Root/Documentation/sysctl/vm.txt

1Documentation 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>
4
5For general info and legal blurb, please look in README.
6
7==============================================================
8
9This file contains the documentation for the sysctl files in
10/proc/sys/vm and is valid for Linux kernel version 2.6.29.
11
12The files in this directory can be used to tune the operation
13of the virtual memory (VM) subsystem of the Linux kernel and
14the writeout of dirty data to disk.
15
16Default values and initialization routines for most of these
17files can be found in mm/swap.c.
18
19Currently, these files are in /proc/sys/vm:
20
21- block_dump
22- dirty_background_bytes
23- dirty_background_ratio
24- dirty_bytes
25- dirty_expire_centisecs
26- dirty_ratio
27- dirty_writeback_centisecs
28- drop_caches
29- hugepages_treat_as_movable
30- hugetlb_shm_group
31- laptop_mode
32- legacy_va_layout
33- lowmem_reserve_ratio
34- max_map_count
35- memory_failure_early_kill
36- memory_failure_recovery
37- min_free_kbytes
38- min_slab_ratio
39- min_unmapped_ratio
40- mmap_min_addr
41- nr_hugepages
42- nr_overcommit_hugepages
43- nr_pdflush_threads
44- nr_trim_pages (only if CONFIG_MMU=n)
45- numa_zonelist_order
46- oom_dump_tasks
47- oom_kill_allocating_task
48- overcommit_memory
49- overcommit_ratio
50- page-cluster
51- panic_on_oom
52- percpu_pagelist_fraction
53- stat_interval
54- swappiness
55- vfs_cache_pressure
56- zone_reclaim_mode
57
58==============================================================
59
60block_dump
61
62block_dump enables block I/O debugging when set to a nonzero value. More
63information on block I/O debugging is in Documentation/laptops/laptop-mode.txt.
64
65==============================================================
66
67dirty_background_bytes
68
69Contains the amount of dirty memory at which the pdflush background writeback
70daemon will start writeback.
71
72If dirty_background_bytes is written, dirty_background_ratio becomes a function
73of its value (dirty_background_bytes / the amount of dirtyable system memory).
74
75==============================================================
76
77dirty_background_ratio
78
79Contains, as a percentage of total system memory, the number of pages at which
80the pdflush background writeback daemon will start writing out dirty data.
81
82==============================================================
83
84dirty_bytes
85
86Contains the amount of dirty memory at which a process generating disk writes
87will itself start writeback.
88
89If dirty_bytes is written, dirty_ratio becomes a function of its value
90(dirty_bytes / the amount of dirtyable system memory).
91
92Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
93value lower than this limit will be ignored and the old configuration will be
94retained.
95
96==============================================================
97
98dirty_expire_centisecs
99
100This tunable is used to define when dirty data is old enough to be eligible
101for writeout by the pdflush daemons. It is expressed in 100'ths of a second.
102Data which has been dirty in-memory for longer than this interval will be
103written out next time a pdflush daemon wakes up.
104
105==============================================================
106
107dirty_ratio
108
109Contains, as a percentage of total system memory, the number of pages at which
110a process which is generating disk writes will itself start writing out dirty
111data.
112
113==============================================================
114
115dirty_writeback_centisecs
116
117The pdflush writeback daemons will periodically wake up and write `old' data
118out to disk. This tunable expresses the interval between those wakeups, in
119100'ths of a second.
120
121Setting this to zero disables periodic writeback altogether.
122
123==============================================================
124
125drop_caches
126
127Writing to this will cause the kernel to drop clean caches, dentries and
128inodes from memory, causing that memory to become free.
129
130To free pagecache:
131    echo 1 > /proc/sys/vm/drop_caches
132To free dentries and inodes:
133    echo 2 > /proc/sys/vm/drop_caches
134To free pagecache, dentries and inodes:
135    echo 3 > /proc/sys/vm/drop_caches
136
137As this is a non-destructive operation and dirty objects are not freeable, the
138user should run `sync' first.
139
140==============================================================
141
142hugepages_treat_as_movable
143
144This parameter is only useful when kernelcore= is specified at boot time to
145create ZONE_MOVABLE for pages that may be reclaimed or migrated. Huge pages
146are not movable so are not normally allocated from ZONE_MOVABLE. A non-zero
147value written to hugepages_treat_as_movable allows huge pages to be allocated
148from ZONE_MOVABLE.
149
150Once enabled, the ZONE_MOVABLE is treated as an area of memory the huge
151pages pool can easily grow or shrink within. Assuming that applications are
152not running that mlock() a lot of memory, it is likely the huge pages pool
153can grow to the size of ZONE_MOVABLE by repeatedly entering the desired value
154into nr_hugepages and triggering page reclaim.
155
156==============================================================
157
158hugetlb_shm_group
159
160hugetlb_shm_group contains group id that is allowed to create SysV
161shared memory segment using hugetlb page.
162
163==============================================================
164
165laptop_mode
166
167laptop_mode is a knob that controls "laptop mode". All the things that are
168controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt.
169
170==============================================================
171
172legacy_va_layout
173
174If non-zero, this sysctl disables the new 32-bit mmap mmap layout - the kernel
175will use the legacy (2.4) layout for all processes.
176
177==============================================================
178
179lowmem_reserve_ratio
180
181For some specialised workloads on highmem machines it is dangerous for
182the kernel to allow process memory to be allocated from the "lowmem"
183zone. This is because that memory could then be pinned via the mlock()
184system call, or by unavailability of swapspace.
185
186And on large highmem machines this lack of reclaimable lowmem memory
187can be fatal.
188
189So the Linux page allocator has a mechanism which prevents allocations
190which _could_ use highmem from using too much lowmem. This means that
191a certain amount of lowmem is defended from the possibility of being
192captured into pinned user memory.
193
194(The same argument applies to the old 16 megabyte ISA DMA region. This
195mechanism will also defend that region from allocations which could use
196highmem or lowmem).
197
198The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is
199in defending these lower zones.
200
201If you have a machine which uses highmem or ISA DMA and your
202applications are using mlock(), or if you are running with no swap then
203you probably should change the lowmem_reserve_ratio setting.
204
205The lowmem_reserve_ratio is an array. You can see them by reading this file.
206-
207% cat /proc/sys/vm/lowmem_reserve_ratio
208256 256 32
209-
210Note: # of this elements is one fewer than number of zones. Because the highest
211      zone's value is not necessary for following calculation.
212
213But, these values are not used directly. The kernel calculates # of protection
214pages for each zones from them. These are shown as array of protection pages
215in /proc/zoneinfo like followings. (This is an example of x86-64 box).
216Each zone has an array of protection pages like this.
217
218-
219Node 0, zone DMA
220  pages free 1355
221        min 3
222        low 3
223        high 4
224    :
225    :
226    numa_other 0
227        protection: (0, 2004, 2004, 2004)
228    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
229  pagesets
230    cpu: 0 pcp: 0
231        :
232-
233These protections are added to score to judge whether this zone should be used
234for page allocation or should be reclaimed.
235
236In this example, if normal pages (index=2) are required to this DMA zone and
237watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
238not be used because pages_free(1355) is smaller than watermark + protection[2]
239(4 + 2004 = 2008). If this protection value is 0, this zone would be used for
240normal page requirement. If requirement is DMA zone(index=0), protection[0]
241(=0) is used.
242
243zone[i]'s protection[j] is calculated by following expression.
244
245(i < j):
246  zone[i]->protection[j]
247  = (total sums of present_pages from zone[i+1] to zone[j] on the node)
248    / lowmem_reserve_ratio[i];
249(i = j):
250   (should not be protected. = 0;
251(i > j):
252   (not necessary, but looks 0)
253
254The default values of lowmem_reserve_ratio[i] are
255    256 (if zone[i] means DMA or DMA32 zone)
256    32 (others).
257As above expression, they are reciprocal number of ratio.
258256 means 1/256. # of protection pages becomes about "0.39%" of total present
259pages of higher zones on the node.
260
261If you would like to protect more pages, smaller values are effective.
262The minimum value is 1 (1/1 -> 100%).
263
264==============================================================
265
266max_map_count:
267
268This file contains the maximum number of memory map areas a process
269may have. Memory map areas are used as a side-effect of calling
270malloc, directly by mmap and mprotect, and also when loading shared
271libraries.
272
273While most applications need less than a thousand maps, certain
274programs, particularly malloc debuggers, may consume lots of them,
275e.g., up to one or two maps per allocation.
276
277The default value is 65536.
278
279=============================================================
280
281memory_failure_early_kill:
282
283Control how to kill processes when uncorrected memory error (typically
284a 2bit error in a memory module) is detected in the background by hardware
285that cannot be handled by the kernel. In some cases (like the page
286still having a valid copy on disk) the kernel will handle the failure
287transparently without affecting any applications. But if there is
288no other uptodate copy of the data it will kill to prevent any data
289corruptions from propagating.
290
2911: Kill all processes that have the corrupted and not reloadable page mapped
292as soon as the corruption is detected. Note this is not supported
293for a few types of pages, like kernel internally allocated data or
294the swap cache, but works for the majority of user pages.
295
2960: Only unmap the corrupted page from all processes and only kill a process
297who tries to access it.
298
299The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
300handle this if they want to.
301
302This is only active on architectures/platforms with advanced machine
303check handling and depends on the hardware capabilities.
304
305Applications can override this setting individually with the PR_MCE_KILL prctl
306
307==============================================================
308
309memory_failure_recovery
310
311Enable memory failure recovery (when supported by the platform)
312
3131: Attempt recovery.
314
3150: Always panic on a memory failure.
316
317==============================================================
318
319min_free_kbytes:
320
321This is used to force the Linux VM to keep a minimum number
322of kilobytes free. The VM uses this number to compute a
323watermark[WMARK_MIN] value for each lowmem zone in the system.
324Each lowmem zone gets a number of reserved free pages based
325proportionally on its size.
326
327Some minimal amount of memory is needed to satisfy PF_MEMALLOC
328allocations; if you set this to lower than 1024KB, your system will
329become subtly broken, and prone to deadlock under high loads.
330
331Setting this too high will OOM your machine instantly.
332
333=============================================================
334
335min_slab_ratio:
336
337This is available only on NUMA kernels.
338
339A percentage of the total pages in each zone. On Zone reclaim
340(fallback from the local zone occurs) slabs will be reclaimed if more
341than this percentage of pages in a zone are reclaimable slab pages.
342This insures that the slab growth stays under control even in NUMA
343systems that rarely perform global reclaim.
344
345The default is 5 percent.
346
347Note that slab reclaim is triggered in a per zone / node fashion.
348The process of reclaiming slab memory is currently not node specific
349and may not be fast.
350
351=============================================================
352
353min_unmapped_ratio:
354
355This is available only on NUMA kernels.
356
357This is a percentage of the total pages in each zone. Zone reclaim will
358only occur if more than this percentage of pages are in a state that
359zone_reclaim_mode allows to be reclaimed.
360
361If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
362against all file-backed unmapped pages including swapcache pages and tmpfs
363files. Otherwise, only unmapped pages backed by normal files but not tmpfs
364files and similar are considered.
365
366The default is 1 percent.
367
368==============================================================
369
370mmap_min_addr
371
372This file indicates the amount of address space which a user process will
373be restricted from mmapping. Since kernel null dereference bugs could
374accidentally operate based on the information in the first couple of pages
375of memory userspace processes should not be allowed to write to them. By
376default this value is set to 0 and no protections will be enforced by the
377security module. Setting this value to something like 64k will allow the
378vast majority of applications to work correctly and provide defense in depth
379against future potential kernel bugs.
380
381==============================================================
382
383nr_hugepages
384
385Change the minimum size of the hugepage pool.
386
387See Documentation/vm/hugetlbpage.txt
388
389==============================================================
390
391nr_overcommit_hugepages
392
393Change the maximum size of the hugepage pool. The maximum is
394nr_hugepages + nr_overcommit_hugepages.
395
396See Documentation/vm/hugetlbpage.txt
397
398==============================================================
399
400nr_pdflush_threads
401
402The current number of pdflush threads. This value is read-only.
403The value changes according to the number of dirty pages in the system.
404
405When necessary, additional pdflush threads are created, one per second, up to
406nr_pdflush_threads_max.
407
408==============================================================
409
410nr_trim_pages
411
412This is available only on NOMMU kernels.
413
414This value adjusts the excess page trimming behaviour of power-of-2 aligned
415NOMMU mmap allocations.
416
417A value of 0 disables trimming of allocations entirely, while a value of 1
418trims excess pages aggressively. Any value >= 1 acts as the watermark where
419trimming of allocations is initiated.
420
421The default value is 1.
422
423See Documentation/nommu-mmap.txt for more information.
424
425==============================================================
426
427numa_zonelist_order
428
429This sysctl is only for NUMA.
430'where the memory is allocated from' is controlled by zonelists.
431(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
432 you may be able to read ZONE_DMA as ZONE_DMA32...)
433
434In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
435ZONE_NORMAL -> ZONE_DMA
436This means that a memory allocation request for GFP_KERNEL will
437get memory from ZONE_DMA only when ZONE_NORMAL is not available.
438
439In NUMA case, you can think of following 2 types of order.
440Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL
441
442(A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
443(B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
444
445Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
446will be used before ZONE_NORMAL exhaustion. This increases possibility of
447out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
448
449Type(B) cannot offer the best locality but is more robust against OOM of
450the DMA zone.
451
452Type(A) is called as "Node" order. Type (B) is "Zone" order.
453
454"Node order" orders the zonelists by node, then by zone within each node.
455Specify "[Nn]ode" for zone order
456
457"Zone Order" orders the zonelists by zone type, then by node within each
458zone. Specify "[Zz]one"for zode order.
459
460Specify "[Dd]efault" to request automatic configuration. Autoconfiguration
461will select "node" order in following case.
462(1) if the DMA zone does not exist or
463(2) if the DMA zone comprises greater than 50% of the available memory or
464(3) if any node's DMA zone comprises greater than 60% of its local memory and
465    the amount of local memory is big enough.
466
467Otherwise, "zone" order will be selected. Default order is recommended unless
468this is causing problems for your system/application.
469
470==============================================================
471
472oom_dump_tasks
473
474Enables a system-wide task dump (excluding kernel threads) to be
475produced when the kernel performs an OOM-killing and includes such
476information as pid, uid, tgid, vm size, rss, cpu, oom_adj score, and
477name. This is helpful to determine why the OOM killer was invoked
478and to identify the rogue task that caused it.
479
480If this is set to zero, this information is suppressed. On very
481large systems with thousands of tasks it may not be feasible to dump
482the memory state information for each one. Such systems should not
483be forced to incur a performance penalty in OOM conditions when the
484information may not be desired.
485
486If this is set to non-zero, this information is shown whenever the
487OOM killer actually kills a memory-hogging task.
488
489The default value is 0.
490
491==============================================================
492
493oom_kill_allocating_task
494
495This enables or disables killing the OOM-triggering task in
496out-of-memory situations.
497
498If this is set to zero, the OOM killer will scan through the entire
499tasklist and select a task based on heuristics to kill. This normally
500selects a rogue memory-hogging task that frees up a large amount of
501memory when killed.
502
503If this is set to non-zero, the OOM killer simply kills the task that
504triggered the out-of-memory condition. This avoids the expensive
505tasklist scan.
506
507If panic_on_oom is selected, it takes precedence over whatever value
508is used in oom_kill_allocating_task.
509
510The default value is 0.
511
512==============================================================
513
514overcommit_memory:
515
516This value contains a flag that enables memory overcommitment.
517
518When this flag is 0, the kernel attempts to estimate the amount
519of free memory left when userspace requests more memory.
520
521When this flag is 1, the kernel pretends there is always enough
522memory until it actually runs out.
523
524When this flag is 2, the kernel uses a "never overcommit"
525policy that attempts to prevent any overcommit of memory.
526
527This feature can be very useful because there are a lot of
528programs that malloc() huge amounts of memory "just-in-case"
529and don't use much of it.
530
531The default value is 0.
532
533See Documentation/vm/overcommit-accounting and
534security/commoncap.c::cap_vm_enough_memory() for more information.
535
536==============================================================
537
538overcommit_ratio:
539
540When overcommit_memory is set to 2, the committed address
541space is not permitted to exceed swap plus this percentage
542of physical RAM. See above.
543
544==============================================================
545
546page-cluster
547
548page-cluster controls the number of pages which are written to swap in
549a single attempt. The swap I/O size.
550
551It is a logarithmic value - setting it to zero means "1 page", setting
552it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
553
554The default value is three (eight pages at a time). There may be some
555small benefits in tuning this to a different value if your workload is
556swap-intensive.
557
558=============================================================
559
560panic_on_oom
561
562This enables or disables panic on out-of-memory feature.
563
564If this is set to 0, the kernel will kill some rogue process,
565called oom_killer. Usually, oom_killer can kill rogue processes and
566system will survive.
567
568If this is set to 1, the kernel panics when out-of-memory happens.
569However, if a process limits using nodes by mempolicy/cpusets,
570and those nodes become memory exhaustion status, one process
571may be killed by oom-killer. No panic occurs in this case.
572Because other nodes' memory may be free. This means system total status
573may be not fatal yet.
574
575If this is set to 2, the kernel panics compulsorily even on the
576above-mentioned. Even oom happens under memory cgroup, the whole
577system panics.
578
579The default value is 0.
5801 and 2 are for failover of clustering. Please select either
581according to your policy of failover.
582panic_on_oom=2+kdump gives you very strong tool to investigate
583why oom happens. You can get snapshot.
584
585=============================================================
586
587percpu_pagelist_fraction
588
589This is the fraction of pages at most (high mark pcp->high) in each zone that
590are allocated for each per cpu page list. The min value for this is 8. It
591means that we don't allow more than 1/8th of pages in each zone to be
592allocated in any single per_cpu_pagelist. This entry only changes the value
593of hot per cpu pagelists. User can specify a number like 100 to allocate
5941/100th of each zone to each per cpu page list.
595
596The batch value of each per cpu pagelist is also updated as a result. It is
597set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8)
598
599The initial value is zero. Kernel does not use this value at boot time to set
600the high water marks for each per cpu page list.
601
602==============================================================
603
604stat_interval
605
606The time interval between which vm statistics are updated. The default
607is 1 second.
608
609==============================================================
610
611swappiness
612
613This control is used to define how aggressive the kernel will swap
614memory pages. Higher values will increase agressiveness, lower values
615decrease the amount of swap.
616
617The default value is 60.
618
619==============================================================
620
621vfs_cache_pressure
622------------------
623
624Controls the tendency of the kernel to reclaim the memory which is used for
625caching of directory and inode objects.
626
627At the default value of vfs_cache_pressure=100 the kernel will attempt to
628reclaim dentries and inodes at a "fair" rate with respect to pagecache and
629swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer
630to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
631never reclaim dentries and inodes due to memory pressure and this can easily
632lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
633causes the kernel to prefer to reclaim dentries and inodes.
634
635==============================================================
636
637zone_reclaim_mode:
638
639Zone_reclaim_mode allows someone to set more or less aggressive approaches to
640reclaim memory when a zone runs out of memory. If it is set to zero then no
641zone reclaim occurs. Allocations will be satisfied from other zones / nodes
642in the system.
643
644This is value ORed together of
645
6461 = Zone reclaim on
6472 = Zone reclaim writes dirty pages out
6484 = Zone reclaim swaps pages
649
650zone_reclaim_mode is set during bootup to 1 if it is determined that pages
651from remote zones will cause a measurable performance reduction. The
652page allocator will then reclaim easily reusable pages (those page
653cache pages that are currently not used) before allocating off node pages.
654
655It may be beneficial to switch off zone reclaim if the system is
656used for a file server and all of memory should be used for caching files
657from disk. In that case the caching effect is more important than
658data locality.
659
660Allowing zone reclaim to write out pages stops processes that are
661writing large amounts of data from dirtying pages on other nodes. Zone
662reclaim will write out dirty pages if a zone fills up and so effectively
663throttle the process. This may decrease the performance of a single process
664since it cannot use all of system memory to buffer the outgoing writes
665anymore but it preserve the memory on other nodes so that the performance
666of other processes running on other nodes will not be affected.
667
668Allowing regular swap effectively restricts allocations to the local
669node unless explicitly overridden by memory policies or cpuset
670configurations.
671
672============ End of Document =================================
673

Archive Download this file



interactive