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(): _BSD_SOURCE
16
18 The madvise() system call advises the kernel about how to handle paging
19 input/output in the address range beginning at address addr and with
20 size length bytes. It allows an application to tell the kernel how it
21 expects to use some mapped or shared memory areas, so that the kernel
22 can choose appropriate read-ahead and caching techniques. This call
23 does not influence the semantics of the application (except in the case
24 of MADV_DONTNEED), but may influence its performance. The kernel is
25 free to ignore the advice.
26 The advice is indicated in the advice argument which can be
28 MADV_NORMAL
30 No special treatment. This is the default.
31 MADV_RANDOM
33 Expect page references in random order. (Hence, read ahead may
34 be less useful than normally.)
35 MADV_SEQUENTIAL
37 Expect page references in sequential order. (Hence, pages in
38 the given range can be aggressively read ahead, and may be freed
39 soon after they are accessed.)
40 MADV_WILLNEED
42 Expect access in the near future. (Hence, it might be a good
43 idea to read some pages ahead.)
44 MADV_DONTNEED
46 Do not expect access in the near future. (For the time being,
47 the application is finished with the given range, so the kernel
48 can free resources associated with it.) Subsequent accesses of
49 pages in this range will succeed, but will result either in
50 reloading of the memory contents from the underlying mapped file
51 (see mmap(2)) or zero-fill-on-demand pages for mappings without
52 an underlying file.
53 MADV_REMOVE (Since Linux 2.6.16)
55 Free up a given range of pages and its associated backing store.
56 Currently, only shmfs/tmpfs supports this; other file systems
57 return with the error ENOSYS.
58 MADV_DONTFORK (Since Linux 2.6.16)
60 Do not make the pages in this range available to the child after
61 a fork(2). This is useful to prevent copy-on-write semantics
62 from changing the physical location of a page(s) if the parent
63 writes to it after a fork(2). (Such page relocations cause
64 problems for hardware that DMAs into the page(s).)
65 MADV_DOFORK (Since Linux 2.6.16)
67 Undo the effect of MADV_DONTFORK, restoring the default behav‐
68 ior, whereby a mapping is inherited across fork(2).
69 MADV_HWPOISON (Since Linux 2.6.32)
71 Poison a page and handle it like a hardware memory corruption.
72 This operation is available only for privileged (CAP_SYS_ADMIN)
73 processes. This operation may result in the calling process
74 receiving a SIGBUS and the page being unmapped. This feature is
75 intended for testing of memory error-handling code; it is avail‐
76 able only if the kernel was configured with CONFIG_MEMORY_FAIL‐
77 URE.
78 MADV_SOFT_OFFLINE (Since Linux 2.6.33)
80 Soft offline the pages in the range specified by addr and
81 length. The memory of each page in the specified range is pre‐
82 served (i.e., when next accessed, the same content will be visi‐
83 ble, but in a new physical page frame), and the original page is
84 offlined (i.e., no longer used, and taken out of normal memory
85 management). The effect of the MADV_SOFT_OFFLINE operation is
86 invisible to (i.e., does not change the semantics of) the call‐
87 ing process. This feature is intended for testing of memory
88 error-handling code; it is available only if the kernel was con‐
89 figured with CONFIG_MEMORY_FAILURE.
90 MADV_MERGEABLE (since Linux 2.6.32)
92 Enable Kernel Samepage Merging (KSM) for the pages in the range
93 specified by addr and length. The kernel regularly scans those
94 areas of user memory that have been marked as mergeable, looking
95 for pages with identical content. These are replaced by a sin‐
96 gle write-protected page (which is automatically copied if a
97 process later wants to update the content of the page). KSM
98 merges only private anonymous pages (see mmap(2)). The KSM fea‐
99 ture is intended for applications that generate many instances
100 of the same data (e.g., virtualization systems such as KVM). It
101 can consume a lot of processing power; use with care. See the
102 Linux kernel source file Documentation/vm/ksm.txt for more
103 details. The MADV_MERGEABLE and MADV_UNMERGEABLE operations are
104 available only if the kernel was configured with CONFIG_KSM.
105 MADV_UNMERGEABLE (since Linux 2.6.32)
107 Undo the effect of an earlier MADV_MERGEABLE operation on the
108 specified address range; KSM unmerges whatever pages it had
109 merged in the address range specified by addr and length.
110 MADV_HUGEPAGE (since Linux 2.6.38)
112 Enables Transparent Huge Pages (THP) for pages in the range
113 specified by addr and length. Currently, Transparent Huge Pages
114 work only with private anonymous pages (see mmap(2)). The ker‐
115 nel will regularly scan the areas marked as huge page candidates
116 to replace them with huge pages. The kernel will also allocate
117 huge pages directly when the region is naturally aligned to the
118 huge page size (see posix_memalign(2)). This feature is primar‐
119 ily aimed at applications that use large mappings of data and
120 access large regions of that memory at a time (e.g., virtualiza‐
121 tion systems such as QEMU). It can very easily waste memory
122 (e.g., a 2MB mapping that only ever accesses 1 byte will result
123 in 2MB of wired memory instead of one 4KB page). See the Linux
124 kernel source file Documentation/vm/transhuge.txt for more
125 details. The MADV_HUGEPAGE and MADV_NOHUGEPAGE operations are
126 available only if the kernel was configured with CONFIG_TRANS‐
127 PARENT_HUGEPAGE.
128 MADV_NOHUGEPAGE (since Linux 2.6.38)
130 Ensures that memory in the address range specified by addr and
131 length will not be collapsed into huge pages.
132 MADV_DONTDUMP (since Linux 3.4)
134 Exclude from a core dump those pages in the range specified by
135 addr and length. This is useful in applications that have large
136 areas of memory that are known not to be useful in a core dump.
137 The effect of MADV_DONTDUMP takes precedence over the bit mask
138 that is set via the /proc/PID/coredump_filter file (see
139 core(5)).
140 MADV_DODUMP (since Linux 3.4)
142 Undo the effect of an earlier MADV_DONTDUMP.
143
145 On success madvise() returns zero. On error, it returns -1 and errno
146 is set appropriately.
147
149 EAGAIN A kernel resource was temporarily unavailable.
150 EBADF The map exists, but the area maps something that isn't a file.
152 EINVAL This error can occur for the following reasons:
154 * The value len is negative.
156 * addr is not page-aligned.
158 * advice is not a valid value
160 * The application is attempting to release locked or shared
162 pages (with MADV_DONTNEED).
163 * MADV_MERGEABLE or MADV_UNMERGEABLE was specified in advice,
165 but the kernel was not configured with CONFIG_KSM.
166 EIO (for MADV_WILLNEED) Paging in this area would exceed the
168 process's maximum resident set size.
169 ENOMEM (for MADV_WILLNEED) Not enough memory: paging in failed.
171 ENOMEM Addresses in the specified range are not currently mapped, or
173 are outside the address space of the process.
174
176 POSIX.1b. POSIX.1-2001 describes posix_madvise(3) with constants
177 POSIX_MADV_NORMAL, etc., with a behavior close to that described here.
178 There is a similar posix_fadvise(2) for file access.
179 MADV_REMOVE, MADV_DONTFORK, MADV_DOFORK, MADV_HWPOISON, MADV_MERGEABLE,
181 and MADV_UNMERGEABLE are Linux-specific.
182
184 Linux notes
185 The current Linux implementation (2.4.0) views this system call more as
186 a command than as advice and hence may return an error when it cannot
187 do what it usually would do in response to this advice. (See the
188 ERRORS description above.) This is nonstandard behavior.
189 The Linux implementation requires that the address addr be page-
191 aligned, and allows length to be zero. If there are some parts of the
192 specified address range that are not mapped, the Linux version of mad‐
193 vise() ignores them and applies the call to the rest (but returns
194 ENOMEM from the system call, as it should).
195
197 getrlimit(2), mincore(2), mmap(2), mprotect(2), msync(2), munmap(2),
198 core(5)
199
201 This page is part of release 3.53 of the Linux man-pages project. A
202 description of the project, and information about reporting bugs, can
203 be found at http://www.kernel.org/doc/man-pages/.
204Linux 2012-04-28 MADVISE(2)