1 .\" Copyright (C) 2001 David Gómez <davidge@jazzfree.com>
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.
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.
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14 .\" manual page may be incorrect or out-of-date.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
21 .\" Formatted or processed versions of this manual, if unaccompanied by
22 .\" the source, must acknowledge the copyright and authors of this work.
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
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
35 .TH MADVISE 2 2021-03-22 "Linux" "Linux Programmer's Manual"
37 madvise \- give advice about use of memory
40 .B #include <sys/mman.h>
42 .BI "int madvise(void *" addr ", size_t " length ", int " advice );
46 Feature Test Macro Requirements for glibc (see
47 .BR feature_test_macros (7)):
54 Up to and including glibc 2.19:
60 system call is used to give advice or directions to the kernel
61 about the address range beginning at address
67 the goal of such advice is to improve system or application performance.
69 Initially, the system call supported a set of "conventional"
71 values, which are also available on several other implementations.
74 is not specified in POSIX.)
75 Subsequently, a number of Linux-specific
77 values have been added.
79 .\" ======================================================================
81 .SS Conventional advice values
85 allow an application to tell the kernel how it expects to use
86 some mapped or shared memory areas, so that the kernel can choose
87 appropriate read-ahead and caching techniques.
90 values do not influence the semantics of the application
91 (except in the case of
93 but may influence its performance.
96 values listed here have analogs in the POSIX-specified
98 function, and the values have the same meanings, with the exception of
101 The advice is indicated in the
103 argument, which is one of the following:
106 No special treatment.
110 Expect page references in random order.
111 (Hence, read ahead may be less useful than normally.)
114 Expect page references in sequential order.
115 (Hence, pages in the given range can be aggressively read ahead,
116 and may be freed soon after they are accessed.)
119 Expect access in the near future.
120 (Hence, it might be a good idea to read some pages ahead.)
123 Do not expect access in the near future.
124 (For the time being, the application is finished with the given range,
125 so the kernel can free resources associated with it.)
130 the semantics of memory access in the specified region are changed:
131 subsequent accesses of pages in the range will succeed, but will result
132 in either repopulating the memory contents from the
133 up-to-date contents of the underlying mapped file
134 (for shared file mappings, shared anonymous mappings,
135 and shmem-based techniques such as System V shared memory segments)
136 or zero-fill-on-demand pages for anonymous private mappings.
138 Note that, when applied to shared mappings,
140 might not lead to immediate freeing of the pages in the range.
141 The kernel is free to delay freeing the pages until an appropriate moment.
142 The resident set size (RSS) of the calling process will be immediately
146 cannot be applied to locked pages, Huge TLB pages, or
149 (Pages marked with the kernel-internal
151 .\" http://lwn.net/Articles/162860/
152 flag are special memory areas that are not managed
153 by the virtual memory subsystem.
154 Such pages are typically created by device drivers that
155 map the pages into user space.)
157 .\" ======================================================================
159 .SS Linux-specific advice values
160 The following Linux-specific
162 values have no counterparts in the POSIX-specified
163 .BR posix_madvise (3),
164 and may or may not have counterparts in the
166 interface available on other implementations.
167 Note that some of these operations change the semantics of memory accesses.
169 .BR MADV_REMOVE " (since Linux 2.6.16)"
170 .\" commit f6b3ec238d12c8cc6cc71490c6e3127988460349
171 Free up a given range of pages
172 and its associated backing store.
173 This is equivalent to punching a hole in the corresponding byte
174 range of the backing store (see
176 Subsequent accesses in the specified address range will see
177 bytes containing zero.
178 .\" Databases want to use this feature to drop a section of their
179 .\" bufferpool (shared memory segments) - without writing back to
180 .\" disk/swap space. This feature is also useful for supporting
181 .\" hot-plug memory on UML.
183 The specified address range must be mapped shared and writable.
184 This flag cannot be applied to locked pages, Huge TLB pages, or
188 In the initial implementation, only
193 .\" commit 3f31d07571eeea18a7d34db9af21d2285b807a17
194 any filesystem which supports the
196 .BR FALLOC_FL_PUNCH_HOLE
199 Hugetlbfs fails with the error
201 and other filesystems fail with the error
204 .BR MADV_DONTFORK " (since Linux 2.6.16)"
205 .\" commit f822566165dd46ff5de9bf895cfa6c51f53bb0c4
206 .\" See http://lwn.net/Articles/171941/
207 Do not make the pages in this range available to the child after a
209 This is useful to prevent copy-on-write semantics from changing
210 the physical location of a page if the parent writes to it after a
212 (Such page relocations cause problems for hardware that
214 .\" [PATCH] madvise MADV_DONTFORK/MADV_DOFORK
215 .\" Currently, copy-on-write may change the physical address of
216 .\" a page even if the user requested that the page is pinned in
217 .\" memory (either by mlock or by get_user_pages). This happens
218 .\" if the process forks meanwhile, and the parent writes to that
219 .\" page. As a result, the page is orphaned: in case of
220 .\" get_user_pages, the application will never see any data hardware
221 .\" DMA's into this page after the COW. In case of mlock'd memory,
222 .\" the parent is not getting the realtime/security benefits of mlock.
224 .\" In particular, this affects the Infiniband modules which do DMA from
225 .\" and into user pages all the time.
227 .\" This patch adds madvise options to control whether memory range is
228 .\" inherited across fork. Useful e.g. for when hardware is doing DMA
229 .\" from/into these pages. Could also be useful to an application
230 .\" wanting to speed up its forks by cutting large areas out of
233 .\" SEE ALSO: http://lwn.net/Articles/171941/
234 .\" "Tweaks to madvise() and posix_fadvise()", 14 Feb 2006
236 .BR MADV_DOFORK " (since Linux 2.6.16)"
239 restoring the default behavior, whereby a mapping is inherited across
242 .BR MADV_HWPOISON " (since Linux 2.6.32)"
243 .\" commit 9893e49d64a4874ea67849ee2cfbf3f3d6817573
244 Poison the pages in the range specified by
248 and handle subsequent references to those pages
249 like a hardware memory corruption.
250 This operation is available only for privileged
251 .RB ( CAP_SYS_ADMIN )
253 This operation may result in the calling process receiving a
255 and the page being unmapped.
257 This feature is intended for testing of memory error-handling code;
258 it is available only if the kernel was configured with
259 .BR CONFIG_MEMORY_FAILURE .
261 .BR MADV_MERGEABLE " (since Linux 2.6.32)"
262 .\" commit f8af4da3b4c14e7267c4ffb952079af3912c51c5
263 Enable Kernel Samepage Merging (KSM) for the pages in the range specified by
267 The kernel regularly scans those areas of user memory that have
268 been marked as mergeable,
269 looking for pages with identical content.
270 These are replaced by a single write-protected page (which is automatically
271 copied if a process later wants to update the content of the page).
272 KSM merges only private anonymous pages (see
275 The KSM feature is intended for applications that generate many
276 instances of the same data (e.g., virtualization systems such as KVM).
277 It can consume a lot of processing power; use with care.
278 See the Linux kernel source file
279 .I Documentation/admin\-guide/mm/ksm.rst
286 operations are available only if the kernel was configured with
289 .BR MADV_UNMERGEABLE " (since Linux 2.6.32)"
290 Undo the effect of an earlier
292 operation on the specified address range;
293 KSM unmerges whatever pages it had merged in the address range specified by
298 .BR MADV_SOFT_OFFLINE " (since Linux 2.6.33)"
299 .\" commit afcf938ee0aac4ef95b1a23bac704c6fbeb26de6
300 Soft offline the pages in the range specified by
304 The memory of each page in the specified range is preserved
305 (i.e., when next accessed, the same content will be visible,
306 but in a new physical page frame),
307 and the original page is offlined
308 (i.e., no longer used, and taken out of normal memory management).
311 operation is invisible to (i.e., does not change the semantics of)
314 This feature is intended for testing of memory error-handling code;
315 it is available only if the kernel was configured with
316 .BR CONFIG_MEMORY_FAILURE .
318 .BR MADV_HUGEPAGE " (since Linux 2.6.38)"
319 .\" commit 0af4e98b6b095c74588af04872f83d333c958c32
320 .\" http://lwn.net/Articles/358904/
321 .\" https://lwn.net/Articles/423584/
322 Enable Transparent Huge Pages (THP) for pages in the range specified by
326 Currently, Transparent Huge Pages work only with private anonymous pages (see
328 The kernel will regularly scan the areas marked as huge page candidates
329 to replace them with huge pages.
330 The kernel will also allocate huge pages directly when the region is
331 naturally aligned to the huge page size (see
332 .BR posix_memalign (2)).
334 This feature is primarily aimed at applications that use large mappings of
335 data and access large regions of that memory at a time (e.g., virtualization
336 systems such as QEMU).
337 It can very easily waste memory (e.g., a 2\ MB mapping that only ever accesses
338 1 byte will result in 2\ MB of wired memory instead of one 4\ KB page).
339 See the Linux kernel source file
340 .I Documentation/admin\-guide/mm/transhuge.rst
343 Most common kernels configurations provide
344 .BR MADV_HUGEPAGE -style
345 behavior by default, and thus
347 is normally not necessary.
348 It is mostly intended for embedded systems, where
349 .BR MADV_HUGEPAGE -style
350 behavior may not be enabled by default in the kernel.
352 this flag can be used in order to selectively enable THP.
355 is used, it should always be in regions of memory with
356 an access pattern that the developer knows in advance won't risk
357 to increase the memory footprint of the application when transparent
358 hugepages are enabled.
364 operations are available only if the kernel was configured with
365 .BR CONFIG_TRANSPARENT_HUGEPAGE .
367 .BR MADV_NOHUGEPAGE " (since Linux 2.6.38)"
368 Ensures that memory in the address range specified by
372 will not be backed by transparent hugepages.
374 .BR MADV_DONTDUMP " (since Linux 3.4)"
375 .\" commit 909af768e88867016f427264ae39d27a57b6a8ed
376 .\" commit accb61fe7bb0f5c2a4102239e4981650f9048519
377 Exclude from a core dump those pages in the range specified by
381 This is useful in applications that have large areas of memory
382 that are known not to be useful in a core dump.
385 takes precedence over the bit mask that is set via the
386 .I /proc/[pid]/coredump_filter
390 .BR MADV_DODUMP " (since Linux 3.4)"
391 Undo the effect of an earlier
394 .BR MADV_FREE " (since Linux 4.5)"
395 The application no longer requires the pages in the range specified by
399 The kernel can thus free these pages,
400 but the freeing could be delayed until memory pressure occurs.
401 For each of the pages that has been marked to be freed
402 but has not yet been freed,
403 the free operation will be canceled if the caller writes into the page.
406 operation, any stale data (i.e., dirty, unwritten pages) will be lost
407 when the kernel frees the pages.
408 However, subsequent writes to pages in the range will succeed
409 and then kernel cannot free those dirtied pages,
410 so that the caller can always see just written data.
411 If there is no subsequent write,
412 the kernel can free the pages at any time.
413 Once pages in the range have been freed, the caller will
414 see zero-fill-on-demand pages upon subsequent page references.
419 can be applied only to private anonymous pages (see
421 In Linux before version 4.12,
422 .\" commit 93e06c7a645343d222c9a838834a51042eebbbf7
423 when freeing pages on a swapless system,
424 the pages in the given range are freed instantly,
425 regardless of memory pressure.
427 .BR MADV_WIPEONFORK " (since Linux 4.14)"
428 .\" commit d2cd9ede6e193dd7d88b6d27399e96229a551b19
429 Present the child process with zero-filled memory in this range after a
431 This is useful in forking servers in order to ensure
432 that sensitive per-process data
433 (for example, PRNG seeds, cryptographic secrets, and so on)
434 is not handed to child processes.
438 operation can be applied only to private anonymous pages (see
441 Within the child created by
445 setting remains in place on the specified address range.
446 This setting is cleared during
449 .BR MADV_KEEPONFORK " (since Linux 4.14)"
450 .\" commit d2cd9ede6e193dd7d88b6d27399e96229a551b19
451 Undo the effect of an earlier
452 .BR MADV_WIPEONFORK .
454 .BR MADV_COLD " (since Linux 5.4)"
455 .\" commit 9c276cc65a58faf98be8e56962745ec99ab87636
456 Deactivate a given range of pages.
457 This will make the pages a more probable
458 reclaim target should there be a memory pressure.
459 This is a nondestructive operation.
460 The advice might be ignored for some pages in the range when it is not
463 .BR MADV_PAGEOUT " (since Linux 5.4)"
464 .\" commit 1a4e58cce84ee88129d5d49c064bd2852b481357
465 Reclaim a given range of pages.
466 This is done to free up memory occupied by these pages.
467 If a page is anonymous, it will be swapped out.
468 If a page is file-backed and dirty, it will be written back to the backing
470 The advice might be ignored for some pages in the range when it is not
476 On error, it returns \-1 and
478 is set to indicate the error.
485 but the specified address range is not a shared writable mapping.
488 A kernel resource was temporarily unavailable.
491 The map exists, but the area maps something that isn't a file.
495 is not page-aligned or
511 and the specified address range includes locked, Huge TLB pages, or
520 .BR MADV_UNMERGEABLE ,
521 but the kernel was not configured with
530 but the specified address range includes file, Huge TLB,
539 Paging in this area would exceed the process's
540 maximum resident set size.
545 Not enough memory: paging in failed.
548 Addresses in the specified range are not currently
549 mapped, or are outside the address space of the process.
555 but the caller does not have the
560 .\" commit d3ac21cacc24790eb45d735769f35753f5b56ceb
561 support for this system call is optional,
562 depending on the setting of the
563 .B CONFIG_ADVISE_SYSCALLS
564 configuration option.
567 is not specified by any standards.
568 Versions of this system call, implementing a wide variety of
570 values, exist on many other implementations.
571 Other implementations typically implement at least the flags listed
573 .IR "Conventional advice flags" ,
574 albeit with some variation in semantics.
576 POSIX.1-2001 describes
577 .BR posix_madvise (3)
579 .BR POSIX_MADV_NORMAL ,
580 .BR POSIX_MADV_RANDOM ,
581 .BR POSIX_MADV_SEQUENTIAL ,
582 .BR POSIX_MADV_WILLNEED ,
584 .BR POSIX_MADV_DONTNEED ,
585 and so on, with behavior close to the similarly named flags listed above.
588 The Linux implementation requires that the address
590 be page-aligned, and allows
593 If there are some parts of the specified address range
594 that are not mapped, the Linux version of
596 ignores them and applies the call to the rest (but returns
598 from the system call, as it should).
602 .\" function first appeared in 4.4BSD.
611 .BR process_madvise (2),
612 .BR posix_madvise (3),