1===============================
2Documentation for /proc/sys/vm/
3===============================
4
5kernel version 2.6.29
6
7Copyright (c) 1998, 1999,  Rik van Riel <riel@nl.linux.org>
8
9Copyright (c) 2008         Peter W. Morreale <pmorreale@novell.com>
10
11For general info and legal blurb, please look in index.rst.
12
13------------------------------------------------------------------------------
14
15This file contains the documentation for the sysctl files in
16/proc/sys/vm and is valid for Linux kernel version 2.6.29.
17
18The files in this directory can be used to tune the operation
19of the virtual memory (VM) subsystem of the Linux kernel and
20the writeout of dirty data to disk.
21
22Default values and initialization routines for most of these
23files can be found in mm/swap.c.
24
25Currently, these files are in /proc/sys/vm:
26
27- admin_reserve_kbytes
28- compact_memory
29- compaction_proactiveness
30- compact_unevictable_allowed
31- dirty_background_bytes
32- dirty_background_ratio
33- dirty_bytes
34- dirty_expire_centisecs
35- dirty_ratio
36- dirtytime_expire_seconds
37- dirty_writeback_centisecs
38- drop_caches
39- extfrag_threshold
40- highmem_is_dirtyable
41- hugetlb_shm_group
42- laptop_mode
43- legacy_va_layout
44- lowmem_reserve_ratio
45- max_map_count
46- memory_failure_early_kill
47- memory_failure_recovery
48- min_free_kbytes
49- min_slab_ratio
50- min_unmapped_ratio
51- mmap_min_addr
52- mmap_rnd_bits
53- mmap_rnd_compat_bits
54- nr_hugepages
55- nr_hugepages_mempolicy
56- nr_overcommit_hugepages
57- nr_trim_pages         (only if CONFIG_MMU=n)
58- numa_zonelist_order
59- oom_dump_tasks
60- oom_kill_allocating_task
61- overcommit_kbytes
62- overcommit_memory
63- overcommit_ratio
64- page-cluster
65- page_lock_unfairness
66- panic_on_oom
67- percpu_pagelist_high_fraction
68- stat_interval
69- stat_refresh
70- numa_stat
71- swappiness
72- unprivileged_userfaultfd
73- user_reserve_kbytes
74- vfs_cache_pressure
75- watermark_boost_factor
76- watermark_scale_factor
77- zone_reclaim_mode
78
79
80admin_reserve_kbytes
81====================
82
83The amount of free memory in the system that should be reserved for users
84with the capability cap_sys_admin.
85
86admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
87
88That should provide enough for the admin to log in and kill a process,
89if necessary, under the default overcommit 'guess' mode.
90
91Systems running under overcommit 'never' should increase this to account
92for the full Virtual Memory Size of programs used to recover. Otherwise,
93root may not be able to log in to recover the system.
94
95How do you calculate a minimum useful reserve?
96
97sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
98
99For overcommit 'guess', we can sum resident set sizes (RSS).
100On x86_64 this is about 8MB.
101
102For overcommit 'never', we can take the max of their virtual sizes (VSZ)
103and add the sum of their RSS.
104On x86_64 this is about 128MB.
105
106Changing this takes effect whenever an application requests memory.
107
108
109compact_memory
110==============
111
112Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
113all zones are compacted such that free memory is available in contiguous
114blocks where possible. This can be important for example in the allocation of
115huge pages although processes will also directly compact memory as required.
116
117compaction_proactiveness
118========================
119
120This tunable takes a value in the range [0, 100] with a default value of
12120. This tunable determines how aggressively compaction is done in the
122background. Write of a non zero value to this tunable will immediately
123trigger the proactive compaction. Setting it to 0 disables proactive compaction.
124
125Note that compaction has a non-trivial system-wide impact as pages
126belonging to different processes are moved around, which could also lead
127to latency spikes in unsuspecting applications. The kernel employs
128various heuristics to avoid wasting CPU cycles if it detects that
129proactive compaction is not being effective.
130
131Be careful when setting it to extreme values like 100, as that may
132cause excessive background compaction activity.
133
134compact_unevictable_allowed
135===========================
136
137Available only when CONFIG_COMPACTION is set. When set to 1, compaction is
138allowed to examine the unevictable lru (mlocked pages) for pages to compact.
139This should be used on systems where stalls for minor page faults are an
140acceptable trade for large contiguous free memory.  Set to 0 to prevent
141compaction from moving pages that are unevictable.  Default value is 1.
142On CONFIG_PREEMPT_RT the default value is 0 in order to avoid a page fault, due
143to compaction, which would block the task from becoming active until the fault
144is resolved.
145
146
147dirty_background_bytes
148======================
149
150Contains the amount of dirty memory at which the background kernel
151flusher threads will start writeback.
152
153Note:
154  dirty_background_bytes is the counterpart of dirty_background_ratio. Only
155  one of them may be specified at a time. When one sysctl is written it is
156  immediately taken into account to evaluate the dirty memory limits and the
157  other appears as 0 when read.
158
159
160dirty_background_ratio
161======================
162
163Contains, as a percentage of total available memory that contains free pages
164and reclaimable pages, the number of pages at which the background kernel
165flusher threads will start writing out dirty data.
166
167The total available memory is not equal to total system memory.
168
169
170dirty_bytes
171===========
172
173Contains the amount of dirty memory at which a process generating disk writes
174will itself start writeback.
175
176Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
177specified at a time. When one sysctl is written it is immediately taken into
178account to evaluate the dirty memory limits and the other appears as 0 when
179read.
180
181Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
182value lower than this limit will be ignored and the old configuration will be
183retained.
184
185
186dirty_expire_centisecs
187======================
188
189This tunable is used to define when dirty data is old enough to be eligible
190for writeout by the kernel flusher threads.  It is expressed in 100'ths
191of a second.  Data which has been dirty in-memory for longer than this
192interval will be written out next time a flusher thread wakes up.
193
194
195dirty_ratio
196===========
197
198Contains, as a percentage of total available memory that contains free pages
199and reclaimable pages, the number of pages at which a process which is
200generating disk writes will itself start writing out dirty data.
201
202The total available memory is not equal to total system memory.
203
204
205dirtytime_expire_seconds
206========================
207
208When a lazytime inode is constantly having its pages dirtied, the inode with
209an updated timestamp will never get chance to be written out.  And, if the
210only thing that has happened on the file system is a dirtytime inode caused
211by an atime update, a worker will be scheduled to make sure that inode
212eventually gets pushed out to disk.  This tunable is used to define when dirty
213inode is old enough to be eligible for writeback by the kernel flusher threads.
214And, it is also used as the interval to wakeup dirtytime_writeback thread.
215
216
217dirty_writeback_centisecs
218=========================
219
220The kernel flusher threads will periodically wake up and write `old` data
221out to disk.  This tunable expresses the interval between those wakeups, in
222100'ths of a second.
223
224Setting this to zero disables periodic writeback altogether.
225
226
227drop_caches
228===========
229
230Writing to this will cause the kernel to drop clean caches, as well as
231reclaimable slab objects like dentries and inodes.  Once dropped, their
232memory becomes free.
233
234To free pagecache::
235
236	echo 1 > /proc/sys/vm/drop_caches
237
238To free reclaimable slab objects (includes dentries and inodes)::
239
240	echo 2 > /proc/sys/vm/drop_caches
241
242To free slab objects and pagecache::
243
244	echo 3 > /proc/sys/vm/drop_caches
245
246This is a non-destructive operation and will not free any dirty objects.
247To increase the number of objects freed by this operation, the user may run
248`sync` prior to writing to /proc/sys/vm/drop_caches.  This will minimize the
249number of dirty objects on the system and create more candidates to be
250dropped.
251
252This file is not a means to control the growth of the various kernel caches
253(inodes, dentries, pagecache, etc...)  These objects are automatically
254reclaimed by the kernel when memory is needed elsewhere on the system.
255
256Use of this file can cause performance problems.  Since it discards cached
257objects, it may cost a significant amount of I/O and CPU to recreate the
258dropped objects, especially if they were under heavy use.  Because of this,
259use outside of a testing or debugging environment is not recommended.
260
261You may see informational messages in your kernel log when this file is
262used::
263
264	cat (1234): drop_caches: 3
265
266These are informational only.  They do not mean that anything is wrong
267with your system.  To disable them, echo 4 (bit 2) into drop_caches.
268
269
270extfrag_threshold
271=================
272
273This parameter affects whether the kernel will compact memory or direct
274reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in
275debugfs shows what the fragmentation index for each order is in each zone in
276the system. Values tending towards 0 imply allocations would fail due to lack
277of memory, values towards 1000 imply failures are due to fragmentation and -1
278implies that the allocation will succeed as long as watermarks are met.
279
280The kernel will not compact memory in a zone if the
281fragmentation index is <= extfrag_threshold. The default value is 500.
282
283
284highmem_is_dirtyable
285====================
286
287Available only for systems with CONFIG_HIGHMEM enabled (32b systems).
288
289This parameter controls whether the high memory is considered for dirty
290writers throttling.  This is not the case by default which means that
291only the amount of memory directly visible/usable by the kernel can
292be dirtied. As a result, on systems with a large amount of memory and
293lowmem basically depleted writers might be throttled too early and
294streaming writes can get very slow.
295
296Changing the value to non zero would allow more memory to be dirtied
297and thus allow writers to write more data which can be flushed to the
298storage more effectively. Note this also comes with a risk of pre-mature
299OOM killer because some writers (e.g. direct block device writes) can
300only use the low memory and they can fill it up with dirty data without
301any throttling.
302
303
304hugetlb_shm_group
305=================
306
307hugetlb_shm_group contains group id that is allowed to create SysV
308shared memory segment using hugetlb page.
309
310
311laptop_mode
312===========
313
314laptop_mode is a knob that controls "laptop mode". All the things that are
315controlled by this knob are discussed in Documentation/admin-guide/laptops/laptop-mode.rst.
316
317
318legacy_va_layout
319================
320
321If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
322will use the legacy (2.4) layout for all processes.
323
324
325lowmem_reserve_ratio
326====================
327
328For some specialised workloads on highmem machines it is dangerous for
329the kernel to allow process memory to be allocated from the "lowmem"
330zone.  This is because that memory could then be pinned via the mlock()
331system call, or by unavailability of swapspace.
332
333And on large highmem machines this lack of reclaimable lowmem memory
334can be fatal.
335
336So the Linux page allocator has a mechanism which prevents allocations
337which *could* use highmem from using too much lowmem.  This means that
338a certain amount of lowmem is defended from the possibility of being
339captured into pinned user memory.
340
341(The same argument applies to the old 16 megabyte ISA DMA region.  This
342mechanism will also defend that region from allocations which could use
343highmem or lowmem).
344
345The `lowmem_reserve_ratio` tunable determines how aggressive the kernel is
346in defending these lower zones.
347
348If you have a machine which uses highmem or ISA DMA and your
349applications are using mlock(), or if you are running with no swap then
350you probably should change the lowmem_reserve_ratio setting.
351
352The lowmem_reserve_ratio is an array. You can see them by reading this file::
353
354	% cat /proc/sys/vm/lowmem_reserve_ratio
355	256     256     32
356
357But, these values are not used directly. The kernel calculates # of protection
358pages for each zones from them. These are shown as array of protection pages
359in /proc/zoneinfo like followings. (This is an example of x86-64 box).
360Each zone has an array of protection pages like this::
361
362  Node 0, zone      DMA
363    pages free     1355
364          min      3
365          low      3
366          high     4
367	:
368	:
369      numa_other   0
370          protection: (0, 2004, 2004, 2004)
371	^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
372    pagesets
373      cpu: 0 pcp: 0
374          :
375
376These protections are added to score to judge whether this zone should be used
377for page allocation or should be reclaimed.
378
379In this example, if normal pages (index=2) are required to this DMA zone and
380watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
381not be used because pages_free(1355) is smaller than watermark + protection[2]
382(4 + 2004 = 2008). If this protection value is 0, this zone would be used for
383normal page requirement. If requirement is DMA zone(index=0), protection[0]
384(=0) is used.
385
386zone[i]'s protection[j] is calculated by following expression::
387
388  (i < j):
389    zone[i]->protection[j]
390    = (total sums of managed_pages from zone[i+1] to zone[j] on the node)
391      / lowmem_reserve_ratio[i];
392  (i = j):
393     (should not be protected. = 0;
394  (i > j):
395     (not necessary, but looks 0)
396
397The default values of lowmem_reserve_ratio[i] are
398
399    === ====================================
400    256 (if zone[i] means DMA or DMA32 zone)
401    32  (others)
402    === ====================================
403
404As above expression, they are reciprocal number of ratio.
405256 means 1/256. # of protection pages becomes about "0.39%" of total managed
406pages of higher zones on the node.
407
408If you would like to protect more pages, smaller values are effective.
409The minimum value is 1 (1/1 -> 100%). The value less than 1 completely
410disables protection of the pages.
411
412
413max_map_count:
414==============
415
416This file contains the maximum number of memory map areas a process
417may have. Memory map areas are used as a side-effect of calling
418malloc, directly by mmap, mprotect, and madvise, and also when loading
419shared libraries.
420
421While most applications need less than a thousand maps, certain
422programs, particularly malloc debuggers, may consume lots of them,
423e.g., up to one or two maps per allocation.
424
425The default value is 65530.
426
427
428memory_failure_early_kill:
429==========================
430
431Control how to kill processes when uncorrected memory error (typically
432a 2bit error in a memory module) is detected in the background by hardware
433that cannot be handled by the kernel. In some cases (like the page
434still having a valid copy on disk) the kernel will handle the failure
435transparently without affecting any applications. But if there is
436no other uptodate copy of the data it will kill to prevent any data
437corruptions from propagating.
438
4391: Kill all processes that have the corrupted and not reloadable page mapped
440as soon as the corruption is detected.  Note this is not supported
441for a few types of pages, like kernel internally allocated data or
442the swap cache, but works for the majority of user pages.
443
4440: Only unmap the corrupted page from all processes and only kill a process
445who tries to access it.
446
447The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
448handle this if they want to.
449
450This is only active on architectures/platforms with advanced machine
451check handling and depends on the hardware capabilities.
452
453Applications can override this setting individually with the PR_MCE_KILL prctl
454
455
456memory_failure_recovery
457=======================
458
459Enable memory failure recovery (when supported by the platform)
460
4611: Attempt recovery.
462
4630: Always panic on a memory failure.
464
465
466min_free_kbytes
467===============
468
469This is used to force the Linux VM to keep a minimum number
470of kilobytes free.  The VM uses this number to compute a
471watermark[WMARK_MIN] value for each lowmem zone in the system.
472Each lowmem zone gets a number of reserved free pages based
473proportionally on its size.
474
475Some minimal amount of memory is needed to satisfy PF_MEMALLOC
476allocations; if you set this to lower than 1024KB, your system will
477become subtly broken, and prone to deadlock under high loads.
478
479Setting this too high will OOM your machine instantly.
480
481
482min_slab_ratio
483==============
484
485This is available only on NUMA kernels.
486
487A percentage of the total pages in each zone.  On Zone reclaim
488(fallback from the local zone occurs) slabs will be reclaimed if more
489than this percentage of pages in a zone are reclaimable slab pages.
490This insures that the slab growth stays under control even in NUMA
491systems that rarely perform global reclaim.
492
493The default is 5 percent.
494
495Note that slab reclaim is triggered in a per zone / node fashion.
496The process of reclaiming slab memory is currently not node specific
497and may not be fast.
498
499
500min_unmapped_ratio
501==================
502
503This is available only on NUMA kernels.
504
505This is a percentage of the total pages in each zone. Zone reclaim will
506only occur if more than this percentage of pages are in a state that
507zone_reclaim_mode allows to be reclaimed.
508
509If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
510against all file-backed unmapped pages including swapcache pages and tmpfs
511files. Otherwise, only unmapped pages backed by normal files but not tmpfs
512files and similar are considered.
513
514The default is 1 percent.
515
516
517mmap_min_addr
518=============
519
520This file indicates the amount of address space  which a user process will
521be restricted from mmapping.  Since kernel null dereference bugs could
522accidentally operate based on the information in the first couple of pages
523of memory userspace processes should not be allowed to write to them.  By
524default this value is set to 0 and no protections will be enforced by the
525security module.  Setting this value to something like 64k will allow the
526vast majority of applications to work correctly and provide defense in depth
527against future potential kernel bugs.
528
529
530mmap_rnd_bits
531=============
532
533This value can be used to select the number of bits to use to
534determine the random offset to the base address of vma regions
535resulting from mmap allocations on architectures which support
536tuning address space randomization.  This value will be bounded
537by the architecture's minimum and maximum supported values.
538
539This value can be changed after boot using the
540/proc/sys/vm/mmap_rnd_bits tunable
541
542
543mmap_rnd_compat_bits
544====================
545
546This value can be used to select the number of bits to use to
547determine the random offset to the base address of vma regions
548resulting from mmap allocations for applications run in
549compatibility mode on architectures which support tuning address
550space randomization.  This value will be bounded by the
551architecture's minimum and maximum supported values.
552
553This value can be changed after boot using the
554/proc/sys/vm/mmap_rnd_compat_bits tunable
555
556
557nr_hugepages
558============
559
560Change the minimum size of the hugepage pool.
561
562See Documentation/admin-guide/mm/hugetlbpage.rst
563
564
565hugetlb_optimize_vmemmap
566========================
567
568This knob is not available when memory_hotplug.memmap_on_memory (kernel parameter)
569is configured or the size of 'struct page' (a structure defined in
570include/linux/mm_types.h) is not power of two (an unusual system config could
571result in this).
572
573Enable (set to 1) or disable (set to 0) the feature of optimizing vmemmap pages
574associated with each HugeTLB page.
575
576Once enabled, the vmemmap pages of subsequent allocation of HugeTLB pages from
577buddy allocator will be optimized (7 pages per 2MB HugeTLB page and 4095 pages
578per 1GB HugeTLB page), whereas already allocated HugeTLB pages will not be
579optimized.  When those optimized HugeTLB pages are freed from the HugeTLB pool
580to the buddy allocator, the vmemmap pages representing that range needs to be
581remapped again and the vmemmap pages discarded earlier need to be rellocated
582again.  If your use case is that HugeTLB pages are allocated 'on the fly' (e.g.
583never explicitly allocating HugeTLB pages with 'nr_hugepages' but only set
584'nr_overcommit_hugepages', those overcommitted HugeTLB pages are allocated 'on
585the fly') instead of being pulled from the HugeTLB pool, you should weigh the
586benefits of memory savings against the more overhead (~2x slower than before)
587of allocation or freeing HugeTLB pages between the HugeTLB pool and the buddy
588allocator.  Another behavior to note is that if the system is under heavy memory
589pressure, it could prevent the user from freeing HugeTLB pages from the HugeTLB
590pool to the buddy allocator since the allocation of vmemmap pages could be
591failed, you have to retry later if your system encounter this situation.
592
593Once disabled, the vmemmap pages of subsequent allocation of HugeTLB pages from
594buddy allocator will not be optimized meaning the extra overhead at allocation
595time from buddy allocator disappears, whereas already optimized HugeTLB pages
596will not be affected.  If you want to make sure there are no optimized HugeTLB
597pages, you can set "nr_hugepages" to 0 first and then disable this.  Note that
598writing 0 to nr_hugepages will make any "in use" HugeTLB pages become surplus
599pages.  So, those surplus pages are still optimized until they are no longer
600in use.  You would need to wait for those surplus pages to be released before
601there are no optimized pages in the system.
602
603
604nr_hugepages_mempolicy
605======================
606
607Change the size of the hugepage pool at run-time on a specific
608set of NUMA nodes.
609
610See Documentation/admin-guide/mm/hugetlbpage.rst
611
612
613nr_overcommit_hugepages
614=======================
615
616Change the maximum size of the hugepage pool. The maximum is
617nr_hugepages + nr_overcommit_hugepages.
618
619See Documentation/admin-guide/mm/hugetlbpage.rst
620
621
622nr_trim_pages
623=============
624
625This is available only on NOMMU kernels.
626
627This value adjusts the excess page trimming behaviour of power-of-2 aligned
628NOMMU mmap allocations.
629
630A value of 0 disables trimming of allocations entirely, while a value of 1
631trims excess pages aggressively. Any value >= 1 acts as the watermark where
632trimming of allocations is initiated.
633
634The default value is 1.
635
636See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
637
638
639numa_zonelist_order
640===================
641
642This sysctl is only for NUMA and it is deprecated. Anything but
643Node order will fail!
644
645'where the memory is allocated from' is controlled by zonelists.
646
647(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
648you may be able to read ZONE_DMA as ZONE_DMA32...)
649
650In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
651ZONE_NORMAL -> ZONE_DMA
652This means that a memory allocation request for GFP_KERNEL will
653get memory from ZONE_DMA only when ZONE_NORMAL is not available.
654
655In NUMA case, you can think of following 2 types of order.
656Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL::
657
658  (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
659  (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
660
661Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
662will be used before ZONE_NORMAL exhaustion. This increases possibility of
663out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
664
665Type(B) cannot offer the best locality but is more robust against OOM of
666the DMA zone.
667
668Type(A) is called as "Node" order. Type (B) is "Zone" order.
669
670"Node order" orders the zonelists by node, then by zone within each node.
671Specify "[Nn]ode" for node order
672
673"Zone Order" orders the zonelists by zone type, then by node within each
674zone.  Specify "[Zz]one" for zone order.
675
676Specify "[Dd]efault" to request automatic configuration.
677
678On 32-bit, the Normal zone needs to be preserved for allocations accessible
679by the kernel, so "zone" order will be selected.
680
681On 64-bit, devices that require DMA32/DMA are relatively rare, so "node"
682order will be selected.
683
684Default order is recommended unless this is causing problems for your
685system/application.
686
687
688oom_dump_tasks
689==============
690
691Enables a system-wide task dump (excluding kernel threads) to be produced
692when the kernel performs an OOM-killing and includes such information as
693pid, uid, tgid, vm size, rss, pgtables_bytes, swapents, oom_score_adj
694score, and name.  This is helpful to determine why the OOM killer was
695invoked, to identify the rogue task that caused it, and to determine why
696the OOM killer chose the task it did to kill.
697
698If this is set to zero, this information is suppressed.  On very
699large systems with thousands of tasks it may not be feasible to dump
700the memory state information for each one.  Such systems should not
701be forced to incur a performance penalty in OOM conditions when the
702information may not be desired.
703
704If this is set to non-zero, this information is shown whenever the
705OOM killer actually kills a memory-hogging task.
706
707The default value is 1 (enabled).
708
709
710oom_kill_allocating_task
711========================
712
713This enables or disables killing the OOM-triggering task in
714out-of-memory situations.
715
716If this is set to zero, the OOM killer will scan through the entire
717tasklist and select a task based on heuristics to kill.  This normally
718selects a rogue memory-hogging task that frees up a large amount of
719memory when killed.
720
721If this is set to non-zero, the OOM killer simply kills the task that
722triggered the out-of-memory condition.  This avoids the expensive
723tasklist scan.
724
725If panic_on_oom is selected, it takes precedence over whatever value
726is used in oom_kill_allocating_task.
727
728The default value is 0.
729
730
731overcommit_kbytes
732=================
733
734When overcommit_memory is set to 2, the committed address space is not
735permitted to exceed swap plus this amount of physical RAM. See below.
736
737Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one
738of them may be specified at a time. Setting one disables the other (which
739then appears as 0 when read).
740
741
742overcommit_memory
743=================
744
745This value contains a flag that enables memory overcommitment.
746
747When this flag is 0, the kernel attempts to estimate the amount
748of free memory left when userspace requests more memory.
749
750When this flag is 1, the kernel pretends there is always enough
751memory until it actually runs out.
752
753When this flag is 2, the kernel uses a "never overcommit"
754policy that attempts to prevent any overcommit of memory.
755Note that user_reserve_kbytes affects this policy.
756
757This feature can be very useful because there are a lot of
758programs that malloc() huge amounts of memory "just-in-case"
759and don't use much of it.
760
761The default value is 0.
762
763See Documentation/vm/overcommit-accounting.rst and
764mm/util.c::__vm_enough_memory() for more information.
765
766
767overcommit_ratio
768================
769
770When overcommit_memory is set to 2, the committed address
771space is not permitted to exceed swap plus this percentage
772of physical RAM.  See above.
773
774
775page-cluster
776============
777
778page-cluster controls the number of pages up to which consecutive pages
779are read in from swap in a single attempt. This is the swap counterpart
780to page cache readahead.
781The mentioned consecutivity is not in terms of virtual/physical addresses,
782but consecutive on swap space - that means they were swapped out together.
783
784It is a logarithmic value - setting it to zero means "1 page", setting
785it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
786Zero disables swap readahead completely.
787
788The default value is three (eight pages at a time).  There may be some
789small benefits in tuning this to a different value if your workload is
790swap-intensive.
791
792Lower values mean lower latencies for initial faults, but at the same time
793extra faults and I/O delays for following faults if they would have been part of
794that consecutive pages readahead would have brought in.
795
796
797page_lock_unfairness
798====================
799
800This value determines the number of times that the page lock can be
801stolen from under a waiter. After the lock is stolen the number of times
802specified in this file (default is 5), the "fair lock handoff" semantics
803will apply, and the waiter will only be awakened if the lock can be taken.
804
805panic_on_oom
806============
807
808This enables or disables panic on out-of-memory feature.
809
810If this is set to 0, the kernel will kill some rogue process,
811called oom_killer.  Usually, oom_killer can kill rogue processes and
812system will survive.
813
814If this is set to 1, the kernel panics when out-of-memory happens.
815However, if a process limits using nodes by mempolicy/cpusets,
816and those nodes become memory exhaustion status, one process
817may be killed by oom-killer. No panic occurs in this case.
818Because other nodes' memory may be free. This means system total status
819may be not fatal yet.
820
821If this is set to 2, the kernel panics compulsorily even on the
822above-mentioned. Even oom happens under memory cgroup, the whole
823system panics.
824
825The default value is 0.
826
8271 and 2 are for failover of clustering. Please select either
828according to your policy of failover.
829
830panic_on_oom=2+kdump gives you very strong tool to investigate
831why oom happens. You can get snapshot.
832
833
834percpu_pagelist_high_fraction
835=============================
836
837This is the fraction of pages in each zone that are can be stored to
838per-cpu page lists. It is an upper boundary that is divided depending
839on the number of online CPUs. The min value for this is 8 which means
840that we do not allow more than 1/8th of pages in each zone to be stored
841on per-cpu page lists. This entry only changes the value of hot per-cpu
842page lists. A user can specify a number like 100 to allocate 1/100th of
843each zone between per-cpu lists.
844
845The batch value of each per-cpu page list remains the same regardless of
846the value of the high fraction so allocation latencies are unaffected.
847
848The initial value is zero. Kernel uses this value to set the high pcp->high
849mark based on the low watermark for the zone and the number of local
850online CPUs.  If the user writes '0' to this sysctl, it will revert to
851this default behavior.
852
853
854stat_interval
855=============
856
857The time interval between which vm statistics are updated.  The default
858is 1 second.
859
860
861stat_refresh
862============
863
864Any read or write (by root only) flushes all the per-cpu vm statistics
865into their global totals, for more accurate reports when testing
866e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo
867
868As a side-effect, it also checks for negative totals (elsewhere reported
869as 0) and "fails" with EINVAL if any are found, with a warning in dmesg.
870(At time of writing, a few stats are known sometimes to be found negative,
871with no ill effects: errors and warnings on these stats are suppressed.)
872
873
874numa_stat
875=========
876
877This interface allows runtime configuration of numa statistics.
878
879When page allocation performance becomes a bottleneck and you can tolerate
880some possible tool breakage and decreased numa counter precision, you can
881do::
882
883	echo 0 > /proc/sys/vm/numa_stat
884
885When page allocation performance is not a bottleneck and you want all
886tooling to work, you can do::
887
888	echo 1 > /proc/sys/vm/numa_stat
889
890
891swappiness
892==========
893
894This control is used to define the rough relative IO cost of swapping
895and filesystem paging, as a value between 0 and 200. At 100, the VM
896assumes equal IO cost and will thus apply memory pressure to the page
897cache and swap-backed pages equally; lower values signify more
898expensive swap IO, higher values indicates cheaper.
899
900Keep in mind that filesystem IO patterns under memory pressure tend to
901be more efficient than swap's random IO. An optimal value will require
902experimentation and will also be workload-dependent.
903
904The default value is 60.
905
906For in-memory swap, like zram or zswap, as well as hybrid setups that
907have swap on faster devices than the filesystem, values beyond 100 can
908be considered. For example, if the random IO against the swap device
909is on average 2x faster than IO from the filesystem, swappiness should
910be 133 (x + 2x = 200, 2x = 133.33).
911
912At 0, the kernel will not initiate swap until the amount of free and
913file-backed pages is less than the high watermark in a zone.
914
915
916unprivileged_userfaultfd
917========================
918
919This flag controls the mode in which unprivileged users can use the
920userfaultfd system calls. Set this to 0 to restrict unprivileged users
921to handle page faults in user mode only. In this case, users without
922SYS_CAP_PTRACE must pass UFFD_USER_MODE_ONLY in order for userfaultfd to
923succeed. Prohibiting use of userfaultfd for handling faults from kernel
924mode may make certain vulnerabilities more difficult to exploit.
925
926Set this to 1 to allow unprivileged users to use the userfaultfd system
927calls without any restrictions.
928
929The default value is 0.
930
931
932user_reserve_kbytes
933===================
934
935When overcommit_memory is set to 2, "never overcommit" mode, reserve
936min(3% of current process size, user_reserve_kbytes) of free memory.
937This is intended to prevent a user from starting a single memory hogging
938process, such that they cannot recover (kill the hog).
939
940user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
941
942If this is reduced to zero, then the user will be allowed to allocate
943all free memory with a single process, minus admin_reserve_kbytes.
944Any subsequent attempts to execute a command will result in
945"fork: Cannot allocate memory".
946
947Changing this takes effect whenever an application requests memory.
948
949
950vfs_cache_pressure
951==================
952
953This percentage value controls the tendency of the kernel to reclaim
954the memory which is used for caching of directory and inode objects.
955
956At the default value of vfs_cache_pressure=100 the kernel will attempt to
957reclaim dentries and inodes at a "fair" rate with respect to pagecache and
958swapcache reclaim.  Decreasing vfs_cache_pressure causes the kernel to prefer
959to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
960never reclaim dentries and inodes due to memory pressure and this can easily
961lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
962causes the kernel to prefer to reclaim dentries and inodes.
963
964Increasing vfs_cache_pressure significantly beyond 100 may have negative
965performance impact. Reclaim code needs to take various locks to find freeable
966directory and inode objects. With vfs_cache_pressure=1000, it will look for
967ten times more freeable objects than there are.
968
969
970watermark_boost_factor
971======================
972
973This factor controls the level of reclaim when memory is being fragmented.
974It defines the percentage of the high watermark of a zone that will be
975reclaimed if pages of different mobility are being mixed within pageblocks.
976The intent is that compaction has less work to do in the future and to
977increase the success rate of future high-order allocations such as SLUB
978allocations, THP and hugetlbfs pages.
979
980To make it sensible with respect to the watermark_scale_factor
981parameter, the unit is in fractions of 10,000. The default value of
98215,000 means that up to 150% of the high watermark will be reclaimed in the
983event of a pageblock being mixed due to fragmentation. The level of reclaim
984is determined by the number of fragmentation events that occurred in the
985recent past. If this value is smaller than a pageblock then a pageblocks
986worth of pages will be reclaimed (e.g.  2MB on 64-bit x86). A boost factor
987of 0 will disable the feature.
988
989
990watermark_scale_factor
991======================
992
993This factor controls the aggressiveness of kswapd. It defines the
994amount of memory left in a node/system before kswapd is woken up and
995how much memory needs to be free before kswapd goes back to sleep.
996
997The unit is in fractions of 10,000. The default value of 10 means the
998distances between watermarks are 0.1% of the available memory in the
999node/system. The maximum value is 3000, or 30% of memory.
1000
1001A high rate of threads entering direct reclaim (allocstall) or kswapd
1002going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate
1003that the number of free pages kswapd maintains for latency reasons is
1004too small for the allocation bursts occurring in the system. This knob
1005can then be used to tune kswapd aggressiveness accordingly.
1006
1007
1008zone_reclaim_mode
1009=================
1010
1011Zone_reclaim_mode allows someone to set more or less aggressive approaches to
1012reclaim memory when a zone runs out of memory. If it is set to zero then no
1013zone reclaim occurs. Allocations will be satisfied from other zones / nodes
1014in the system.
1015
1016This is value OR'ed together of
1017
1018=	===================================
10191	Zone reclaim on
10202	Zone reclaim writes dirty pages out
10214	Zone reclaim swaps pages
1022=	===================================
1023
1024zone_reclaim_mode is disabled by default.  For file servers or workloads
1025that benefit from having their data cached, zone_reclaim_mode should be
1026left disabled as the caching effect is likely to be more important than
1027data locality.
1028
1029Consider enabling one or more zone_reclaim mode bits if it's known that the
1030workload is partitioned such that each partition fits within a NUMA node
1031and that accessing remote memory would cause a measurable performance
1032reduction.  The page allocator will take additional actions before
1033allocating off node pages.
1034
1035Allowing zone reclaim to write out pages stops processes that are
1036writing large amounts of data from dirtying pages on other nodes. Zone
1037reclaim will write out dirty pages if a zone fills up and so effectively
1038throttle the process. This may decrease the performance of a single process
1039since it cannot use all of system memory to buffer the outgoing writes
1040anymore but it preserve the memory on other nodes so that the performance
1041of other processes running on other nodes will not be affected.
1042
1043Allowing regular swap effectively restricts allocations to the local
1044node unless explicitly overridden by memory policies or cpuset
1045configurations.
1046