Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
1.. _overcommit_accounting:
2
3=====================
4Overcommit Accounting
5=====================
6
7The Linux kernel supports the following overcommit handling modes
8
90
10 Heuristic overcommit handling. Obvious overcommits of address
11 space are refused. Used for a typical system. It ensures a
12 seriously wild allocation fails while allowing overcommit to
13 reduce swap usage. root is allowed to allocate slightly more
14 memory in this mode. This is the default.
15
161
17 Always overcommit. Appropriate for some scientific
18 applications. Classic example is code using sparse arrays and
19 just relying on the virtual memory consisting almost entirely
20 of zero pages.
21
222
23 Don't overcommit. The total address space commit for the
24 system is not permitted to exceed swap + a configurable amount
25 (default is 50%) of physical RAM. Depending on the amount you
26 use, in most situations this means a process will not be
27 killed while accessing pages but will receive errors on memory
28 allocation as appropriate.
29
30 Useful for applications that want to guarantee their memory
31 allocations will be available in the future without having to
32 initialize every page.
33
34The overcommit policy is set via the sysctl ``vm.overcommit_memory``.
35
36The overcommit amount can be set via ``vm.overcommit_ratio`` (percentage)
37or ``vm.overcommit_kbytes`` (absolute value). These only have an effect
38when ``vm.overcommit_memory`` is set to 2.
39
40The current overcommit limit and amount committed are viewable in
41``/proc/meminfo`` as CommitLimit and Committed_AS respectively.
42
43Gotchas
44=======
45
46The C language stack growth does an implicit mremap. If you want absolute
47guarantees and run close to the edge you MUST mmap your stack for the
48largest size you think you will need. For typical stack usage this does
49not matter much but it's a corner case if you really really care
50
51In mode 2 the MAP_NORESERVE flag is ignored.
52
53
54How It Works
55============
56
57The overcommit is based on the following rules
58
59For a file backed map
60 | SHARED or READ-only - 0 cost (the file is the map not swap)
61 | PRIVATE WRITABLE - size of mapping per instance
62
63For an anonymous or ``/dev/zero`` map
64 | SHARED - size of mapping
65 | PRIVATE READ-only - 0 cost (but of little use)
66 | PRIVATE WRITABLE - size of mapping per instance
67
68Additional accounting
69 | Pages made writable copies by mmap
70 | shmfs memory drawn from the same pool
71
72Status
73======
74
75* We account mmap memory mappings
76* We account mprotect changes in commit
77* We account mremap changes in size
78* We account brk
79* We account munmap
80* We report the commit status in /proc
81* Account and check on fork
82* Review stack handling/building on exec
83* SHMfs accounting
84* Implement actual limit enforcement
85
86To Do
87=====
88* Account ptrace pages (this is hard)