Based on kernel version 3.9. Page generated on 2013-05-02 23:15 EST.
1 Documentation for /proc/sys/vm/* kernel version 2.6.29 2 (c) 1998, 1999, Rik van Riel <riel@nl.linux.org> 3 (c) 2008 Peter W. Morreale <pmorreale@novell.com> 4 5 For general info and legal blurb, please look in README. 6 7 ============================================================== 8 9 This file contains the documentation for the sysctl files in 10 /proc/sys/vm and is valid for Linux kernel version 2.6.29. 11 12 The files in this directory can be used to tune the operation 13 of the virtual memory (VM) subsystem of the Linux kernel and 14 the writeout of dirty data to disk. 15 16 Default values and initialization routines for most of these 17 files can be found in mm/swap.c. 18 19 Currently, these files are in /proc/sys/vm: 20 21 - block_dump 22 - compact_memory 23 - dirty_background_bytes 24 - dirty_background_ratio 25 - dirty_bytes 26 - dirty_expire_centisecs 27 - dirty_ratio 28 - dirty_writeback_centisecs 29 - drop_caches 30 - extfrag_threshold 31 - hugepages_treat_as_movable 32 - hugetlb_shm_group 33 - laptop_mode 34 - legacy_va_layout 35 - lowmem_reserve_ratio 36 - max_map_count 37 - memory_failure_early_kill 38 - memory_failure_recovery 39 - min_free_kbytes 40 - min_slab_ratio 41 - min_unmapped_ratio 42 - mmap_min_addr 43 - nr_hugepages 44 - nr_overcommit_hugepages 45 - nr_trim_pages (only if CONFIG_MMU=n) 46 - numa_zonelist_order 47 - oom_dump_tasks 48 - oom_kill_allocating_task 49 - overcommit_memory 50 - overcommit_ratio 51 - page-cluster 52 - panic_on_oom 53 - percpu_pagelist_fraction 54 - stat_interval 55 - swappiness 56 - vfs_cache_pressure 57 - zone_reclaim_mode 58 59 ============================================================== 60 61 block_dump 62 63 block_dump enables block I/O debugging when set to a nonzero value. More 64 information on block I/O debugging is in Documentation/laptops/laptop-mode.txt. 65 66 ============================================================== 67 68 compact_memory 69 70 Available only when CONFIG_COMPACTION is set. When 1 is written to the file, 71 all zones are compacted such that free memory is available in contiguous 72 blocks where possible. This can be important for example in the allocation of 73 huge pages although processes will also directly compact memory as required. 74 75 ============================================================== 76 77 dirty_background_bytes 78 79 Contains the amount of dirty memory at which the background kernel 80 flusher threads will start writeback. 81 82 Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only 83 one of them may be specified at a time. When one sysctl is written it is 84 immediately taken into account to evaluate the dirty memory limits and the 85 other appears as 0 when read. 86 87 ============================================================== 88 89 dirty_background_ratio 90 91 Contains, as a percentage of total system memory, the number of pages at which 92 the background kernel flusher threads will start writing out dirty data. 93 94 ============================================================== 95 96 dirty_bytes 97 98 Contains the amount of dirty memory at which a process generating disk writes 99 will itself start writeback. 100 101 Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be 102 specified at a time. When one sysctl is written it is immediately taken into 103 account to evaluate the dirty memory limits and the other appears as 0 when 104 read. 105 106 Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any 107 value lower than this limit will be ignored and the old configuration will be 108 retained. 109 110 ============================================================== 111 112 dirty_expire_centisecs 113 114 This tunable is used to define when dirty data is old enough to be eligible 115 for writeout by the kernel flusher threads. It is expressed in 100'ths 116 of a second. Data which has been dirty in-memory for longer than this 117 interval will be written out next time a flusher thread wakes up. 118 119 ============================================================== 120 121 dirty_ratio 122 123 Contains, as a percentage of total system memory, the number of pages at which 124 a process which is generating disk writes will itself start writing out dirty 125 data. 126 127 ============================================================== 128 129 dirty_writeback_centisecs 130 131 The kernel flusher threads will periodically wake up and write `old' data 132 out to disk. This tunable expresses the interval between those wakeups, in 133 100'ths of a second. 134 135 Setting this to zero disables periodic writeback altogether. 136 137 ============================================================== 138 139 drop_caches 140 141 Writing to this will cause the kernel to drop clean caches, dentries and 142 inodes from memory, causing that memory to become free. 143 144 To free pagecache: 145 echo 1 > /proc/sys/vm/drop_caches 146 To free dentries and inodes: 147 echo 2 > /proc/sys/vm/drop_caches 148 To free pagecache, dentries and inodes: 149 echo 3 > /proc/sys/vm/drop_caches 150 151 As this is a non-destructive operation and dirty objects are not freeable, the 152 user should run `sync' first. 153 154 ============================================================== 155 156 extfrag_threshold 157 158 This parameter affects whether the kernel will compact memory or direct 159 reclaim to satisfy a high-order allocation. /proc/extfrag_index shows what 160 the fragmentation index for each order is in each zone in the system. Values 161 tending towards 0 imply allocations would fail due to lack of memory, 162 values towards 1000 imply failures are due to fragmentation and -1 implies 163 that the allocation will succeed as long as watermarks are met. 164 165 The kernel will not compact memory in a zone if the 166 fragmentation index is <= extfrag_threshold. The default value is 500. 167 168 ============================================================== 169 170 hugepages_treat_as_movable 171 172 This parameter is only useful when kernelcore= is specified at boot time to 173 create ZONE_MOVABLE for pages that may be reclaimed or migrated. Huge pages 174 are not movable so are not normally allocated from ZONE_MOVABLE. A non-zero 175 value written to hugepages_treat_as_movable allows huge pages to be allocated 176 from ZONE_MOVABLE. 177 178 Once enabled, the ZONE_MOVABLE is treated as an area of memory the huge 179 pages pool can easily grow or shrink within. Assuming that applications are 180 not running that mlock() a lot of memory, it is likely the huge pages pool 181 can grow to the size of ZONE_MOVABLE by repeatedly entering the desired value 182 into nr_hugepages and triggering page reclaim. 183 184 ============================================================== 185 186 hugetlb_shm_group 187 188 hugetlb_shm_group contains group id that is allowed to create SysV 189 shared memory segment using hugetlb page. 190 191 ============================================================== 192 193 laptop_mode 194 195 laptop_mode is a knob that controls "laptop mode". All the things that are 196 controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt. 197 198 ============================================================== 199 200 legacy_va_layout 201 202 If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel 203 will use the legacy (2.4) layout for all processes. 204 205 ============================================================== 206 207 lowmem_reserve_ratio 208 209 For some specialised workloads on highmem machines it is dangerous for 210 the kernel to allow process memory to be allocated from the "lowmem" 211 zone. This is because that memory could then be pinned via the mlock() 212 system call, or by unavailability of swapspace. 213 214 And on large highmem machines this lack of reclaimable lowmem memory 215 can be fatal. 216 217 So the Linux page allocator has a mechanism which prevents allocations 218 which _could_ use highmem from using too much lowmem. This means that 219 a certain amount of lowmem is defended from the possibility of being 220 captured into pinned user memory. 221 222 (The same argument applies to the old 16 megabyte ISA DMA region. This 223 mechanism will also defend that region from allocations which could use 224 highmem or lowmem). 225 226 The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is 227 in defending these lower zones. 228 229 If you have a machine which uses highmem or ISA DMA and your 230 applications are using mlock(), or if you are running with no swap then 231 you probably should change the lowmem_reserve_ratio setting. 232 233 The lowmem_reserve_ratio is an array. You can see them by reading this file. 234 - 235 % cat /proc/sys/vm/lowmem_reserve_ratio 236 256 256 32 237 - 238 Note: # of this elements is one fewer than number of zones. Because the highest 239 zone's value is not necessary for following calculation. 240 241 But, these values are not used directly. The kernel calculates # of protection 242 pages for each zones from them. These are shown as array of protection pages 243 in /proc/zoneinfo like followings. (This is an example of x86-64 box). 244 Each zone has an array of protection pages like this. 245 246 - 247 Node 0, zone DMA 248 pages free 1355 249 min 3 250 low 3 251 high 4 252 : 253 : 254 numa_other 0 255 protection: (0, 2004, 2004, 2004) 256 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 257 pagesets 258 cpu: 0 pcp: 0 259 : 260 - 261 These protections are added to score to judge whether this zone should be used 262 for page allocation or should be reclaimed. 263 264 In this example, if normal pages (index=2) are required to this DMA zone and 265 watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should 266 not be used because pages_free(1355) is smaller than watermark + protection[2] 267 (4 + 2004 = 2008). If this protection value is 0, this zone would be used for 268 normal page requirement. If requirement is DMA zone(index=0), protection[0] 269 (=0) is used. 270 271 zone[i]'s protection[j] is calculated by following expression. 272 273 (i < j): 274 zone[i]->protection[j] 275 = (total sums of present_pages from zone[i+1] to zone[j] on the node) 276 / lowmem_reserve_ratio[i]; 277 (i = j): 278 (should not be protected. = 0; 279 (i > j): 280 (not necessary, but looks 0) 281 282 The default values of lowmem_reserve_ratio[i] are 283 256 (if zone[i] means DMA or DMA32 zone) 284 32 (others). 285 As above expression, they are reciprocal number of ratio. 286 256 means 1/256. # of protection pages becomes about "0.39%" of total present 287 pages of higher zones on the node. 288 289 If you would like to protect more pages, smaller values are effective. 290 The minimum value is 1 (1/1 -> 100%). 291 292 ============================================================== 293 294 max_map_count: 295 296 This file contains the maximum number of memory map areas a process 297 may have. Memory map areas are used as a side-effect of calling 298 malloc, directly by mmap and mprotect, and also when loading shared 299 libraries. 300 301 While most applications need less than a thousand maps, certain 302 programs, particularly malloc debuggers, may consume lots of them, 303 e.g., up to one or two maps per allocation. 304 305 The default value is 65536. 306 307 ============================================================= 308 309 memory_failure_early_kill: 310 311 Control how to kill processes when uncorrected memory error (typically 312 a 2bit error in a memory module) is detected in the background by hardware 313 that cannot be handled by the kernel. In some cases (like the page 314 still having a valid copy on disk) the kernel will handle the failure 315 transparently without affecting any applications. But if there is 316 no other uptodate copy of the data it will kill to prevent any data 317 corruptions from propagating. 318 319 1: Kill all processes that have the corrupted and not reloadable page mapped 320 as soon as the corruption is detected. Note this is not supported 321 for a few types of pages, like kernel internally allocated data or 322 the swap cache, but works for the majority of user pages. 323 324 0: Only unmap the corrupted page from all processes and only kill a process 325 who tries to access it. 326 327 The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can 328 handle this if they want to. 329 330 This is only active on architectures/platforms with advanced machine 331 check handling and depends on the hardware capabilities. 332 333 Applications can override this setting individually with the PR_MCE_KILL prctl 334 335 ============================================================== 336 337 memory_failure_recovery 338 339 Enable memory failure recovery (when supported by the platform) 340 341 1: Attempt recovery. 342 343 0: Always panic on a memory failure. 344 345 ============================================================== 346 347 min_free_kbytes: 348 349 This is used to force the Linux VM to keep a minimum number 350 of kilobytes free. The VM uses this number to compute a 351 watermark[WMARK_MIN] value for each lowmem zone in the system. 352 Each lowmem zone gets a number of reserved free pages based 353 proportionally on its size. 354 355 Some minimal amount of memory is needed to satisfy PF_MEMALLOC 356 allocations; if you set this to lower than 1024KB, your system will 357 become subtly broken, and prone to deadlock under high loads. 358 359 Setting this too high will OOM your machine instantly. 360 361 ============================================================= 362 363 min_slab_ratio: 364 365 This is available only on NUMA kernels. 366 367 A percentage of the total pages in each zone. On Zone reclaim 368 (fallback from the local zone occurs) slabs will be reclaimed if more 369 than this percentage of pages in a zone are reclaimable slab pages. 370 This insures that the slab growth stays under control even in NUMA 371 systems that rarely perform global reclaim. 372 373 The default is 5 percent. 374 375 Note that slab reclaim is triggered in a per zone / node fashion. 376 The process of reclaiming slab memory is currently not node specific 377 and may not be fast. 378 379 ============================================================= 380 381 min_unmapped_ratio: 382 383 This is available only on NUMA kernels. 384 385 This is a percentage of the total pages in each zone. Zone reclaim will 386 only occur if more than this percentage of pages are in a state that 387 zone_reclaim_mode allows to be reclaimed. 388 389 If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared 390 against all file-backed unmapped pages including swapcache pages and tmpfs 391 files. Otherwise, only unmapped pages backed by normal files but not tmpfs 392 files and similar are considered. 393 394 The default is 1 percent. 395 396 ============================================================== 397 398 mmap_min_addr 399 400 This file indicates the amount of address space which a user process will 401 be restricted from mmapping. Since kernel null dereference bugs could 402 accidentally operate based on the information in the first couple of pages 403 of memory userspace processes should not be allowed to write to them. By 404 default this value is set to 0 and no protections will be enforced by the 405 security module. Setting this value to something like 64k will allow the 406 vast majority of applications to work correctly and provide defense in depth 407 against future potential kernel bugs. 408 409 ============================================================== 410 411 nr_hugepages 412 413 Change the minimum size of the hugepage pool. 414 415 See Documentation/vm/hugetlbpage.txt 416 417 ============================================================== 418 419 nr_overcommit_hugepages 420 421 Change the maximum size of the hugepage pool. The maximum is 422 nr_hugepages + nr_overcommit_hugepages. 423 424 See Documentation/vm/hugetlbpage.txt 425 426 ============================================================== 427 428 nr_trim_pages 429 430 This is available only on NOMMU kernels. 431 432 This value adjusts the excess page trimming behaviour of power-of-2 aligned 433 NOMMU mmap allocations. 434 435 A value of 0 disables trimming of allocations entirely, while a value of 1 436 trims excess pages aggressively. Any value >= 1 acts as the watermark where 437 trimming of allocations is initiated. 438 439 The default value is 1. 440 441 See Documentation/nommu-mmap.txt for more information. 442 443 ============================================================== 444 445 numa_zonelist_order 446 447 This sysctl is only for NUMA. 448 'where the memory is allocated from' is controlled by zonelists. 449 (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation. 450 you may be able to read ZONE_DMA as ZONE_DMA32...) 451 452 In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following. 453 ZONE_NORMAL -> ZONE_DMA 454 This means that a memory allocation request for GFP_KERNEL will 455 get memory from ZONE_DMA only when ZONE_NORMAL is not available. 456 457 In NUMA case, you can think of following 2 types of order. 458 Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL 459 460 (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL 461 (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA. 462 463 Type(A) offers the best locality for processes on Node(0), but ZONE_DMA 464 will be used before ZONE_NORMAL exhaustion. This increases possibility of 465 out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small. 466 467 Type(B) cannot offer the best locality but is more robust against OOM of 468 the DMA zone. 469 470 Type(A) is called as "Node" order. Type (B) is "Zone" order. 471 472 "Node order" orders the zonelists by node, then by zone within each node. 473 Specify "[Nn]ode" for node order 474 475 "Zone Order" orders the zonelists by zone type, then by node within each 476 zone. Specify "[Zz]one" for zone order. 477 478 Specify "[Dd]efault" to request automatic configuration. Autoconfiguration 479 will select "node" order in following case. 480 (1) if the DMA zone does not exist or 481 (2) if the DMA zone comprises greater than 50% of the available memory or 482 (3) if any node's DMA zone comprises greater than 60% of its local memory and 483 the amount of local memory is big enough. 484 485 Otherwise, "zone" order will be selected. Default order is recommended unless 486 this is causing problems for your system/application. 487 488 ============================================================== 489 490 oom_dump_tasks 491 492 Enables a system-wide task dump (excluding kernel threads) to be 493 produced when the kernel performs an OOM-killing and includes such 494 information as pid, uid, tgid, vm size, rss, nr_ptes, swapents, 495 oom_score_adj score, and name. This is helpful to determine why the 496 OOM killer was invoked, to identify the rogue task that caused it, 497 and to determine why the OOM killer chose the task it did to kill. 498 499 If this is set to zero, this information is suppressed. On very 500 large systems with thousands of tasks it may not be feasible to dump 501 the memory state information for each one. Such systems should not 502 be forced to incur a performance penalty in OOM conditions when the 503 information may not be desired. 504 505 If this is set to non-zero, this information is shown whenever the 506 OOM killer actually kills a memory-hogging task. 507 508 The default value is 1 (enabled). 509 510 ============================================================== 511 512 oom_kill_allocating_task 513 514 This enables or disables killing the OOM-triggering task in 515 out-of-memory situations. 516 517 If this is set to zero, the OOM killer will scan through the entire 518 tasklist and select a task based on heuristics to kill. This normally 519 selects a rogue memory-hogging task that frees up a large amount of 520 memory when killed. 521 522 If this is set to non-zero, the OOM killer simply kills the task that 523 triggered the out-of-memory condition. This avoids the expensive 524 tasklist scan. 525 526 If panic_on_oom is selected, it takes precedence over whatever value 527 is used in oom_kill_allocating_task. 528 529 The default value is 0. 530 531 ============================================================== 532 533 overcommit_memory: 534 535 This value contains a flag that enables memory overcommitment. 536 537 When this flag is 0, the kernel attempts to estimate the amount 538 of free memory left when userspace requests more memory. 539 540 When this flag is 1, the kernel pretends there is always enough 541 memory until it actually runs out. 542 543 When this flag is 2, the kernel uses a "never overcommit" 544 policy that attempts to prevent any overcommit of memory. 545 546 This feature can be very useful because there are a lot of 547 programs that malloc() huge amounts of memory "just-in-case" 548 and don't use much of it. 549 550 The default value is 0. 551 552 See Documentation/vm/overcommit-accounting and 553 security/commoncap.c::cap_vm_enough_memory() for more information. 554 555 ============================================================== 556 557 overcommit_ratio: 558 559 When overcommit_memory is set to 2, the committed address 560 space is not permitted to exceed swap plus this percentage 561 of physical RAM. See above. 562 563 ============================================================== 564 565 page-cluster 566 567 page-cluster controls the number of pages up to which consecutive pages 568 are read in from swap in a single attempt. This is the swap counterpart 569 to page cache readahead. 570 The mentioned consecutivity is not in terms of virtual/physical addresses, 571 but consecutive on swap space - that means they were swapped out together. 572 573 It is a logarithmic value - setting it to zero means "1 page", setting 574 it to 1 means "2 pages", setting it to 2 means "4 pages", etc. 575 Zero disables swap readahead completely. 576 577 The default value is three (eight pages at a time). There may be some 578 small benefits in tuning this to a different value if your workload is 579 swap-intensive. 580 581 Lower values mean lower latencies for initial faults, but at the same time 582 extra faults and I/O delays for following faults if they would have been part of 583 that consecutive pages readahead would have brought in. 584 585 ============================================================= 586 587 panic_on_oom 588 589 This enables or disables panic on out-of-memory feature. 590 591 If this is set to 0, the kernel will kill some rogue process, 592 called oom_killer. Usually, oom_killer can kill rogue processes and 593 system will survive. 594 595 If this is set to 1, the kernel panics when out-of-memory happens. 596 However, if a process limits using nodes by mempolicy/cpusets, 597 and those nodes become memory exhaustion status, one process 598 may be killed by oom-killer. No panic occurs in this case. 599 Because other nodes' memory may be free. This means system total status 600 may be not fatal yet. 601 602 If this is set to 2, the kernel panics compulsorily even on the 603 above-mentioned. Even oom happens under memory cgroup, the whole 604 system panics. 605 606 The default value is 0. 607 1 and 2 are for failover of clustering. Please select either 608 according to your policy of failover. 609 panic_on_oom=2+kdump gives you very strong tool to investigate 610 why oom happens. You can get snapshot. 611 612 ============================================================= 613 614 percpu_pagelist_fraction 615 616 This is the fraction of pages at most (high mark pcp->high) in each zone that 617 are allocated for each per cpu page list. The min value for this is 8. It 618 means that we don't allow more than 1/8th of pages in each zone to be 619 allocated in any single per_cpu_pagelist. This entry only changes the value 620 of hot per cpu pagelists. User can specify a number like 100 to allocate 621 1/100th of each zone to each per cpu page list. 622 623 The batch value of each per cpu pagelist is also updated as a result. It is 624 set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8) 625 626 The initial value is zero. Kernel does not use this value at boot time to set 627 the high water marks for each per cpu page list. 628 629 ============================================================== 630 631 stat_interval 632 633 The time interval between which vm statistics are updated. The default 634 is 1 second. 635 636 ============================================================== 637 638 swappiness 639 640 This control is used to define how aggressive the kernel will swap 641 memory pages. Higher values will increase agressiveness, lower values 642 decrease the amount of swap. 643 644 The default value is 60. 645 646 ============================================================== 647 648 vfs_cache_pressure 649 ------------------ 650 651 Controls the tendency of the kernel to reclaim the memory which is used for 652 caching of directory and inode objects. 653 654 At the default value of vfs_cache_pressure=100 the kernel will attempt to 655 reclaim dentries and inodes at a "fair" rate with respect to pagecache and 656 swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer 657 to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will 658 never reclaim dentries and inodes due to memory pressure and this can easily 659 lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100 660 causes the kernel to prefer to reclaim dentries and inodes. 661 662 ============================================================== 663 664 zone_reclaim_mode: 665 666 Zone_reclaim_mode allows someone to set more or less aggressive approaches to 667 reclaim memory when a zone runs out of memory. If it is set to zero then no 668 zone reclaim occurs. Allocations will be satisfied from other zones / nodes 669 in the system. 670 671 This is value ORed together of 672 673 1 = Zone reclaim on 674 2 = Zone reclaim writes dirty pages out 675 4 = Zone reclaim swaps pages 676 677 zone_reclaim_mode is set during bootup to 1 if it is determined that pages 678 from remote zones will cause a measurable performance reduction. The 679 page allocator will then reclaim easily reusable pages (those page 680 cache pages that are currently not used) before allocating off node pages. 681 682 It may be beneficial to switch off zone reclaim if the system is 683 used for a file server and all of memory should be used for caching files 684 from disk. In that case the caching effect is more important than 685 data locality. 686 687 Allowing zone reclaim to write out pages stops processes that are 688 writing large amounts of data from dirtying pages on other nodes. Zone 689 reclaim will write out dirty pages if a zone fills up and so effectively 690 throttle the process. This may decrease the performance of a single process 691 since it cannot use all of system memory to buffer the outgoing writes 692 anymore but it preserve the memory on other nodes so that the performance 693 of other processes running on other nodes will not be affected. 694 695 Allowing regular swap effectively restricts allocations to the local 696 node unless explicitly overridden by memory policies or cpuset 697 configurations. 698 699 ============ End of Document =================================