blob: 8eaa2fc4b8fae253930a798f38394438198dbf5a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001The Linux kernel supports the following overcommit handling modes
2
30 - Heuristic overcommit handling. Obvious overcommits of
4 address space are refused. Used for a typical system. It
5 ensures a seriously wild allocation fails while allowing
6 overcommit to reduce swap usage. root is allowed to
Lucas De Marchi25985ed2011-03-30 22:57:33 -03007 allocate slightly more memory in this mode. This is the
Linus Torvalds1da177e2005-04-16 15:20:36 -07008 default.
9
101 - Always overcommit. Appropriate for some scientific
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -070011 applications. Classic example is code using sparse arrays
12 and just relying on the virtual memory consisting almost
13 entirely of zero pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014
152 - Don't overcommit. The total address space commit
16 for the system is not permitted to exceed swap + a
17 configurable percentage (default is 50) of physical RAM.
18 Depending on the percentage you use, in most situations
19 this means a process will not be killed while accessing
20 pages but will receive errors on memory allocation as
21 appropriate.
22
Andrew Shewmakerc9b1d092013-04-29 15:08:10 -070023 Useful for applications that want to guarantee their
24 memory allocations will be available in the future
25 without having to initialize every page.
26
Linus Torvalds1da177e2005-04-16 15:20:36 -070027The overcommit policy is set via the sysctl `vm.overcommit_memory'.
28
29The overcommit percentage is set via `vm.overcommit_ratio'.
30
31The current overcommit limit and amount committed are viewable in
32/proc/meminfo as CommitLimit and Committed_AS respectively.
33
34Gotchas
35-------
36
37The C language stack growth does an implicit mremap. If you want absolute
38guarantees and run close to the edge you MUST mmap your stack for the
39largest size you think you will need. For typical stack usage this does
40not matter much but it's a corner case if you really really care
41
42In mode 2 the MAP_NORESERVE flag is ignored.
43
44
45How It Works
46------------
47
48The overcommit is based on the following rules
49
50For a file backed map
51 SHARED or READ-only - 0 cost (the file is the map not swap)
52 PRIVATE WRITABLE - size of mapping per instance
53
54For an anonymous or /dev/zero map
55 SHARED - size of mapping
56 PRIVATE READ-only - 0 cost (but of little use)
57 PRIVATE WRITABLE - size of mapping per instance
58
59Additional accounting
60 Pages made writable copies by mmap
61 shmfs memory drawn from the same pool
62
63Status
64------
65
66o We account mmap memory mappings
67o We account mprotect changes in commit
68o We account mremap changes in size
69o We account brk
70o We account munmap
71o We report the commit status in /proc
72o Account and check on fork
73o Review stack handling/building on exec
74o SHMfs accounting
75o Implement actual limit enforcement
76
77To Do
78-----
79o Account ptrace pages (this is hard)