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  14 .\" manual page may be incorrect or out-of-date.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 2020-04-11 "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 .PP
  41 .BI "int madvise(void *" addr ", size_t " length ", int " advice );
  42 .PP
  43 .RS -4
  44 Feature Test Macro Requirements for glibc (see
  45 .BR feature_test_macros (7)):
  46 .RE
  47 .PP
  48 .BR madvise ():
  49 .PD 0
  50 .RS 4
  51 .TP 4
  52 Since glibc 2.19:
  53 _DEFAULT_SOURCE
  54 .TP
  55 Up to and including glibc 2.19:
  56 _BSD_SOURCE
  57 .RE
  58 .PD
  59 .SH DESCRIPTION
  60 The
  61 .BR madvise ()
  62 system call is used to give advice or directions to the kernel
  63 about the address range beginning at address
  64 .I addr
  65 and with size
  66 .I length
  67 bytes
  68 In most cases,
  69 the goal of such advice is to improve system or application performance.
  70 .PP
  71 Initially, the system call supported a set of "conventional"
  72 .I advice
  73 values, which are also available on several other implementations.
  74 (Note, though, that
  75 .BR madvise ()
  76 is not specified in POSIX.)
  77 Subsequently, a number of Linux-specific
  78 .IR advice
  79 values have been added.
  80 .\"
  81 .\" ======================================================================
  82 .\"
  83 .SS Conventional advice values
  84 The
  85 .I advice
  86 values listed below
  87 allow an application to tell the kernel how it expects to use
  88 some mapped or shared memory areas, so that the kernel can choose
  89 appropriate read-ahead and caching techniques.
  90 These
  91 .I advice
  92 values do not influence the semantics of the application
  93 (except in the case of
  94 .BR MADV_DONTNEED ),
  95 but may influence its performance.
  96 All of the
  97 .I advice
  98 values listed here have analogs in the POSIX-specified
  99 .BR posix_madvise (3)
 100 function, and the values have the same meanings, with the exception of
 101 .BR MADV_DONTNEED .
 102 .PP
 103 The advice is indicated in the
 104 .I advice
 105 argument, which is one of the following:
 106 .TP
 107 .B MADV_NORMAL
 108 No special treatment.
 109 This is the default.
 110 .TP
 111 .B MADV_RANDOM
 112 Expect page references in random order.
 113 (Hence, read ahead may be less useful than normally.)
 114 .TP
 115 .B MADV_SEQUENTIAL
 116 Expect page references in sequential order.
 117 (Hence, pages in the given range can be aggressively read ahead,
 118 and may be freed soon after they are accessed.)
 119 .TP
 120 .B MADV_WILLNEED
 121 Expect access in the near future.
 122 (Hence, it might be a good idea to read some pages ahead.)
 123 .TP
 124 .B MADV_DONTNEED
 125 Do not expect access in the near future.
 126 (For the time being, the application is finished with the given range,
 127 so the kernel can free resources associated with it.)
 128 .IP
 129 After a successful
 130 .B MADV_DONTNEED
 131 operation,
 132 the semantics of memory access in the specified region are changed:
 133 subsequent accesses of pages in the range will succeed, but will result
 134 in either repopulating the memory contents from the
 135 up-to-date contents of the underlying mapped file
 136 (for shared file mappings, shared anonymous mappings,
 137 and shmem-based techniques such as System V shared memory segments)
 138 or zero-fill-on-demand pages for anonymous private mappings.
 139 .IP
 140 Note that, when applied to shared mappings,
 141 .BR MADV_DONTNEED
 142 might not lead to immediate freeing of the pages in the range.
 143 The kernel is free to delay freeing the pages until an appropriate moment.
 144 The resident set size (RSS) of the calling process will be immediately
 145 reduced however.
 146 .IP
 147 .B MADV_DONTNEED
 148 cannot be applied to locked pages, Huge TLB pages, or
 149 .BR VM_PFNMAP
 150 pages.
 151 (Pages marked with the kernel-internal
 152 .B VM_PFNMAP
 153 .\" http://lwn.net/Articles/162860/
 154 flag are special memory areas that are not managed
 155 by the virtual memory subsystem.
 156 Such pages are typically created by device drivers that
 157 map the pages into user space.)
 158 .\"
 159 .\" ======================================================================
 160 .\"
 161 .SS Linux-specific advice values
 162 The following Linux-specific
 163 .I advice
 164 values have no counterparts in the POSIX-specified
 165 .BR posix_madvise (3),
 166 and may or may not have counterparts in the
 167 .BR madvise ()
 168 interface available on other implementations.
 169 Note that some of these operations change the semantics of memory accesses.
 170 .TP
 171 .BR MADV_REMOVE " (since Linux 2.6.16)"
 172 .\" commit f6b3ec238d12c8cc6cc71490c6e3127988460349
 173 Free up a given range of pages
 174 and its associated backing store.
 175 This is equivalent to punching a hole in the corresponding byte
 176 range of the backing store (see
 177 .BR fallocate (2)).
 178 Subsequent accesses in the specified address range will see
 179 bytes containing zero.
 180 .\" Databases want to use this feature to drop a section of their
 181 .\" bufferpool (shared memory segments) - without writing back to
 182 .\" disk/swap space.  This feature is also useful for supporting
 183 .\" hot-plug memory on UML.
 184 .IP
 185 The specified address range must be mapped shared and writable.
 186 This flag cannot be applied to locked pages, Huge TLB pages, or
 187 .BR VM_PFNMAP
 188 pages.
 189 .IP
 190 In the initial implementation, only
 191 .BR tmpfs (5)
 192 was supported
 193 .BR MADV_REMOVE ;
 194 but since Linux 3.5,
 195 .\" commit 3f31d07571eeea18a7d34db9af21d2285b807a17
 196 any filesystem which supports the
 197 .BR fallocate (2)
 198 .BR FALLOC_FL_PUNCH_HOLE
 199 mode also supports
 200 .BR MADV_REMOVE .
 201 Hugetlbfs fails with the error
 202 .BR EINVAL
 203 and other filesystems fail with the error
 204 .BR EOPNOTSUPP .
 205 .TP
 206 .BR MADV_DONTFORK " (since Linux 2.6.16)"
 207 .\" commit f822566165dd46ff5de9bf895cfa6c51f53bb0c4
 208 .\" See http://lwn.net/Articles/171941/
 209 Do not make the pages in this range available to the child after a
 210 .BR fork (2).
 211 This is useful to prevent copy-on-write semantics from changing
 212 the physical location of a page if the parent writes to it after a
 213 .BR fork (2).
 214 (Such page relocations cause problems for hardware that
 215 DMAs into the page.)
 216 .\" [PATCH] madvise MADV_DONTFORK/MADV_DOFORK
 217 .\" Currently, copy-on-write may change the physical address of
 218 .\" a page even if the user requested that the page is pinned in
 219 .\" memory (either by mlock or by get_user_pages).  This happens
 220 .\" if the process forks meanwhile, and the parent writes to that
 221 .\" page.  As a result, the page is orphaned: in case of
 222 .\" get_user_pages, the application will never see any data hardware
 223 .\" DMA's into this page after the COW.  In case of mlock'd memory,
 224 .\" the parent is not getting the realtime/security benefits of mlock.
 225 .\"
 226 .\" In particular, this affects the Infiniband modules which do DMA from
 227 .\" and into user pages all the time.
 228 .\"
 229 .\" This patch adds madvise options to control whether memory range is
 230 .\" inherited across fork. Useful e.g. for when hardware is doing DMA
 231 .\" from/into these pages.  Could also be useful to an application
 232 .\" wanting to speed up its forks by cutting large areas out of
 233 .\" consideration.
 234 .\"
 235 .\" SEE ALSO: http://lwn.net/Articles/171941/
 236 .\" "Tweaks to madvise() and posix_fadvise()", 14 Feb 2006
 237 .TP
 238 .BR MADV_DOFORK " (since Linux 2.6.16)"
 239 Undo the effect of
 240 .BR MADV_DONTFORK ,
 241 restoring the default behavior, whereby a mapping is inherited across
 242 .BR fork (2).
 243 .TP
 244 .BR MADV_HWPOISON " (since Linux 2.6.32)"
 245 .\" commit 9893e49d64a4874ea67849ee2cfbf3f3d6817573
 246 Poison the pages in the range specified by
 247 .I addr
 248 and
 249 .IR length
 250 and handle subsequent references to those pages
 251 like a hardware memory corruption.
 252 This operation is available only for privileged
 253 .RB ( CAP_SYS_ADMIN )
 254 processes.
 255 This operation may result in the calling process receiving a
 256 .B SIGBUS
 257 and the page being unmapped.
 258 .IP
 259 This feature is intended for testing of memory error-handling code;
 260 it is available only if the kernel was configured with
 261 .BR CONFIG_MEMORY_FAILURE .
 262 .TP
 263 .BR MADV_MERGEABLE " (since Linux 2.6.32)"
 264 .\" commit f8af4da3b4c14e7267c4ffb952079af3912c51c5
 265 Enable Kernel Samepage Merging (KSM) for the pages in the range specified by
 266 .I addr
 267 and
 268 .IR length .
 269 The kernel regularly scans those areas of user memory that have
 270 been marked as mergeable,
 271 looking for pages with identical content.
 272 These are replaced by a single write-protected page (which is automatically
 273 copied if a process later wants to update the content of the page).
 274 KSM merges only private anonymous pages (see
 275 .BR mmap (2)).
 276 .IP
 277 The KSM feature is intended for applications that generate many
 278 instances of the same data (e.g., virtualization systems such as KVM).
 279 It can consume a lot of processing power; use with care.
 280 See the Linux kernel source file
 281 .I Documentation/admin-guide/mm/ksm.rst
 282 for more details.
 283 .IP
 284 The
 285 .BR MADV_MERGEABLE
 286 and
 287 .BR MADV_UNMERGEABLE
 288 operations are available only if the kernel was configured with
 289 .BR CONFIG_KSM .
 290 .TP
 291 .BR MADV_UNMERGEABLE " (since Linux 2.6.32)"
 292 Undo the effect of an earlier
 293 .BR MADV_MERGEABLE
 294 operation on the specified address range;
 295 KSM unmerges whatever pages it had merged in the address range specified by
 296 .IR addr
 297 and
 298 .IR length .
 299 .TP
 300 .BR MADV_SOFT_OFFLINE " (since Linux 2.6.33)"
 301 .\" commit afcf938ee0aac4ef95b1a23bac704c6fbeb26de6
 302 Soft offline the pages in the range specified by
 303 .I addr
 304 and
 305 .IR length .
 306 The memory of each page in the specified range is preserved
 307 (i.e., when next accessed, the same content will be visible,
 308 but in a new physical page frame),
 309 and the original page is offlined
 310 (i.e., no longer used, and taken out of normal memory management).
 311 The effect of the
 312 .B MADV_SOFT_OFFLINE
 313 operation is invisible to (i.e., does not change the semantics of)
 314 the calling process.
 315 .IP
 316 This feature is intended for testing of memory error-handling code;
 317 it is available only if the kernel was configured with
 318 .BR CONFIG_MEMORY_FAILURE .
 319 .TP
 320 .BR MADV_HUGEPAGE " (since Linux 2.6.38)"
 321 .\" commit 0af4e98b6b095c74588af04872f83d333c958c32
 322 .\" http://lwn.net/Articles/358904/
 323 .\" https://lwn.net/Articles/423584/
 324 Enable Transparent Huge Pages (THP) for pages in the range specified by
 325 .I addr
 326 and
 327 .IR length .
 328 Currently, Transparent Huge Pages work only with private anonymous pages (see
 329 .BR mmap (2)).
 330 The kernel will regularly scan the areas marked as huge page candidates
 331 to replace them with huge pages.
 332 The kernel will also allocate huge pages directly when the region is
 333 naturally aligned to the huge page size (see
 334 .BR posix_memalign (2)).
 335 .IP
 336 This feature is primarily aimed at applications that use large mappings of
 337 data and access large regions of that memory at a time (e.g., virtualization
 338 systems such as QEMU).
 339 It can very easily waste memory (e.g., a 2\ MB mapping that only ever accesses
 340 1 byte will result in 2\ MB of wired memory instead of one 4\ KB page).
 341 See the Linux kernel source file
 342 .I Documentation/admin-guide/mm/transhuge.rst
 343 for more details.
 344 .IP
 345 Most common kernels configurations provide
 346 .BR MADV_HUGEPAGE -style
 347 behavior by default, and thus
 348 .BR MADV_HUGEPAGE
 349 is normally not necessary.
 350 It is mostly intended for embedded systems, where
 351 .BR MADV_HUGEPAGE -style
 352 behavior may not be enabled by default in the kernel.
 353 On such systems,
 354 this flag can be used in order to selectively enable THP.
 355 Whenever
 356 .BR MADV_HUGEPAGE
 357 is used, it should always be in regions of memory with
 358 an access pattern that the developer knows in advance won't risk
 359 to increase the memory footprint of the application when transparent
 360 hugepages are enabled.
 361 .IP
 362 The
 363 .BR MADV_HUGEPAGE
 364 and
 365 .BR MADV_NOHUGEPAGE
 366 operations are available only if the kernel was configured with
 367 .BR CONFIG_TRANSPARENT_HUGEPAGE .
 368 .TP
 369 .BR MADV_NOHUGEPAGE " (since Linux 2.6.38)"
 370 Ensures that memory in the address range specified by
 371 .IR addr
 372 and
 373 .IR length
 374 will not be backed by transparent hugepages.
 375 .TP
 376 .BR MADV_DONTDUMP " (since Linux 3.4)"
 377 .\" commit 909af768e88867016f427264ae39d27a57b6a8ed
 378 .\" commit accb61fe7bb0f5c2a4102239e4981650f9048519
 379 Exclude from a core dump those pages in the range specified by
 380 .I addr
 381 and
 382 .IR length .
 383 This is useful in applications that have large areas of memory
 384 that are known not to be useful in a core dump.
 385 The effect of
 386 .BR MADV_DONTDUMP
 387 takes precedence over the bit mask that is set via the
 388 .I /proc/[pid]/coredump_filter
 389 file (see
 390 .BR core (5)).
 391 .TP
 392 .BR MADV_DODUMP " (since Linux 3.4)"
 393 Undo the effect of an earlier
 394 .BR MADV_DONTDUMP .
 395 .TP
 396 .BR MADV_FREE " (since Linux 4.5)"
 397 The application no longer requires the pages in the range specified by
 398 .IR addr
 399 and
 400 .IR len .
 401 The kernel can thus free these pages,
 402 but the freeing could be delayed until memory pressure occurs.
 403 For each of the pages that has been marked to be freed
 404 but has not yet been freed,
 405 the free operation will be canceled if the caller writes into the page.
 406 After a successful
 407 .B MADV_FREE
 408 operation, any stale data (i.e., dirty, unwritten pages) will be lost
 409 when the kernel frees the pages.
 410 However, subsequent writes to pages in the range will succeed
 411 and then kernel cannot free those dirtied pages,
 412 so that the caller can always see just written data.
 413 If there is no subsequent write,
 414 the kernel can free the pages at any time.
 415 Once pages in the range have been freed, the caller will
 416 see zero-fill-on-demand pages upon subsequent page references.
 417 .IP
 418 The
 419 .B MADV_FREE
 420 operation
 421 can be applied only to private anonymous pages (see
 422 .BR mmap (2)).
 423 In Linux before version 4.12,
 424 .\" commit 93e06c7a645343d222c9a838834a51042eebbbf7
 425 when freeing pages on a swapless system,
 426 the pages in the given range are freed instantly,
 427 regardless of memory pressure.
 428 .TP
 429 .BR MADV_WIPEONFORK " (since Linux 4.14)"
 430 .\" commit d2cd9ede6e193dd7d88b6d27399e96229a551b19
 431 Present the child process with zero-filled memory in this range after a
 432 .BR fork (2).
 433 This is useful in forking servers in order to ensure
 434 that sensitive per-process data
 435 (for example, PRNG seeds, cryptographic secrets, and so on)
 436 is not handed to child processes.
 437 .IP
 438 The
 439 .B MADV_WIPEONFORK
 440 operation can be applied only to private anonymous pages (see
 441 .BR mmap (2)).
 442 .IP
 443 Within the child created by
 444 .BR fork (2),
 445 the
 446 .B MADV_WIPEONFORK
 447 setting remains in place on the specified address range.
 448 This setting is cleared during
 449 .BR execve (2).
 450 .TP
 451 .BR MADV_KEEPONFORK " (since Linux 4.14)"
 452 .\" commit d2cd9ede6e193dd7d88b6d27399e96229a551b19
 453 Undo the effect of an earlier
 454 .BR MADV_WIPEONFORK .
 455 .SH RETURN VALUE
 456 On success,
 457 .BR madvise ()
 458 returns zero.
 459 On error, it returns \-1 and
 460 .I errno
 461 is set appropriately.
 462 .SH ERRORS
 463 .TP
 464 .B EACCES
 465 .I advice
 466 is
 467 .BR MADV_REMOVE ,
 468 but the specified address range is not a shared writable mapping.
 469 .TP
 470 .B EAGAIN
 471 A kernel resource was temporarily unavailable.
 472 .TP
 473 .B EBADF
 474 The map exists, but the area maps something that isn't a file.
 475 .TP
 476 .B EINVAL
 477 .I addr
 478 is not page-aligned or
 479 .I length
 480 is negative.
 481 .\" .I length
 482 .\" is zero,
 483 .TP
 484 .B EINVAL
 485 .I advice
 486 is not a valid.
 487 .TP
 488 .B EINVAL
 489 .I advice
 490 is
 491 .B MADV_DONTNEED
 492 or
 493 .BR MADV_REMOVE
 494 and the specified address range includes locked, Huge TLB pages, or
 495 .B VM_PFNMAP
 496 pages.
 497 .TP
 498 .B EINVAL
 499 .I advice
 500 is
 501 .BR MADV_MERGEABLE
 502 or
 503 .BR MADV_UNMERGEABLE ,
 504 but the kernel was not configured with
 505 .BR CONFIG_KSM .
 506 .TP
 507 .B EINVAL
 508 .I advice
 509 is
 510 .BR MADV_FREE
 511 or
 512 .BR MADV_WIPEONFORK
 513 but the specified address range includes file, Huge TLB,
 514 .BR MAP_SHARED ,
 515 or
 516 .BR VM_PFNMAP
 517 ranges.
 518 .TP
 519 .B EIO
 520 (for
 521 .BR MADV_WILLNEED )
 522 Paging in this area would exceed the process's
 523 maximum resident set size.
 524 .TP
 525 .B ENOMEM
 526 (for
 527 .BR MADV_WILLNEED )
 528 Not enough memory: paging in failed.
 529 .TP
 530 .B ENOMEM
 531 Addresses in the specified range are not currently
 532 mapped, or are outside the address space of the process.
 533 .TP
 534 .B EPERM
 535 .I advice
 536 is
 537 .BR MADV_HWPOISON ,
 538 but the caller does not have the
 539 .B CAP_SYS_ADMIN
 540 capability.
 541 .SH VERSIONS
 542 Since Linux 3.18,
 543 .\" commit d3ac21cacc24790eb45d735769f35753f5b56ceb
 544 support for this system call is optional,
 545 depending on the setting of the
 546 .B CONFIG_ADVISE_SYSCALLS
 547 configuration option.
 548 .SH CONFORMING TO
 549 .BR madvise ()
 550 is not specified by any standards.
 551 Versions of this system call, implementing a wide variety of
 552 .I advice
 553 values, exist on many other implementations.
 554 Other implementations typically implement at least the flags listed
 555 above under
 556 .IR "Conventional advice flags" ,
 557 albeit with some variation in semantics.
 558 .PP
 559 POSIX.1-2001 describes
 560 .BR posix_madvise (3)
 561 with constants
 562 .BR POSIX_MADV_NORMAL ,
 563 .BR POSIX_MADV_RANDOM ,
 564 .BR POSIX_MADV_SEQUENTIAL ,
 565 .BR POSIX_MADV_WILLNEED ,
 566 and
 567 .BR POSIX_MADV_DONTNEED ,
 568 and so on, with behavior close to the similarly named flags listed above.
 569 .SH NOTES
 570 .SS Linux notes
 571 The Linux implementation requires that the address
 572 .I addr
 573 be page-aligned, and allows
 574 .I length
 575 to be zero.
 576 If there are some parts of the specified address range
 577 that are not mapped, the Linux version of
 578 .BR madvise ()
 579 ignores them and applies the call to the rest (but returns
 580 .B ENOMEM
 581 from the system call, as it should).
 582 .\" .SH HISTORY
 583 .\" The
 584 .\" .BR madvise ()
 585 .\" function first appeared in 4.4BSD.
 586 .SH SEE ALSO
 587 .BR getrlimit (2),
 588 .BR mincore (2),
 589 .BR mmap (2),
 590 .BR mprotect (2),
 591 .BR msync (2),
 592 .BR munmap (2),
 593 .BR prctl (2),
 594 .BR posix_madvise (3),
 595 .BR core (5)