1101 lines · plain
1===============================2Documentation for /proc/sys/vm/3===============================4 5kernel version 2.6.296 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 in16/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 operation19of the virtual memory (VM) subsystem of the Linux kernel and20the writeout of dirty data to disk.21 22Default values and initialization routines for most of these23files can be found in mm/swap.c.24 25Currently, these files are in /proc/sys/vm:26 27- admin_reserve_kbytes28- compact_memory29- compaction_proactiveness30- compact_unevictable_allowed31- dirty_background_bytes32- dirty_background_ratio33- dirty_bytes34- dirty_expire_centisecs35- dirty_ratio36- dirtytime_expire_seconds37- dirty_writeback_centisecs38- drop_caches39- enable_soft_offline40- extfrag_threshold41- highmem_is_dirtyable42- hugetlb_shm_group43- laptop_mode44- legacy_va_layout45- lowmem_reserve_ratio46- max_map_count47- mem_profiling (only if CONFIG_MEM_ALLOC_PROFILING=y)48- memory_failure_early_kill49- memory_failure_recovery50- min_free_kbytes51- min_slab_ratio52- min_unmapped_ratio53- mmap_min_addr54- mmap_rnd_bits55- mmap_rnd_compat_bits56- nr_hugepages57- nr_hugepages_mempolicy58- nr_overcommit_hugepages59- nr_trim_pages (only if CONFIG_MMU=n)60- numa_zonelist_order61- oom_dump_tasks62- oom_kill_allocating_task63- overcommit_kbytes64- overcommit_memory65- overcommit_ratio66- page-cluster67- page_lock_unfairness68- panic_on_oom69- percpu_pagelist_high_fraction70- stat_interval71- stat_refresh72- numa_stat73- swappiness74- unprivileged_userfaultfd75- user_reserve_kbytes76- vfs_cache_pressure77- watermark_boost_factor78- watermark_scale_factor79- zone_reclaim_mode80 81 82admin_reserve_kbytes83====================84 85The amount of free memory in the system that should be reserved for users86with the capability cap_sys_admin.87 88admin_reserve_kbytes defaults to min(3% of free pages, 8MB)89 90That should provide enough for the admin to log in and kill a process,91if necessary, under the default overcommit 'guess' mode.92 93Systems running under overcommit 'never' should increase this to account94for the full Virtual Memory Size of programs used to recover. Otherwise,95root may not be able to log in to recover the system.96 97How do you calculate a minimum useful reserve?98 99sshd or login + bash (or some other shell) + top (or ps, kill, etc.)100 101For overcommit 'guess', we can sum resident set sizes (RSS).102On x86_64 this is about 8MB.103 104For overcommit 'never', we can take the max of their virtual sizes (VSZ)105and add the sum of their RSS.106On x86_64 this is about 128MB.107 108Changing this takes effect whenever an application requests memory.109 110 111compact_memory112==============113 114Available only when CONFIG_COMPACTION is set. When 1 is written to the file,115all zones are compacted such that free memory is available in contiguous116blocks where possible. This can be important for example in the allocation of117huge pages although processes will also directly compact memory as required.118 119compaction_proactiveness120========================121 122This tunable takes a value in the range [0, 100] with a default value of12320. This tunable determines how aggressively compaction is done in the124background. Write of a non zero value to this tunable will immediately125trigger the proactive compaction. Setting it to 0 disables proactive compaction.126 127Note that compaction has a non-trivial system-wide impact as pages128belonging to different processes are moved around, which could also lead129to latency spikes in unsuspecting applications. The kernel employs130various heuristics to avoid wasting CPU cycles if it detects that131proactive compaction is not being effective.132 133Be careful when setting it to extreme values like 100, as that may134cause excessive background compaction activity.135 136compact_unevictable_allowed137===========================138 139Available only when CONFIG_COMPACTION is set. When set to 1, compaction is140allowed to examine the unevictable lru (mlocked pages) for pages to compact.141This should be used on systems where stalls for minor page faults are an142acceptable trade for large contiguous free memory. Set to 0 to prevent143compaction from moving pages that are unevictable. Default value is 1.144On CONFIG_PREEMPT_RT the default value is 0 in order to avoid a page fault, due145to compaction, which would block the task from becoming active until the fault146is resolved.147 148 149dirty_background_bytes150======================151 152Contains the amount of dirty memory at which the background kernel153flusher threads will start writeback.154 155Note:156 dirty_background_bytes is the counterpart of dirty_background_ratio. Only157 one of them may be specified at a time. When one sysctl is written it is158 immediately taken into account to evaluate the dirty memory limits and the159 other appears as 0 when read.160 161 162dirty_background_ratio163======================164 165Contains, as a percentage of total available memory that contains free pages166and reclaimable pages, the number of pages at which the background kernel167flusher threads will start writing out dirty data.168 169The total available memory is not equal to total system memory.170 171 172dirty_bytes173===========174 175Contains the amount of dirty memory at which a process generating disk writes176will itself start writeback.177 178Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be179specified at a time. When one sysctl is written it is immediately taken into180account to evaluate the dirty memory limits and the other appears as 0 when181read.182 183Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any184value lower than this limit will be ignored and the old configuration will be185retained.186 187 188dirty_expire_centisecs189======================190 191This tunable is used to define when dirty data is old enough to be eligible192for writeout by the kernel flusher threads. It is expressed in 100'ths193of a second. Data which has been dirty in-memory for longer than this194interval will be written out next time a flusher thread wakes up.195 196 197dirty_ratio198===========199 200Contains, as a percentage of total available memory that contains free pages201and reclaimable pages, the number of pages at which a process which is202generating disk writes will itself start writing out dirty data.203 204The total available memory is not equal to total system memory.205 206 207dirtytime_expire_seconds208========================209 210When a lazytime inode is constantly having its pages dirtied, the inode with211an updated timestamp will never get chance to be written out. And, if the212only thing that has happened on the file system is a dirtytime inode caused213by an atime update, a worker will be scheduled to make sure that inode214eventually gets pushed out to disk. This tunable is used to define when dirty215inode is old enough to be eligible for writeback by the kernel flusher threads.216And, it is also used as the interval to wakeup dirtytime_writeback thread.217 218 219dirty_writeback_centisecs220=========================221 222The kernel flusher threads will periodically wake up and write `old` data223out to disk. This tunable expresses the interval between those wakeups, in224100'ths of a second.225 226Setting this to zero disables periodic writeback altogether.227 228 229drop_caches230===========231 232Writing to this will cause the kernel to drop clean caches, as well as233reclaimable slab objects like dentries and inodes. Once dropped, their234memory becomes free.235 236To free pagecache::237 238 echo 1 > /proc/sys/vm/drop_caches239 240To free reclaimable slab objects (includes dentries and inodes)::241 242 echo 2 > /proc/sys/vm/drop_caches243 244To free slab objects and pagecache::245 246 echo 3 > /proc/sys/vm/drop_caches247 248This is a non-destructive operation and will not free any dirty objects.249To increase the number of objects freed by this operation, the user may run250`sync` prior to writing to /proc/sys/vm/drop_caches. This will minimize the251number of dirty objects on the system and create more candidates to be252dropped.253 254This file is not a means to control the growth of the various kernel caches255(inodes, dentries, pagecache, etc...) These objects are automatically256reclaimed by the kernel when memory is needed elsewhere on the system.257 258Use of this file can cause performance problems. Since it discards cached259objects, it may cost a significant amount of I/O and CPU to recreate the260dropped objects, especially if they were under heavy use. Because of this,261use outside of a testing or debugging environment is not recommended.262 263You may see informational messages in your kernel log when this file is264used::265 266 cat (1234): drop_caches: 3267 268These are informational only. They do not mean that anything is wrong269with your system. To disable them, echo 4 (bit 2) into drop_caches.270 271enable_soft_offline272===================273Correctable memory errors are very common on servers. Soft-offline is kernel's274solution for memory pages having (excessive) corrected memory errors.275 276For different types of page, soft-offline has different behaviors / costs.277 278- For a raw error page, soft-offline migrates the in-use page's content to279 a new raw page.280 281- For a page that is part of a transparent hugepage, soft-offline splits the282 transparent hugepage into raw pages, then migrates only the raw error page.283 As a result, user is transparently backed by 1 less hugepage, impacting284 memory access performance.285 286- For a page that is part of a HugeTLB hugepage, soft-offline first migrates287 the entire HugeTLB hugepage, during which a free hugepage will be consumed288 as migration target. Then the original hugepage is dissolved into raw289 pages without compensation, reducing the capacity of the HugeTLB pool by 1.290 291It is user's call to choose between reliability (staying away from fragile292physical memory) vs performance / capacity implications in transparent and293HugeTLB cases.294 295For all architectures, enable_soft_offline controls whether to soft offline296memory pages. When set to 1, kernel attempts to soft offline the pages297whenever it thinks needed. When set to 0, kernel returns EOPNOTSUPP to298the request to soft offline the pages. Its default value is 1.299 300It is worth mentioning that after setting enable_soft_offline to 0, the301following requests to soft offline pages will not be performed:302 303- Request to soft offline pages from RAS Correctable Errors Collector.304 305- On ARM, the request to soft offline pages from GHES driver.306 307- On PARISC, the request to soft offline pages from Page Deallocation Table.308 309extfrag_threshold310=================311 312This parameter affects whether the kernel will compact memory or direct313reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in314debugfs shows what the fragmentation index for each order is in each zone in315the system. Values tending towards 0 imply allocations would fail due to lack316of memory, values towards 1000 imply failures are due to fragmentation and -1317implies that the allocation will succeed as long as watermarks are met.318 319The kernel will not compact memory in a zone if the320fragmentation index is <= extfrag_threshold. The default value is 500.321 322 323highmem_is_dirtyable324====================325 326Available only for systems with CONFIG_HIGHMEM enabled (32b systems).327 328This parameter controls whether the high memory is considered for dirty329writers throttling. This is not the case by default which means that330only the amount of memory directly visible/usable by the kernel can331be dirtied. As a result, on systems with a large amount of memory and332lowmem basically depleted writers might be throttled too early and333streaming writes can get very slow.334 335Changing the value to non zero would allow more memory to be dirtied336and thus allow writers to write more data which can be flushed to the337storage more effectively. Note this also comes with a risk of pre-mature338OOM killer because some writers (e.g. direct block device writes) can339only use the low memory and they can fill it up with dirty data without340any throttling.341 342 343hugetlb_shm_group344=================345 346hugetlb_shm_group contains group id that is allowed to create SysV347shared memory segment using hugetlb page.348 349 350laptop_mode351===========352 353laptop_mode is a knob that controls "laptop mode". All the things that are354controlled by this knob are discussed in Documentation/admin-guide/laptops/laptop-mode.rst.355 356 357legacy_va_layout358================359 360If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel361will use the legacy (2.4) layout for all processes.362 363 364lowmem_reserve_ratio365====================366 367For some specialised workloads on highmem machines it is dangerous for368the kernel to allow process memory to be allocated from the "lowmem"369zone. This is because that memory could then be pinned via the mlock()370system call, or by unavailability of swapspace.371 372And on large highmem machines this lack of reclaimable lowmem memory373can be fatal.374 375So the Linux page allocator has a mechanism which prevents allocations376which *could* use highmem from using too much lowmem. This means that377a certain amount of lowmem is defended from the possibility of being378captured into pinned user memory.379 380(The same argument applies to the old 16 megabyte ISA DMA region. This381mechanism will also defend that region from allocations which could use382highmem or lowmem).383 384The `lowmem_reserve_ratio` tunable determines how aggressive the kernel is385in defending these lower zones.386 387If you have a machine which uses highmem or ISA DMA and your388applications are using mlock(), or if you are running with no swap then389you probably should change the lowmem_reserve_ratio setting.390 391The lowmem_reserve_ratio is an array. You can see them by reading this file::392 393 % cat /proc/sys/vm/lowmem_reserve_ratio394 256 256 32395 396But, these values are not used directly. The kernel calculates # of protection397pages for each zones from them. These are shown as array of protection pages398in /proc/zoneinfo like the following. (This is an example of x86-64 box).399Each zone has an array of protection pages like this::400 401 Node 0, zone DMA402 pages free 1355403 min 3404 low 3405 high 4406 :407 :408 numa_other 0409 protection: (0, 2004, 2004, 2004)410 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^411 pagesets412 cpu: 0 pcp: 0413 :414 415These protections are added to score to judge whether this zone should be used416for page allocation or should be reclaimed.417 418In this example, if normal pages (index=2) are required to this DMA zone and419watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should420not be used because pages_free(1355) is smaller than watermark + protection[2]421(4 + 2004 = 2008). If this protection value is 0, this zone would be used for422normal page requirement. If requirement is DMA zone(index=0), protection[0]423(=0) is used.424 425zone[i]'s protection[j] is calculated by following expression::426 427 (i < j):428 zone[i]->protection[j]429 = (total sums of managed_pages from zone[i+1] to zone[j] on the node)430 / lowmem_reserve_ratio[i];431 (i = j):432 (should not be protected. = 0;433 (i > j):434 (not necessary, but looks 0)435 436The default values of lowmem_reserve_ratio[i] are437 438 === ====================================439 256 (if zone[i] means DMA or DMA32 zone)440 32 (others)441 === ====================================442 443As above expression, they are reciprocal number of ratio.444256 means 1/256. # of protection pages becomes about "0.39%" of total managed445pages of higher zones on the node.446 447If you would like to protect more pages, smaller values are effective.448The minimum value is 1 (1/1 -> 100%). The value less than 1 completely449disables protection of the pages.450 451 452max_map_count:453==============454 455This file contains the maximum number of memory map areas a process456may have. Memory map areas are used as a side-effect of calling457malloc, directly by mmap, mprotect, and madvise, and also when loading458shared libraries.459 460While most applications need less than a thousand maps, certain461programs, particularly malloc debuggers, may consume lots of them,462e.g., up to one or two maps per allocation.463 464The default value is 65530.465 466 467mem_profiling468==============469 470Enable memory profiling (when CONFIG_MEM_ALLOC_PROFILING=y)471 4721: Enable memory profiling.473 4740: Disable memory profiling.475 476Enabling memory profiling introduces a small performance overhead for all477memory allocations.478 479The default value depends on CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT.480 481 482memory_failure_early_kill:483==========================484 485Control how to kill processes when uncorrected memory error (typically486a 2bit error in a memory module) is detected in the background by hardware487that cannot be handled by the kernel. In some cases (like the page488still having a valid copy on disk) the kernel will handle the failure489transparently without affecting any applications. But if there is490no other up-to-date copy of the data it will kill to prevent any data491corruptions from propagating.492 4931: Kill all processes that have the corrupted and not reloadable page mapped494as soon as the corruption is detected. Note this is not supported495for a few types of pages, like kernel internally allocated data or496the swap cache, but works for the majority of user pages.497 4980: Only unmap the corrupted page from all processes and only kill a process499who tries to access it.500 501The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can502handle this if they want to.503 504This is only active on architectures/platforms with advanced machine505check handling and depends on the hardware capabilities.506 507Applications can override this setting individually with the PR_MCE_KILL prctl508 509 510memory_failure_recovery511=======================512 513Enable memory failure recovery (when supported by the platform)514 5151: Attempt recovery.516 5170: Always panic on a memory failure.518 519 520min_free_kbytes521===============522 523This is used to force the Linux VM to keep a minimum number524of kilobytes free. The VM uses this number to compute a525watermark[WMARK_MIN] value for each lowmem zone in the system.526Each lowmem zone gets a number of reserved free pages based527proportionally on its size.528 529Some minimal amount of memory is needed to satisfy PF_MEMALLOC530allocations; if you set this to lower than 1024KB, your system will531become subtly broken, and prone to deadlock under high loads.532 533Setting this too high will OOM your machine instantly.534 535 536min_slab_ratio537==============538 539This is available only on NUMA kernels.540 541A percentage of the total pages in each zone. On Zone reclaim542(fallback from the local zone occurs) slabs will be reclaimed if more543than this percentage of pages in a zone are reclaimable slab pages.544This insures that the slab growth stays under control even in NUMA545systems that rarely perform global reclaim.546 547The default is 5 percent.548 549Note that slab reclaim is triggered in a per zone / node fashion.550The process of reclaiming slab memory is currently not node specific551and may not be fast.552 553 554min_unmapped_ratio555==================556 557This is available only on NUMA kernels.558 559This is a percentage of the total pages in each zone. Zone reclaim will560only occur if more than this percentage of pages are in a state that561zone_reclaim_mode allows to be reclaimed.562 563If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared564against all file-backed unmapped pages including swapcache pages and tmpfs565files. Otherwise, only unmapped pages backed by normal files but not tmpfs566files and similar are considered.567 568The default is 1 percent.569 570 571mmap_min_addr572=============573 574This file indicates the amount of address space which a user process will575be restricted from mmapping. Since kernel null dereference bugs could576accidentally operate based on the information in the first couple of pages577of memory userspace processes should not be allowed to write to them. By578default this value is set to 0 and no protections will be enforced by the579security module. Setting this value to something like 64k will allow the580vast majority of applications to work correctly and provide defense in depth581against future potential kernel bugs.582 583 584mmap_rnd_bits585=============586 587This value can be used to select the number of bits to use to588determine the random offset to the base address of vma regions589resulting from mmap allocations on architectures which support590tuning address space randomization. This value will be bounded591by the architecture's minimum and maximum supported values.592 593This value can be changed after boot using the594/proc/sys/vm/mmap_rnd_bits tunable595 596 597mmap_rnd_compat_bits598====================599 600This value can be used to select the number of bits to use to601determine the random offset to the base address of vma regions602resulting from mmap allocations for applications run in603compatibility mode on architectures which support tuning address604space randomization. This value will be bounded by the605architecture's minimum and maximum supported values.606 607This value can be changed after boot using the608/proc/sys/vm/mmap_rnd_compat_bits tunable609 610 611nr_hugepages612============613 614Change the minimum size of the hugepage pool.615 616See Documentation/admin-guide/mm/hugetlbpage.rst617 618 619hugetlb_optimize_vmemmap620========================621 622This knob is not available when the size of 'struct page' (a structure defined623in include/linux/mm_types.h) is not power of two (an unusual system config could624result in this).625 626Enable (set to 1) or disable (set to 0) HugeTLB Vmemmap Optimization (HVO).627 628Once enabled, the vmemmap pages of subsequent allocation of HugeTLB pages from629buddy allocator will be optimized (7 pages per 2MB HugeTLB page and 4095 pages630per 1GB HugeTLB page), whereas already allocated HugeTLB pages will not be631optimized. When those optimized HugeTLB pages are freed from the HugeTLB pool632to the buddy allocator, the vmemmap pages representing that range needs to be633remapped again and the vmemmap pages discarded earlier need to be rellocated634again. If your use case is that HugeTLB pages are allocated 'on the fly' (e.g.635never explicitly allocating HugeTLB pages with 'nr_hugepages' but only set636'nr_overcommit_hugepages', those overcommitted HugeTLB pages are allocated 'on637the fly') instead of being pulled from the HugeTLB pool, you should weigh the638benefits of memory savings against the more overhead (~2x slower than before)639of allocation or freeing HugeTLB pages between the HugeTLB pool and the buddy640allocator. Another behavior to note is that if the system is under heavy memory641pressure, it could prevent the user from freeing HugeTLB pages from the HugeTLB642pool to the buddy allocator since the allocation of vmemmap pages could be643failed, you have to retry later if your system encounter this situation.644 645Once disabled, the vmemmap pages of subsequent allocation of HugeTLB pages from646buddy allocator will not be optimized meaning the extra overhead at allocation647time from buddy allocator disappears, whereas already optimized HugeTLB pages648will not be affected. If you want to make sure there are no optimized HugeTLB649pages, you can set "nr_hugepages" to 0 first and then disable this. Note that650writing 0 to nr_hugepages will make any "in use" HugeTLB pages become surplus651pages. So, those surplus pages are still optimized until they are no longer652in use. You would need to wait for those surplus pages to be released before653there are no optimized pages in the system.654 655 656nr_hugepages_mempolicy657======================658 659Change the size of the hugepage pool at run-time on a specific660set of NUMA nodes.661 662See Documentation/admin-guide/mm/hugetlbpage.rst663 664 665nr_overcommit_hugepages666=======================667 668Change the maximum size of the hugepage pool. The maximum is669nr_hugepages + nr_overcommit_hugepages.670 671See Documentation/admin-guide/mm/hugetlbpage.rst672 673 674nr_trim_pages675=============676 677This is available only on NOMMU kernels.678 679This value adjusts the excess page trimming behaviour of power-of-2 aligned680NOMMU mmap allocations.681 682A value of 0 disables trimming of allocations entirely, while a value of 1683trims excess pages aggressively. Any value >= 1 acts as the watermark where684trimming of allocations is initiated.685 686The default value is 1.687 688See Documentation/admin-guide/mm/nommu-mmap.rst for more information.689 690 691numa_zonelist_order692===================693 694This sysctl is only for NUMA and it is deprecated. Anything but695Node order will fail!696 697'where the memory is allocated from' is controlled by zonelists.698 699(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.700you may be able to read ZONE_DMA as ZONE_DMA32...)701 702In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.703ZONE_NORMAL -> ZONE_DMA704This means that a memory allocation request for GFP_KERNEL will705get memory from ZONE_DMA only when ZONE_NORMAL is not available.706 707In NUMA case, you can think of following 2 types of order.708Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL::709 710 (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL711 (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.712 713Type(A) offers the best locality for processes on Node(0), but ZONE_DMA714will be used before ZONE_NORMAL exhaustion. This increases possibility of715out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.716 717Type(B) cannot offer the best locality but is more robust against OOM of718the DMA zone.719 720Type(A) is called as "Node" order. Type (B) is "Zone" order.721 722"Node order" orders the zonelists by node, then by zone within each node.723Specify "[Nn]ode" for node order724 725"Zone Order" orders the zonelists by zone type, then by node within each726zone. Specify "[Zz]one" for zone order.727 728Specify "[Dd]efault" to request automatic configuration.729 730On 32-bit, the Normal zone needs to be preserved for allocations accessible731by the kernel, so "zone" order will be selected.732 733On 64-bit, devices that require DMA32/DMA are relatively rare, so "node"734order will be selected.735 736Default order is recommended unless this is causing problems for your737system/application.738 739 740oom_dump_tasks741==============742 743Enables a system-wide task dump (excluding kernel threads) to be produced744when the kernel performs an OOM-killing and includes such information as745pid, uid, tgid, vm size, rss, pgtables_bytes, swapents, oom_score_adj746score, and name. This is helpful to determine why the OOM killer was747invoked, to identify the rogue task that caused it, and to determine why748the OOM killer chose the task it did to kill.749 750If this is set to zero, this information is suppressed. On very751large systems with thousands of tasks it may not be feasible to dump752the memory state information for each one. Such systems should not753be forced to incur a performance penalty in OOM conditions when the754information may not be desired.755 756If this is set to non-zero, this information is shown whenever the757OOM killer actually kills a memory-hogging task.758 759The default value is 1 (enabled).760 761 762oom_kill_allocating_task763========================764 765This enables or disables killing the OOM-triggering task in766out-of-memory situations.767 768If this is set to zero, the OOM killer will scan through the entire769tasklist and select a task based on heuristics to kill. This normally770selects a rogue memory-hogging task that frees up a large amount of771memory when killed.772 773If this is set to non-zero, the OOM killer simply kills the task that774triggered the out-of-memory condition. This avoids the expensive775tasklist scan.776 777If panic_on_oom is selected, it takes precedence over whatever value778is used in oom_kill_allocating_task.779 780The default value is 0.781 782 783overcommit_kbytes784=================785 786When overcommit_memory is set to 2, the committed address space is not787permitted to exceed swap plus this amount of physical RAM. See below.788 789Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one790of them may be specified at a time. Setting one disables the other (which791then appears as 0 when read).792 793 794overcommit_memory795=================796 797This value contains a flag that enables memory overcommitment.798 799When this flag is 0, the kernel compares the userspace memory request800size against total memory plus swap and rejects obvious overcommits.801 802When this flag is 1, the kernel pretends there is always enough803memory until it actually runs out.804 805When this flag is 2, the kernel uses a "never overcommit"806policy that attempts to prevent any overcommit of memory.807Note that user_reserve_kbytes affects this policy.808 809This feature can be very useful because there are a lot of810programs that malloc() huge amounts of memory "just-in-case"811and don't use much of it.812 813The default value is 0.814 815See Documentation/mm/overcommit-accounting.rst and816mm/util.c::__vm_enough_memory() for more information.817 818 819overcommit_ratio820================821 822When overcommit_memory is set to 2, the committed address823space is not permitted to exceed swap plus this percentage824of physical RAM. See above.825 826 827page-cluster828============829 830page-cluster controls the number of pages up to which consecutive pages831are read in from swap in a single attempt. This is the swap counterpart832to page cache readahead.833The mentioned consecutivity is not in terms of virtual/physical addresses,834but consecutive on swap space - that means they were swapped out together.835 836It is a logarithmic value - setting it to zero means "1 page", setting837it to 1 means "2 pages", setting it to 2 means "4 pages", etc.838Zero disables swap readahead completely.839 840The default value is three (eight pages at a time). There may be some841small benefits in tuning this to a different value if your workload is842swap-intensive.843 844Lower values mean lower latencies for initial faults, but at the same time845extra faults and I/O delays for following faults if they would have been part of846that consecutive pages readahead would have brought in.847 848 849page_lock_unfairness850====================851 852This value determines the number of times that the page lock can be853stolen from under a waiter. After the lock is stolen the number of times854specified in this file (default is 5), the "fair lock handoff" semantics855will apply, and the waiter will only be awakened if the lock can be taken.856 857panic_on_oom858============859 860This enables or disables panic on out-of-memory feature.861 862If this is set to 0, the kernel will kill some rogue process,863called oom_killer. Usually, oom_killer can kill rogue processes and864system will survive.865 866If this is set to 1, the kernel panics when out-of-memory happens.867However, if a process limits using nodes by mempolicy/cpusets,868and those nodes become memory exhaustion status, one process869may be killed by oom-killer. No panic occurs in this case.870Because other nodes' memory may be free. This means system total status871may be not fatal yet.872 873If this is set to 2, the kernel panics compulsorily even on the874above-mentioned. Even oom happens under memory cgroup, the whole875system panics.876 877The default value is 0.878 8791 and 2 are for failover of clustering. Please select either880according to your policy of failover.881 882panic_on_oom=2+kdump gives you very strong tool to investigate883why oom happens. You can get snapshot.884 885 886percpu_pagelist_high_fraction887=============================888 889This is the fraction of pages in each zone that are can be stored to890per-cpu page lists. It is an upper boundary that is divided depending891on the number of online CPUs. The min value for this is 8 which means892that we do not allow more than 1/8th of pages in each zone to be stored893on per-cpu page lists. This entry only changes the value of hot per-cpu894page lists. A user can specify a number like 100 to allocate 1/100th of895each zone between per-cpu lists.896 897The batch value of each per-cpu page list remains the same regardless of898the value of the high fraction so allocation latencies are unaffected.899 900The initial value is zero. Kernel uses this value to set the high pcp->high901mark based on the low watermark for the zone and the number of local902online CPUs. If the user writes '0' to this sysctl, it will revert to903this default behavior.904 905 906stat_interval907=============908 909The time interval between which vm statistics are updated. The default910is 1 second.911 912 913stat_refresh914============915 916Any read or write (by root only) flushes all the per-cpu vm statistics917into their global totals, for more accurate reports when testing918e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo919 920As a side-effect, it also checks for negative totals (elsewhere reported921as 0) and "fails" with EINVAL if any are found, with a warning in dmesg.922(At time of writing, a few stats are known sometimes to be found negative,923with no ill effects: errors and warnings on these stats are suppressed.)924 925 926numa_stat927=========928 929This interface allows runtime configuration of numa statistics.930 931When page allocation performance becomes a bottleneck and you can tolerate932some possible tool breakage and decreased numa counter precision, you can933do::934 935 echo 0 > /proc/sys/vm/numa_stat936 937When page allocation performance is not a bottleneck and you want all938tooling to work, you can do::939 940 echo 1 > /proc/sys/vm/numa_stat941 942 943swappiness944==========945 946This control is used to define the rough relative IO cost of swapping947and filesystem paging, as a value between 0 and 200. At 100, the VM948assumes equal IO cost and will thus apply memory pressure to the page949cache and swap-backed pages equally; lower values signify more950expensive swap IO, higher values indicates cheaper.951 952Keep in mind that filesystem IO patterns under memory pressure tend to953be more efficient than swap's random IO. An optimal value will require954experimentation and will also be workload-dependent.955 956The default value is 60.957 958For in-memory swap, like zram or zswap, as well as hybrid setups that959have swap on faster devices than the filesystem, values beyond 100 can960be considered. For example, if the random IO against the swap device961is on average 2x faster than IO from the filesystem, swappiness should962be 133 (x + 2x = 200, 2x = 133.33).963 964At 0, the kernel will not initiate swap until the amount of free and965file-backed pages is less than the high watermark in a zone.966 967 968unprivileged_userfaultfd969========================970 971This flag controls the mode in which unprivileged users can use the972userfaultfd system calls. Set this to 0 to restrict unprivileged users973to handle page faults in user mode only. In this case, users without974SYS_CAP_PTRACE must pass UFFD_USER_MODE_ONLY in order for userfaultfd to975succeed. Prohibiting use of userfaultfd for handling faults from kernel976mode may make certain vulnerabilities more difficult to exploit.977 978Set this to 1 to allow unprivileged users to use the userfaultfd system979calls without any restrictions.980 981The default value is 0.982 983Another way to control permissions for userfaultfd is to use984/dev/userfaultfd instead of userfaultfd(2). See985Documentation/admin-guide/mm/userfaultfd.rst.986 987user_reserve_kbytes988===================989 990When overcommit_memory is set to 2, "never overcommit" mode, reserve991min(3% of current process size, user_reserve_kbytes) of free memory.992This is intended to prevent a user from starting a single memory hogging993process, such that they cannot recover (kill the hog).994 995user_reserve_kbytes defaults to min(3% of the current process size, 128MB).996 997If this is reduced to zero, then the user will be allowed to allocate998all free memory with a single process, minus admin_reserve_kbytes.999Any subsequent attempts to execute a command will result in1000"fork: Cannot allocate memory".1001 1002Changing this takes effect whenever an application requests memory.1003 1004 1005vfs_cache_pressure1006==================1007 1008This percentage value controls the tendency of the kernel to reclaim1009the memory which is used for caching of directory and inode objects.1010 1011At the default value of vfs_cache_pressure=100 the kernel will attempt to1012reclaim dentries and inodes at a "fair" rate with respect to pagecache and1013swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer1014to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will1015never reclaim dentries and inodes due to memory pressure and this can easily1016lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 1001017causes the kernel to prefer to reclaim dentries and inodes.1018 1019Increasing vfs_cache_pressure significantly beyond 100 may have negative1020performance impact. Reclaim code needs to take various locks to find freeable1021directory and inode objects. With vfs_cache_pressure=1000, it will look for1022ten times more freeable objects than there are.1023 1024 1025watermark_boost_factor1026======================1027 1028This factor controls the level of reclaim when memory is being fragmented.1029It defines the percentage of the high watermark of a zone that will be1030reclaimed if pages of different mobility are being mixed within pageblocks.1031The intent is that compaction has less work to do in the future and to1032increase the success rate of future high-order allocations such as SLUB1033allocations, THP and hugetlbfs pages.1034 1035To make it sensible with respect to the watermark_scale_factor1036parameter, the unit is in fractions of 10,000. The default value of103715,000 means that up to 150% of the high watermark will be reclaimed in the1038event of a pageblock being mixed due to fragmentation. The level of reclaim1039is determined by the number of fragmentation events that occurred in the1040recent past. If this value is smaller than a pageblock then a pageblocks1041worth of pages will be reclaimed (e.g. 2MB on 64-bit x86). A boost factor1042of 0 will disable the feature.1043 1044 1045watermark_scale_factor1046======================1047 1048This factor controls the aggressiveness of kswapd. It defines the1049amount of memory left in a node/system before kswapd is woken up and1050how much memory needs to be free before kswapd goes back to sleep.1051 1052The unit is in fractions of 10,000. The default value of 10 means the1053distances between watermarks are 0.1% of the available memory in the1054node/system. The maximum value is 3000, or 30% of memory.1055 1056A high rate of threads entering direct reclaim (allocstall) or kswapd1057going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate1058that the number of free pages kswapd maintains for latency reasons is1059too small for the allocation bursts occurring in the system. This knob1060can then be used to tune kswapd aggressiveness accordingly.1061 1062 1063zone_reclaim_mode1064=================1065 1066Zone_reclaim_mode allows someone to set more or less aggressive approaches to1067reclaim memory when a zone runs out of memory. If it is set to zero then no1068zone reclaim occurs. Allocations will be satisfied from other zones / nodes1069in the system.1070 1071This is value OR'ed together of1072 1073= ===================================10741 Zone reclaim on10752 Zone reclaim writes dirty pages out10764 Zone reclaim swaps pages1077= ===================================1078 1079zone_reclaim_mode is disabled by default. For file servers or workloads1080that benefit from having their data cached, zone_reclaim_mode should be1081left disabled as the caching effect is likely to be more important than1082data locality.1083 1084Consider enabling one or more zone_reclaim mode bits if it's known that the1085workload is partitioned such that each partition fits within a NUMA node1086and that accessing remote memory would cause a measurable performance1087reduction. The page allocator will take additional actions before1088allocating off node pages.1089 1090Allowing zone reclaim to write out pages stops processes that are1091writing large amounts of data from dirtying pages on other nodes. Zone1092reclaim will write out dirty pages if a zone fills up and so effectively1093throttle the process. This may decrease the performance of a single process1094since it cannot use all of system memory to buffer the outgoing writes1095anymore but it preserve the memory on other nodes so that the performance1096of other processes running on other nodes will not be affected.1097 1098Allowing regular swap effectively restricts allocations to the local1099node unless explicitly overridden by memory policies or cpuset1100configurations.1101