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- compact_memory
23- dirty_background_bytes
24- dirty_background_ratio
25- dirty_bytes
26- dirty_expire_centisecs
27- dirty_ratio
28- dirty_writeback_centisecs
29- drop_caches
30- extfrag_threshold
31- hugepages_treat_as_movable
32- hugetlb_shm_group
33- laptop_mode
34- legacy_va_layout
35- lowmem_reserve_ratio
36- max_map_count
37- memory_failure_early_kill
38- memory_failure_recovery
39- min_free_kbytes
40- min_slab_ratio
41- min_unmapped_ratio
42- mmap_min_addr
43- nr_hugepages
44- nr_overcommit_hugepages
45- nr_pdflush_threads
46- nr_trim_pages (only if CONFIG_MMU=n)
47- numa_zonelist_order
48- oom_dump_tasks
49- oom_kill_allocating_task
50- overcommit_memory
51- overcommit_ratio
52- page-cluster
53- panic_on_oom
54- percpu_pagelist_fraction
55- stat_interval
56- swappiness
57- vfs_cache_pressure
58- zone_reclaim_mode
59
60==============================================================
61
62block_dump
63
64block_dump enables block I/O debugging when set to a nonzero value. More
65information on block I/O debugging is in Documentation/laptops/laptop-mode.txt.
66
67==============================================================
68
69compact_memory
70
71Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
72all zones are compacted such that free memory is available in contiguous
73blocks where possible. This can be important for example in the allocation of
74huge pages although processes will also directly compact memory as required.
75
76==============================================================
77
78dirty_background_bytes
79
80Contains the amount of dirty memory at which the pdflush background writeback
81daemon will start writeback.
82
83Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only
84one of them may be specified at a time. When one sysctl is written it is
85immediately taken into account to evaluate the dirty memory limits and the
86other appears as 0 when read.
87
88==============================================================
89
90dirty_background_ratio
91
92Contains, as a percentage of total system memory, the number of pages at which
93the pdflush background writeback daemon will start writing out dirty data.
94
95==============================================================
96
97dirty_bytes
98
99Contains the amount of dirty memory at which a process generating disk writes
100will itself start writeback.
101
102Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
103specified at a time. When one sysctl is written it is immediately taken into
104account to evaluate the dirty memory limits and the other appears as 0 when
105read.
106
107Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
108value lower than this limit will be ignored and the old configuration will be
109retained.
110
111==============================================================
112
113dirty_expire_centisecs
114
115This tunable is used to define when dirty data is old enough to be eligible
116for writeout by the pdflush daemons. It is expressed in 100'ths of a second.
117Data which has been dirty in-memory for longer than this interval will be
118written out next time a pdflush daemon wakes up.
119
120==============================================================
121
122dirty_ratio
123
124Contains, as a percentage of total system memory, the number of pages at which
125a process which is generating disk writes will itself start writing out dirty
126data.
127
128==============================================================
129
130dirty_writeback_centisecs
131
132The pdflush writeback daemons will periodically wake up and write `old' data
133out to disk. This tunable expresses the interval between those wakeups, in
134100'ths of a second.
135
136Setting this to zero disables periodic writeback altogether.
137
138==============================================================
139
140drop_caches
141
142Writing to this will cause the kernel to drop clean caches, dentries and
143inodes from memory, causing that memory to become free.
144
145To free pagecache:
146    echo 1 > /proc/sys/vm/drop_caches
147To free dentries and inodes:
148    echo 2 > /proc/sys/vm/drop_caches
149To free pagecache, dentries and inodes:
150    echo 3 > /proc/sys/vm/drop_caches
151
152As this is a non-destructive operation and dirty objects are not freeable, the
153user should run `sync' first.
154
155==============================================================
156
157extfrag_threshold
158
159This parameter affects whether the kernel will compact memory or direct
160reclaim to satisfy a high-order allocation. /proc/extfrag_index shows what
161the fragmentation index for each order is in each zone in the system. Values
162tending towards 0 imply allocations would fail due to lack of memory,
163values towards 1000 imply failures are due to fragmentation and -1 implies
164that the allocation will succeed as long as watermarks are met.
165
166The kernel will not compact memory in a zone if the
167fragmentation index is <= extfrag_threshold. The default value is 500.
168
169==============================================================
170
171hugepages_treat_as_movable
172
173This parameter is only useful when kernelcore= is specified at boot time to
174create ZONE_MOVABLE for pages that may be reclaimed or migrated. Huge pages
175are not movable so are not normally allocated from ZONE_MOVABLE. A non-zero
176value written to hugepages_treat_as_movable allows huge pages to be allocated
177from ZONE_MOVABLE.
178
179Once enabled, the ZONE_MOVABLE is treated as an area of memory the huge
180pages pool can easily grow or shrink within. Assuming that applications are
181not running that mlock() a lot of memory, it is likely the huge pages pool
182can grow to the size of ZONE_MOVABLE by repeatedly entering the desired value
183into nr_hugepages and triggering page reclaim.
184
185==============================================================
186
187hugetlb_shm_group
188
189hugetlb_shm_group contains group id that is allowed to create SysV
190shared memory segment using hugetlb page.
191
192==============================================================
193
194laptop_mode
195
196laptop_mode is a knob that controls "laptop mode". All the things that are
197controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt.
198
199==============================================================
200
201legacy_va_layout
202
203If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
204will use the legacy (2.4) layout for all processes.
205
206==============================================================
207
208lowmem_reserve_ratio
209
210For some specialised workloads on highmem machines it is dangerous for
211the kernel to allow process memory to be allocated from the "lowmem"
212zone. This is because that memory could then be pinned via the mlock()
213system call, or by unavailability of swapspace.
214
215And on large highmem machines this lack of reclaimable lowmem memory
216can be fatal.
217
218So the Linux page allocator has a mechanism which prevents allocations
219which _could_ use highmem from using too much lowmem. This means that
220a certain amount of lowmem is defended from the possibility of being
221captured into pinned user memory.
222
223(The same argument applies to the old 16 megabyte ISA DMA region. This
224mechanism will also defend that region from allocations which could use
225highmem or lowmem).
226
227The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is
228in defending these lower zones.
229
230If you have a machine which uses highmem or ISA DMA and your
231applications are using mlock(), or if you are running with no swap then
232you probably should change the lowmem_reserve_ratio setting.
233
234The lowmem_reserve_ratio is an array. You can see them by reading this file.
235-
236% cat /proc/sys/vm/lowmem_reserve_ratio
237256 256 32
238-
239Note: # of this elements is one fewer than number of zones. Because the highest
240      zone's value is not necessary for following calculation.
241
242But, these values are not used directly. The kernel calculates # of protection
243pages for each zones from them. These are shown as array of protection pages
244in /proc/zoneinfo like followings. (This is an example of x86-64 box).
245Each zone has an array of protection pages like this.
246
247-
248Node 0, zone DMA
249  pages free 1355
250        min 3
251        low 3
252        high 4
253    :
254    :
255    numa_other 0
256        protection: (0, 2004, 2004, 2004)
257    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
258  pagesets
259    cpu: 0 pcp: 0
260        :
261-
262These protections are added to score to judge whether this zone should be used
263for page allocation or should be reclaimed.
264
265In this example, if normal pages (index=2) are required to this DMA zone and
266watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
267not be used because pages_free(1355) is smaller than watermark + protection[2]
268(4 + 2004 = 2008). If this protection value is 0, this zone would be used for
269normal page requirement. If requirement is DMA zone(index=0), protection[0]
270(=0) is used.
271
272zone[i]'s protection[j] is calculated by following expression.
273
274(i < j):
275  zone[i]->protection[j]
276  = (total sums of present_pages from zone[i+1] to zone[j] on the node)
277    / lowmem_reserve_ratio[i];
278(i = j):
279   (should not be protected. = 0;
280(i > j):
281   (not necessary, but looks 0)
282
283The default values of lowmem_reserve_ratio[i] are
284    256 (if zone[i] means DMA or DMA32 zone)
285    32 (others).
286As above expression, they are reciprocal number of ratio.
287256 means 1/256. # of protection pages becomes about "0.39%" of total present
288pages of higher zones on the node.
289
290If you would like to protect more pages, smaller values are effective.
291The minimum value is 1 (1/1 -> 100%).
292
293==============================================================
294
295max_map_count:
296
297This file contains the maximum number of memory map areas a process
298may have. Memory map areas are used as a side-effect of calling
299malloc, directly by mmap and mprotect, and also when loading shared
300libraries.
301
302While most applications need less than a thousand maps, certain
303programs, particularly malloc debuggers, may consume lots of them,
304e.g., up to one or two maps per allocation.
305
306The default value is 65536.
307
308=============================================================
309
310memory_failure_early_kill:
311
312Control how to kill processes when uncorrected memory error (typically
313a 2bit error in a memory module) is detected in the background by hardware
314that cannot be handled by the kernel. In some cases (like the page
315still having a valid copy on disk) the kernel will handle the failure
316transparently without affecting any applications. But if there is
317no other uptodate copy of the data it will kill to prevent any data
318corruptions from propagating.
319
3201: Kill all processes that have the corrupted and not reloadable page mapped
321as soon as the corruption is detected. Note this is not supported
322for a few types of pages, like kernel internally allocated data or
323the swap cache, but works for the majority of user pages.
324
3250: Only unmap the corrupted page from all processes and only kill a process
326who tries to access it.
327
328The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
329handle this if they want to.
330
331This is only active on architectures/platforms with advanced machine
332check handling and depends on the hardware capabilities.
333
334Applications can override this setting individually with the PR_MCE_KILL prctl
335
336==============================================================
337
338memory_failure_recovery
339
340Enable memory failure recovery (when supported by the platform)
341
3421: Attempt recovery.
343
3440: Always panic on a memory failure.
345
346==============================================================
347
348min_free_kbytes:
349
350This is used to force the Linux VM to keep a minimum number
351of kilobytes free. The VM uses this number to compute a
352watermark[WMARK_MIN] value for each lowmem zone in the system.
353Each lowmem zone gets a number of reserved free pages based
354proportionally on its size.
355
356Some minimal amount of memory is needed to satisfy PF_MEMALLOC
357allocations; if you set this to lower than 1024KB, your system will
358become subtly broken, and prone to deadlock under high loads.
359
360Setting this too high will OOM your machine instantly.
361
362=============================================================
363
364min_slab_ratio:
365
366This is available only on NUMA kernels.
367
368A percentage of the total pages in each zone. On Zone reclaim
369(fallback from the local zone occurs) slabs will be reclaimed if more
370than this percentage of pages in a zone are reclaimable slab pages.
371This insures that the slab growth stays under control even in NUMA
372systems that rarely perform global reclaim.
373
374The default is 5 percent.
375
376Note that slab reclaim is triggered in a per zone / node fashion.
377The process of reclaiming slab memory is currently not node specific
378and may not be fast.
379
380=============================================================
381
382min_unmapped_ratio:
383
384This is available only on NUMA kernels.
385
386This is a percentage of the total pages in each zone. Zone reclaim will
387only occur if more than this percentage of pages are in a state that
388zone_reclaim_mode allows to be reclaimed.
389
390If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
391against all file-backed unmapped pages including swapcache pages and tmpfs
392files. Otherwise, only unmapped pages backed by normal files but not tmpfs
393files and similar are considered.
394
395The default is 1 percent.
396
397==============================================================
398
399mmap_min_addr
400
401This file indicates the amount of address space which a user process will
402be restricted from mmapping. Since kernel null dereference bugs could
403accidentally operate based on the information in the first couple of pages
404of memory userspace processes should not be allowed to write to them. By
405default this value is set to 0 and no protections will be enforced by the
406security module. Setting this value to something like 64k will allow the
407vast majority of applications to work correctly and provide defense in depth
408against future potential kernel bugs.
409
410==============================================================
411
412nr_hugepages
413
414Change the minimum size of the hugepage pool.
415
416See Documentation/vm/hugetlbpage.txt
417
418==============================================================
419
420nr_overcommit_hugepages
421
422Change the maximum size of the hugepage pool. The maximum is
423nr_hugepages + nr_overcommit_hugepages.
424
425See Documentation/vm/hugetlbpage.txt
426
427==============================================================
428
429nr_pdflush_threads
430
431The current number of pdflush threads. This value is read-only.
432The value changes according to the number of dirty pages in the system.
433
434When necessary, additional pdflush threads are created, one per second, up to
435nr_pdflush_threads_max.
436
437==============================================================
438
439nr_trim_pages
440
441This is available only on NOMMU kernels.
442
443This value adjusts the excess page trimming behaviour of power-of-2 aligned
444NOMMU mmap allocations.
445
446A value of 0 disables trimming of allocations entirely, while a value of 1
447trims excess pages aggressively. Any value >= 1 acts as the watermark where
448trimming of allocations is initiated.
449
450The default value is 1.
451
452See Documentation/nommu-mmap.txt for more information.
453
454==============================================================
455
456numa_zonelist_order
457
458This sysctl is only for NUMA.
459'where the memory is allocated from' is controlled by zonelists.
460(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
461 you may be able to read ZONE_DMA as ZONE_DMA32...)
462
463In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
464ZONE_NORMAL -> ZONE_DMA
465This means that a memory allocation request for GFP_KERNEL will
466get memory from ZONE_DMA only when ZONE_NORMAL is not available.
467
468In NUMA case, you can think of following 2 types of order.
469Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL
470
471(A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
472(B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
473
474Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
475will be used before ZONE_NORMAL exhaustion. This increases possibility of
476out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
477
478Type(B) cannot offer the best locality but is more robust against OOM of
479the DMA zone.
480
481Type(A) is called as "Node" order. Type (B) is "Zone" order.
482
483"Node order" orders the zonelists by node, then by zone within each node.
484Specify "[Nn]ode" for zone order
485
486"Zone Order" orders the zonelists by zone type, then by node within each
487zone. Specify "[Zz]one"for zode order.
488
489Specify "[Dd]efault" to request automatic configuration. Autoconfiguration
490will select "node" order in following case.
491(1) if the DMA zone does not exist or
492(2) if the DMA zone comprises greater than 50% of the available memory or
493(3) if any node's DMA zone comprises greater than 60% of its local memory and
494    the amount of local memory is big enough.
495
496Otherwise, "zone" order will be selected. Default order is recommended unless
497this is causing problems for your system/application.
498
499==============================================================
500
501oom_dump_tasks
502
503Enables a system-wide task dump (excluding kernel threads) to be
504produced when the kernel performs an OOM-killing and includes such
505information as pid, uid, tgid, vm size, rss, cpu, oom_adj score, and
506name. This is helpful to determine why the OOM killer was invoked
507and to identify the rogue task that caused it.
508
509If this is set to zero, this information is suppressed. On very
510large systems with thousands of tasks it may not be feasible to dump
511the memory state information for each one. Such systems should not
512be forced to incur a performance penalty in OOM conditions when the
513information may not be desired.
514
515If this is set to non-zero, this information is shown whenever the
516OOM killer actually kills a memory-hogging task.
517
518The default value is 1 (enabled).
519
520==============================================================
521
522oom_kill_allocating_task
523
524This enables or disables killing the OOM-triggering task in
525out-of-memory situations.
526
527If this is set to zero, the OOM killer will scan through the entire
528tasklist and select a task based on heuristics to kill. This normally
529selects a rogue memory-hogging task that frees up a large amount of
530memory when killed.
531
532If this is set to non-zero, the OOM killer simply kills the task that
533triggered the out-of-memory condition. This avoids the expensive
534tasklist scan.
535
536If panic_on_oom is selected, it takes precedence over whatever value
537is used in oom_kill_allocating_task.
538
539The default value is 0.
540
541==============================================================
542
543overcommit_memory:
544
545This value contains a flag that enables memory overcommitment.
546
547When this flag is 0, the kernel attempts to estimate the amount
548of free memory left when userspace requests more memory.
549
550When this flag is 1, the kernel pretends there is always enough
551memory until it actually runs out.
552
553When this flag is 2, the kernel uses a "never overcommit"
554policy that attempts to prevent any overcommit of memory.
555
556This feature can be very useful because there are a lot of
557programs that malloc() huge amounts of memory "just-in-case"
558and don't use much of it.
559
560The default value is 0.
561
562See Documentation/vm/overcommit-accounting and
563security/commoncap.c::cap_vm_enough_memory() for more information.
564
565==============================================================
566
567overcommit_ratio:
568
569When overcommit_memory is set to 2, the committed address
570space is not permitted to exceed swap plus this percentage
571of physical RAM. See above.
572
573==============================================================
574
575page-cluster
576
577page-cluster controls the number of pages which are written to swap in
578a single attempt. The swap I/O size.
579
580It is a logarithmic value - setting it to zero means "1 page", setting
581it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
582
583The default value is three (eight pages at a time). There may be some
584small benefits in tuning this to a different value if your workload is
585swap-intensive.
586
587=============================================================
588
589panic_on_oom
590
591This enables or disables panic on out-of-memory feature.
592
593If this is set to 0, the kernel will kill some rogue process,
594called oom_killer. Usually, oom_killer can kill rogue processes and
595system will survive.
596
597If this is set to 1, the kernel panics when out-of-memory happens.
598However, if a process limits using nodes by mempolicy/cpusets,
599and those nodes become memory exhaustion status, one process
600may be killed by oom-killer. No panic occurs in this case.
601Because other nodes' memory may be free. This means system total status
602may be not fatal yet.
603
604If this is set to 2, the kernel panics compulsorily even on the
605above-mentioned. Even oom happens under memory cgroup, the whole
606system panics.
607
608The default value is 0.
6091 and 2 are for failover of clustering. Please select either
610according to your policy of failover.
611panic_on_oom=2+kdump gives you very strong tool to investigate
612why oom happens. You can get snapshot.
613
614=============================================================
615
616percpu_pagelist_fraction
617
618This is the fraction of pages at most (high mark pcp->high) in each zone that
619are allocated for each per cpu page list. The min value for this is 8. It
620means that we don't allow more than 1/8th of pages in each zone to be
621allocated in any single per_cpu_pagelist. This entry only changes the value
622of hot per cpu pagelists. User can specify a number like 100 to allocate
6231/100th of each zone to each per cpu page list.
624
625The batch value of each per cpu pagelist is also updated as a result. It is
626set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8)
627
628The initial value is zero. Kernel does not use this value at boot time to set
629the high water marks for each per cpu page list.
630
631==============================================================
632
633stat_interval
634
635The time interval between which vm statistics are updated. The default
636is 1 second.
637
638==============================================================
639
640swappiness
641
642This control is used to define how aggressive the kernel will swap
643memory pages. Higher values will increase agressiveness, lower values
644decrease the amount of swap.
645
646The default value is 60.
647
648==============================================================
649
650vfs_cache_pressure
651------------------
652
653Controls the tendency of the kernel to reclaim the memory which is used for
654caching of directory and inode objects.
655
656At the default value of vfs_cache_pressure=100 the kernel will attempt to
657reclaim dentries and inodes at a "fair" rate with respect to pagecache and
658swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer
659to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
660never reclaim dentries and inodes due to memory pressure and this can easily
661lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
662causes the kernel to prefer to reclaim dentries and inodes.
663
664==============================================================
665
666zone_reclaim_mode:
667
668Zone_reclaim_mode allows someone to set more or less aggressive approaches to
669reclaim memory when a zone runs out of memory. If it is set to zero then no
670zone reclaim occurs. Allocations will be satisfied from other zones / nodes
671in the system.
672
673This is value ORed together of
674
6751 = Zone reclaim on
6762 = Zone reclaim writes dirty pages out
6774 = Zone reclaim swaps pages
678
679zone_reclaim_mode is set during bootup to 1 if it is determined that pages
680from remote zones will cause a measurable performance reduction. The
681page allocator will then reclaim easily reusable pages (those page
682cache pages that are currently not used) before allocating off node pages.
683
684It may be beneficial to switch off zone reclaim if the system is
685used for a file server and all of memory should be used for caching files
686from disk. In that case the caching effect is more important than
687data locality.
688
689Allowing zone reclaim to write out pages stops processes that are
690writing large amounts of data from dirtying pages on other nodes. Zone
691reclaim will write out dirty pages if a zone fills up and so effectively
692throttle the process. This may decrease the performance of a single process
693since it cannot use all of system memory to buffer the outgoing writes
694anymore but it preserve the memory on other nodes so that the performance
695of other processes running on other nodes will not be affected.
696
697Allowing regular swap effectively restricts allocations to the local
698node unless explicitly overridden by memory policies or cpuset
699configurations.
700
701============ End of Document =================================
702

Archive Download this file



interactive