1 ===============================
2 Documentation for /proc/sys/vm/
3 ===============================
7 Copyright (c) 1998, 1999, Rik van Riel <riel@nl.linux.org>
9 Copyright (c) 2008 Peter W. Morreale <pmorreale@novell.com>
11 For general info and legal blurb, please look in index.rst.
13 ------------------------------------------------------------------------------
15 This file contains the documentation for the sysctl files in
16 /proc/sys/vm and is valid for Linux kernel version 2.6.29.
18 The files in this directory can be used to tune the operation
19 of the virtual memory (VM) subsystem of the Linux kernel and
20 the writeout of dirty data to disk.
22 Default values and initialization routines for most of these
23 files can be found in mm/swap.c.
25 Currently, these files are in /proc/sys/vm:
27 - admin_reserve_kbytes
30 - compaction_proactiveness
31 - compact_unevictable_allowed
32 - dirty_background_bytes
33 - dirty_background_ratio
35 - dirty_expire_centisecs
37 - dirtytime_expire_seconds
38 - dirty_writeback_centisecs
41 - highmem_is_dirtyable
45 - lowmem_reserve_ratio
47 - memory_failure_early_kill
48 - memory_failure_recovery
54 - mmap_rnd_compat_bits
56 - nr_hugepages_mempolicy
57 - nr_overcommit_hugepages
58 - nr_trim_pages (only if CONFIG_MMU=n)
61 - oom_kill_allocating_task
67 - percpu_pagelist_fraction
72 - unprivileged_userfaultfd
75 - watermark_boost_factor
76 - watermark_scale_factor
83 The amount of free memory in the system that should be reserved for users
84 with the capability cap_sys_admin.
86 admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
88 That should provide enough for the admin to log in and kill a process,
89 if necessary, under the default overcommit 'guess' mode.
91 Systems running under overcommit 'never' should increase this to account
92 for the full Virtual Memory Size of programs used to recover. Otherwise,
93 root may not be able to log in to recover the system.
95 How do you calculate a minimum useful reserve?
97 sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
99 For overcommit 'guess', we can sum resident set sizes (RSS).
100 On x86_64 this is about 8MB.
102 For overcommit 'never', we can take the max of their virtual sizes (VSZ)
103 and add the sum of their RSS.
104 On x86_64 this is about 128MB.
106 Changing this takes effect whenever an application requests memory.
112 block_dump enables block I/O debugging when set to a nonzero value. More
113 information on block I/O debugging is in Documentation/admin-guide/laptops/laptop-mode.rst.
119 Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
120 all zones are compacted such that free memory is available in contiguous
121 blocks where possible. This can be important for example in the allocation of
122 huge pages although processes will also directly compact memory as required.
124 compaction_proactiveness
125 ========================
127 This tunable takes a value in the range [0, 100] with a default value of
128 20. This tunable determines how aggressively compaction is done in the
129 background. Setting it to 0 disables proactive compaction.
131 Note that compaction has a non-trivial system-wide impact as pages
132 belonging to different processes are moved around, which could also lead
133 to latency spikes in unsuspecting applications. The kernel employs
134 various heuristics to avoid wasting CPU cycles if it detects that
135 proactive compaction is not being effective.
137 Be careful when setting it to extreme values like 100, as that may
138 cause excessive background compaction activity.
140 compact_unevictable_allowed
141 ===========================
143 Available only when CONFIG_COMPACTION is set. When set to 1, compaction is
144 allowed to examine the unevictable lru (mlocked pages) for pages to compact.
145 This should be used on systems where stalls for minor page faults are an
146 acceptable trade for large contiguous free memory. Set to 0 to prevent
147 compaction from moving pages that are unevictable. Default value is 1.
148 On CONFIG_PREEMPT_RT the default value is 0 in order to avoid a page fault, due
149 to compaction, which would block the task from becoming active until the fault
153 dirty_background_bytes
154 ======================
156 Contains the amount of dirty memory at which the background kernel
157 flusher threads will start writeback.
160 dirty_background_bytes is the counterpart of dirty_background_ratio. Only
161 one of them may be specified at a time. When one sysctl is written it is
162 immediately taken into account to evaluate the dirty memory limits and the
163 other appears as 0 when read.
166 dirty_background_ratio
167 ======================
169 Contains, as a percentage of total available memory that contains free pages
170 and reclaimable pages, the number of pages at which the background kernel
171 flusher threads will start writing out dirty data.
173 The total available memory is not equal to total system memory.
179 Contains the amount of dirty memory at which a process generating disk writes
180 will itself start writeback.
182 Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
183 specified at a time. When one sysctl is written it is immediately taken into
184 account to evaluate the dirty memory limits and the other appears as 0 when
187 Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
188 value lower than this limit will be ignored and the old configuration will be
192 dirty_expire_centisecs
193 ======================
195 This tunable is used to define when dirty data is old enough to be eligible
196 for writeout by the kernel flusher threads. It is expressed in 100'ths
197 of a second. Data which has been dirty in-memory for longer than this
198 interval will be written out next time a flusher thread wakes up.
204 Contains, as a percentage of total available memory that contains free pages
205 and reclaimable pages, the number of pages at which a process which is
206 generating disk writes will itself start writing out dirty data.
208 The total available memory is not equal to total system memory.
211 dirtytime_expire_seconds
212 ========================
214 When a lazytime inode is constantly having its pages dirtied, the inode with
215 an updated timestamp will never get chance to be written out. And, if the
216 only thing that has happened on the file system is a dirtytime inode caused
217 by an atime update, a worker will be scheduled to make sure that inode
218 eventually gets pushed out to disk. This tunable is used to define when dirty
219 inode is old enough to be eligible for writeback by the kernel flusher threads.
220 And, it is also used as the interval to wakeup dirtytime_writeback thread.
223 dirty_writeback_centisecs
224 =========================
226 The kernel flusher threads will periodically wake up and write `old` data
227 out to disk. This tunable expresses the interval between those wakeups, in
230 Setting this to zero disables periodic writeback altogether.
236 Writing to this will cause the kernel to drop clean caches, as well as
237 reclaimable slab objects like dentries and inodes. Once dropped, their
242 echo 1 > /proc/sys/vm/drop_caches
244 To free reclaimable slab objects (includes dentries and inodes)::
246 echo 2 > /proc/sys/vm/drop_caches
248 To free slab objects and pagecache::
250 echo 3 > /proc/sys/vm/drop_caches
252 This is a non-destructive operation and will not free any dirty objects.
253 To increase the number of objects freed by this operation, the user may run
254 `sync` prior to writing to /proc/sys/vm/drop_caches. This will minimize the
255 number of dirty objects on the system and create more candidates to be
258 This file is not a means to control the growth of the various kernel caches
259 (inodes, dentries, pagecache, etc...) These objects are automatically
260 reclaimed by the kernel when memory is needed elsewhere on the system.
262 Use of this file can cause performance problems. Since it discards cached
263 objects, it may cost a significant amount of I/O and CPU to recreate the
264 dropped objects, especially if they were under heavy use. Because of this,
265 use outside of a testing or debugging environment is not recommended.
267 You may see informational messages in your kernel log when this file is
270 cat (1234): drop_caches: 3
272 These are informational only. They do not mean that anything is wrong
273 with your system. To disable them, echo 4 (bit 2) into drop_caches.
279 This parameter affects whether the kernel will compact memory or direct
280 reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in
281 debugfs shows what the fragmentation index for each order is in each zone in
282 the system. Values tending towards 0 imply allocations would fail due to lack
283 of memory, values towards 1000 imply failures are due to fragmentation and -1
284 implies that the allocation will succeed as long as watermarks are met.
286 The kernel will not compact memory in a zone if the
287 fragmentation index is <= extfrag_threshold. The default value is 500.
293 Available only for systems with CONFIG_HIGHMEM enabled (32b systems).
295 This parameter controls whether the high memory is considered for dirty
296 writers throttling. This is not the case by default which means that
297 only the amount of memory directly visible/usable by the kernel can
298 be dirtied. As a result, on systems with a large amount of memory and
299 lowmem basically depleted writers might be throttled too early and
300 streaming writes can get very slow.
302 Changing the value to non zero would allow more memory to be dirtied
303 and thus allow writers to write more data which can be flushed to the
304 storage more effectively. Note this also comes with a risk of pre-mature
305 OOM killer because some writers (e.g. direct block device writes) can
306 only use the low memory and they can fill it up with dirty data without
313 hugetlb_shm_group contains group id that is allowed to create SysV
314 shared memory segment using hugetlb page.
320 laptop_mode is a knob that controls "laptop mode". All the things that are
321 controlled by this knob are discussed in Documentation/admin-guide/laptops/laptop-mode.rst.
327 If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
328 will use the legacy (2.4) layout for all processes.
334 For some specialised workloads on highmem machines it is dangerous for
335 the kernel to allow process memory to be allocated from the "lowmem"
336 zone. This is because that memory could then be pinned via the mlock()
337 system call, or by unavailability of swapspace.
339 And on large highmem machines this lack of reclaimable lowmem memory
342 So the Linux page allocator has a mechanism which prevents allocations
343 which *could* use highmem from using too much lowmem. This means that
344 a certain amount of lowmem is defended from the possibility of being
345 captured into pinned user memory.
347 (The same argument applies to the old 16 megabyte ISA DMA region. This
348 mechanism will also defend that region from allocations which could use
351 The `lowmem_reserve_ratio` tunable determines how aggressive the kernel is
352 in defending these lower zones.
354 If you have a machine which uses highmem or ISA DMA and your
355 applications are using mlock(), or if you are running with no swap then
356 you probably should change the lowmem_reserve_ratio setting.
358 The lowmem_reserve_ratio is an array. You can see them by reading this file::
360 % cat /proc/sys/vm/lowmem_reserve_ratio
363 But, these values are not used directly. The kernel calculates # of protection
364 pages for each zones from them. These are shown as array of protection pages
365 in /proc/zoneinfo like followings. (This is an example of x86-64 box).
366 Each zone has an array of protection pages like this::
376 protection: (0, 2004, 2004, 2004)
377 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
382 These protections are added to score to judge whether this zone should be used
383 for page allocation or should be reclaimed.
385 In this example, if normal pages (index=2) are required to this DMA zone and
386 watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
387 not be used because pages_free(1355) is smaller than watermark + protection[2]
388 (4 + 2004 = 2008). If this protection value is 0, this zone would be used for
389 normal page requirement. If requirement is DMA zone(index=0), protection[0]
392 zone[i]'s protection[j] is calculated by following expression::
395 zone[i]->protection[j]
396 = (total sums of managed_pages from zone[i+1] to zone[j] on the node)
397 / lowmem_reserve_ratio[i];
399 (should not be protected. = 0;
401 (not necessary, but looks 0)
403 The default values of lowmem_reserve_ratio[i] are
405 === ====================================
406 256 (if zone[i] means DMA or DMA32 zone)
408 === ====================================
410 As above expression, they are reciprocal number of ratio.
411 256 means 1/256. # of protection pages becomes about "0.39%" of total managed
412 pages of higher zones on the node.
414 If you would like to protect more pages, smaller values are effective.
415 The minimum value is 1 (1/1 -> 100%). The value less than 1 completely
416 disables protection of the pages.
422 This file contains the maximum number of memory map areas a process
423 may have. Memory map areas are used as a side-effect of calling
424 malloc, directly by mmap, mprotect, and madvise, and also when loading
427 While most applications need less than a thousand maps, certain
428 programs, particularly malloc debuggers, may consume lots of them,
429 e.g., up to one or two maps per allocation.
431 The default value is 65536.
434 memory_failure_early_kill:
435 ==========================
437 Control how to kill processes when uncorrected memory error (typically
438 a 2bit error in a memory module) is detected in the background by hardware
439 that cannot be handled by the kernel. In some cases (like the page
440 still having a valid copy on disk) the kernel will handle the failure
441 transparently without affecting any applications. But if there is
442 no other uptodate copy of the data it will kill to prevent any data
443 corruptions from propagating.
445 1: Kill all processes that have the corrupted and not reloadable page mapped
446 as soon as the corruption is detected. Note this is not supported
447 for a few types of pages, like kernel internally allocated data or
448 the swap cache, but works for the majority of user pages.
450 0: Only unmap the corrupted page from all processes and only kill a process
451 who tries to access it.
453 The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
454 handle this if they want to.
456 This is only active on architectures/platforms with advanced machine
457 check handling and depends on the hardware capabilities.
459 Applications can override this setting individually with the PR_MCE_KILL prctl
462 memory_failure_recovery
463 =======================
465 Enable memory failure recovery (when supported by the platform)
469 0: Always panic on a memory failure.
475 This is used to force the Linux VM to keep a minimum number
476 of kilobytes free. The VM uses this number to compute a
477 watermark[WMARK_MIN] value for each lowmem zone in the system.
478 Each lowmem zone gets a number of reserved free pages based
479 proportionally on its size.
481 Some minimal amount of memory is needed to satisfy PF_MEMALLOC
482 allocations; if you set this to lower than 1024KB, your system will
483 become subtly broken, and prone to deadlock under high loads.
485 Setting this too high will OOM your machine instantly.
491 This is available only on NUMA kernels.
493 A percentage of the total pages in each zone. On Zone reclaim
494 (fallback from the local zone occurs) slabs will be reclaimed if more
495 than this percentage of pages in a zone are reclaimable slab pages.
496 This insures that the slab growth stays under control even in NUMA
497 systems that rarely perform global reclaim.
499 The default is 5 percent.
501 Note that slab reclaim is triggered in a per zone / node fashion.
502 The process of reclaiming slab memory is currently not node specific
509 This is available only on NUMA kernels.
511 This is a percentage of the total pages in each zone. Zone reclaim will
512 only occur if more than this percentage of pages are in a state that
513 zone_reclaim_mode allows to be reclaimed.
515 If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
516 against all file-backed unmapped pages including swapcache pages and tmpfs
517 files. Otherwise, only unmapped pages backed by normal files but not tmpfs
518 files and similar are considered.
520 The default is 1 percent.
526 This file indicates the amount of address space which a user process will
527 be restricted from mmapping. Since kernel null dereference bugs could
528 accidentally operate based on the information in the first couple of pages
529 of memory userspace processes should not be allowed to write to them. By
530 default this value is set to 0 and no protections will be enforced by the
531 security module. Setting this value to something like 64k will allow the
532 vast majority of applications to work correctly and provide defense in depth
533 against future potential kernel bugs.
539 This value can be used to select the number of bits to use to
540 determine the random offset to the base address of vma regions
541 resulting from mmap allocations on architectures which support
542 tuning address space randomization. This value will be bounded
543 by the architecture's minimum and maximum supported values.
545 This value can be changed after boot using the
546 /proc/sys/vm/mmap_rnd_bits tunable
552 This value can be used to select the number of bits to use to
553 determine the random offset to the base address of vma regions
554 resulting from mmap allocations for applications run in
555 compatibility mode on architectures which support tuning address
556 space randomization. This value will be bounded by the
557 architecture's minimum and maximum supported values.
559 This value can be changed after boot using the
560 /proc/sys/vm/mmap_rnd_compat_bits tunable
566 Change the minimum size of the hugepage pool.
568 See Documentation/admin-guide/mm/hugetlbpage.rst
571 nr_hugepages_mempolicy
572 ======================
574 Change the size of the hugepage pool at run-time on a specific
577 See Documentation/admin-guide/mm/hugetlbpage.rst
580 nr_overcommit_hugepages
581 =======================
583 Change the maximum size of the hugepage pool. The maximum is
584 nr_hugepages + nr_overcommit_hugepages.
586 See Documentation/admin-guide/mm/hugetlbpage.rst
592 This is available only on NOMMU kernels.
594 This value adjusts the excess page trimming behaviour of power-of-2 aligned
595 NOMMU mmap allocations.
597 A value of 0 disables trimming of allocations entirely, while a value of 1
598 trims excess pages aggressively. Any value >= 1 acts as the watermark where
599 trimming of allocations is initiated.
601 The default value is 1.
603 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
609 This sysctl is only for NUMA and it is deprecated. Anything but
610 Node order will fail!
612 'where the memory is allocated from' is controlled by zonelists.
614 (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
615 you may be able to read ZONE_DMA as ZONE_DMA32...)
617 In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
618 ZONE_NORMAL -> ZONE_DMA
619 This means that a memory allocation request for GFP_KERNEL will
620 get memory from ZONE_DMA only when ZONE_NORMAL is not available.
622 In NUMA case, you can think of following 2 types of order.
623 Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL::
625 (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
626 (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
628 Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
629 will be used before ZONE_NORMAL exhaustion. This increases possibility of
630 out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
632 Type(B) cannot offer the best locality but is more robust against OOM of
635 Type(A) is called as "Node" order. Type (B) is "Zone" order.
637 "Node order" orders the zonelists by node, then by zone within each node.
638 Specify "[Nn]ode" for node order
640 "Zone Order" orders the zonelists by zone type, then by node within each
641 zone. Specify "[Zz]one" for zone order.
643 Specify "[Dd]efault" to request automatic configuration.
645 On 32-bit, the Normal zone needs to be preserved for allocations accessible
646 by the kernel, so "zone" order will be selected.
648 On 64-bit, devices that require DMA32/DMA are relatively rare, so "node"
649 order will be selected.
651 Default order is recommended unless this is causing problems for your
658 Enables a system-wide task dump (excluding kernel threads) to be produced
659 when the kernel performs an OOM-killing and includes such information as
660 pid, uid, tgid, vm size, rss, pgtables_bytes, swapents, oom_score_adj
661 score, and name. This is helpful to determine why the OOM killer was
662 invoked, to identify the rogue task that caused it, and to determine why
663 the OOM killer chose the task it did to kill.
665 If this is set to zero, this information is suppressed. On very
666 large systems with thousands of tasks it may not be feasible to dump
667 the memory state information for each one. Such systems should not
668 be forced to incur a performance penalty in OOM conditions when the
669 information may not be desired.
671 If this is set to non-zero, this information is shown whenever the
672 OOM killer actually kills a memory-hogging task.
674 The default value is 1 (enabled).
677 oom_kill_allocating_task
678 ========================
680 This enables or disables killing the OOM-triggering task in
681 out-of-memory situations.
683 If this is set to zero, the OOM killer will scan through the entire
684 tasklist and select a task based on heuristics to kill. This normally
685 selects a rogue memory-hogging task that frees up a large amount of
688 If this is set to non-zero, the OOM killer simply kills the task that
689 triggered the out-of-memory condition. This avoids the expensive
692 If panic_on_oom is selected, it takes precedence over whatever value
693 is used in oom_kill_allocating_task.
695 The default value is 0.
701 When overcommit_memory is set to 2, the committed address space is not
702 permitted to exceed swap plus this amount of physical RAM. See below.
704 Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one
705 of them may be specified at a time. Setting one disables the other (which
706 then appears as 0 when read).
712 This value contains a flag that enables memory overcommitment.
714 When this flag is 0, the kernel attempts to estimate the amount
715 of free memory left when userspace requests more memory.
717 When this flag is 1, the kernel pretends there is always enough
718 memory until it actually runs out.
720 When this flag is 2, the kernel uses a "never overcommit"
721 policy that attempts to prevent any overcommit of memory.
722 Note that user_reserve_kbytes affects this policy.
724 This feature can be very useful because there are a lot of
725 programs that malloc() huge amounts of memory "just-in-case"
726 and don't use much of it.
728 The default value is 0.
730 See Documentation/vm/overcommit-accounting.rst and
731 mm/util.c::__vm_enough_memory() for more information.
737 When overcommit_memory is set to 2, the committed address
738 space is not permitted to exceed swap plus this percentage
739 of physical RAM. See above.
745 page-cluster controls the number of pages up to which consecutive pages
746 are read in from swap in a single attempt. This is the swap counterpart
747 to page cache readahead.
748 The mentioned consecutivity is not in terms of virtual/physical addresses,
749 but consecutive on swap space - that means they were swapped out together.
751 It is a logarithmic value - setting it to zero means "1 page", setting
752 it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
753 Zero disables swap readahead completely.
755 The default value is three (eight pages at a time). There may be some
756 small benefits in tuning this to a different value if your workload is
759 Lower values mean lower latencies for initial faults, but at the same time
760 extra faults and I/O delays for following faults if they would have been part of
761 that consecutive pages readahead would have brought in.
767 This enables or disables panic on out-of-memory feature.
769 If this is set to 0, the kernel will kill some rogue process,
770 called oom_killer. Usually, oom_killer can kill rogue processes and
773 If this is set to 1, the kernel panics when out-of-memory happens.
774 However, if a process limits using nodes by mempolicy/cpusets,
775 and those nodes become memory exhaustion status, one process
776 may be killed by oom-killer. No panic occurs in this case.
777 Because other nodes' memory may be free. This means system total status
778 may be not fatal yet.
780 If this is set to 2, the kernel panics compulsorily even on the
781 above-mentioned. Even oom happens under memory cgroup, the whole
784 The default value is 0.
786 1 and 2 are for failover of clustering. Please select either
787 according to your policy of failover.
789 panic_on_oom=2+kdump gives you very strong tool to investigate
790 why oom happens. You can get snapshot.
793 percpu_pagelist_fraction
794 ========================
796 This is the fraction of pages at most (high mark pcp->high) in each zone that
797 are allocated for each per cpu page list. The min value for this is 8. It
798 means that we don't allow more than 1/8th of pages in each zone to be
799 allocated in any single per_cpu_pagelist. This entry only changes the value
800 of hot per cpu pagelists. User can specify a number like 100 to allocate
801 1/100th of each zone to each per cpu page list.
803 The batch value of each per cpu pagelist is also updated as a result. It is
804 set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8)
806 The initial value is zero. Kernel does not use this value at boot time to set
807 the high water marks for each per cpu page list. If the user writes '0' to this
808 sysctl, it will revert to this default behavior.
814 The time interval between which vm statistics are updated. The default
821 Any read or write (by root only) flushes all the per-cpu vm statistics
822 into their global totals, for more accurate reports when testing
823 e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo
825 As a side-effect, it also checks for negative totals (elsewhere reported
826 as 0) and "fails" with EINVAL if any are found, with a warning in dmesg.
827 (At time of writing, a few stats are known sometimes to be found negative,
828 with no ill effects: errors and warnings on these stats are suppressed.)
834 This interface allows runtime configuration of numa statistics.
836 When page allocation performance becomes a bottleneck and you can tolerate
837 some possible tool breakage and decreased numa counter precision, you can
840 echo 0 > /proc/sys/vm/numa_stat
842 When page allocation performance is not a bottleneck and you want all
843 tooling to work, you can do::
845 echo 1 > /proc/sys/vm/numa_stat
851 This control is used to define the rough relative IO cost of swapping
852 and filesystem paging, as a value between 0 and 200. At 100, the VM
853 assumes equal IO cost and will thus apply memory pressure to the page
854 cache and swap-backed pages equally; lower values signify more
855 expensive swap IO, higher values indicates cheaper.
857 Keep in mind that filesystem IO patterns under memory pressure tend to
858 be more efficient than swap's random IO. An optimal value will require
859 experimentation and will also be workload-dependent.
861 The default value is 60.
863 For in-memory swap, like zram or zswap, as well as hybrid setups that
864 have swap on faster devices than the filesystem, values beyond 100 can
865 be considered. For example, if the random IO against the swap device
866 is on average 2x faster than IO from the filesystem, swappiness should
867 be 133 (x + 2x = 200, 2x = 133.33).
869 At 0, the kernel will not initiate swap until the amount of free and
870 file-backed pages is less than the high watermark in a zone.
873 unprivileged_userfaultfd
874 ========================
876 This flag controls the mode in which unprivileged users can use the
877 userfaultfd system calls. Set this to 0 to restrict unprivileged users
878 to handle page faults in user mode only. In this case, users without
879 SYS_CAP_PTRACE must pass UFFD_USER_MODE_ONLY in order for userfaultfd to
880 succeed. Prohibiting use of userfaultfd for handling faults from kernel
881 mode may make certain vulnerabilities more difficult to exploit.
883 Set this to 1 to allow unprivileged users to use the userfaultfd system
884 calls without any restrictions.
886 The default value is 0.
892 When overcommit_memory is set to 2, "never overcommit" mode, reserve
893 min(3% of current process size, user_reserve_kbytes) of free memory.
894 This is intended to prevent a user from starting a single memory hogging
895 process, such that they cannot recover (kill the hog).
897 user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
899 If this is reduced to zero, then the user will be allowed to allocate
900 all free memory with a single process, minus admin_reserve_kbytes.
901 Any subsequent attempts to execute a command will result in
902 "fork: Cannot allocate memory".
904 Changing this takes effect whenever an application requests memory.
910 This percentage value controls the tendency of the kernel to reclaim
911 the memory which is used for caching of directory and inode objects.
913 At the default value of vfs_cache_pressure=100 the kernel will attempt to
914 reclaim dentries and inodes at a "fair" rate with respect to pagecache and
915 swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer
916 to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
917 never reclaim dentries and inodes due to memory pressure and this can easily
918 lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
919 causes the kernel to prefer to reclaim dentries and inodes.
921 Increasing vfs_cache_pressure significantly beyond 100 may have negative
922 performance impact. Reclaim code needs to take various locks to find freeable
923 directory and inode objects. With vfs_cache_pressure=1000, it will look for
924 ten times more freeable objects than there are.
927 watermark_boost_factor
928 ======================
930 This factor controls the level of reclaim when memory is being fragmented.
931 It defines the percentage of the high watermark of a zone that will be
932 reclaimed if pages of different mobility are being mixed within pageblocks.
933 The intent is that compaction has less work to do in the future and to
934 increase the success rate of future high-order allocations such as SLUB
935 allocations, THP and hugetlbfs pages.
937 To make it sensible with respect to the watermark_scale_factor
938 parameter, the unit is in fractions of 10,000. The default value of
939 15,000 on !DISCONTIGMEM configurations means that up to 150% of the high
940 watermark will be reclaimed in the event of a pageblock being mixed due
941 to fragmentation. The level of reclaim is determined by the number of
942 fragmentation events that occurred in the recent past. If this value is
943 smaller than a pageblock then a pageblocks worth of pages will be reclaimed
944 (e.g. 2MB on 64-bit x86). A boost factor of 0 will disable the feature.
947 watermark_scale_factor
948 ======================
950 This factor controls the aggressiveness of kswapd. It defines the
951 amount of memory left in a node/system before kswapd is woken up and
952 how much memory needs to be free before kswapd goes back to sleep.
954 The unit is in fractions of 10,000. The default value of 10 means the
955 distances between watermarks are 0.1% of the available memory in the
956 node/system. The maximum value is 1000, or 10% of memory.
958 A high rate of threads entering direct reclaim (allocstall) or kswapd
959 going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate
960 that the number of free pages kswapd maintains for latency reasons is
961 too small for the allocation bursts occurring in the system. This knob
962 can then be used to tune kswapd aggressiveness accordingly.
968 Zone_reclaim_mode allows someone to set more or less aggressive approaches to
969 reclaim memory when a zone runs out of memory. If it is set to zero then no
970 zone reclaim occurs. Allocations will be satisfied from other zones / nodes
973 This is value OR'ed together of
975 = ===================================
977 2 Zone reclaim writes dirty pages out
978 4 Zone reclaim swaps pages
979 = ===================================
981 zone_reclaim_mode is disabled by default. For file servers or workloads
982 that benefit from having their data cached, zone_reclaim_mode should be
983 left disabled as the caching effect is likely to be more important than
986 zone_reclaim may be enabled if it's known that the workload is partitioned
987 such that each partition fits within a NUMA node and that accessing remote
988 memory would cause a measurable performance reduction. The page allocator
989 will then reclaim easily reusable pages (those page cache pages that are
990 currently not used) before allocating off node pages.
992 Allowing zone reclaim to write out pages stops processes that are
993 writing large amounts of data from dirtying pages on other nodes. Zone
994 reclaim will write out dirty pages if a zone fills up and so effectively
995 throttle the process. This may decrease the performance of a single process
996 since it cannot use all of system memory to buffer the outgoing writes
997 anymore but it preserve the memory on other nodes so that the performance
998 of other processes running on other nodes will not be affected.
1000 Allowing regular swap effectively restricts allocations to the local
1001 node unless explicitly overridden by memory policies or cpuset