1=============================== 2Documentation for /proc/sys/vm/ 3=============================== 4 5kernel version 2.6.29 6 7Copyright (c) 1998, 1999, Rik van Riel <riel@nl.linux.org> 8 9Copyright (c) 2008 Peter W. Morreale <pmorreale@novell.com> 10 11For general info and legal blurb, please look in index.rst. 12 13------------------------------------------------------------------------------ 14 15This file contains the documentation for the sysctl files in 16/proc/sys/vm and is valid for Linux kernel version 2.6.29. 17 18The files in this directory can be used to tune the operation 19of the virtual memory (VM) subsystem of the Linux kernel and 20the writeout of dirty data to disk. 21 22Default values and initialization routines for most of these 23files can be found in mm/swap.c. 24 25Currently, these files are in /proc/sys/vm: 26 27- admin_reserve_kbytes 28- block_dump 29- compact_memory 30- compaction_proactiveness 31- compact_unevictable_allowed 32- dirty_background_bytes 33- dirty_background_ratio 34- dirty_bytes 35- dirty_expire_centisecs 36- dirty_ratio 37- dirtytime_expire_seconds 38- dirty_writeback_centisecs 39- drop_caches 40- extfrag_threshold 41- highmem_is_dirtyable 42- hugetlb_shm_group 43- laptop_mode 44- legacy_va_layout 45- lowmem_reserve_ratio 46- max_map_count 47- memory_failure_early_kill 48- memory_failure_recovery 49- min_free_kbytes 50- min_slab_ratio 51- min_unmapped_ratio 52- mmap_min_addr 53- mmap_rnd_bits 54- mmap_rnd_compat_bits 55- nr_hugepages 56- nr_hugepages_mempolicy 57- nr_overcommit_hugepages 58- nr_trim_pages (only if CONFIG_MMU=n) 59- numa_zonelist_order 60- oom_dump_tasks 61- oom_kill_allocating_task 62- overcommit_kbytes 63- overcommit_memory 64- overcommit_ratio 65- page-cluster 66- panic_on_oom 67- percpu_pagelist_fraction 68- stat_interval 69- stat_refresh 70- numa_stat 71- swappiness 72- unprivileged_userfaultfd 73- user_reserve_kbytes 74- vfs_cache_pressure 75- watermark_boost_factor 76- watermark_scale_factor 77- zone_reclaim_mode 78 79 80admin_reserve_kbytes 81==================== 82 83The amount of free memory in the system that should be reserved for users 84with the capability cap_sys_admin. 85 86admin_reserve_kbytes defaults to min(3% of free pages, 8MB) 87 88That should provide enough for the admin to log in and kill a process, 89if necessary, under the default overcommit 'guess' mode. 90 91Systems running under overcommit 'never' should increase this to account 92for the full Virtual Memory Size of programs used to recover. Otherwise, 93root may not be able to log in to recover the system. 94 95How do you calculate a minimum useful reserve? 96 97sshd or login + bash (or some other shell) + top (or ps, kill, etc.) 98 99For overcommit 'guess', we can sum resident set sizes (RSS). 100On x86_64 this is about 8MB. 101 102For overcommit 'never', we can take the max of their virtual sizes (VSZ) 103and add the sum of their RSS. 104On x86_64 this is about 128MB. 105 106Changing this takes effect whenever an application requests memory. 107 108 109block_dump 110========== 111 112block_dump enables block I/O debugging when set to a nonzero value. More 113information on block I/O debugging is in Documentation/admin-guide/laptops/laptop-mode.rst. 114 115 116compact_memory 117============== 118 119Available only when CONFIG_COMPACTION is set. When 1 is written to the file, 120all zones are compacted such that free memory is available in contiguous 121blocks where possible. This can be important for example in the allocation of 122huge pages although processes will also directly compact memory as required. 123 124compaction_proactiveness 125======================== 126 127This tunable takes a value in the range [0, 100] with a default value of 12820. This tunable determines how aggressively compaction is done in the 129background. Setting it to 0 disables proactive compaction. 130 131Note that compaction has a non-trivial system-wide impact as pages 132belonging to different processes are moved around, which could also lead 133to latency spikes in unsuspecting applications. The kernel employs 134various heuristics to avoid wasting CPU cycles if it detects that 135proactive compaction is not being effective. 136 137Be careful when setting it to extreme values like 100, as that may 138cause excessive background compaction activity. 139 140compact_unevictable_allowed 141=========================== 142 143Available only when CONFIG_COMPACTION is set. When set to 1, compaction is 144allowed to examine the unevictable lru (mlocked pages) for pages to compact. 145This should be used on systems where stalls for minor page faults are an 146acceptable trade for large contiguous free memory. Set to 0 to prevent 147compaction from moving pages that are unevictable. Default value is 1. 148On CONFIG_PREEMPT_RT the default value is 0 in order to avoid a page fault, due 149to compaction, which would block the task from becoming active until the fault 150is resolved. 151 152 153dirty_background_bytes 154====================== 155 156Contains the amount of dirty memory at which the background kernel 157flusher threads will start writeback. 158 159Note: 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. 164 165 166dirty_background_ratio 167====================== 168 169Contains, as a percentage of total available memory that contains free pages 170and reclaimable pages, the number of pages at which the background kernel 171flusher threads will start writing out dirty data. 172 173The total available memory is not equal to total system memory. 174 175 176dirty_bytes 177=========== 178 179Contains the amount of dirty memory at which a process generating disk writes 180will itself start writeback. 181 182Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be 183specified at a time. When one sysctl is written it is immediately taken into 184account to evaluate the dirty memory limits and the other appears as 0 when 185read. 186 187Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any 188value lower than this limit will be ignored and the old configuration will be 189retained. 190 191 192dirty_expire_centisecs 193====================== 194 195This tunable is used to define when dirty data is old enough to be eligible 196for writeout by the kernel flusher threads. It is expressed in 100'ths 197of a second. Data which has been dirty in-memory for longer than this 198interval will be written out next time a flusher thread wakes up. 199 200 201dirty_ratio 202=========== 203 204Contains, as a percentage of total available memory that contains free pages 205and reclaimable pages, the number of pages at which a process which is 206generating disk writes will itself start writing out dirty data. 207 208The total available memory is not equal to total system memory. 209 210 211dirtytime_expire_seconds 212======================== 213 214When a lazytime inode is constantly having its pages dirtied, the inode with 215an updated timestamp will never get chance to be written out. And, if the 216only thing that has happened on the file system is a dirtytime inode caused 217by an atime update, a worker will be scheduled to make sure that inode 218eventually gets pushed out to disk. This tunable is used to define when dirty 219inode is old enough to be eligible for writeback by the kernel flusher threads. 220And, it is also used as the interval to wakeup dirtytime_writeback thread. 221 222 223dirty_writeback_centisecs 224========================= 225 226The kernel flusher threads will periodically wake up and write `old` data 227out to disk. This tunable expresses the interval between those wakeups, in 228100'ths of a second. 229 230Setting this to zero disables periodic writeback altogether. 231 232 233drop_caches 234=========== 235 236Writing to this will cause the kernel to drop clean caches, as well as 237reclaimable slab objects like dentries and inodes. Once dropped, their 238memory becomes free. 239 240To free pagecache:: 241 242 echo 1 > /proc/sys/vm/drop_caches 243 244To free reclaimable slab objects (includes dentries and inodes):: 245 246 echo 2 > /proc/sys/vm/drop_caches 247 248To free slab objects and pagecache:: 249 250 echo 3 > /proc/sys/vm/drop_caches 251 252This is a non-destructive operation and will not free any dirty objects. 253To 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 255number of dirty objects on the system and create more candidates to be 256dropped. 257 258This file is not a means to control the growth of the various kernel caches 259(inodes, dentries, pagecache, etc...) These objects are automatically 260reclaimed by the kernel when memory is needed elsewhere on the system. 261 262Use of this file can cause performance problems. Since it discards cached 263objects, it may cost a significant amount of I/O and CPU to recreate the 264dropped objects, especially if they were under heavy use. Because of this, 265use outside of a testing or debugging environment is not recommended. 266 267You may see informational messages in your kernel log when this file is 268used:: 269 270 cat (1234): drop_caches: 3 271 272These are informational only. They do not mean that anything is wrong 273with your system. To disable them, echo 4 (bit 2) into drop_caches. 274 275 276extfrag_threshold 277================= 278 279This parameter affects whether the kernel will compact memory or direct 280reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in 281debugfs shows what the fragmentation index for each order is in each zone in 282the system. Values tending towards 0 imply allocations would fail due to lack 283of memory, values towards 1000 imply failures are due to fragmentation and -1 284implies that the allocation will succeed as long as watermarks are met. 285 286The kernel will not compact memory in a zone if the 287fragmentation index is <= extfrag_threshold. The default value is 500. 288 289 290highmem_is_dirtyable 291==================== 292 293Available only for systems with CONFIG_HIGHMEM enabled (32b systems). 294 295This parameter controls whether the high memory is considered for dirty 296writers throttling. This is not the case by default which means that 297only the amount of memory directly visible/usable by the kernel can 298be dirtied. As a result, on systems with a large amount of memory and 299lowmem basically depleted writers might be throttled too early and 300streaming writes can get very slow. 301 302Changing the value to non zero would allow more memory to be dirtied 303and thus allow writers to write more data which can be flushed to the 304storage more effectively. Note this also comes with a risk of pre-mature 305OOM killer because some writers (e.g. direct block device writes) can 306only use the low memory and they can fill it up with dirty data without 307any throttling. 308 309 310hugetlb_shm_group 311================= 312 313hugetlb_shm_group contains group id that is allowed to create SysV 314shared memory segment using hugetlb page. 315 316 317laptop_mode 318=========== 319 320laptop_mode is a knob that controls "laptop mode". All the things that are 321controlled by this knob are discussed in Documentation/admin-guide/laptops/laptop-mode.rst. 322 323 324legacy_va_layout 325================ 326 327If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel 328will use the legacy (2.4) layout for all processes. 329 330 331lowmem_reserve_ratio 332==================== 333 334For some specialised workloads on highmem machines it is dangerous for 335the kernel to allow process memory to be allocated from the "lowmem" 336zone. This is because that memory could then be pinned via the mlock() 337system call, or by unavailability of swapspace. 338 339And on large highmem machines this lack of reclaimable lowmem memory 340can be fatal. 341 342So the Linux page allocator has a mechanism which prevents allocations 343which *could* use highmem from using too much lowmem. This means that 344a certain amount of lowmem is defended from the possibility of being 345captured into pinned user memory. 346 347(The same argument applies to the old 16 megabyte ISA DMA region. This 348mechanism will also defend that region from allocations which could use 349highmem or lowmem). 350 351The `lowmem_reserve_ratio` tunable determines how aggressive the kernel is 352in defending these lower zones. 353 354If you have a machine which uses highmem or ISA DMA and your 355applications are using mlock(), or if you are running with no swap then 356you probably should change the lowmem_reserve_ratio setting. 357 358The lowmem_reserve_ratio is an array. You can see them by reading this file:: 359 360 % cat /proc/sys/vm/lowmem_reserve_ratio 361 256 256 32 362 363But, these values are not used directly. The kernel calculates # of protection 364pages for each zones from them. These are shown as array of protection pages 365in /proc/zoneinfo like followings. (This is an example of x86-64 box). 366Each zone has an array of protection pages like this:: 367 368 Node 0, zone DMA 369 pages free 1355 370 min 3 371 low 3 372 high 4 373 : 374 : 375 numa_other 0 376 protection: (0, 2004, 2004, 2004) 377 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 378 pagesets 379 cpu: 0 pcp: 0 380 : 381 382These protections are added to score to judge whether this zone should be used 383for page allocation or should be reclaimed. 384 385In this example, if normal pages (index=2) are required to this DMA zone and 386watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should 387not 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 389normal page requirement. If requirement is DMA zone(index=0), protection[0] 390(=0) is used. 391 392zone[i]'s protection[j] is calculated by following expression:: 393 394 (i < j): 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]; 398 (i = j): 399 (should not be protected. = 0; 400 (i > j): 401 (not necessary, but looks 0) 402 403The default values of lowmem_reserve_ratio[i] are 404 405 === ==================================== 406 256 (if zone[i] means DMA or DMA32 zone) 407 32 (others) 408 === ==================================== 409 410As above expression, they are reciprocal number of ratio. 411256 means 1/256. # of protection pages becomes about "0.39%" of total managed 412pages of higher zones on the node. 413 414If you would like to protect more pages, smaller values are effective. 415The minimum value is 1 (1/1 -> 100%). The value less than 1 completely 416disables protection of the pages. 417 418 419max_map_count: 420============== 421 422This file contains the maximum number of memory map areas a process 423may have. Memory map areas are used as a side-effect of calling 424malloc, directly by mmap, mprotect, and madvise, and also when loading 425shared libraries. 426 427While most applications need less than a thousand maps, certain 428programs, particularly malloc debuggers, may consume lots of them, 429e.g., up to one or two maps per allocation. 430 431The default value is 65530. 432 433 434memory_failure_early_kill: 435========================== 436 437Control how to kill processes when uncorrected memory error (typically 438a 2bit error in a memory module) is detected in the background by hardware 439that cannot be handled by the kernel. In some cases (like the page 440still having a valid copy on disk) the kernel will handle the failure 441transparently without affecting any applications. But if there is 442no other uptodate copy of the data it will kill to prevent any data 443corruptions from propagating. 444 4451: Kill all processes that have the corrupted and not reloadable page mapped 446as soon as the corruption is detected. Note this is not supported 447for a few types of pages, like kernel internally allocated data or 448the swap cache, but works for the majority of user pages. 449 4500: Only unmap the corrupted page from all processes and only kill a process 451who tries to access it. 452 453The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can 454handle this if they want to. 455 456This is only active on architectures/platforms with advanced machine 457check handling and depends on the hardware capabilities. 458 459Applications can override this setting individually with the PR_MCE_KILL prctl 460 461 462memory_failure_recovery 463======================= 464 465Enable memory failure recovery (when supported by the platform) 466 4671: Attempt recovery. 468 4690: Always panic on a memory failure. 470 471 472min_free_kbytes 473=============== 474 475This is used to force the Linux VM to keep a minimum number 476of kilobytes free. The VM uses this number to compute a 477watermark[WMARK_MIN] value for each lowmem zone in the system. 478Each lowmem zone gets a number of reserved free pages based 479proportionally on its size. 480 481Some minimal amount of memory is needed to satisfy PF_MEMALLOC 482allocations; if you set this to lower than 1024KB, your system will 483become subtly broken, and prone to deadlock under high loads. 484 485Setting this too high will OOM your machine instantly. 486 487 488min_slab_ratio 489============== 490 491This is available only on NUMA kernels. 492 493A percentage of the total pages in each zone. On Zone reclaim 494(fallback from the local zone occurs) slabs will be reclaimed if more 495than this percentage of pages in a zone are reclaimable slab pages. 496This insures that the slab growth stays under control even in NUMA 497systems that rarely perform global reclaim. 498 499The default is 5 percent. 500 501Note that slab reclaim is triggered in a per zone / node fashion. 502The process of reclaiming slab memory is currently not node specific 503and may not be fast. 504 505 506min_unmapped_ratio 507================== 508 509This is available only on NUMA kernels. 510 511This is a percentage of the total pages in each zone. Zone reclaim will 512only occur if more than this percentage of pages are in a state that 513zone_reclaim_mode allows to be reclaimed. 514 515If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared 516against all file-backed unmapped pages including swapcache pages and tmpfs 517files. Otherwise, only unmapped pages backed by normal files but not tmpfs 518files and similar are considered. 519 520The default is 1 percent. 521 522 523mmap_min_addr 524============= 525 526This file indicates the amount of address space which a user process will 527be restricted from mmapping. Since kernel null dereference bugs could 528accidentally operate based on the information in the first couple of pages 529of memory userspace processes should not be allowed to write to them. By 530default this value is set to 0 and no protections will be enforced by the 531security module. Setting this value to something like 64k will allow the 532vast majority of applications to work correctly and provide defense in depth 533against future potential kernel bugs. 534 535 536mmap_rnd_bits 537============= 538 539This value can be used to select the number of bits to use to 540determine the random offset to the base address of vma regions 541resulting from mmap allocations on architectures which support 542tuning address space randomization. This value will be bounded 543by the architecture's minimum and maximum supported values. 544 545This value can be changed after boot using the 546/proc/sys/vm/mmap_rnd_bits tunable 547 548 549mmap_rnd_compat_bits 550==================== 551 552This value can be used to select the number of bits to use to 553determine the random offset to the base address of vma regions 554resulting from mmap allocations for applications run in 555compatibility mode on architectures which support tuning address 556space randomization. This value will be bounded by the 557architecture's minimum and maximum supported values. 558 559This value can be changed after boot using the 560/proc/sys/vm/mmap_rnd_compat_bits tunable 561 562 563nr_hugepages 564============ 565 566Change the minimum size of the hugepage pool. 567 568See Documentation/admin-guide/mm/hugetlbpage.rst 569 570 571nr_hugepages_mempolicy 572====================== 573 574Change the size of the hugepage pool at run-time on a specific 575set of NUMA nodes. 576 577See Documentation/admin-guide/mm/hugetlbpage.rst 578 579 580nr_overcommit_hugepages 581======================= 582 583Change the maximum size of the hugepage pool. The maximum is 584nr_hugepages + nr_overcommit_hugepages. 585 586See Documentation/admin-guide/mm/hugetlbpage.rst 587 588 589nr_trim_pages 590============= 591 592This is available only on NOMMU kernels. 593 594This value adjusts the excess page trimming behaviour of power-of-2 aligned 595NOMMU mmap allocations. 596 597A value of 0 disables trimming of allocations entirely, while a value of 1 598trims excess pages aggressively. Any value >= 1 acts as the watermark where 599trimming of allocations is initiated. 600 601The default value is 1. 602 603See Documentation/admin-guide/mm/nommu-mmap.rst for more information. 604 605 606numa_zonelist_order 607=================== 608 609This sysctl is only for NUMA and it is deprecated. Anything but 610Node order will fail! 611 612'where the memory is allocated from' is controlled by zonelists. 613 614(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation. 615you may be able to read ZONE_DMA as ZONE_DMA32...) 616 617In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following. 618ZONE_NORMAL -> ZONE_DMA 619This means that a memory allocation request for GFP_KERNEL will 620get memory from ZONE_DMA only when ZONE_NORMAL is not available. 621 622In NUMA case, you can think of following 2 types of order. 623Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL:: 624 625 (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. 627 628Type(A) offers the best locality for processes on Node(0), but ZONE_DMA 629will be used before ZONE_NORMAL exhaustion. This increases possibility of 630out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small. 631 632Type(B) cannot offer the best locality but is more robust against OOM of 633the DMA zone. 634 635Type(A) is called as "Node" order. Type (B) is "Zone" order. 636 637"Node order" orders the zonelists by node, then by zone within each node. 638Specify "[Nn]ode" for node order 639 640"Zone Order" orders the zonelists by zone type, then by node within each 641zone. Specify "[Zz]one" for zone order. 642 643Specify "[Dd]efault" to request automatic configuration. 644 645On 32-bit, the Normal zone needs to be preserved for allocations accessible 646by the kernel, so "zone" order will be selected. 647 648On 64-bit, devices that require DMA32/DMA are relatively rare, so "node" 649order will be selected. 650 651Default order is recommended unless this is causing problems for your 652system/application. 653 654 655oom_dump_tasks 656============== 657 658Enables a system-wide task dump (excluding kernel threads) to be produced 659when the kernel performs an OOM-killing and includes such information as 660pid, uid, tgid, vm size, rss, pgtables_bytes, swapents, oom_score_adj 661score, and name. This is helpful to determine why the OOM killer was 662invoked, to identify the rogue task that caused it, and to determine why 663the OOM killer chose the task it did to kill. 664 665If this is set to zero, this information is suppressed. On very 666large systems with thousands of tasks it may not be feasible to dump 667the memory state information for each one. Such systems should not 668be forced to incur a performance penalty in OOM conditions when the 669information may not be desired. 670 671If this is set to non-zero, this information is shown whenever the 672OOM killer actually kills a memory-hogging task. 673 674The default value is 1 (enabled). 675 676 677oom_kill_allocating_task 678======================== 679 680This enables or disables killing the OOM-triggering task in 681out-of-memory situations. 682 683If this is set to zero, the OOM killer will scan through the entire 684tasklist and select a task based on heuristics to kill. This normally 685selects a rogue memory-hogging task that frees up a large amount of 686memory when killed. 687 688If this is set to non-zero, the OOM killer simply kills the task that 689triggered the out-of-memory condition. This avoids the expensive 690tasklist scan. 691 692If panic_on_oom is selected, it takes precedence over whatever value 693is used in oom_kill_allocating_task. 694 695The default value is 0. 696 697 698overcommit_kbytes 699================= 700 701When overcommit_memory is set to 2, the committed address space is not 702permitted to exceed swap plus this amount of physical RAM. See below. 703 704Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one 705of them may be specified at a time. Setting one disables the other (which 706then appears as 0 when read). 707 708 709overcommit_memory 710================= 711 712This value contains a flag that enables memory overcommitment. 713 714When this flag is 0, the kernel attempts to estimate the amount 715of free memory left when userspace requests more memory. 716 717When this flag is 1, the kernel pretends there is always enough 718memory until it actually runs out. 719 720When this flag is 2, the kernel uses a "never overcommit" 721policy that attempts to prevent any overcommit of memory. 722Note that user_reserve_kbytes affects this policy. 723 724This feature can be very useful because there are a lot of 725programs that malloc() huge amounts of memory "just-in-case" 726and don't use much of it. 727 728The default value is 0. 729 730See Documentation/vm/overcommit-accounting.rst and 731mm/util.c::__vm_enough_memory() for more information. 732 733 734overcommit_ratio 735================ 736 737When overcommit_memory is set to 2, the committed address 738space is not permitted to exceed swap plus this percentage 739of physical RAM. See above. 740 741 742page-cluster 743============ 744 745page-cluster controls the number of pages up to which consecutive pages 746are read in from swap in a single attempt. This is the swap counterpart 747to page cache readahead. 748The mentioned consecutivity is not in terms of virtual/physical addresses, 749but consecutive on swap space - that means they were swapped out together. 750 751It is a logarithmic value - setting it to zero means "1 page", setting 752it to 1 means "2 pages", setting it to 2 means "4 pages", etc. 753Zero disables swap readahead completely. 754 755The default value is three (eight pages at a time). There may be some 756small benefits in tuning this to a different value if your workload is 757swap-intensive. 758 759Lower values mean lower latencies for initial faults, but at the same time 760extra faults and I/O delays for following faults if they would have been part of 761that consecutive pages readahead would have brought in. 762 763 764panic_on_oom 765============ 766 767This enables or disables panic on out-of-memory feature. 768 769If this is set to 0, the kernel will kill some rogue process, 770called oom_killer. Usually, oom_killer can kill rogue processes and 771system will survive. 772 773If this is set to 1, the kernel panics when out-of-memory happens. 774However, if a process limits using nodes by mempolicy/cpusets, 775and those nodes become memory exhaustion status, one process 776may be killed by oom-killer. No panic occurs in this case. 777Because other nodes' memory may be free. This means system total status 778may be not fatal yet. 779 780If this is set to 2, the kernel panics compulsorily even on the 781above-mentioned. Even oom happens under memory cgroup, the whole 782system panics. 783 784The default value is 0. 785 7861 and 2 are for failover of clustering. Please select either 787according to your policy of failover. 788 789panic_on_oom=2+kdump gives you very strong tool to investigate 790why oom happens. You can get snapshot. 791 792 793percpu_pagelist_fraction 794======================== 795 796This is the fraction of pages at most (high mark pcp->high) in each zone that 797are allocated for each per cpu page list. The min value for this is 8. It 798means that we don't allow more than 1/8th of pages in each zone to be 799allocated in any single per_cpu_pagelist. This entry only changes the value 800of hot per cpu pagelists. User can specify a number like 100 to allocate 8011/100th of each zone to each per cpu page list. 802 803The batch value of each per cpu pagelist is also updated as a result. It is 804set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8) 805 806The initial value is zero. Kernel does not use this value at boot time to set 807the high water marks for each per cpu page list. If the user writes '0' to this 808sysctl, it will revert to this default behavior. 809 810 811stat_interval 812============= 813 814The time interval between which vm statistics are updated. The default 815is 1 second. 816 817 818stat_refresh 819============ 820 821Any read or write (by root only) flushes all the per-cpu vm statistics 822into their global totals, for more accurate reports when testing 823e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo 824 825As a side-effect, it also checks for negative totals (elsewhere reported 826as 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, 828with no ill effects: errors and warnings on these stats are suppressed.) 829 830 831numa_stat 832========= 833 834This interface allows runtime configuration of numa statistics. 835 836When page allocation performance becomes a bottleneck and you can tolerate 837some possible tool breakage and decreased numa counter precision, you can 838do:: 839 840 echo 0 > /proc/sys/vm/numa_stat 841 842When page allocation performance is not a bottleneck and you want all 843tooling to work, you can do:: 844 845 echo 1 > /proc/sys/vm/numa_stat 846 847 848swappiness 849========== 850 851This control is used to define the rough relative IO cost of swapping 852and filesystem paging, as a value between 0 and 200. At 100, the VM 853assumes equal IO cost and will thus apply memory pressure to the page 854cache and swap-backed pages equally; lower values signify more 855expensive swap IO, higher values indicates cheaper. 856 857Keep in mind that filesystem IO patterns under memory pressure tend to 858be more efficient than swap's random IO. An optimal value will require 859experimentation and will also be workload-dependent. 860 861The default value is 60. 862 863For in-memory swap, like zram or zswap, as well as hybrid setups that 864have swap on faster devices than the filesystem, values beyond 100 can 865be considered. For example, if the random IO against the swap device 866is on average 2x faster than IO from the filesystem, swappiness should 867be 133 (x + 2x = 200, 2x = 133.33). 868 869At 0, the kernel will not initiate swap until the amount of free and 870file-backed pages is less than the high watermark in a zone. 871 872 873unprivileged_userfaultfd 874======================== 875 876This flag controls the mode in which unprivileged users can use the 877userfaultfd system calls. Set this to 0 to restrict unprivileged users 878to handle page faults in user mode only. In this case, users without 879SYS_CAP_PTRACE must pass UFFD_USER_MODE_ONLY in order for userfaultfd to 880succeed. Prohibiting use of userfaultfd for handling faults from kernel 881mode may make certain vulnerabilities more difficult to exploit. 882 883Set this to 1 to allow unprivileged users to use the userfaultfd system 884calls without any restrictions. 885 886The default value is 0. 887 888 889user_reserve_kbytes 890=================== 891 892When overcommit_memory is set to 2, "never overcommit" mode, reserve 893min(3% of current process size, user_reserve_kbytes) of free memory. 894This is intended to prevent a user from starting a single memory hogging 895process, such that they cannot recover (kill the hog). 896 897user_reserve_kbytes defaults to min(3% of the current process size, 128MB). 898 899If this is reduced to zero, then the user will be allowed to allocate 900all free memory with a single process, minus admin_reserve_kbytes. 901Any subsequent attempts to execute a command will result in 902"fork: Cannot allocate memory". 903 904Changing this takes effect whenever an application requests memory. 905 906 907vfs_cache_pressure 908================== 909 910This percentage value controls the tendency of the kernel to reclaim 911the memory which is used for caching of directory and inode objects. 912 913At the default value of vfs_cache_pressure=100 the kernel will attempt to 914reclaim dentries and inodes at a "fair" rate with respect to pagecache and 915swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer 916to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will 917never reclaim dentries and inodes due to memory pressure and this can easily 918lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100 919causes the kernel to prefer to reclaim dentries and inodes. 920 921Increasing vfs_cache_pressure significantly beyond 100 may have negative 922performance impact. Reclaim code needs to take various locks to find freeable 923directory and inode objects. With vfs_cache_pressure=1000, it will look for 924ten times more freeable objects than there are. 925 926 927watermark_boost_factor 928====================== 929 930This factor controls the level of reclaim when memory is being fragmented. 931It defines the percentage of the high watermark of a zone that will be 932reclaimed if pages of different mobility are being mixed within pageblocks. 933The intent is that compaction has less work to do in the future and to 934increase the success rate of future high-order allocations such as SLUB 935allocations, THP and hugetlbfs pages. 936 937To make it sensible with respect to the watermark_scale_factor 938parameter, the unit is in fractions of 10,000. The default value of 93915,000 on !DISCONTIGMEM configurations means that up to 150% of the high 940watermark will be reclaimed in the event of a pageblock being mixed due 941to fragmentation. The level of reclaim is determined by the number of 942fragmentation events that occurred in the recent past. If this value is 943smaller 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. 945 946 947watermark_scale_factor 948====================== 949 950This factor controls the aggressiveness of kswapd. It defines the 951amount of memory left in a node/system before kswapd is woken up and 952how much memory needs to be free before kswapd goes back to sleep. 953 954The unit is in fractions of 10,000. The default value of 10 means the 955distances between watermarks are 0.1% of the available memory in the 956node/system. The maximum value is 1000, or 10% of memory. 957 958A high rate of threads entering direct reclaim (allocstall) or kswapd 959going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate 960that the number of free pages kswapd maintains for latency reasons is 961too small for the allocation bursts occurring in the system. This knob 962can then be used to tune kswapd aggressiveness accordingly. 963 964 965zone_reclaim_mode 966================= 967 968Zone_reclaim_mode allows someone to set more or less aggressive approaches to 969reclaim memory when a zone runs out of memory. If it is set to zero then no 970zone reclaim occurs. Allocations will be satisfied from other zones / nodes 971in the system. 972 973This is value OR'ed together of 974 975= =================================== 9761 Zone reclaim on 9772 Zone reclaim writes dirty pages out 9784 Zone reclaim swaps pages 979= =================================== 980 981zone_reclaim_mode is disabled by default. For file servers or workloads 982that benefit from having their data cached, zone_reclaim_mode should be 983left disabled as the caching effect is likely to be more important than 984data locality. 985 986zone_reclaim may be enabled if it's known that the workload is partitioned 987such that each partition fits within a NUMA node and that accessing remote 988memory would cause a measurable performance reduction. The page allocator 989will then reclaim easily reusable pages (those page cache pages that are 990currently not used) before allocating off node pages. 991 992Allowing zone reclaim to write out pages stops processes that are 993writing large amounts of data from dirtying pages on other nodes. Zone 994reclaim will write out dirty pages if a zone fills up and so effectively 995throttle the process. This may decrease the performance of a single process 996since it cannot use all of system memory to buffer the outgoing writes 997anymore but it preserve the memory on other nodes so that the performance 998of other processes running on other nodes will not be affected. 999 1000Allowing regular swap effectively restricts allocations to the local 1001node unless explicitly overridden by memory policies or cpuset 1002configurations. 1003