/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
 83If dirty_background_bytes is written, dirty_background_ratio becomes a function
 84of its value (dirty_background_bytes / the amount of dirtyable system memory).
 85
 86==============================================================
 87
 88dirty_background_ratio
 89
 90Contains, as a percentage of total system memory, the number of pages at which
 91the pdflush background writeback daemon will start writing out dirty data.
 92
 93==============================================================
 94
 95dirty_bytes
 96
 97Contains the amount of dirty memory at which a process generating disk writes
 98will itself start writeback.
 99
100If dirty_bytes is written, dirty_ratio becomes a function of its value
101(dirty_bytes / the amount of dirtyable system memory).
102
103Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
104value lower than this limit will be ignored and the old configuration will be
105retained.
106
107==============================================================
108
109dirty_expire_centisecs
110
111This tunable is used to define when dirty data is old enough to be eligible
112for writeout by the pdflush daemons.  It is expressed in 100'ths of a second.
113Data which has been dirty in-memory for longer than this interval will be
114written out next time a pdflush daemon wakes up.
115
116==============================================================
117
118dirty_ratio
119
120Contains, as a percentage of total system memory, the number of pages at which
121a process which is generating disk writes will itself start writing out dirty
122data.
123
124==============================================================
125
126dirty_writeback_centisecs
127
128The pdflush writeback daemons will periodically wake up and write `old' data
129out to disk.  This tunable expresses the interval between those wakeups, in
130100'ths of a second.
131
132Setting this to zero disables periodic writeback altogether.
133
134==============================================================
135
136drop_caches
137
138Writing to this will cause the kernel to drop clean caches, dentries and
139inodes from memory, causing that memory to become free.
140
141To free pagecache:
142	echo 1 > /proc/sys/vm/drop_caches
143To free dentries and inodes:
144	echo 2 > /proc/sys/vm/drop_caches
145To free pagecache, dentries and inodes:
146	echo 3 > /proc/sys/vm/drop_caches
147
148As this is a non-destructive operation and dirty objects are not freeable, the
149user should run `sync' first.
150
151==============================================================
152
153extfrag_threshold
154
155This parameter affects whether the kernel will compact memory or direct
156reclaim to satisfy a high-order allocation. /proc/extfrag_index shows what
157the fragmentation index for each order is in each zone in the system. Values
158tending towards 0 imply allocations would fail due to lack of memory,
159values towards 1000 imply failures are due to fragmentation and -1 implies
160that the allocation will succeed as long as watermarks are met.
161
162The kernel will not compact memory in a zone if the
163fragmentation index is <= extfrag_threshold. The default value is 500.
164
165==============================================================
166
167hugepages_treat_as_movable
168
169This parameter is only useful when kernelcore= is specified at boot time to
170create ZONE_MOVABLE for pages that may be reclaimed or migrated. Huge pages
171are not movable so are not normally allocated from ZONE_MOVABLE. A non-zero
172value written to hugepages_treat_as_movable allows huge pages to be allocated
173from ZONE_MOVABLE.
174
175Once enabled, the ZONE_MOVABLE is treated as an area of memory the huge
176pages pool can easily grow or shrink within. Assuming that applications are
177not running that mlock() a lot of memory, it is likely the huge pages pool
178can grow to the size of ZONE_MOVABLE by repeatedly entering the desired value
179into nr_hugepages and triggering page reclaim.
180
181==============================================================
182
183hugetlb_shm_group
184
185hugetlb_shm_group contains group id that is allowed to create SysV
186shared memory segment using hugetlb page.
187
188==============================================================
189
190laptop_mode
191
192laptop_mode is a knob that controls "laptop mode". All the things that are
193controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt.
194
195==============================================================
196
197legacy_va_layout
198
199If non-zero, this sysctl disables the new 32-bit mmap mmap layout - the kernel
200will use the legacy (2.4) layout for all processes.
201
202==============================================================
203
204lowmem_reserve_ratio
205
206For some specialised workloads on highmem machines it is dangerous for
207the kernel to allow process memory to be allocated from the "lowmem"
208zone.  This is because that memory could then be pinned via the mlock()
209system call, or by unavailability of swapspace.
210
211And on large highmem machines this lack of reclaimable lowmem memory
212can be fatal.
213
214So the Linux page allocator has a mechanism which prevents allocations
215which _could_ use highmem from using too much lowmem.  This means that
216a certain amount of lowmem is defended from the possibility of being
217captured into pinned user memory.
218
219(The same argument applies to the old 16 megabyte ISA DMA region.  This
220mechanism will also defend that region from allocations which could use
221highmem or lowmem).
222
223The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is
224in defending these lower zones.
225
226If you have a machine which uses highmem or ISA DMA and your
227applications are using mlock(), or if you are running with no swap then
228you probably should change the lowmem_reserve_ratio setting.
229
230The lowmem_reserve_ratio is an array. You can see them by reading this file.
231-
232% cat /proc/sys/vm/lowmem_reserve_ratio
233256     256     32
234-
235Note: # of this elements is one fewer than number of zones. Because the highest
236      zone's value is not necessary for following calculation.
237
238But, these values are not used directly. The kernel calculates # of protection
239pages for each zones from them. These are shown as array of protection pages
240in /proc/zoneinfo like followings. (This is an example of x86-64 box).
241Each zone has an array of protection pages like this.
242
243-
244Node 0, zone      DMA
245  pages free     1355
246        min      3
247        low      3
248        high     4
249	:
250	:
251    numa_other   0
252        protection: (0, 2004, 2004, 2004)
253	^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
254  pagesets
255    cpu: 0 pcp: 0
256        :
257-
258These protections are added to score to judge whether this zone should be used
259for page allocation or should be reclaimed.
260
261In this example, if normal pages (index=2) are required to this DMA zone and
262watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
263not be used because pages_free(1355) is smaller than watermark + protection[2]
264(4 + 2004 = 2008). If this protection value is 0, this zone would be used for
265normal page requirement. If requirement is DMA zone(index=0), protection[0]
266(=0) is used.
267
268zone[i]'s protection[j] is calculated by following expression.
269
270(i < j):
271  zone[i]->protection[j]
272  = (total sums of present_pages from zone[i+1] to zone[j] on the node)
273    / lowmem_reserve_ratio[i];
274(i = j):
275   (should not be protected. = 0;
276(i > j):
277   (not necessary, but looks 0)
278
279The default values of lowmem_reserve_ratio[i] are
280    256 (if zone[i] means DMA or DMA32 zone)
281    32  (others).
282As above expression, they are reciprocal number of ratio.
283256 means 1/256. # of protection pages becomes about "0.39%" of total present
284pages of higher zones on the node.
285
286If you would like to protect more pages, smaller values are effective.
287The minimum value is 1 (1/1 -> 100%).
288
289==============================================================
290
291max_map_count:
292
293This file contains the maximum number of memory map areas a process
294may have. Memory map areas are used as a side-effect of calling
295malloc, directly by mmap and mprotect, and also when loading shared
296libraries.
297
298While most applications need less than a thousand maps, certain
299programs, particularly malloc debuggers, may consume lots of them,
300e.g., up to one or two maps per allocation.
301
302The default value is 65536.
303
304=============================================================
305
306memory_failure_early_kill:
307
308Control how to kill processes when uncorrected memory error (typically
309a 2bit error in a memory module) is detected in the background by hardware
310that cannot be handled by the kernel. In some cases (like the page
311still having a valid copy on disk) the kernel will handle the failure
312transparently without affecting any applications. But if there is
313no other uptodate copy of the data it will kill to prevent any data
314corruptions from propagating.
315
3161: Kill all processes that have the corrupted and not reloadable page mapped
317as soon as the corruption is detected.  Note this is not supported
318for a few types of pages, like kernel internally allocated data or
319the swap cache, but works for the majority of user pages.
320
3210: Only unmap the corrupted page from all processes and only kill a process
322who tries to access it.
323
324The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
325handle this if they want to.
326
327This is only active on architectures/platforms with advanced machine
328check handling and depends on the hardware capabilities.
329
330Applications can override this setting individually with the PR_MCE_KILL prctl
331
332==============================================================
333
334memory_failure_recovery
335
336Enable memory failure recovery (when supported by the platform)
337
3381: Attempt recovery.
339
3400: Always panic on a memory failure.
341
342==============================================================
343
344min_free_kbytes:
345
346This is used to force the Linux VM to keep a minimum number
347of kilobytes free.  The VM uses this number to compute a
348watermark[WMARK_MIN] value for each lowmem zone in the system.
349Each lowmem zone gets a number of reserved free pages based
350proportionally on its size.
351
352Some minimal amount of memory is needed to satisfy PF_MEMALLOC
353allocations; if you set this to lower than 1024KB, your system will
354become subtly broken, and prone to deadlock under high loads.
355
356Setting this too high will OOM your machine instantly.
357
358=============================================================
359
360min_slab_ratio:
361
362This is available only on NUMA kernels.
363
364A percentage of the total pages in each zone.  On Zone reclaim
365(fallback from the local zone occurs) slabs will be reclaimed if more
366than this percentage of pages in a zone are reclaimable slab pages.
367This insures that the slab growth stays under control even in NUMA
368systems that rarely perform global reclaim.
369
370The default is 5 percent.
371
372Note that slab reclaim is triggered in a per zone / node fashion.
373The process of reclaiming slab memory is currently not node specific
374and may not be fast.
375
376=============================================================
377
378min_unmapped_ratio:
379
380This is available only on NUMA kernels.
381
382This is a percentage of the total pages in each zone. Zone reclaim will
383only occur if more than this percentage of pages are in a state that
384zone_reclaim_mode allows to be reclaimed.
385
386If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
387against all file-backed unmapped pages including swapcache pages and tmpfs
388files. Otherwise, only unmapped pages backed by normal files but not tmpfs
389files and similar are considered.
390
391The default is 1 percent.
392
393==============================================================
394
395mmap_min_addr
396
397This file indicates the amount of address space  which a user process will
398be restricted from mmapping.  Since kernel null dereference bugs could
399accidentally operate based on the information in the first couple of pages
400of memory userspace processes should not be allowed to write to them.  By
401default this value is set to 0 and no protections will be enforced by the
402security module.  Setting this value to something like 64k will allow the
403vast majority of applications to work correctly and provide defense in depth
404against future potential kernel bugs.
405
406==============================================================
407
408nr_hugepages
409
410Change the minimum size of the hugepage pool.
411
412See Documentation/vm/hugetlbpage.txt
413
414==============================================================
415
416nr_overcommit_hugepages
417
418Change the maximum size of the hugepage pool. The maximum is
419nr_hugepages + nr_overcommit_hugepages.
420
421See Documentation/vm/hugetlbpage.txt
422
423==============================================================
424
425nr_pdflush_threads
426
427The current number of pdflush threads.  This value is read-only.
428The value changes according to the number of dirty pages in the system.
429
430When necessary, additional pdflush threads are created, one per second, up to
431nr_pdflush_threads_max.
432
433==============================================================
434
435nr_trim_pages
436
437This is available only on NOMMU kernels.
438
439This value adjusts the excess page trimming behaviour of power-of-2 aligned
440NOMMU mmap allocations.
441
442A value of 0 disables trimming of allocations entirely, while a value of 1
443trims excess pages aggressively. Any value >= 1 acts as the watermark where
444trimming of allocations is initiated.
445
446The default value is 1.
447
448See Documentation/nommu-mmap.txt for more information.
449
450==============================================================
451
452numa_zonelist_order
453
454This sysctl is only for NUMA.
455'where the memory is allocated from' is controlled by zonelists.
456(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
457 you may be able to read ZONE_DMA as ZONE_DMA32...)
458
459In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
460ZONE_NORMAL -> ZONE_DMA
461This means that a memory allocation request for GFP_KERNEL will
462get memory from ZONE_DMA only when ZONE_NORMAL is not available.
463
464In NUMA case, you can think of following 2 types of order.
465Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL
466
467(A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
468(B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
469
470Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
471will be used before ZONE_NORMAL exhaustion. This increases possibility of
472out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
473
474Type(B) cannot offer the best locality but is more robust against OOM of
475the DMA zone.
476
477Type(A) is called as "Node" order. Type (B) is "Zone" order.
478
479"Node order" orders the zonelists by node, then by zone within each node.
480Specify "[Nn]ode" for zone order
481
482"Zone Order" orders the zonelists by zone type, then by node within each
483zone.  Specify "[Zz]one"for zode order.
484
485Specify "[Dd]efault" to request automatic configuration.  Autoconfiguration
486will select "node" order in following case.
487(1) if the DMA zone does not exist or
488(2) if the DMA zone comprises greater than 50% of the available memory or
489(3) if any node's DMA zone comprises greater than 60% of its local memory and
490    the amount of local memory is big enough.
491
492Otherwise, "zone" order will be selected. Default order is recommended unless
493this is causing problems for your system/application.
494
495==============================================================
496
497oom_dump_tasks
498
499Enables a system-wide task dump (excluding kernel threads) to be
500produced when the kernel performs an OOM-killing and includes such
501information as pid, uid, tgid, vm size, rss, cpu, oom_adj score, and
502name.  This is helpful to determine why the OOM killer was invoked
503and to identify the rogue task that caused it.
504
505If this is set to zero, this information is suppressed.  On very
506large systems with thousands of tasks it may not be feasible to dump
507the memory state information for each one.  Such systems should not
508be forced to incur a performance penalty in OOM conditions when the
509information may not be desired.
510
511If this is set to non-zero, this information is shown whenever the
512OOM killer actually kills a memory-hogging task.
513
514The default value is 0.
515
516==============================================================
517
518oom_kill_allocating_task
519
520This enables or disables killing the OOM-triggering task in
521out-of-memory situations.
522
523If this is set to zero, the OOM killer will scan through the entire
524tasklist and select a task based on heuristics to kill.  This normally
525selects a rogue memory-hogging task that frees up a large amount of
526memory when killed.
527
528If this is set to non-zero, the OOM killer simply kills the task that
529triggered the out-of-memory condition.  This avoids the expensive
530tasklist scan.
531
532If panic_on_oom is selected, it takes precedence over whatever value
533is used in oom_kill_allocating_task.
534
535The default value is 0.
536
537==============================================================
538
539overcommit_memory:
540
541This value contains a flag that enables memory overcommitment.
542
543When this flag is 0, the kernel attempts to estimate the amount
544of free memory left when userspace requests more memory.
545
546When this flag is 1, the kernel pretends there is always enough
547memory until it actually runs out.
548
549When this flag is 2, the kernel uses a "never overcommit"
550policy that attempts to prevent any overcommit of memory.
551
552This feature can be very useful because there are a lot of
553programs that malloc() huge amounts of memory "just-in-case"
554and don't use much of it.
555
556The default value is 0.
557
558See Documentation/vm/overcommit-accounting and
559security/commoncap.c::cap_vm_enough_memory() for more information.
560
561==============================================================
562
563overcommit_ratio:
564
565When overcommit_memory is set to 2, the committed address
566space is not permitted to exceed swap plus this percentage
567of physical RAM.  See above.
568
569==============================================================
570
571page-cluster
572
573page-cluster controls the number of pages which are written to swap in
574a single attempt.  The swap I/O size.
575
576It is a logarithmic value - setting it to zero means "1 page", setting
577it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
578
579The default value is three (eight pages at a time).  There may be some
580small benefits in tuning this to a different value if your workload is
581swap-intensive.
582
583=============================================================
584
585panic_on_oom
586
587This enables or disables panic on out-of-memory feature.
588
589If this is set to 0, the kernel will kill some rogue process,
590called oom_killer.  Usually, oom_killer can kill rogue processes and
591system will survive.
592
593If this is set to 1, the kernel panics when out-of-memory happens.
594However, if a process limits using nodes by mempolicy/cpusets,
595and those nodes become memory exhaustion status, one process
596may be killed by oom-killer. No panic occurs in this case.
597Because other nodes' memory may be free. This means system total status
598may be not fatal yet.
599
600If this is set to 2, the kernel panics compulsorily even on the
601above-mentioned. Even oom happens under memory cgroup, the whole
602system panics.
603
604The default value is 0.
6051 and 2 are for failover of clustering. Please select either
606according to your policy of failover.
607panic_on_oom=2+kdump gives you very strong tool to investigate
608why oom happens. You can get snapshot.
609
610=============================================================
611
612percpu_pagelist_fraction
613
614This is the fraction of pages at most (high mark pcp->high) in each zone that
615are allocated for each per cpu page list.  The min value for this is 8.  It
616means that we don't allow more than 1/8th of pages in each zone to be
617allocated in any single per_cpu_pagelist.  This entry only changes the value
618of hot per cpu pagelists.  User can specify a number like 100 to allocate
6191/100th of each zone to each per cpu page list.
620
621The batch value of each per cpu pagelist is also updated as a result.  It is
622set to pcp->high/4.  The upper limit of batch is (PAGE_SHIFT * 8)
623
624The initial value is zero.  Kernel does not use this value at boot time to set
625the high water marks for each per cpu page list.
626
627==============================================================
628
629stat_interval
630
631The time interval between which vm statistics are updated.  The default
632is 1 second.
633
634==============================================================
635
636swappiness
637
638This control is used to define how aggressive the kernel will swap
639memory pages.  Higher values will increase agressiveness, lower values
640decrease the amount of swap.
641
642The default value is 60.
643
644==============================================================
645
646vfs_cache_pressure
647------------------
648
649Controls the tendency of the kernel to reclaim the memory which is used for
650caching of directory and inode objects.
651
652At the default value of vfs_cache_pressure=100 the kernel will attempt to
653reclaim dentries and inodes at a "fair" rate with respect to pagecache and
654swapcache reclaim.  Decreasing vfs_cache_pressure causes the kernel to prefer
655to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
656never reclaim dentries and inodes due to memory pressure and this can easily
657lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
658causes the kernel to prefer to reclaim dentries and inodes.
659
660==============================================================
661
662zone_reclaim_mode:
663
664Zone_reclaim_mode allows someone to set more or less aggressive approaches to
665reclaim memory when a zone runs out of memory. If it is set to zero then no
666zone reclaim occurs. Allocations will be satisfied from other zones / nodes
667in the system.
668
669This is value ORed together of
670
6711	= Zone reclaim on
6722	= Zone reclaim writes dirty pages out
6734	= Zone reclaim swaps pages
674
675zone_reclaim_mode is set during bootup to 1 if it is determined that pages
676from remote zones will cause a measurable performance reduction. The
677page allocator will then reclaim easily reusable pages (those page
678cache pages that are currently not used) before allocating off node pages.
679
680It may be beneficial to switch off zone reclaim if the system is
681used for a file server and all of memory should be used for caching files
682from disk. In that case the caching effect is more important than
683data locality.
684
685Allowing zone reclaim to write out pages stops processes that are
686writing large amounts of data from dirtying pages on other nodes. Zone
687reclaim will write out dirty pages if a zone fills up and so effectively
688throttle the process. This may decrease the performance of a single process
689since it cannot use all of system memory to buffer the outgoing writes
690anymore but it preserve the memory on other nodes so that the performance
691of other processes running on other nodes will not be affected.
692
693Allowing regular swap effectively restricts allocations to the local
694node unless explicitly overridden by memory policies or cpuset
695configurations.
696
697============ End of Document =================================