1MADVISE(2) Linux Programmer's Manual MADVISE(2)
2
6 madvise - give advice about use of memory
7
9 #include <sys/mman.h>
10 int madvise(void *addr, size_t length, int advice);
12 Feature Test Macro Requirements for glibc (see feature_test_macros(7)):
14 madvise():
16 Since glibc 2.19:
17 _DEFAULT_SOURCE
18 Up to and including glibc 2.19:
19 _BSD_SOURCE
20
22 The madvise() system call is used to give advice or directions to the
23 kernel about the address range beginning at address addr and with size
24 length bytes In most cases, the goal of such advice is to improve sys‐
25 tem or application performance.
26 Initially, the system call supported a set of "conventional" advice
28 values, which are also available on several other implementations.
29 (Note, though, that madvise() is not specified in POSIX.) Subse‐
30 quently, a number of Linux-specific advice values have been added.
31 Conventional advice values
33 The advice values listed below allow an application to tell the kernel
34 how it expects to use some mapped or shared memory areas, so that the
35 kernel can choose appropriate read-ahead and caching techniques. These
36 advice values do not influence the semantics of the application (except
37 in the case of MADV_DONTNEED), but may influence its performance. All
38 of the advice values listed here have analogs in the POSIX-specified
39 posix_madvise(3) function, and the values have the same meanings, with
40 the exception of MADV_DONTNEED.
41 The advice is indicated in the advice argument, which is one of the
43 following:
44 MADV_NORMAL
46 No special treatment. This is the default.
47 MADV_RANDOM
49 Expect page references in random order. (Hence, read ahead may
50 be less useful than normally.)
51 MADV_SEQUENTIAL
53 Expect page references in sequential order. (Hence, pages in
54 the given range can be aggressively read ahead, and may be freed
55 soon after they are accessed.)
56 MADV_WILLNEED
58 Expect access in the near future. (Hence, it might be a good
59 idea to read some pages ahead.)
60 MADV_DONTNEED
62 Do not expect access in the near future. (For the time being,
63 the application is finished with the given range, so the kernel
64 can free resources associated with it.)
65 After a successful MADV_DONTNEED operation, the semantics of
67 memory access in the specified region are changed: subsequent
68 accesses of pages in the range will succeed, but will result in
69 either repopulating the memory contents from the up-to-date con‐
70 tents of the underlying mapped file (for shared file mappings,
71 shared anonymous mappings, and shmem-based techniques such as
72 System V shared memory segments) or zero-fill-on-demand pages
73 for anonymous private mappings.
74 Note that, when applied to shared mappings, MADV_DONTNEED might
76 not lead to immediate freeing of the pages in the range. The
77 kernel is free to delay freeing the pages until an appropriate
78 moment. The resident set size (RSS) of the calling process will
79 be immediately reduced however.
80 MADV_DONTNEED cannot be applied to locked pages, Huge TLB pages,
82 or VM_PFNMAP pages. (Pages marked with the kernel-internal
83 VM_PFNMAP flag are special memory areas that are not managed by
84 the virtual memory subsystem. Such pages are typically created
85 by device drivers that map the pages into user space.)
86 Linux-specific advice values
88 The following Linux-specific advice values have no counterparts in the
89 POSIX-specified posix_madvise(3), and may or may not have counterparts
90 in the madvise() interface available on other implementations. Note
91 that some of these operations change the semantics of memory accesses.
92 MADV_REMOVE (since Linux 2.6.16)
94 Free up a given range of pages and its associated backing store.
95 This is equivalent to punching a hole in the corresponding byte
96 range of the backing store (see fallocate(2)). Subsequent ac‐
97 cesses in the specified address range will see bytes containing
98 zero.
99 The specified address range must be mapped shared and writable.
101 This flag cannot be applied to locked pages, Huge TLB pages, or
102 VM_PFNMAP pages.
103 In the initial implementation, only tmpfs(5) was supported
105 MADV_REMOVE; but since Linux 3.5, any filesystem which supports
106 the fallocate(2) FALLOC_FL_PUNCH_HOLE mode also supports
107 MADV_REMOVE. Hugetlbfs fails with the error EINVAL and other
108 filesystems fail with the error EOPNOTSUPP.
109 MADV_DONTFORK (since Linux 2.6.16)
111 Do not make the pages in this range available to the child after
112 a fork(2). This is useful to prevent copy-on-write semantics
113 from changing the physical location of a page if the parent
114 writes to it after a fork(2). (Such page relocations cause
115 problems for hardware that DMAs into the page.)
116 MADV_DOFORK (since Linux 2.6.16)
118 Undo the effect of MADV_DONTFORK, restoring the default behav‐
119 ior, whereby a mapping is inherited across fork(2).
120 MADV_HWPOISON (since Linux 2.6.32)
122 Poison the pages in the range specified by addr and length and
123 handle subsequent references to those pages like a hardware mem‐
124 ory corruption. This operation is available only for privileged
125 (CAP_SYS_ADMIN) processes. This operation may result in the
126 calling process receiving a SIGBUS and the page being unmapped.
127 This feature is intended for testing of memory error-handling
129 code; it is available only if the kernel was configured with
130 CONFIG_MEMORY_FAILURE.
131 MADV_MERGEABLE (since Linux 2.6.32)
133 Enable Kernel Samepage Merging (KSM) for the pages in the range
134 specified by addr and length. The kernel regularly scans those
135 areas of user memory that have been marked as mergeable, looking
136 for pages with identical content. These are replaced by a sin‐
137 gle write-protected page (which is automatically copied if a
138 process later wants to update the content of the page). KSM
139 merges only private anonymous pages (see mmap(2)).
140 The KSM feature is intended for applications that generate many
142 instances of the same data (e.g., virtualization systems such as
143 KVM). It can consume a lot of processing power; use with care.
144 See the Linux kernel source file Documentation/ad‐
145 min-guide/mm/ksm.rst for more details.
146 The MADV_MERGEABLE and MADV_UNMERGEABLE operations are available
148 only if the kernel was configured with CONFIG_KSM.
149 MADV_UNMERGEABLE (since Linux 2.6.32)
151 Undo the effect of an earlier MADV_MERGEABLE operation on the
152 specified address range; KSM unmerges whatever pages it had
153 merged in the address range specified by addr and length.
154 MADV_SOFT_OFFLINE (since Linux 2.6.33)
156 Soft offline the pages in the range specified by addr and
157 length. The memory of each page in the specified range is pre‐
158 served (i.e., when next accessed, the same content will be visi‐
159 ble, but in a new physical page frame), and the original page is
160 offlined (i.e., no longer used, and taken out of normal memory
161 management). The effect of the MADV_SOFT_OFFLINE operation is
162 invisible to (i.e., does not change the semantics of) the call‐
163 ing process.
164 This feature is intended for testing of memory error-handling
166 code; it is available only if the kernel was configured with
167 CONFIG_MEMORY_FAILURE.
168 MADV_HUGEPAGE (since Linux 2.6.38)
170 Enable Transparent Huge Pages (THP) for pages in the range spec‐
171 ified by addr and length. Currently, Transparent Huge Pages
172 work only with private anonymous pages (see mmap(2)). The ker‐
173 nel will regularly scan the areas marked as huge page candidates
174 to replace them with huge pages. The kernel will also allocate
175 huge pages directly when the region is naturally aligned to the
176 huge page size (see posix_memalign(2)).
177 This feature is primarily aimed at applications that use large
179 mappings of data and access large regions of that memory at a
180 time (e.g., virtualization systems such as QEMU). It can very
181 easily waste memory (e.g., a 2 MB mapping that only ever ac‐
182 cesses 1 byte will result in 2 MB of wired memory instead of one
183 4 KB page). See the Linux kernel source file Documentation/ad‐
184 min-guide/mm/transhuge.rst for more details.
185 Most common kernels configurations provide MADV_HUGEPAGE-style
187 behavior by default, and thus MADV_HUGEPAGE is normally not nec‐
188 essary. It is mostly intended for embedded systems, where
189 MADV_HUGEPAGE-style behavior may not be enabled by default in
190 the kernel. On such systems, this flag can be used in order to
191 selectively enable THP. Whenever MADV_HUGEPAGE is used, it
192 should always be in regions of memory with an access pattern
193 that the developer knows in advance won't risk to increase the
194 memory footprint of the application when transparent hugepages
195 are enabled.
196 The MADV_HUGEPAGE and MADV_NOHUGEPAGE operations are available
198 only if the kernel was configured with CONFIG_TRANSPAR‐
199 ENT_HUGEPAGE.
200 MADV_NOHUGEPAGE (since Linux 2.6.38)
202 Ensures that memory in the address range specified by addr and
203 length will not be backed by transparent hugepages.
204 MADV_DONTDUMP (since Linux 3.4)
206 Exclude from a core dump those pages in the range specified by
207 addr and length. This is useful in applications that have large
208 areas of memory that are known not to be useful in a core dump.
209 The effect of MADV_DONTDUMP takes precedence over the bit mask
210 that is set via the /proc/[pid]/coredump_filter file (see
211 core(5)).
212 MADV_DODUMP (since Linux 3.4)
214 Undo the effect of an earlier MADV_DONTDUMP.
215 MADV_FREE (since Linux 4.5)
217 The application no longer requires the pages in the range speci‐
218 fied by addr and len. The kernel can thus free these pages, but
219 the freeing could be delayed until memory pressure occurs. For
220 each of the pages that has been marked to be freed but has not
221 yet been freed, the free operation will be canceled if the
222 caller writes into the page. After a successful MADV_FREE oper‐
223 ation, any stale data (i.e., dirty, unwritten pages) will be
224 lost when the kernel frees the pages. However, subsequent
225 writes to pages in the range will succeed and then kernel cannot
226 free those dirtied pages, so that the caller can always see just
227 written data. If there is no subsequent write, the kernel can
228 free the pages at any time. Once pages in the range have been
229 freed, the caller will see zero-fill-on-demand pages upon subse‐
230 quent page references.
231 The MADV_FREE operation can be applied only to private anonymous
233 pages (see mmap(2)). In Linux before version 4.12, when freeing
234 pages on a swapless system, the pages in the given range are
235 freed instantly, regardless of memory pressure.
236 MADV_WIPEONFORK (since Linux 4.14)
238 Present the child process with zero-filled memory in this range
239 after a fork(2). This is useful in forking servers in order to
240 ensure that sensitive per-process data (for example, PRNG seeds,
241 cryptographic secrets, and so on) is not handed to child pro‐
242 cesses.
243 The MADV_WIPEONFORK operation can be applied only to private
245 anonymous pages (see mmap(2)).
246 Within the child created by fork(2), the MADV_WIPEONFORK setting
248 remains in place on the specified address range. This setting
249 is cleared during execve(2).
250 MADV_KEEPONFORK (since Linux 4.14)
252 Undo the effect of an earlier MADV_WIPEONFORK.
253 MADV_COLD (since Linux 5.4)
255 Deactivate a given range of pages. This will make the pages a
256 more probable reclaim target should there be a memory pressure.
257 This is a nondestructive operation. The advice might be ignored
258 for some pages in the range when it is not applicable.
259 MADV_PAGEOUT (since Linux 5.4)
261 Reclaim a given range of pages. This is done to free up memory
262 occupied by these pages. If a page is anonymous, it will be
263 swapped out. If a page is file-backed and dirty, it will be
264 written back to the backing storage. The advice might be ig‐
265 nored for some pages in the range when it is not applicable.
266
268 On success, madvise() returns zero. On error, it returns -1 and errno
269 is set to indicate the error.
270
272 EACCES advice is MADV_REMOVE, but the specified address range is not a
273 shared writable mapping.
274 EAGAIN A kernel resource was temporarily unavailable.
276 EBADF The map exists, but the area maps something that isn't a file.
278 EINVAL addr is not page-aligned or length is negative.
280 EINVAL advice is not a valid.
282 EINVAL advice is MADV_DONTNEED or MADV_REMOVE and the specified address
284 range includes locked, Huge TLB pages, or VM_PFNMAP pages.
285 EINVAL advice is MADV_MERGEABLE or MADV_UNMERGEABLE, but the kernel was
287 not configured with CONFIG_KSM.
288 EINVAL advice is MADV_FREE or MADV_WIPEONFORK but the specified address
290 range includes file, Huge TLB, MAP_SHARED, or VM_PFNMAP ranges.
291 EIO (for MADV_WILLNEED) Paging in this area would exceed the
293 process's maximum resident set size.
294 ENOMEM (for MADV_WILLNEED) Not enough memory: paging in failed.
296 ENOMEM Addresses in the specified range are not currently mapped, or
298 are outside the address space of the process.
299 EPERM advice is MADV_HWPOISON, but the caller does not have the
301 CAP_SYS_ADMIN capability.
302
304 Since Linux 3.18, support for this system call is optional, depending
305 on the setting of the CONFIG_ADVISE_SYSCALLS configuration option.
306
308 madvise() is not specified by any standards. Versions of this system
309 call, implementing a wide variety of advice values, exist on many other
310 implementations. Other implementations typically implement at least
311 the flags listed above under Conventional advice flags, albeit with
312 some variation in semantics.
313 POSIX.1-2001 describes posix_madvise(3) with constants POSIX_MADV_NOR‐
315 MAL, POSIX_MADV_RANDOM, POSIX_MADV_SEQUENTIAL, POSIX_MADV_WILLNEED, and
316 POSIX_MADV_DONTNEED, and so on, with behavior close to the similarly
317 named flags listed above.
318
320 Linux notes
321 The Linux implementation requires that the address addr be page-
322 aligned, and allows length to be zero. If there are some parts of the
323 specified address range that are not mapped, the Linux version of mad‐
324 vise() ignores them and applies the call to the rest (but returns
325 ENOMEM from the system call, as it should).
326
328 getrlimit(2), mincore(2), mmap(2), mprotect(2), msync(2), munmap(2),
329 prctl(2), process_madvise(2), posix_madvise(3), core(5)
330
332 This page is part of release 5.13 of the Linux man-pages project. A
333 description of the project, information about reporting bugs, and the
334 latest version of this page, can be found at
335 https://www.kernel.org/doc/man-pages/.
336Linux 2021-03-22 MADVISE(2)