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
29 - compaction_proactiveness
30 - compact_unevictable_allowed
31 - dirty_background_bytes
32 - dirty_background_ratio
34 - dirty_expire_centisecs
36 - dirtytime_expire_seconds
37 - dirty_writeback_centisecs
40 - highmem_is_dirtyable
44 - lowmem_reserve_ratio
46 - memory_failure_early_kill
47 - memory_failure_recovery
53 - mmap_rnd_compat_bits
55 - nr_hugepages_mempolicy
56 - nr_overcommit_hugepages
57 - nr_trim_pages (only if CONFIG_MMU=n)
60 - oom_kill_allocating_task
66 - percpu_pagelist_high_fraction
71 - unprivileged_userfaultfd
74 - watermark_boost_factor
75 - watermark_scale_factor
82 The amount of free memory in the system that should be reserved for users
83 with the capability cap_sys_admin.
85 admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
87 That should provide enough for the admin to log in and kill a process,
88 if necessary, under the default overcommit 'guess' mode.
90 Systems running under overcommit 'never' should increase this to account
91 for the full Virtual Memory Size of programs used to recover. Otherwise,
92 root may not be able to log in to recover the system.
94 How do you calculate a minimum useful reserve?
96 sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
98 For overcommit 'guess', we can sum resident set sizes (RSS).
99 On x86_64 this is about 8MB.
101 For overcommit 'never', we can take the max of their virtual sizes (VSZ)
102 and add the sum of their RSS.
103 On x86_64 this is about 128MB.
105 Changing this takes effect whenever an application requests memory.
111 Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
112 all zones are compacted such that free memory is available in contiguous
113 blocks where possible. This can be important for example in the allocation of
114 huge pages although processes will also directly compact memory as required.
116 compaction_proactiveness
117 ========================
119 This tunable takes a value in the range [0, 100] with a default value of
120 20. This tunable determines how aggressively compaction is done in the
121 background. Write of a non zero value to this tunable will immediately
122 trigger the proactive compaction. Setting it to 0 disables proactive compaction.
124 Note that compaction has a non-trivial system-wide impact as pages
125 belonging to different processes are moved around, which could also lead
126 to latency spikes in unsuspecting applications. The kernel employs
127 various heuristics to avoid wasting CPU cycles if it detects that
128 proactive compaction is not being effective.
130 Be careful when setting it to extreme values like 100, as that may
131 cause excessive background compaction activity.
133 compact_unevictable_allowed
134 ===========================
136 Available only when CONFIG_COMPACTION is set. When set to 1, compaction is
137 allowed to examine the unevictable lru (mlocked pages) for pages to compact.
138 This should be used on systems where stalls for minor page faults are an
139 acceptable trade for large contiguous free memory. Set to 0 to prevent
140 compaction from moving pages that are unevictable. Default value is 1.
141 On CONFIG_PREEMPT_RT the default value is 0 in order to avoid a page fault, due
142 to compaction, which would block the task from becoming active until the fault
146 dirty_background_bytes
147 ======================
149 Contains the amount of dirty memory at which the background kernel
150 flusher threads will start writeback.
153 dirty_background_bytes is the counterpart of dirty_background_ratio. Only
154 one of them may be specified at a time. When one sysctl is written it is
155 immediately taken into account to evaluate the dirty memory limits and the
156 other appears as 0 when read.
159 dirty_background_ratio
160 ======================
162 Contains, as a percentage of total available memory that contains free pages
163 and reclaimable pages, the number of pages at which the background kernel
164 flusher threads will start writing out dirty data.
166 The total available memory is not equal to total system memory.
172 Contains the amount of dirty memory at which a process generating disk writes
173 will itself start writeback.
175 Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
176 specified at a time. When one sysctl is written it is immediately taken into
177 account to evaluate the dirty memory limits and the other appears as 0 when
180 Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
181 value lower than this limit will be ignored and the old configuration will be
185 dirty_expire_centisecs
186 ======================
188 This tunable is used to define when dirty data is old enough to be eligible
189 for writeout by the kernel flusher threads. It is expressed in 100'ths
190 of a second. Data which has been dirty in-memory for longer than this
191 interval will be written out next time a flusher thread wakes up.
197 Contains, as a percentage of total available memory that contains free pages
198 and reclaimable pages, the number of pages at which a process which is
199 generating disk writes will itself start writing out dirty data.
201 The total available memory is not equal to total system memory.
204 dirtytime_expire_seconds
205 ========================
207 When a lazytime inode is constantly having its pages dirtied, the inode with
208 an updated timestamp will never get chance to be written out. And, if the
209 only thing that has happened on the file system is a dirtytime inode caused
210 by an atime update, a worker will be scheduled to make sure that inode
211 eventually gets pushed out to disk. This tunable is used to define when dirty
212 inode is old enough to be eligible for writeback by the kernel flusher threads.
213 And, it is also used as the interval to wakeup dirtytime_writeback thread.
216 dirty_writeback_centisecs
217 =========================
219 The kernel flusher threads will periodically wake up and write `old` data
220 out to disk. This tunable expresses the interval between those wakeups, in
223 Setting this to zero disables periodic writeback altogether.
229 Writing to this will cause the kernel to drop clean caches, as well as
230 reclaimable slab objects like dentries and inodes. Once dropped, their
235 echo 1 > /proc/sys/vm/drop_caches
237 To free reclaimable slab objects (includes dentries and inodes)::
239 echo 2 > /proc/sys/vm/drop_caches
241 To free slab objects and pagecache::
243 echo 3 > /proc/sys/vm/drop_caches
245 This is a non-destructive operation and will not free any dirty objects.
246 To increase the number of objects freed by this operation, the user may run
247 `sync` prior to writing to /proc/sys/vm/drop_caches. This will minimize the
248 number of dirty objects on the system and create more candidates to be
251 This file is not a means to control the growth of the various kernel caches
252 (inodes, dentries, pagecache, etc...) These objects are automatically
253 reclaimed by the kernel when memory is needed elsewhere on the system.
255 Use of this file can cause performance problems. Since it discards cached
256 objects, it may cost a significant amount of I/O and CPU to recreate the
257 dropped objects, especially if they were under heavy use. Because of this,
258 use outside of a testing or debugging environment is not recommended.
260 You may see informational messages in your kernel log when this file is
263 cat (1234): drop_caches: 3
265 These are informational only. They do not mean that anything is wrong
266 with your system. To disable them, echo 4 (bit 2) into drop_caches.
272 This parameter affects whether the kernel will compact memory or direct
273 reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in
274 debugfs shows what the fragmentation index for each order is in each zone in
275 the system. Values tending towards 0 imply allocations would fail due to lack
276 of memory, values towards 1000 imply failures are due to fragmentation and -1
277 implies that the allocation will succeed as long as watermarks are met.
279 The kernel will not compact memory in a zone if the
280 fragmentation index is <= extfrag_threshold. The default value is 500.
286 Available only for systems with CONFIG_HIGHMEM enabled (32b systems).
288 This parameter controls whether the high memory is considered for dirty
289 writers throttling. This is not the case by default which means that
290 only the amount of memory directly visible/usable by the kernel can
291 be dirtied. As a result, on systems with a large amount of memory and
292 lowmem basically depleted writers might be throttled too early and
293 streaming writes can get very slow.
295 Changing the value to non zero would allow more memory to be dirtied
296 and thus allow writers to write more data which can be flushed to the
297 storage more effectively. Note this also comes with a risk of pre-mature
298 OOM killer because some writers (e.g. direct block device writes) can
299 only use the low memory and they can fill it up with dirty data without
306 hugetlb_shm_group contains group id that is allowed to create SysV
307 shared memory segment using hugetlb page.
313 laptop_mode is a knob that controls "laptop mode". All the things that are
314 controlled by this knob are discussed in Documentation/admin-guide/laptops/laptop-mode.rst.
320 If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
321 will use the legacy (2.4) layout for all processes.
327 For some specialised workloads on highmem machines it is dangerous for
328 the kernel to allow process memory to be allocated from the "lowmem"
329 zone. This is because that memory could then be pinned via the mlock()
330 system call, or by unavailability of swapspace.
332 And on large highmem machines this lack of reclaimable lowmem memory
335 So the Linux page allocator has a mechanism which prevents allocations
336 which *could* use highmem from using too much lowmem. This means that
337 a certain amount of lowmem is defended from the possibility of being
338 captured into pinned user memory.
340 (The same argument applies to the old 16 megabyte ISA DMA region. This
341 mechanism will also defend that region from allocations which could use
344 The `lowmem_reserve_ratio` tunable determines how aggressive the kernel is
345 in defending these lower zones.
347 If you have a machine which uses highmem or ISA DMA and your
348 applications are using mlock(), or if you are running with no swap then
349 you probably should change the lowmem_reserve_ratio setting.
351 The lowmem_reserve_ratio is an array. You can see them by reading this file::
353 % cat /proc/sys/vm/lowmem_reserve_ratio
356 But, these values are not used directly. The kernel calculates # of protection
357 pages for each zones from them. These are shown as array of protection pages
358 in /proc/zoneinfo like followings. (This is an example of x86-64 box).
359 Each zone has an array of protection pages like this::
369 protection: (0, 2004, 2004, 2004)
370 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
375 These protections are added to score to judge whether this zone should be used
376 for page allocation or should be reclaimed.
378 In this example, if normal pages (index=2) are required to this DMA zone and
379 watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
380 not be used because pages_free(1355) is smaller than watermark + protection[2]
381 (4 + 2004 = 2008). If this protection value is 0, this zone would be used for
382 normal page requirement. If requirement is DMA zone(index=0), protection[0]
385 zone[i]'s protection[j] is calculated by following expression::
388 zone[i]->protection[j]
389 = (total sums of managed_pages from zone[i+1] to zone[j] on the node)
390 / lowmem_reserve_ratio[i];
392 (should not be protected. = 0;
394 (not necessary, but looks 0)
396 The default values of lowmem_reserve_ratio[i] are
398 === ====================================
399 256 (if zone[i] means DMA or DMA32 zone)
401 === ====================================
403 As above expression, they are reciprocal number of ratio.
404 256 means 1/256. # of protection pages becomes about "0.39%" of total managed
405 pages of higher zones on the node.
407 If you would like to protect more pages, smaller values are effective.
408 The minimum value is 1 (1/1 -> 100%). The value less than 1 completely
409 disables protection of the pages.
415 This file contains the maximum number of memory map areas a process
416 may have. Memory map areas are used as a side-effect of calling
417 malloc, directly by mmap, mprotect, and madvise, and also when loading
420 While most applications need less than a thousand maps, certain
421 programs, particularly malloc debuggers, may consume lots of them,
422 e.g., up to one or two maps per allocation.
424 The default value is 65530.
427 memory_failure_early_kill:
428 ==========================
430 Control how to kill processes when uncorrected memory error (typically
431 a 2bit error in a memory module) is detected in the background by hardware
432 that cannot be handled by the kernel. In some cases (like the page
433 still having a valid copy on disk) the kernel will handle the failure
434 transparently without affecting any applications. But if there is
435 no other uptodate copy of the data it will kill to prevent any data
436 corruptions from propagating.
438 1: Kill all processes that have the corrupted and not reloadable page mapped
439 as soon as the corruption is detected. Note this is not supported
440 for a few types of pages, like kernel internally allocated data or
441 the swap cache, but works for the majority of user pages.
443 0: Only unmap the corrupted page from all processes and only kill a process
444 who tries to access it.
446 The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
447 handle this if they want to.
449 This is only active on architectures/platforms with advanced machine
450 check handling and depends on the hardware capabilities.
452 Applications can override this setting individually with the PR_MCE_KILL prctl
455 memory_failure_recovery
456 =======================
458 Enable memory failure recovery (when supported by the platform)
462 0: Always panic on a memory failure.
468 This is used to force the Linux VM to keep a minimum number
469 of kilobytes free. The VM uses this number to compute a
470 watermark[WMARK_MIN] value for each lowmem zone in the system.
471 Each lowmem zone gets a number of reserved free pages based
472 proportionally on its size.
474 Some minimal amount of memory is needed to satisfy PF_MEMALLOC
475 allocations; if you set this to lower than 1024KB, your system will
476 become subtly broken, and prone to deadlock under high loads.
478 Setting this too high will OOM your machine instantly.
484 This is available only on NUMA kernels.
486 A percentage of the total pages in each zone. On Zone reclaim
487 (fallback from the local zone occurs) slabs will be reclaimed if more
488 than this percentage of pages in a zone are reclaimable slab pages.
489 This insures that the slab growth stays under control even in NUMA
490 systems that rarely perform global reclaim.
492 The default is 5 percent.
494 Note that slab reclaim is triggered in a per zone / node fashion.
495 The process of reclaiming slab memory is currently not node specific
502 This is available only on NUMA kernels.
504 This is a percentage of the total pages in each zone. Zone reclaim will
505 only occur if more than this percentage of pages are in a state that
506 zone_reclaim_mode allows to be reclaimed.
508 If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
509 against all file-backed unmapped pages including swapcache pages and tmpfs
510 files. Otherwise, only unmapped pages backed by normal files but not tmpfs
511 files and similar are considered.
513 The default is 1 percent.
519 This file indicates the amount of address space which a user process will
520 be restricted from mmapping. Since kernel null dereference bugs could
521 accidentally operate based on the information in the first couple of pages
522 of memory userspace processes should not be allowed to write to them. By
523 default this value is set to 0 and no protections will be enforced by the
524 security module. Setting this value to something like 64k will allow the
525 vast majority of applications to work correctly and provide defense in depth
526 against future potential kernel bugs.
532 This value can be used to select the number of bits to use to
533 determine the random offset to the base address of vma regions
534 resulting from mmap allocations on architectures which support
535 tuning address space randomization. This value will be bounded
536 by the architecture's minimum and maximum supported values.
538 This value can be changed after boot using the
539 /proc/sys/vm/mmap_rnd_bits tunable
545 This value can be used to select the number of bits to use to
546 determine the random offset to the base address of vma regions
547 resulting from mmap allocations for applications run in
548 compatibility mode on architectures which support tuning address
549 space randomization. This value will be bounded by the
550 architecture's minimum and maximum supported values.
552 This value can be changed after boot using the
553 /proc/sys/vm/mmap_rnd_compat_bits tunable
559 Change the minimum size of the hugepage pool.
561 See Documentation/admin-guide/mm/hugetlbpage.rst
564 nr_hugepages_mempolicy
565 ======================
567 Change the size of the hugepage pool at run-time on a specific
570 See Documentation/admin-guide/mm/hugetlbpage.rst
573 nr_overcommit_hugepages
574 =======================
576 Change the maximum size of the hugepage pool. The maximum is
577 nr_hugepages + nr_overcommit_hugepages.
579 See Documentation/admin-guide/mm/hugetlbpage.rst
585 This is available only on NOMMU kernels.
587 This value adjusts the excess page trimming behaviour of power-of-2 aligned
588 NOMMU mmap allocations.
590 A value of 0 disables trimming of allocations entirely, while a value of 1
591 trims excess pages aggressively. Any value >= 1 acts as the watermark where
592 trimming of allocations is initiated.
594 The default value is 1.
596 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
602 This sysctl is only for NUMA and it is deprecated. Anything but
603 Node order will fail!
605 'where the memory is allocated from' is controlled by zonelists.
607 (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
608 you may be able to read ZONE_DMA as ZONE_DMA32...)
610 In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
611 ZONE_NORMAL -> ZONE_DMA
612 This means that a memory allocation request for GFP_KERNEL will
613 get memory from ZONE_DMA only when ZONE_NORMAL is not available.
615 In NUMA case, you can think of following 2 types of order.
616 Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL::
618 (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
619 (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
621 Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
622 will be used before ZONE_NORMAL exhaustion. This increases possibility of
623 out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
625 Type(B) cannot offer the best locality but is more robust against OOM of
628 Type(A) is called as "Node" order. Type (B) is "Zone" order.
630 "Node order" orders the zonelists by node, then by zone within each node.
631 Specify "[Nn]ode" for node order
633 "Zone Order" orders the zonelists by zone type, then by node within each
634 zone. Specify "[Zz]one" for zone order.
636 Specify "[Dd]efault" to request automatic configuration.
638 On 32-bit, the Normal zone needs to be preserved for allocations accessible
639 by the kernel, so "zone" order will be selected.
641 On 64-bit, devices that require DMA32/DMA are relatively rare, so "node"
642 order will be selected.
644 Default order is recommended unless this is causing problems for your
651 Enables a system-wide task dump (excluding kernel threads) to be produced
652 when the kernel performs an OOM-killing and includes such information as
653 pid, uid, tgid, vm size, rss, pgtables_bytes, swapents, oom_score_adj
654 score, and name. This is helpful to determine why the OOM killer was
655 invoked, to identify the rogue task that caused it, and to determine why
656 the OOM killer chose the task it did to kill.
658 If this is set to zero, this information is suppressed. On very
659 large systems with thousands of tasks it may not be feasible to dump
660 the memory state information for each one. Such systems should not
661 be forced to incur a performance penalty in OOM conditions when the
662 information may not be desired.
664 If this is set to non-zero, this information is shown whenever the
665 OOM killer actually kills a memory-hogging task.
667 The default value is 1 (enabled).
670 oom_kill_allocating_task
671 ========================
673 This enables or disables killing the OOM-triggering task in
674 out-of-memory situations.
676 If this is set to zero, the OOM killer will scan through the entire
677 tasklist and select a task based on heuristics to kill. This normally
678 selects a rogue memory-hogging task that frees up a large amount of
681 If this is set to non-zero, the OOM killer simply kills the task that
682 triggered the out-of-memory condition. This avoids the expensive
685 If panic_on_oom is selected, it takes precedence over whatever value
686 is used in oom_kill_allocating_task.
688 The default value is 0.
694 When overcommit_memory is set to 2, the committed address space is not
695 permitted to exceed swap plus this amount of physical RAM. See below.
697 Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one
698 of them may be specified at a time. Setting one disables the other (which
699 then appears as 0 when read).
705 This value contains a flag that enables memory overcommitment.
707 When this flag is 0, the kernel attempts to estimate the amount
708 of free memory left when userspace requests more memory.
710 When this flag is 1, the kernel pretends there is always enough
711 memory until it actually runs out.
713 When this flag is 2, the kernel uses a "never overcommit"
714 policy that attempts to prevent any overcommit of memory.
715 Note that user_reserve_kbytes affects this policy.
717 This feature can be very useful because there are a lot of
718 programs that malloc() huge amounts of memory "just-in-case"
719 and don't use much of it.
721 The default value is 0.
723 See Documentation/vm/overcommit-accounting.rst and
724 mm/util.c::__vm_enough_memory() for more information.
730 When overcommit_memory is set to 2, the committed address
731 space is not permitted to exceed swap plus this percentage
732 of physical RAM. See above.
738 page-cluster controls the number of pages up to which consecutive pages
739 are read in from swap in a single attempt. This is the swap counterpart
740 to page cache readahead.
741 The mentioned consecutivity is not in terms of virtual/physical addresses,
742 but consecutive on swap space - that means they were swapped out together.
744 It is a logarithmic value - setting it to zero means "1 page", setting
745 it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
746 Zero disables swap readahead completely.
748 The default value is three (eight pages at a time). There may be some
749 small benefits in tuning this to a different value if your workload is
752 Lower values mean lower latencies for initial faults, but at the same time
753 extra faults and I/O delays for following faults if they would have been part of
754 that consecutive pages readahead would have brought in.
760 This enables or disables panic on out-of-memory feature.
762 If this is set to 0, the kernel will kill some rogue process,
763 called oom_killer. Usually, oom_killer can kill rogue processes and
766 If this is set to 1, the kernel panics when out-of-memory happens.
767 However, if a process limits using nodes by mempolicy/cpusets,
768 and those nodes become memory exhaustion status, one process
769 may be killed by oom-killer. No panic occurs in this case.
770 Because other nodes' memory may be free. This means system total status
771 may be not fatal yet.
773 If this is set to 2, the kernel panics compulsorily even on the
774 above-mentioned. Even oom happens under memory cgroup, the whole
777 The default value is 0.
779 1 and 2 are for failover of clustering. Please select either
780 according to your policy of failover.
782 panic_on_oom=2+kdump gives you very strong tool to investigate
783 why oom happens. You can get snapshot.
786 percpu_pagelist_high_fraction
787 =============================
789 This is the fraction of pages in each zone that are can be stored to
790 per-cpu page lists. It is an upper boundary that is divided depending
791 on the number of online CPUs. The min value for this is 8 which means
792 that we do not allow more than 1/8th of pages in each zone to be stored
793 on per-cpu page lists. This entry only changes the value of hot per-cpu
794 page lists. A user can specify a number like 100 to allocate 1/100th of
795 each zone between per-cpu lists.
797 The batch value of each per-cpu page list remains the same regardless of
798 the value of the high fraction so allocation latencies are unaffected.
800 The initial value is zero. Kernel uses this value to set the high pcp->high
801 mark based on the low watermark for the zone and the number of local
802 online CPUs. If the user writes '0' to this sysctl, it will revert to
803 this default behavior.
809 The time interval between which vm statistics are updated. The default
816 Any read or write (by root only) flushes all the per-cpu vm statistics
817 into their global totals, for more accurate reports when testing
818 e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo
820 As a side-effect, it also checks for negative totals (elsewhere reported
821 as 0) and "fails" with EINVAL if any are found, with a warning in dmesg.
822 (At time of writing, a few stats are known sometimes to be found negative,
823 with no ill effects: errors and warnings on these stats are suppressed.)
829 This interface allows runtime configuration of numa statistics.
831 When page allocation performance becomes a bottleneck and you can tolerate
832 some possible tool breakage and decreased numa counter precision, you can
835 echo 0 > /proc/sys/vm/numa_stat
837 When page allocation performance is not a bottleneck and you want all
838 tooling to work, you can do::
840 echo 1 > /proc/sys/vm/numa_stat
846 This control is used to define the rough relative IO cost of swapping
847 and filesystem paging, as a value between 0 and 200. At 100, the VM
848 assumes equal IO cost and will thus apply memory pressure to the page
849 cache and swap-backed pages equally; lower values signify more
850 expensive swap IO, higher values indicates cheaper.
852 Keep in mind that filesystem IO patterns under memory pressure tend to
853 be more efficient than swap's random IO. An optimal value will require
854 experimentation and will also be workload-dependent.
856 The default value is 60.
858 For in-memory swap, like zram or zswap, as well as hybrid setups that
859 have swap on faster devices than the filesystem, values beyond 100 can
860 be considered. For example, if the random IO against the swap device
861 is on average 2x faster than IO from the filesystem, swappiness should
862 be 133 (x + 2x = 200, 2x = 133.33).
864 At 0, the kernel will not initiate swap until the amount of free and
865 file-backed pages is less than the high watermark in a zone.
868 unprivileged_userfaultfd
869 ========================
871 This flag controls the mode in which unprivileged users can use the
872 userfaultfd system calls. Set this to 0 to restrict unprivileged users
873 to handle page faults in user mode only. In this case, users without
874 SYS_CAP_PTRACE must pass UFFD_USER_MODE_ONLY in order for userfaultfd to
875 succeed. Prohibiting use of userfaultfd for handling faults from kernel
876 mode may make certain vulnerabilities more difficult to exploit.
878 Set this to 1 to allow unprivileged users to use the userfaultfd system
879 calls without any restrictions.
881 The default value is 0.
887 When overcommit_memory is set to 2, "never overcommit" mode, reserve
888 min(3% of current process size, user_reserve_kbytes) of free memory.
889 This is intended to prevent a user from starting a single memory hogging
890 process, such that they cannot recover (kill the hog).
892 user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
894 If this is reduced to zero, then the user will be allowed to allocate
895 all free memory with a single process, minus admin_reserve_kbytes.
896 Any subsequent attempts to execute a command will result in
897 "fork: Cannot allocate memory".
899 Changing this takes effect whenever an application requests memory.
905 This percentage value controls the tendency of the kernel to reclaim
906 the memory which is used for caching of directory and inode objects.
908 At the default value of vfs_cache_pressure=100 the kernel will attempt to
909 reclaim dentries and inodes at a "fair" rate with respect to pagecache and
910 swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer
911 to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
912 never reclaim dentries and inodes due to memory pressure and this can easily
913 lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
914 causes the kernel to prefer to reclaim dentries and inodes.
916 Increasing vfs_cache_pressure significantly beyond 100 may have negative
917 performance impact. Reclaim code needs to take various locks to find freeable
918 directory and inode objects. With vfs_cache_pressure=1000, it will look for
919 ten times more freeable objects than there are.
922 watermark_boost_factor
923 ======================
925 This factor controls the level of reclaim when memory is being fragmented.
926 It defines the percentage of the high watermark of a zone that will be
927 reclaimed if pages of different mobility are being mixed within pageblocks.
928 The intent is that compaction has less work to do in the future and to
929 increase the success rate of future high-order allocations such as SLUB
930 allocations, THP and hugetlbfs pages.
932 To make it sensible with respect to the watermark_scale_factor
933 parameter, the unit is in fractions of 10,000. The default value of
934 15,000 means that up to 150% of the high watermark will be reclaimed in the
935 event of a pageblock being mixed due to fragmentation. The level of reclaim
936 is determined by the number of fragmentation events that occurred in the
937 recent past. If this value is smaller than a pageblock then a pageblocks
938 worth of pages will be reclaimed (e.g. 2MB on 64-bit x86). A boost factor
939 of 0 will disable the feature.
942 watermark_scale_factor
943 ======================
945 This factor controls the aggressiveness of kswapd. It defines the
946 amount of memory left in a node/system before kswapd is woken up and
947 how much memory needs to be free before kswapd goes back to sleep.
949 The unit is in fractions of 10,000. The default value of 10 means the
950 distances between watermarks are 0.1% of the available memory in the
951 node/system. The maximum value is 1000, or 10% of memory.
953 A high rate of threads entering direct reclaim (allocstall) or kswapd
954 going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate
955 that the number of free pages kswapd maintains for latency reasons is
956 too small for the allocation bursts occurring in the system. This knob
957 can then be used to tune kswapd aggressiveness accordingly.
963 Zone_reclaim_mode allows someone to set more or less aggressive approaches to
964 reclaim memory when a zone runs out of memory. If it is set to zero then no
965 zone reclaim occurs. Allocations will be satisfied from other zones / nodes
968 This is value OR'ed together of
970 = ===================================
972 2 Zone reclaim writes dirty pages out
973 4 Zone reclaim swaps pages
974 = ===================================
976 zone_reclaim_mode is disabled by default. For file servers or workloads
977 that benefit from having their data cached, zone_reclaim_mode should be
978 left disabled as the caching effect is likely to be more important than
981 Consider enabling one or more zone_reclaim mode bits if it's known that the
982 workload is partitioned such that each partition fits within a NUMA node
983 and that accessing remote memory would cause a measurable performance
984 reduction. The page allocator will take additional actions before
985 allocating off node pages.
987 Allowing zone reclaim to write out pages stops processes that are
988 writing large amounts of data from dirtying pages on other nodes. Zone
989 reclaim will write out dirty pages if a zone fills up and so effectively
990 throttle the process. This may decrease the performance of a single process
991 since it cannot use all of system memory to buffer the outgoing writes
992 anymore but it preserve the memory on other nodes so that the performance
993 of other processes running on other nodes will not be affected.
995 Allowing regular swap effectively restricts allocations to the local
996 node unless explicitly overridden by memory policies or cpuset