 410 (for
 411 .BR MADV_WILLNEED )
 412 Paging in this area would exceed the process's
 413 maximum resident set size.
 414 .TP
 415 .B ENOMEM
 416 (for
 417 .BR MADV_WILLNEED )
 418 Not enough memory: paging in failed.
 419 .TP
 420 .B ENOMEM
 421 Addresses in the specified range are not currently
 422 mapped, or are outside the address space of the process.
 423 .TP
 424 .B EPERM
 425 .I advice
 426 is
 427 .BR MADV_HWPOISON ,
 428 but the caller does not have the
 429 .B CAP_SYS_ADMIN
 430 capability.
 431 .SH VERSIONS
 432 Since Linux 3.18,
 433 .\" commit d3ac21cacc24790eb45d735769f35753f5b56ceb
 434 support for this system call is optional,
 435 depending on the setting of the
 436 .B CONFIG_ADVISE_SYSCALLS
 437 configuration option.
 438 .SH CONFORMING TO
 439 .BR madvise ()
 440 is not specified by any standards.
 441 Versions of this system call, implementing a wide variety of
 442 .I advice
 443 values, exist on many other implementations.
 444 Other implementations typically implement at least the flags listed
 445 above under
 446 .IR "Conventional advice flags" ,
 447 albeit with some variation in semantics.
 448
 449 POSIX.1-2001 describes
 450 .BR posix_madvise (3)
 451 with constants
 452 .BR POSIX_MADV_NORMAL ,
 453 .BR POSIX_MADV_RANDOM ,
 454 .BR POSIX_MADV_SEQUENTIAL ,
 455 .BR POSIX_MADV_WILLNEED ,
 456 and
 457 .BR POSIX_MADV_DONTNEED ,
 458 and so on, with behavior close to the similarly named flags listed above.
 459 (POSIX.1-2008 adds a further flag,
 460 .BR POSIX_MADV_NOREUSE ,
 461 that has no analog in
 462 .BR madvise (2).)
 463 .SH NOTES
 464 .SS Linux notes
 465 The Linux implementation requires that the address
 466 .I addr
 467 be page-aligned, and allows
 468 .I length
 469 to be zero.
 470 If there are some parts of the specified address range
 471 that are not mapped, the Linux version of
 472 .BR madvise ()
 473 ignores them and applies the call to the rest (but returns
 474 .B ENOMEM
 475 from the system call, as it should).
 476 .\" .SH HISTORY
 477 .\" The
 478 .\" .BR madvise ()
 479 .\" function first appeared in 4.4BSD.
 480 .SH SEE ALSO
 481 .BR getrlimit (2),
 482 .BR mincore (2),
 483 .BR mmap (2),
 484 .BR mprotect (2),
 485 .BR msync (2),
 486 .BR munmap (2),
 487 .BR posix_fadvise (2),
 488 .BR prctl (2),
 489 .BR core (5)