1 Documentation for /proc/sys/vm/* kernel version 2.6.29
2 (c) 1998, 1999, Rik van Riel <riel@nl.linux.org>
3 (c) 2008 Peter W. Morreale <pmorreale@novell.com>
5 For general info and legal blurb, please look in README.
7 ==============================================================
9 This file contains the documentation for the sysctl files in
10 /proc/sys/vm and is valid for Linux kernel version 2.6.29.
12 The files in this directory can be used to tune the operation
13 of the virtual memory (VM) subsystem of the Linux kernel and
14 the writeout of dirty data to disk.
16 Default values and initialization routines for most of these
17 files can be found in mm/swap.c.
19 Currently, these files are in /proc/sys/vm:
21 - admin_reserve_kbytes
24 - dirty_background_bytes
25 - dirty_background_ratio
27 - dirty_expire_centisecs
29 - dirty_writeback_centisecs
33 - hugepages_treat_as_movable
37 - lowmem_reserve_ratio
39 - memory_failure_early_kill
40 - memory_failure_recovery
46 - nr_overcommit_hugepages
47 - nr_trim_pages (only if CONFIG_MMU=n)
50 - oom_kill_allocating_task
55 - percpu_pagelist_fraction
62 ==============================================================
66 The amount of free memory in the system that should be reserved for users
67 with the capability cap_sys_admin.
69 admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
71 That should provide enough for the admin to log in and kill a process,
72 if necessary, under the default overcommit 'guess' mode.
74 Systems running under overcommit 'never' should increase this to account
75 for the full Virtual Memory Size of programs used to recover. Otherwise,
76 root may not be able to log in to recover the system.
78 How do you calculate a minimum useful reserve?
80 sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
82 For overcommit 'guess', we can sum resident set sizes (RSS).
83 On x86_64 this is about 8MB.
85 For overcommit 'never', we can take the max of their virtual sizes (VSZ)
86 and add the sum of their RSS.
87 On x86_64 this is about 128MB.
89 Changing this takes effect whenever an application requests memory.
91 ==============================================================
95 block_dump enables block I/O debugging when set to a nonzero value. More
96 information on block I/O debugging is in Documentation/laptops/laptop-mode.txt.
98 ==============================================================
102 Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
103 all zones are compacted such that free memory is available in contiguous
104 blocks where possible. This can be important for example in the allocation of
105 huge pages although processes will also directly compact memory as required.
107 ==============================================================
109 dirty_background_bytes
111 Contains the amount of dirty memory at which the background kernel
112 flusher threads will start writeback.
114 Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only
115 one of them may be specified at a time. When one sysctl is written it is
116 immediately taken into account to evaluate the dirty memory limits and the
117 other appears as 0 when read.
119 ==============================================================
121 dirty_background_ratio
123 Contains, as a percentage of total system memory, the number of pages at which
124 the background kernel flusher threads will start writing out dirty data.
126 ==============================================================
130 Contains the amount of dirty memory at which a process generating disk writes
131 will itself start writeback.
133 Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
134 specified at a time. When one sysctl is written it is immediately taken into
135 account to evaluate the dirty memory limits and the other appears as 0 when
138 Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
139 value lower than this limit will be ignored and the old configuration will be
142 ==============================================================
144 dirty_expire_centisecs
146 This tunable is used to define when dirty data is old enough to be eligible
147 for writeout by the kernel flusher threads. It is expressed in 100'ths
148 of a second. Data which has been dirty in-memory for longer than this
149 interval will be written out next time a flusher thread wakes up.
151 ==============================================================
155 Contains, as a percentage of total system memory, the number of pages at which
156 a process which is generating disk writes will itself start writing out dirty
159 ==============================================================
161 dirty_writeback_centisecs
163 The kernel flusher threads will periodically wake up and write `old' data
164 out to disk. This tunable expresses the interval between those wakeups, in
167 Setting this to zero disables periodic writeback altogether.
169 ==============================================================
173 Writing to this will cause the kernel to drop clean caches, dentries and
174 inodes from memory, causing that memory to become free.
177 echo 1 > /proc/sys/vm/drop_caches
178 To free dentries and inodes:
179 echo 2 > /proc/sys/vm/drop_caches
180 To free pagecache, dentries and inodes:
181 echo 3 > /proc/sys/vm/drop_caches
183 As this is a non-destructive operation and dirty objects are not freeable, the
184 user should run `sync' first.
186 ==============================================================
190 This parameter affects whether the kernel will compact memory or direct
191 reclaim to satisfy a high-order allocation. /proc/extfrag_index shows what
192 the fragmentation index for each order is in each zone in the system. Values
193 tending towards 0 imply allocations would fail due to lack of memory,
194 values towards 1000 imply failures are due to fragmentation and -1 implies
195 that the allocation will succeed as long as watermarks are met.
197 The kernel will not compact memory in a zone if the
198 fragmentation index is <= extfrag_threshold. The default value is 500.
200 ==============================================================
204 This parameter tells the VM to keep extra free memory between the threshold
205 where background reclaim (kswapd) kicks in, and the threshold where direct
206 reclaim (by allocating processes) kicks in.
208 This is useful for workloads that require low latency memory allocations
209 and have a bounded burstiness in memory allocations, for example a
210 realtime application that receives and transmits network traffic
211 (causing in-kernel memory allocations) with a maximum total message burst
212 size of 200MB may need 200MB of extra free memory to avoid direct reclaim
215 ==============================================================
217 hugepages_treat_as_movable
219 This parameter is only useful when kernelcore= is specified at boot time to
220 create ZONE_MOVABLE for pages that may be reclaimed or migrated. Huge pages
221 are not movable so are not normally allocated from ZONE_MOVABLE. A non-zero
222 value written to hugepages_treat_as_movable allows huge pages to be allocated
225 Once enabled, the ZONE_MOVABLE is treated as an area of memory the huge
226 pages pool can easily grow or shrink within. Assuming that applications are
227 not running that mlock() a lot of memory, it is likely the huge pages pool
228 can grow to the size of ZONE_MOVABLE by repeatedly entering the desired value
229 into nr_hugepages and triggering page reclaim.
231 ==============================================================
235 hugetlb_shm_group contains group id that is allowed to create SysV
236 shared memory segment using hugetlb page.
238 ==============================================================
242 laptop_mode is a knob that controls "laptop mode". All the things that are
243 controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt.
245 ==============================================================
249 If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
250 will use the legacy (2.4) layout for all processes.
252 ==============================================================
256 For some specialised workloads on highmem machines it is dangerous for
257 the kernel to allow process memory to be allocated from the "lowmem"
258 zone. This is because that memory could then be pinned via the mlock()
259 system call, or by unavailability of swapspace.
261 And on large highmem machines this lack of reclaimable lowmem memory
264 So the Linux page allocator has a mechanism which prevents allocations
265 which _could_ use highmem from using too much lowmem. This means that
266 a certain amount of lowmem is defended from the possibility of being
267 captured into pinned user memory.
269 (The same argument applies to the old 16 megabyte ISA DMA region. This
270 mechanism will also defend that region from allocations which could use
273 The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is
274 in defending these lower zones.
276 If you have a machine which uses highmem or ISA DMA and your
277 applications are using mlock(), or if you are running with no swap then
278 you probably should change the lowmem_reserve_ratio setting.
280 The lowmem_reserve_ratio is an array. You can see them by reading this file.
282 % cat /proc/sys/vm/lowmem_reserve_ratio
285 Note: # of this elements is one fewer than number of zones. Because the highest
286 zone's value is not necessary for following calculation.
288 But, these values are not used directly. The kernel calculates # of protection
289 pages for each zones from them. These are shown as array of protection pages
290 in /proc/zoneinfo like followings. (This is an example of x86-64 box).
291 Each zone has an array of protection pages like this.
302 protection: (0, 2004, 2004, 2004)
303 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
308 These protections are added to score to judge whether this zone should be used
309 for page allocation or should be reclaimed.
311 In this example, if normal pages (index=2) are required to this DMA zone and
312 watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
313 not be used because pages_free(1355) is smaller than watermark + protection[2]
314 (4 + 2004 = 2008). If this protection value is 0, this zone would be used for
315 normal page requirement. If requirement is DMA zone(index=0), protection[0]
318 zone[i]'s protection[j] is calculated by following expression.
321 zone[i]->protection[j]
322 = (total sums of present_pages from zone[i+1] to zone[j] on the node)
323 / lowmem_reserve_ratio[i];
325 (should not be protected. = 0;
327 (not necessary, but looks 0)
329 The default values of lowmem_reserve_ratio[i] are
330 256 (if zone[i] means DMA or DMA32 zone)
332 As above expression, they are reciprocal number of ratio.
333 256 means 1/256. # of protection pages becomes about "0.39%" of total present
334 pages of higher zones on the node.
336 If you would like to protect more pages, smaller values are effective.
337 The minimum value is 1 (1/1 -> 100%).
339 ==============================================================
343 This file contains the maximum number of memory map areas a process
344 may have. Memory map areas are used as a side-effect of calling
345 malloc, directly by mmap and mprotect, and also when loading shared
348 While most applications need less than a thousand maps, certain
349 programs, particularly malloc debuggers, may consume lots of them,
350 e.g., up to one or two maps per allocation.
352 The default value is 65536.
354 =============================================================
356 memory_failure_early_kill:
358 Control how to kill processes when uncorrected memory error (typically
359 a 2bit error in a memory module) is detected in the background by hardware
360 that cannot be handled by the kernel. In some cases (like the page
361 still having a valid copy on disk) the kernel will handle the failure
362 transparently without affecting any applications. But if there is
363 no other uptodate copy of the data it will kill to prevent any data
364 corruptions from propagating.
366 1: Kill all processes that have the corrupted and not reloadable page mapped
367 as soon as the corruption is detected. Note this is not supported
368 for a few types of pages, like kernel internally allocated data or
369 the swap cache, but works for the majority of user pages.
371 0: Only unmap the corrupted page from all processes and only kill a process
372 who tries to access it.
374 The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
375 handle this if they want to.
377 This is only active on architectures/platforms with advanced machine
378 check handling and depends on the hardware capabilities.
380 Applications can override this setting individually with the PR_MCE_KILL prctl
382 ==============================================================
384 memory_failure_recovery
386 Enable memory failure recovery (when supported by the platform)
390 0: Always panic on a memory failure.
392 ==============================================================
396 This is used to force the Linux VM to keep a minimum number
397 of kilobytes free. The VM uses this number to compute a
398 watermark[WMARK_MIN] value for each lowmem zone in the system.
399 Each lowmem zone gets a number of reserved free pages based
400 proportionally on its size.
402 Some minimal amount of memory is needed to satisfy PF_MEMALLOC
403 allocations; if you set this to lower than 1024KB, your system will
404 become subtly broken, and prone to deadlock under high loads.
406 Setting this too high will OOM your machine instantly.
408 =============================================================
412 This is available only on NUMA kernels.
414 A percentage of the total pages in each zone. On Zone reclaim
415 (fallback from the local zone occurs) slabs will be reclaimed if more
416 than this percentage of pages in a zone are reclaimable slab pages.
417 This insures that the slab growth stays under control even in NUMA
418 systems that rarely perform global reclaim.
420 The default is 5 percent.
422 Note that slab reclaim is triggered in a per zone / node fashion.
423 The process of reclaiming slab memory is currently not node specific
426 =============================================================
430 This is available only on NUMA kernels.
432 This is a percentage of the total pages in each zone. Zone reclaim will
433 only occur if more than this percentage of pages are in a state that
434 zone_reclaim_mode allows to be reclaimed.
436 If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
437 against all file-backed unmapped pages including swapcache pages and tmpfs
438 files. Otherwise, only unmapped pages backed by normal files but not tmpfs
439 files and similar are considered.
441 The default is 1 percent.
443 ==============================================================
447 This file indicates the amount of address space which a user process will
448 be restricted from mmapping. Since kernel null dereference bugs could
449 accidentally operate based on the information in the first couple of pages
450 of memory userspace processes should not be allowed to write to them. By
451 default this value is set to 0 and no protections will be enforced by the
452 security module. Setting this value to something like 64k will allow the
453 vast majority of applications to work correctly and provide defense in depth
454 against future potential kernel bugs.
456 ==============================================================
460 Change the minimum size of the hugepage pool.
462 See Documentation/vm/hugetlbpage.txt
464 ==============================================================
466 nr_overcommit_hugepages
468 Change the maximum size of the hugepage pool. The maximum is
469 nr_hugepages + nr_overcommit_hugepages.
471 See Documentation/vm/hugetlbpage.txt
473 ==============================================================
477 This is available only on NOMMU kernels.
479 This value adjusts the excess page trimming behaviour of power-of-2 aligned
480 NOMMU mmap allocations.
482 A value of 0 disables trimming of allocations entirely, while a value of 1
483 trims excess pages aggressively. Any value >= 1 acts as the watermark where
484 trimming of allocations is initiated.
486 The default value is 1.
488 See Documentation/nommu-mmap.txt for more information.
490 ==============================================================
494 This sysctl is only for NUMA.
495 'where the memory is allocated from' is controlled by zonelists.
496 (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
497 you may be able to read ZONE_DMA as ZONE_DMA32...)
499 In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
500 ZONE_NORMAL -> ZONE_DMA
501 This means that a memory allocation request for GFP_KERNEL will
502 get memory from ZONE_DMA only when ZONE_NORMAL is not available.
504 In NUMA case, you can think of following 2 types of order.
505 Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL
507 (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
508 (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
510 Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
511 will be used before ZONE_NORMAL exhaustion. This increases possibility of
512 out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
514 Type(B) cannot offer the best locality but is more robust against OOM of
517 Type(A) is called as "Node" order. Type (B) is "Zone" order.
519 "Node order" orders the zonelists by node, then by zone within each node.
520 Specify "[Nn]ode" for node order
522 "Zone Order" orders the zonelists by zone type, then by node within each
523 zone. Specify "[Zz]one" for zone order.
525 Specify "[Dd]efault" to request automatic configuration. Autoconfiguration
526 will select "node" order in following case.
527 (1) if the DMA zone does not exist or
528 (2) if the DMA zone comprises greater than 50% of the available memory or
529 (3) if any node's DMA zone comprises greater than 60% of its local memory and
530 the amount of local memory is big enough.
532 Otherwise, "zone" order will be selected. Default order is recommended unless
533 this is causing problems for your system/application.
535 ==============================================================
539 Enables a system-wide task dump (excluding kernel threads) to be
540 produced when the kernel performs an OOM-killing and includes such
541 information as pid, uid, tgid, vm size, rss, nr_ptes, swapents,
542 oom_score_adj score, and name. This is helpful to determine why the
543 OOM killer was invoked, to identify the rogue task that caused it,
544 and to determine why the OOM killer chose the task it did to kill.
546 If this is set to zero, this information is suppressed. On very
547 large systems with thousands of tasks it may not be feasible to dump
548 the memory state information for each one. Such systems should not
549 be forced to incur a performance penalty in OOM conditions when the
550 information may not be desired.
552 If this is set to non-zero, this information is shown whenever the
553 OOM killer actually kills a memory-hogging task.
555 The default value is 1 (enabled).
557 ==============================================================
559 oom_kill_allocating_task
561 This enables or disables killing the OOM-triggering task in
562 out-of-memory situations.
564 If this is set to zero, the OOM killer will scan through the entire
565 tasklist and select a task based on heuristics to kill. This normally
566 selects a rogue memory-hogging task that frees up a large amount of
569 If this is set to non-zero, the OOM killer simply kills the task that
570 triggered the out-of-memory condition. This avoids the expensive
573 If panic_on_oom is selected, it takes precedence over whatever value
574 is used in oom_kill_allocating_task.
576 The default value is 0.
578 ==============================================================
582 This value contains a flag that enables memory overcommitment.
584 When this flag is 0, the kernel attempts to estimate the amount
585 of free memory left when userspace requests more memory.
587 When this flag is 1, the kernel pretends there is always enough
588 memory until it actually runs out.
590 When this flag is 2, the kernel uses a "never overcommit"
591 policy that attempts to prevent any overcommit of memory.
592 Note that user_reserve_kbytes affects this policy.
594 This feature can be very useful because there are a lot of
595 programs that malloc() huge amounts of memory "just-in-case"
596 and don't use much of it.
598 The default value is 0.
600 See Documentation/vm/overcommit-accounting and
601 security/commoncap.c::cap_vm_enough_memory() for more information.
603 ==============================================================
607 When overcommit_memory is set to 2, the committed address
608 space is not permitted to exceed swap plus this percentage
609 of physical RAM. See above.
611 ==============================================================
615 page-cluster controls the number of pages up to which consecutive pages
616 are read in from swap in a single attempt. This is the swap counterpart
617 to page cache readahead.
618 The mentioned consecutivity is not in terms of virtual/physical addresses,
619 but consecutive on swap space - that means they were swapped out together.
621 It is a logarithmic value - setting it to zero means "1 page", setting
622 it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
623 Zero disables swap readahead completely.
625 The default value is three (eight pages at a time). There may be some
626 small benefits in tuning this to a different value if your workload is
629 Lower values mean lower latencies for initial faults, but at the same time
630 extra faults and I/O delays for following faults if they would have been part of
631 that consecutive pages readahead would have brought in.
633 =============================================================
637 This enables or disables panic on out-of-memory feature.
639 If this is set to 0, the kernel will kill some rogue process,
640 called oom_killer. Usually, oom_killer can kill rogue processes and
643 If this is set to 1, the kernel panics when out-of-memory happens.
644 However, if a process limits using nodes by mempolicy/cpusets,
645 and those nodes become memory exhaustion status, one process
646 may be killed by oom-killer. No panic occurs in this case.
647 Because other nodes' memory may be free. This means system total status
648 may be not fatal yet.
650 If this is set to 2, the kernel panics compulsorily even on the
651 above-mentioned. Even oom happens under memory cgroup, the whole
654 The default value is 0.
655 1 and 2 are for failover of clustering. Please select either
656 according to your policy of failover.
657 panic_on_oom=2+kdump gives you very strong tool to investigate
658 why oom happens. You can get snapshot.
660 =============================================================
662 percpu_pagelist_fraction
664 This is the fraction of pages at most (high mark pcp->high) in each zone that
665 are allocated for each per cpu page list. The min value for this is 8. It
666 means that we don't allow more than 1/8th of pages in each zone to be
667 allocated in any single per_cpu_pagelist. This entry only changes the value
668 of hot per cpu pagelists. User can specify a number like 100 to allocate
669 1/100th of each zone to each per cpu page list.
671 The batch value of each per cpu pagelist is also updated as a result. It is
672 set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8)
674 The initial value is zero. Kernel does not use this value at boot time to set
675 the high water marks for each per cpu page list.
677 ==============================================================
681 The time interval between which vm statistics are updated. The default
684 ==============================================================
688 This control is used to define how aggressive the kernel will swap
689 memory pages. Higher values will increase agressiveness, lower values
690 decrease the amount of swap.
692 The default value is 60.
694 ==============================================================
696 - user_reserve_kbytes
698 When overcommit_memory is set to 2, "never overommit" mode, reserve
699 min(3% of current process size, user_reserve_kbytes) of free memory.
700 This is intended to prevent a user from starting a single memory hogging
701 process, such that they cannot recover (kill the hog).
703 user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
705 If this is reduced to zero, then the user will be allowed to allocate
706 all free memory with a single process, minus admin_reserve_kbytes.
707 Any subsequent attempts to execute a command will result in
708 "fork: Cannot allocate memory".
710 Changing this takes effect whenever an application requests memory.
712 ==============================================================
717 Controls the tendency of the kernel to reclaim the memory which is used for
718 caching of directory and inode objects.
720 At the default value of vfs_cache_pressure=100 the kernel will attempt to
721 reclaim dentries and inodes at a "fair" rate with respect to pagecache and
722 swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer
723 to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
724 never reclaim dentries and inodes due to memory pressure and this can easily
725 lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
726 causes the kernel to prefer to reclaim dentries and inodes.
728 ==============================================================
732 Zone_reclaim_mode allows someone to set more or less aggressive approaches to
733 reclaim memory when a zone runs out of memory. If it is set to zero then no
734 zone reclaim occurs. Allocations will be satisfied from other zones / nodes
737 This is value ORed together of
740 2 = Zone reclaim writes dirty pages out
741 4 = Zone reclaim swaps pages
743 zone_reclaim_mode is set during bootup to 1 if it is determined that pages
744 from remote zones will cause a measurable performance reduction. The
745 page allocator will then reclaim easily reusable pages (those page
746 cache pages that are currently not used) before allocating off node pages.
748 It may be beneficial to switch off zone reclaim if the system is
749 used for a file server and all of memory should be used for caching files
750 from disk. In that case the caching effect is more important than
753 Allowing zone reclaim to write out pages stops processes that are
754 writing large amounts of data from dirtying pages on other nodes. Zone
755 reclaim will write out dirty pages if a zone fills up and so effectively
756 throttle the process. This may decrease the performance of a single process
757 since it cannot use all of system memory to buffer the outgoing writes
758 anymore but it preserve the memory on other nodes so that the performance
759 of other processes running on other nodes will not be affected.
761 Allowing regular swap effectively restricts allocations to the local
762 node unless explicitly overridden by memory policies or cpuset
765 ============ End of Document =================================