1=============================== 2Documentation for /proc/sys/vm/ 3=============================== 4 5kernel version 2.6.29 6 7Copyright (c) 1998, 1999, Rik van Riel <riel@nl.linux.org> 8 9Copyright (c) 2008 Peter W. Morreale <pmorreale@novell.com> 10 11For general info and legal blurb, please look in index.rst. 12 13------------------------------------------------------------------------------ 14 15This file contains the documentation for the sysctl files in 16/proc/sys/vm and is valid for Linux kernel version 2.6.29. 17 18The files in this directory can be used to tune the operation 19of the virtual memory (VM) subsystem of the Linux kernel and 20the writeout of dirty data to disk. 21 22Default values and initialization routines for most of these 23files can be found in mm/swap.c. 24 25Currently, these files are in /proc/sys/vm: 26 27- admin_reserve_kbytes 28- compact_memory 29- compaction_proactiveness 30- compact_unevictable_allowed 31- defrag_mode 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- enable_soft_offline 41- extfrag_threshold 42- highmem_is_dirtyable 43- hugetlb_shm_group 44- laptop_mode 45- legacy_va_layout 46- lowmem_reserve_ratio 47- max_map_count 48- mem_profiling (only if CONFIG_MEM_ALLOC_PROFILING=y) 49- memory_failure_early_kill 50- memory_failure_recovery 51- min_free_kbytes 52- min_slab_ratio 53- min_unmapped_ratio 54- mmap_min_addr 55- mmap_rnd_bits 56- mmap_rnd_compat_bits 57- nr_hugepages 58- nr_hugepages_mempolicy 59- nr_overcommit_hugepages 60- nr_trim_pages (only if CONFIG_MMU=n) 61- numa_zonelist_order 62- oom_dump_tasks 63- oom_kill_allocating_task 64- overcommit_kbytes 65- overcommit_memory 66- overcommit_ratio 67- page-cluster 68- page_lock_unfairness 69- panic_on_oom 70- percpu_pagelist_high_fraction 71- stat_interval 72- stat_refresh 73- numa_stat 74- swappiness 75- unprivileged_userfaultfd 76- user_reserve_kbytes 77- vfs_cache_pressure 78- watermark_boost_factor 79- watermark_scale_factor 80- zone_reclaim_mode 81 82 83admin_reserve_kbytes 84==================== 85 86The amount of free memory in the system that should be reserved for users 87with the capability cap_sys_admin. 88 89admin_reserve_kbytes defaults to min(3% of free pages, 8MB) 90 91That should provide enough for the admin to log in and kill a process, 92if necessary, under the default overcommit 'guess' mode. 93 94Systems running under overcommit 'never' should increase this to account 95for the full Virtual Memory Size of programs used to recover. Otherwise, 96root may not be able to log in to recover the system. 97 98How do you calculate a minimum useful reserve? 99 100sshd or login + bash (or some other shell) + top (or ps, kill, etc.) 101 102For overcommit 'guess', we can sum resident set sizes (RSS). 103On x86_64 this is about 8MB. 104 105For overcommit 'never', we can take the max of their virtual sizes (VSZ) 106and add the sum of their RSS. 107On x86_64 this is about 128MB. 108 109Changing this takes effect whenever an application requests memory. 110 111 112compact_memory 113============== 114 115Available only when CONFIG_COMPACTION is set. When 1 is written to the file, 116all zones are compacted such that free memory is available in contiguous 117blocks where possible. This can be important for example in the allocation of 118huge pages although processes will also directly compact memory as required. 119 120compaction_proactiveness 121======================== 122 123This tunable takes a value in the range [0, 100] with a default value of 12420. This tunable determines how aggressively compaction is done in the 125background. Write of a non zero value to this tunable will immediately 126trigger the proactive compaction. Setting it to 0 disables proactive compaction. 127 128Note that compaction has a non-trivial system-wide impact as pages 129belonging to different processes are moved around, which could also lead 130to latency spikes in unsuspecting applications. The kernel employs 131various heuristics to avoid wasting CPU cycles if it detects that 132proactive compaction is not being effective. 133 134Be careful when setting it to extreme values like 100, as that may 135cause excessive background compaction activity. 136 137compact_unevictable_allowed 138=========================== 139 140Available only when CONFIG_COMPACTION is set. When set to 1, compaction is 141allowed to examine the unevictable lru (mlocked pages) for pages to compact. 142This should be used on systems where stalls for minor page faults are an 143acceptable trade for large contiguous free memory. Set to 0 to prevent 144compaction from moving pages that are unevictable. Default value is 1. 145On CONFIG_PREEMPT_RT the default value is 0 in order to avoid a page fault, due 146to compaction, which would block the task from becoming active until the fault 147is resolved. 148 149defrag_mode 150=========== 151 152When set to 1, the page allocator tries harder to avoid fragmentation 153and maintain the ability to produce huge pages / higher-order pages. 154 155It is recommended to enable this right after boot, as fragmentation, 156once it occurred, can be long-lasting or even permanent. 157 158dirty_background_bytes 159====================== 160 161Contains the amount of dirty memory at which the background kernel 162flusher threads will start writeback. 163 164Note: 165 dirty_background_bytes is the counterpart of dirty_background_ratio. Only 166 one of them may be specified at a time. When one sysctl is written it is 167 immediately taken into account to evaluate the dirty memory limits and the 168 other appears as 0 when read. 169 170 171dirty_background_ratio 172====================== 173 174Contains, as a percentage of total available memory that contains free pages 175and reclaimable pages, the number of pages at which the background kernel 176flusher threads will start writing out dirty data. 177 178The total available memory is not equal to total system memory. 179 180 181dirty_bytes 182=========== 183 184Contains the amount of dirty memory at which a process generating disk writes 185will itself start writeback. 186 187Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be 188specified at a time. When one sysctl is written it is immediately taken into 189account to evaluate the dirty memory limits and the other appears as 0 when 190read. 191 192Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any 193value lower than this limit will be ignored and the old configuration will be 194retained. 195 196 197dirty_expire_centisecs 198====================== 199 200This tunable is used to define when dirty data is old enough to be eligible 201for writeout by the kernel flusher threads. It is expressed in 100'ths 202of a second. Data which has been dirty in-memory for longer than this 203interval will be written out next time a flusher thread wakes up. 204 205 206dirty_ratio 207=========== 208 209Contains, as a percentage of total available memory that contains free pages 210and reclaimable pages, the number of pages at which a process which is 211generating disk writes will itself start writing out dirty data. 212 213The total available memory is not equal to total system memory. 214 215 216dirtytime_expire_seconds 217======================== 218 219When a lazytime inode is constantly having its pages dirtied, the inode with 220an updated timestamp will never get chance to be written out. And, if the 221only thing that has happened on the file system is a dirtytime inode caused 222by an atime update, a worker will be scheduled to make sure that inode 223eventually gets pushed out to disk. This tunable is used to define when dirty 224inode is old enough to be eligible for writeback by the kernel flusher threads. 225And, it is also used as the interval to wakeup dirtytime_writeback thread. 226 227 228dirty_writeback_centisecs 229========================= 230 231The kernel flusher threads will periodically wake up and write `old` data 232out to disk. This tunable expresses the interval between those wakeups, in 233100'ths of a second. 234 235Setting this to zero disables periodic writeback altogether. 236 237 238drop_caches 239=========== 240 241Writing to this will cause the kernel to drop clean caches, as well as 242reclaimable slab objects like dentries and inodes. Once dropped, their 243memory becomes free. 244 245To free pagecache:: 246 247 echo 1 > /proc/sys/vm/drop_caches 248 249To free reclaimable slab objects (includes dentries and inodes):: 250 251 echo 2 > /proc/sys/vm/drop_caches 252 253To free slab objects and pagecache:: 254 255 echo 3 > /proc/sys/vm/drop_caches 256 257This is a non-destructive operation and will not free any dirty objects. 258To increase the number of objects freed by this operation, the user may run 259`sync` prior to writing to /proc/sys/vm/drop_caches. This will minimize the 260number of dirty objects on the system and create more candidates to be 261dropped. 262 263This file is not a means to control the growth of the various kernel caches 264(inodes, dentries, pagecache, etc...) These objects are automatically 265reclaimed by the kernel when memory is needed elsewhere on the system. 266 267Use of this file can cause performance problems. Since it discards cached 268objects, it may cost a significant amount of I/O and CPU to recreate the 269dropped objects, especially if they were under heavy use. Because of this, 270use outside of a testing or debugging environment is not recommended. 271 272You may see informational messages in your kernel log when this file is 273used:: 274 275 cat (1234): drop_caches: 3 276 277These are informational only. They do not mean that anything is wrong 278with your system. To disable them, echo 4 (bit 2) into drop_caches. 279 280enable_soft_offline 281=================== 282Correctable memory errors are very common on servers. Soft-offline is kernel's 283solution for memory pages having (excessive) corrected memory errors. 284 285For different types of page, soft-offline has different behaviors / costs. 286 287- For a raw error page, soft-offline migrates the in-use page's content to 288 a new raw page. 289 290- For a page that is part of a transparent hugepage, soft-offline splits the 291 transparent hugepage into raw pages, then migrates only the raw error page. 292 As a result, user is transparently backed by 1 less hugepage, impacting 293 memory access performance. 294 295- For a page that is part of a HugeTLB hugepage, soft-offline first migrates 296 the entire HugeTLB hugepage, during which a free hugepage will be consumed 297 as migration target. Then the original hugepage is dissolved into raw 298 pages without compensation, reducing the capacity of the HugeTLB pool by 1. 299 300It is user's call to choose between reliability (staying away from fragile 301physical memory) vs performance / capacity implications in transparent and 302HugeTLB cases. 303 304For all architectures, enable_soft_offline controls whether to soft offline 305memory pages. When set to 1, kernel attempts to soft offline the pages 306whenever it thinks needed. When set to 0, kernel returns EOPNOTSUPP to 307the request to soft offline the pages. Its default value is 1. 308 309It is worth mentioning that after setting enable_soft_offline to 0, the 310following requests to soft offline pages will not be performed: 311 312- Request to soft offline pages from RAS Correctable Errors Collector. 313 314- On ARM, the request to soft offline pages from GHES driver. 315 316- On PARISC, the request to soft offline pages from Page Deallocation Table. 317 318extfrag_threshold 319================= 320 321This parameter affects whether the kernel will compact memory or direct 322reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in 323debugfs shows what the fragmentation index for each order is in each zone in 324the system. Values tending towards 0 imply allocations would fail due to lack 325of memory, values towards 1000 imply failures are due to fragmentation and -1 326implies that the allocation will succeed as long as watermarks are met. 327 328The kernel will not compact memory in a zone if the 329fragmentation index is <= extfrag_threshold. The default value is 500. 330 331 332highmem_is_dirtyable 333==================== 334 335Available only for systems with CONFIG_HIGHMEM enabled (32b systems). 336 337This parameter controls whether the high memory is considered for dirty 338writers throttling. This is not the case by default which means that 339only the amount of memory directly visible/usable by the kernel can 340be dirtied. As a result, on systems with a large amount of memory and 341lowmem basically depleted writers might be throttled too early and 342streaming writes can get very slow. 343 344Changing the value to non zero would allow more memory to be dirtied 345and thus allow writers to write more data which can be flushed to the 346storage more effectively. Note this also comes with a risk of pre-mature 347OOM killer because some writers (e.g. direct block device writes) can 348only use the low memory and they can fill it up with dirty data without 349any throttling. 350 351 352hugetlb_shm_group 353================= 354 355hugetlb_shm_group contains group id that is allowed to create SysV 356shared memory segment using hugetlb page. 357 358 359laptop_mode 360=========== 361 362laptop_mode is a knob that controls "laptop mode". All the things that are 363controlled by this knob are discussed in Documentation/admin-guide/laptops/laptop-mode.rst. 364 365 366legacy_va_layout 367================ 368 369If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel 370will use the legacy (2.4) layout for all processes. 371 372 373lowmem_reserve_ratio 374==================== 375 376For some specialised workloads on highmem machines it is dangerous for 377the kernel to allow process memory to be allocated from the "lowmem" 378zone. This is because that memory could then be pinned via the mlock() 379system call, or by unavailability of swapspace. 380 381And on large highmem machines this lack of reclaimable lowmem memory 382can be fatal. 383 384So the Linux page allocator has a mechanism which prevents allocations 385which *could* use highmem from using too much lowmem. This means that 386a certain amount of lowmem is defended from the possibility of being 387captured into pinned user memory. 388 389(The same argument applies to the old 16 megabyte ISA DMA region. This 390mechanism will also defend that region from allocations which could use 391highmem or lowmem). 392 393The `lowmem_reserve_ratio` tunable determines how aggressive the kernel is 394in defending these lower zones. 395 396If you have a machine which uses highmem or ISA DMA and your 397applications are using mlock(), or if you are running with no swap then 398you probably should change the lowmem_reserve_ratio setting. 399 400The lowmem_reserve_ratio is an array. You can see them by reading this file:: 401 402 % cat /proc/sys/vm/lowmem_reserve_ratio 403 256 256 32 404 405But, these values are not used directly. The kernel calculates # of protection 406pages for each zones from them. These are shown as array of protection pages 407in /proc/zoneinfo like the following. (This is an example of x86-64 box). 408Each zone has an array of protection pages like this:: 409 410 Node 0, zone DMA 411 pages free 1355 412 min 3 413 low 3 414 high 4 415 : 416 : 417 numa_other 0 418 protection: (0, 2004, 2004, 2004) 419 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 420 pagesets 421 cpu: 0 pcp: 0 422 : 423 424These protections are added to score to judge whether this zone should be used 425for page allocation or should be reclaimed. 426 427In this example, if normal pages (index=2) are required to this DMA zone and 428watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should 429not be used because pages_free(1355) is smaller than watermark + protection[2] 430(4 + 2004 = 2008). If this protection value is 0, this zone would be used for 431normal page requirement. If requirement is DMA zone(index=0), protection[0] 432(=0) is used. 433 434zone[i]'s protection[j] is calculated by following expression:: 435 436 (i < j): 437 zone[i]->protection[j] 438 = (total sums of managed_pages from zone[i+1] to zone[j] on the node) 439 / lowmem_reserve_ratio[i]; 440 (i = j): 441 (should not be protected. = 0; 442 (i > j): 443 (not necessary, but looks 0) 444 445The default values of lowmem_reserve_ratio[i] are 446 447 === ==================================== 448 256 (if zone[i] means DMA or DMA32 zone) 449 32 (others) 450 === ==================================== 451 452As above expression, they are reciprocal number of ratio. 453256 means 1/256. # of protection pages becomes about "0.39%" of total managed 454pages of higher zones on the node. 455 456If you would like to protect more pages, smaller values are effective. 457The minimum value is 1 (1/1 -> 100%). The value less than 1 completely 458disables protection of the pages. 459 460 461max_map_count: 462============== 463 464This file contains the maximum number of memory map areas a process 465may have. Memory map areas are used as a side-effect of calling 466malloc, directly by mmap, mprotect, and madvise, and also when loading 467shared libraries. 468 469While most applications need less than a thousand maps, certain 470programs, particularly malloc debuggers, may consume lots of them, 471e.g., up to one or two maps per allocation. 472 473The default value is 65530. 474 475 476mem_profiling 477============== 478 479Enable memory profiling (when CONFIG_MEM_ALLOC_PROFILING=y) 480 4811: Enable memory profiling. 482 4830: Disable memory profiling. 484 485Enabling memory profiling introduces a small performance overhead for all 486memory allocations. 487 488The default value depends on CONFIG_MEM_ALLOC_PROFILING_ENABLED_BY_DEFAULT. 489 490 491memory_failure_early_kill: 492========================== 493 494Control how to kill processes when uncorrected memory error (typically 495a 2bit error in a memory module) is detected in the background by hardware 496that cannot be handled by the kernel. In some cases (like the page 497still having a valid copy on disk) the kernel will handle the failure 498transparently without affecting any applications. But if there is 499no other up-to-date copy of the data it will kill to prevent any data 500corruptions from propagating. 501 5021: Kill all processes that have the corrupted and not reloadable page mapped 503as soon as the corruption is detected. Note this is not supported 504for a few types of pages, like kernel internally allocated data or 505the swap cache, but works for the majority of user pages. 506 5070: Only unmap the corrupted page from all processes and only kill a process 508who tries to access it. 509 510The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can 511handle this if they want to. 512 513This is only active on architectures/platforms with advanced machine 514check handling and depends on the hardware capabilities. 515 516Applications can override this setting individually with the PR_MCE_KILL prctl 517 518 519memory_failure_recovery 520======================= 521 522Enable memory failure recovery (when supported by the platform) 523 5241: Attempt recovery. 525 5260: Always panic on a memory failure. 527 528 529min_free_kbytes 530=============== 531 532This is used to force the Linux VM to keep a minimum number 533of kilobytes free. The VM uses this number to compute a 534watermark[WMARK_MIN] value for each lowmem zone in the system. 535Each lowmem zone gets a number of reserved free pages based 536proportionally on its size. 537 538Some minimal amount of memory is needed to satisfy PF_MEMALLOC 539allocations; if you set this to lower than 1024KB, your system will 540become subtly broken, and prone to deadlock under high loads. 541 542Setting this too high will OOM your machine instantly. 543 544 545min_slab_ratio 546============== 547 548This is available only on NUMA kernels. 549 550A percentage of the total pages in each zone. On Zone reclaim 551(fallback from the local zone occurs) slabs will be reclaimed if more 552than this percentage of pages in a zone are reclaimable slab pages. 553This insures that the slab growth stays under control even in NUMA 554systems that rarely perform global reclaim. 555 556The default is 5 percent. 557 558Note that slab reclaim is triggered in a per zone / node fashion. 559The process of reclaiming slab memory is currently not node specific 560and may not be fast. 561 562 563min_unmapped_ratio 564================== 565 566This is available only on NUMA kernels. 567 568This is a percentage of the total pages in each zone. Zone reclaim will 569only occur if more than this percentage of pages are in a state that 570zone_reclaim_mode allows to be reclaimed. 571 572If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared 573against all file-backed unmapped pages including swapcache pages and tmpfs 574files. Otherwise, only unmapped pages backed by normal files but not tmpfs 575files and similar are considered. 576 577The default is 1 percent. 578 579 580mmap_min_addr 581============= 582 583This file indicates the amount of address space which a user process will 584be restricted from mmapping. Since kernel null dereference bugs could 585accidentally operate based on the information in the first couple of pages 586of memory userspace processes should not be allowed to write to them. By 587default this value is set to 0 and no protections will be enforced by the 588security module. Setting this value to something like 64k will allow the 589vast majority of applications to work correctly and provide defense in depth 590against future potential kernel bugs. 591 592 593mmap_rnd_bits 594============= 595 596This value can be used to select the number of bits to use to 597determine the random offset to the base address of vma regions 598resulting from mmap allocations on architectures which support 599tuning address space randomization. This value will be bounded 600by the architecture's minimum and maximum supported values. 601 602This value can be changed after boot using the 603/proc/sys/vm/mmap_rnd_bits tunable 604 605 606mmap_rnd_compat_bits 607==================== 608 609This value can be used to select the number of bits to use to 610determine the random offset to the base address of vma regions 611resulting from mmap allocations for applications run in 612compatibility mode on architectures which support tuning address 613space randomization. This value will be bounded by the 614architecture's minimum and maximum supported values. 615 616This value can be changed after boot using the 617/proc/sys/vm/mmap_rnd_compat_bits tunable 618 619 620nr_hugepages 621============ 622 623Change the minimum size of the hugepage pool. 624 625See Documentation/admin-guide/mm/hugetlbpage.rst 626 627 628hugetlb_optimize_vmemmap 629======================== 630 631This knob is not available when the size of 'struct page' (a structure defined 632in include/linux/mm_types.h) is not power of two (an unusual system config could 633result in this). 634 635Enable (set to 1) or disable (set to 0) HugeTLB Vmemmap Optimization (HVO). 636 637Once enabled, the vmemmap pages of subsequent allocation of HugeTLB pages from 638buddy allocator will be optimized (7 pages per 2MB HugeTLB page and 4095 pages 639per 1GB HugeTLB page), whereas already allocated HugeTLB pages will not be 640optimized. When those optimized HugeTLB pages are freed from the HugeTLB pool 641to the buddy allocator, the vmemmap pages representing that range needs to be 642remapped again and the vmemmap pages discarded earlier need to be rellocated 643again. If your use case is that HugeTLB pages are allocated 'on the fly' (e.g. 644never explicitly allocating HugeTLB pages with 'nr_hugepages' but only set 645'nr_overcommit_hugepages', those overcommitted HugeTLB pages are allocated 'on 646the fly') instead of being pulled from the HugeTLB pool, you should weigh the 647benefits of memory savings against the more overhead (~2x slower than before) 648of allocation or freeing HugeTLB pages between the HugeTLB pool and the buddy 649allocator. Another behavior to note is that if the system is under heavy memory 650pressure, it could prevent the user from freeing HugeTLB pages from the HugeTLB 651pool to the buddy allocator since the allocation of vmemmap pages could be 652failed, you have to retry later if your system encounter this situation. 653 654Once disabled, the vmemmap pages of subsequent allocation of HugeTLB pages from 655buddy allocator will not be optimized meaning the extra overhead at allocation 656time from buddy allocator disappears, whereas already optimized HugeTLB pages 657will not be affected. If you want to make sure there are no optimized HugeTLB 658pages, you can set "nr_hugepages" to 0 first and then disable this. Note that 659writing 0 to nr_hugepages will make any "in use" HugeTLB pages become surplus 660pages. So, those surplus pages are still optimized until they are no longer 661in use. You would need to wait for those surplus pages to be released before 662there are no optimized pages in the system. 663 664 665nr_hugepages_mempolicy 666====================== 667 668Change the size of the hugepage pool at run-time on a specific 669set of NUMA nodes. 670 671See Documentation/admin-guide/mm/hugetlbpage.rst 672 673 674nr_overcommit_hugepages 675======================= 676 677Change the maximum size of the hugepage pool. The maximum is 678nr_hugepages + nr_overcommit_hugepages. 679 680See Documentation/admin-guide/mm/hugetlbpage.rst 681 682 683nr_trim_pages 684============= 685 686This is available only on NOMMU kernels. 687 688This value adjusts the excess page trimming behaviour of power-of-2 aligned 689NOMMU mmap allocations. 690 691A value of 0 disables trimming of allocations entirely, while a value of 1 692trims excess pages aggressively. Any value >= 1 acts as the watermark where 693trimming of allocations is initiated. 694 695The default value is 1. 696 697See Documentation/admin-guide/mm/nommu-mmap.rst for more information. 698 699 700numa_zonelist_order 701=================== 702 703This sysctl is only for NUMA and it is deprecated. Anything but 704Node order will fail! 705 706'where the memory is allocated from' is controlled by zonelists. 707 708(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation. 709you may be able to read ZONE_DMA as ZONE_DMA32...) 710 711In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following. 712ZONE_NORMAL -> ZONE_DMA 713This means that a memory allocation request for GFP_KERNEL will 714get memory from ZONE_DMA only when ZONE_NORMAL is not available. 715 716In NUMA case, you can think of following 2 types of order. 717Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL:: 718 719 (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL 720 (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA. 721 722Type(A) offers the best locality for processes on Node(0), but ZONE_DMA 723will be used before ZONE_NORMAL exhaustion. This increases possibility of 724out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small. 725 726Type(B) cannot offer the best locality but is more robust against OOM of 727the DMA zone. 728 729Type(A) is called as "Node" order. Type (B) is "Zone" order. 730 731"Node order" orders the zonelists by node, then by zone within each node. 732Specify "[Nn]ode" for node order 733 734"Zone Order" orders the zonelists by zone type, then by node within each 735zone. Specify "[Zz]one" for zone order. 736 737Specify "[Dd]efault" to request automatic configuration. 738 739On 32-bit, the Normal zone needs to be preserved for allocations accessible 740by the kernel, so "zone" order will be selected. 741 742On 64-bit, devices that require DMA32/DMA are relatively rare, so "node" 743order will be selected. 744 745Default order is recommended unless this is causing problems for your 746system/application. 747 748 749oom_dump_tasks 750============== 751 752Enables a system-wide task dump (excluding kernel threads) to be produced 753when the kernel performs an OOM-killing and includes such information as 754pid, uid, tgid, vm size, rss, pgtables_bytes, swapents, oom_score_adj 755score, and name. This is helpful to determine why the OOM killer was 756invoked, to identify the rogue task that caused it, and to determine why 757the OOM killer chose the task it did to kill. 758 759If this is set to zero, this information is suppressed. On very 760large systems with thousands of tasks it may not be feasible to dump 761the memory state information for each one. Such systems should not 762be forced to incur a performance penalty in OOM conditions when the 763information may not be desired. 764 765If this is set to non-zero, this information is shown whenever the 766OOM killer actually kills a memory-hogging task. 767 768The default value is 1 (enabled). 769 770 771oom_kill_allocating_task 772======================== 773 774This enables or disables killing the OOM-triggering task in 775out-of-memory situations. 776 777If this is set to zero, the OOM killer will scan through the entire 778tasklist and select a task based on heuristics to kill. This normally 779selects a rogue memory-hogging task that frees up a large amount of 780memory when killed. 781 782If this is set to non-zero, the OOM killer simply kills the task that 783triggered the out-of-memory condition. This avoids the expensive 784tasklist scan. 785 786If panic_on_oom is selected, it takes precedence over whatever value 787is used in oom_kill_allocating_task. 788 789The default value is 0. 790 791 792overcommit_kbytes 793================= 794 795When overcommit_memory is set to 2, the committed address space is not 796permitted to exceed swap plus this amount of physical RAM. See below. 797 798Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one 799of them may be specified at a time. Setting one disables the other (which 800then appears as 0 when read). 801 802 803overcommit_memory 804================= 805 806This value contains a flag that enables memory overcommitment. 807 808When this flag is 0, the kernel compares the userspace memory request 809size against total memory plus swap and rejects obvious overcommits. 810 811When this flag is 1, the kernel pretends there is always enough 812memory until it actually runs out. 813 814When this flag is 2, the kernel uses a "never overcommit" 815policy that attempts to prevent any overcommit of memory. 816Note that user_reserve_kbytes affects this policy. 817 818This feature can be very useful because there are a lot of 819programs that malloc() huge amounts of memory "just-in-case" 820and don't use much of it. 821 822The default value is 0. 823 824See Documentation/mm/overcommit-accounting.rst and 825mm/util.c::__vm_enough_memory() for more information. 826 827 828overcommit_ratio 829================ 830 831When overcommit_memory is set to 2, the committed address 832space is not permitted to exceed swap plus this percentage 833of physical RAM. See above. 834 835 836page-cluster 837============ 838 839page-cluster controls the number of pages up to which consecutive pages 840are read in from swap in a single attempt. This is the swap counterpart 841to page cache readahead. 842The mentioned consecutivity is not in terms of virtual/physical addresses, 843but consecutive on swap space - that means they were swapped out together. 844 845It is a logarithmic value - setting it to zero means "1 page", setting 846it to 1 means "2 pages", setting it to 2 means "4 pages", etc. 847Zero disables swap readahead completely. 848 849The default value is three (eight pages at a time). There may be some 850small benefits in tuning this to a different value if your workload is 851swap-intensive. 852 853Lower values mean lower latencies for initial faults, but at the same time 854extra faults and I/O delays for following faults if they would have been part of 855that consecutive pages readahead would have brought in. 856 857 858page_lock_unfairness 859==================== 860 861This value determines the number of times that the page lock can be 862stolen from under a waiter. After the lock is stolen the number of times 863specified in this file (default is 5), the "fair lock handoff" semantics 864will apply, and the waiter will only be awakened if the lock can be taken. 865 866panic_on_oom 867============ 868 869This enables or disables panic on out-of-memory feature. 870 871If this is set to 0, the kernel will kill some rogue process, 872called oom_killer. Usually, oom_killer can kill rogue processes and 873system will survive. 874 875If this is set to 1, the kernel panics when out-of-memory happens. 876However, if a process limits using nodes by mempolicy/cpusets, 877and those nodes become memory exhaustion status, one process 878may be killed by oom-killer. No panic occurs in this case. 879Because other nodes' memory may be free. This means system total status 880may be not fatal yet. 881 882If this is set to 2, the kernel panics compulsorily even on the 883above-mentioned. Even oom happens under memory cgroup, the whole 884system panics. 885 886The default value is 0. 887 8881 and 2 are for failover of clustering. Please select either 889according to your policy of failover. 890 891panic_on_oom=2+kdump gives you very strong tool to investigate 892why oom happens. You can get snapshot. 893 894 895percpu_pagelist_high_fraction 896============================= 897 898This is the fraction of pages in each zone that are can be stored to 899per-cpu page lists. It is an upper boundary that is divided depending 900on the number of online CPUs. The min value for this is 8 which means 901that we do not allow more than 1/8th of pages in each zone to be stored 902on per-cpu page lists. This entry only changes the value of hot per-cpu 903page lists. A user can specify a number like 100 to allocate 1/100th of 904each zone between per-cpu lists. 905 906The batch value of each per-cpu page list remains the same regardless of 907the value of the high fraction so allocation latencies are unaffected. 908 909The initial value is zero. Kernel uses this value to set the high pcp->high 910mark based on the low watermark for the zone and the number of local 911online CPUs. If the user writes '0' to this sysctl, it will revert to 912this default behavior. 913 914 915stat_interval 916============= 917 918The time interval between which vm statistics are updated. The default 919is 1 second. 920 921 922stat_refresh 923============ 924 925Any read or write (by root only) flushes all the per-cpu vm statistics 926into their global totals, for more accurate reports when testing 927e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo 928 929As a side-effect, it also checks for negative totals (elsewhere reported 930as 0) and "fails" with EINVAL if any are found, with a warning in dmesg. 931(At time of writing, a few stats are known sometimes to be found negative, 932with no ill effects: errors and warnings on these stats are suppressed.) 933 934 935numa_stat 936========= 937 938This interface allows runtime configuration of numa statistics. 939 940When page allocation performance becomes a bottleneck and you can tolerate 941some possible tool breakage and decreased numa counter precision, you can 942do:: 943 944 echo 0 > /proc/sys/vm/numa_stat 945 946When page allocation performance is not a bottleneck and you want all 947tooling to work, you can do:: 948 949 echo 1 > /proc/sys/vm/numa_stat 950 951 952swappiness 953========== 954 955This control is used to define the rough relative IO cost of swapping 956and filesystem paging, as a value between 0 and 200. At 100, the VM 957assumes equal IO cost and will thus apply memory pressure to the page 958cache and swap-backed pages equally; lower values signify more 959expensive swap IO, higher values indicates cheaper. 960 961Keep in mind that filesystem IO patterns under memory pressure tend to 962be more efficient than swap's random IO. An optimal value will require 963experimentation and will also be workload-dependent. 964 965The default value is 60. 966 967For in-memory swap, like zram or zswap, as well as hybrid setups that 968have swap on faster devices than the filesystem, values beyond 100 can 969be considered. For example, if the random IO against the swap device 970is on average 2x faster than IO from the filesystem, swappiness should 971be 133 (x + 2x = 200, 2x = 133.33). 972 973At 0, the kernel will not initiate swap until the amount of free and 974file-backed pages is less than the high watermark in a zone. 975 976 977unprivileged_userfaultfd 978======================== 979 980This flag controls the mode in which unprivileged users can use the 981userfaultfd system calls. Set this to 0 to restrict unprivileged users 982to handle page faults in user mode only. In this case, users without 983SYS_CAP_PTRACE must pass UFFD_USER_MODE_ONLY in order for userfaultfd to 984succeed. Prohibiting use of userfaultfd for handling faults from kernel 985mode may make certain vulnerabilities more difficult to exploit. 986 987Set this to 1 to allow unprivileged users to use the userfaultfd system 988calls without any restrictions. 989 990The default value is 0. 991 992Another way to control permissions for userfaultfd is to use 993/dev/userfaultfd instead of userfaultfd(2). See 994Documentation/admin-guide/mm/userfaultfd.rst. 995 996user_reserve_kbytes 997=================== 998 999When overcommit_memory is set to 2, "never overcommit" mode, reserve 1000min(3% of current process size, user_reserve_kbytes) of free memory. 1001This is intended to prevent a user from starting a single memory hogging 1002process, such that they cannot recover (kill the hog). 1003 1004user_reserve_kbytes defaults to min(3% of the current process size, 128MB). 1005 1006If this is reduced to zero, then the user will be allowed to allocate 1007all free memory with a single process, minus admin_reserve_kbytes. 1008Any subsequent attempts to execute a command will result in 1009"fork: Cannot allocate memory". 1010 1011Changing this takes effect whenever an application requests memory. 1012 1013 1014vfs_cache_pressure 1015================== 1016 1017This percentage value controls the tendency of the kernel to reclaim 1018the memory which is used for caching of directory and inode objects. 1019 1020At the default value of vfs_cache_pressure=100 the kernel will attempt to 1021reclaim dentries and inodes at a "fair" rate with respect to pagecache and 1022swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer 1023to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will 1024never reclaim dentries and inodes due to memory pressure and this can easily 1025lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100 1026causes the kernel to prefer to reclaim dentries and inodes. 1027 1028Increasing vfs_cache_pressure significantly beyond 100 may have negative 1029performance impact. Reclaim code needs to take various locks to find freeable 1030directory and inode objects. With vfs_cache_pressure=1000, it will look for 1031ten times more freeable objects than there are. 1032 1033 1034watermark_boost_factor 1035====================== 1036 1037This factor controls the level of reclaim when memory is being fragmented. 1038It defines the percentage of the high watermark of a zone that will be 1039reclaimed if pages of different mobility are being mixed within pageblocks. 1040The intent is that compaction has less work to do in the future and to 1041increase the success rate of future high-order allocations such as SLUB 1042allocations, THP and hugetlbfs pages. 1043 1044To make it sensible with respect to the watermark_scale_factor 1045parameter, the unit is in fractions of 10,000. The default value of 104615,000 means that up to 150% of the high watermark will be reclaimed in the 1047event of a pageblock being mixed due to fragmentation. The level of reclaim 1048is determined by the number of fragmentation events that occurred in the 1049recent past. If this value is smaller than a pageblock then a pageblocks 1050worth of pages will be reclaimed (e.g. 2MB on 64-bit x86). A boost factor 1051of 0 will disable the feature. 1052 1053 1054watermark_scale_factor 1055====================== 1056 1057This factor controls the aggressiveness of kswapd. It defines the 1058amount of memory left in a node/system before kswapd is woken up and 1059how much memory needs to be free before kswapd goes back to sleep. 1060 1061The unit is in fractions of 10,000. The default value of 10 means the 1062distances between watermarks are 0.1% of the available memory in the 1063node/system. The maximum value is 3000, or 30% of memory. 1064 1065A high rate of threads entering direct reclaim (allocstall) or kswapd 1066going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate 1067that the number of free pages kswapd maintains for latency reasons is 1068too small for the allocation bursts occurring in the system. This knob 1069can then be used to tune kswapd aggressiveness accordingly. 1070 1071 1072zone_reclaim_mode 1073================= 1074 1075Zone_reclaim_mode allows someone to set more or less aggressive approaches to 1076reclaim memory when a zone runs out of memory. If it is set to zero then no 1077zone reclaim occurs. Allocations will be satisfied from other zones / nodes 1078in the system. 1079 1080This is value OR'ed together of 1081 1082= =================================== 10831 Zone reclaim on 10842 Zone reclaim writes dirty pages out 10854 Zone reclaim swaps pages 1086= =================================== 1087 1088zone_reclaim_mode is disabled by default. For file servers or workloads 1089that benefit from having their data cached, zone_reclaim_mode should be 1090left disabled as the caching effect is likely to be more important than 1091data locality. 1092 1093Consider enabling one or more zone_reclaim mode bits if it's known that the 1094workload is partitioned such that each partition fits within a NUMA node 1095and that accessing remote memory would cause a measurable performance 1096reduction. The page allocator will take additional actions before 1097allocating off node pages. 1098 1099Allowing zone reclaim to write out pages stops processes that are 1100writing large amounts of data from dirtying pages on other nodes. Zone 1101reclaim will write out dirty pages if a zone fills up and so effectively 1102throttle the process. This may decrease the performance of a single process 1103since it cannot use all of system memory to buffer the outgoing writes 1104anymore but it preserve the memory on other nodes so that the performance 1105of other processes running on other nodes will not be affected. 1106 1107Allowing regular swap effectively restricts allocations to the local 1108node unless explicitly overridden by memory policies or cpuset 1109configurations. 1110