Commit | Line | Data |
---|---|---|
db0fb184 | 1 | Documentation for /proc/sys/vm/* kernel version 2.6.29 |
1da177e4 | 2 | (c) 1998, 1999, Rik van Riel <riel@nl.linux.org> |
db0fb184 | 3 | (c) 2008 Peter W. Morreale <pmorreale@novell.com> |
1da177e4 LT |
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 | |
db0fb184 | 10 | /proc/sys/vm and is valid for Linux kernel version 2.6.29. |
1da177e4 LT |
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: | |
db0fb184 | 20 | |
4eeab4f5 | 21 | - admin_reserve_kbytes |
db0fb184 | 22 | - block_dump |
76ab0f53 | 23 | - compact_memory |
db0fb184 | 24 | - dirty_background_bytes |
1da177e4 | 25 | - dirty_background_ratio |
db0fb184 | 26 | - dirty_bytes |
1da177e4 | 27 | - dirty_expire_centisecs |
db0fb184 | 28 | - dirty_ratio |
1da177e4 | 29 | - dirty_writeback_centisecs |
db0fb184 | 30 | - drop_caches |
5e771905 | 31 | - extfrag_threshold |
db0fb184 PM |
32 | - hugepages_treat_as_movable |
33 | - hugetlb_shm_group | |
34 | - laptop_mode | |
35 | - legacy_va_layout | |
36 | - lowmem_reserve_ratio | |
1da177e4 | 37 | - max_map_count |
6a46079c AK |
38 | - memory_failure_early_kill |
39 | - memory_failure_recovery | |
1da177e4 | 40 | - min_free_kbytes |
0ff38490 | 41 | - min_slab_ratio |
db0fb184 PM |
42 | - min_unmapped_ratio |
43 | - mmap_min_addr | |
d5dbac87 NA |
44 | - nr_hugepages |
45 | - nr_overcommit_hugepages | |
db0fb184 PM |
46 | - nr_trim_pages (only if CONFIG_MMU=n) |
47 | - numa_zonelist_order | |
48 | - oom_dump_tasks | |
49 | - oom_kill_allocating_task | |
50 | - overcommit_memory | |
51 | - overcommit_ratio | |
52 | - page-cluster | |
53 | - panic_on_oom | |
54 | - percpu_pagelist_fraction | |
55 | - stat_interval | |
56 | - swappiness | |
c9b1d098 | 57 | - user_reserve_kbytes |
db0fb184 PM |
58 | - vfs_cache_pressure |
59 | - zone_reclaim_mode | |
60 | ||
1da177e4 LT |
61 | ============================================================== |
62 | ||
4eeab4f5 AS |
63 | admin_reserve_kbytes |
64 | ||
65 | The amount of free memory in the system that should be reserved for users | |
66 | with the capability cap_sys_admin. | |
67 | ||
68 | admin_reserve_kbytes defaults to min(3% of free pages, 8MB) | |
69 | ||
70 | That should provide enough for the admin to log in and kill a process, | |
71 | if necessary, under the default overcommit 'guess' mode. | |
72 | ||
73 | Systems running under overcommit 'never' should increase this to account | |
74 | for the full Virtual Memory Size of programs used to recover. Otherwise, | |
75 | root may not be able to log in to recover the system. | |
76 | ||
77 | How do you calculate a minimum useful reserve? | |
78 | ||
79 | sshd or login + bash (or some other shell) + top (or ps, kill, etc.) | |
80 | ||
81 | For overcommit 'guess', we can sum resident set sizes (RSS). | |
82 | On x86_64 this is about 8MB. | |
83 | ||
84 | For overcommit 'never', we can take the max of their virtual sizes (VSZ) | |
85 | and add the sum of their RSS. | |
86 | On x86_64 this is about 128MB. | |
87 | ||
88 | Changing this takes effect whenever an application requests memory. | |
89 | ||
90 | ============================================================== | |
91 | ||
db0fb184 | 92 | block_dump |
1da177e4 | 93 | |
db0fb184 PM |
94 | block_dump enables block I/O debugging when set to a nonzero value. More |
95 | information on block I/O debugging is in Documentation/laptops/laptop-mode.txt. | |
1da177e4 LT |
96 | |
97 | ============================================================== | |
98 | ||
76ab0f53 MG |
99 | compact_memory |
100 | ||
101 | Available only when CONFIG_COMPACTION is set. When 1 is written to the file, | |
102 | all zones are compacted such that free memory is available in contiguous | |
103 | blocks where possible. This can be important for example in the allocation of | |
104 | huge pages although processes will also directly compact memory as required. | |
105 | ||
106 | ============================================================== | |
107 | ||
db0fb184 | 108 | dirty_background_bytes |
1da177e4 | 109 | |
6601fac8 AB |
110 | Contains the amount of dirty memory at which the background kernel |
111 | flusher threads will start writeback. | |
1da177e4 | 112 | |
abffc020 AR |
113 | Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only |
114 | one of them may be specified at a time. When one sysctl is written it is | |
115 | immediately taken into account to evaluate the dirty memory limits and the | |
116 | other appears as 0 when read. | |
1da177e4 | 117 | |
db0fb184 | 118 | ============================================================== |
1da177e4 | 119 | |
db0fb184 | 120 | dirty_background_ratio |
1da177e4 | 121 | |
db0fb184 | 122 | Contains, as a percentage of total system memory, the number of pages at which |
6601fac8 | 123 | the background kernel flusher threads will start writing out dirty data. |
1da177e4 | 124 | |
db0fb184 | 125 | ============================================================== |
1da177e4 | 126 | |
db0fb184 PM |
127 | dirty_bytes |
128 | ||
129 | Contains the amount of dirty memory at which a process generating disk writes | |
130 | will itself start writeback. | |
131 | ||
abffc020 AR |
132 | Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be |
133 | specified at a time. When one sysctl is written it is immediately taken into | |
134 | account to evaluate the dirty memory limits and the other appears as 0 when | |
135 | read. | |
1da177e4 | 136 | |
9e4a5bda AR |
137 | Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any |
138 | value lower than this limit will be ignored and the old configuration will be | |
139 | retained. | |
140 | ||
1da177e4 LT |
141 | ============================================================== |
142 | ||
db0fb184 | 143 | dirty_expire_centisecs |
1da177e4 | 144 | |
db0fb184 | 145 | This tunable is used to define when dirty data is old enough to be eligible |
6601fac8 AB |
146 | for writeout by the kernel flusher threads. It is expressed in 100'ths |
147 | of a second. Data which has been dirty in-memory for longer than this | |
148 | interval will be written out next time a flusher thread wakes up. | |
db0fb184 PM |
149 | |
150 | ============================================================== | |
151 | ||
152 | dirty_ratio | |
153 | ||
154 | Contains, as a percentage of total system memory, the number of pages at which | |
155 | a process which is generating disk writes will itself start writing out dirty | |
156 | data. | |
1da177e4 LT |
157 | |
158 | ============================================================== | |
159 | ||
db0fb184 | 160 | dirty_writeback_centisecs |
1da177e4 | 161 | |
6601fac8 | 162 | The kernel flusher threads will periodically wake up and write `old' data |
db0fb184 PM |
163 | out to disk. This tunable expresses the interval between those wakeups, in |
164 | 100'ths of a second. | |
1da177e4 | 165 | |
db0fb184 | 166 | Setting this to zero disables periodic writeback altogether. |
1da177e4 LT |
167 | |
168 | ============================================================== | |
169 | ||
db0fb184 | 170 | drop_caches |
1da177e4 | 171 | |
db0fb184 PM |
172 | Writing to this will cause the kernel to drop clean caches, dentries and |
173 | inodes from memory, causing that memory to become free. | |
1da177e4 | 174 | |
db0fb184 PM |
175 | To free pagecache: |
176 | echo 1 > /proc/sys/vm/drop_caches | |
177 | To free dentries and inodes: | |
178 | echo 2 > /proc/sys/vm/drop_caches | |
179 | To free pagecache, dentries and inodes: | |
180 | echo 3 > /proc/sys/vm/drop_caches | |
1da177e4 | 181 | |
db0fb184 PM |
182 | As this is a non-destructive operation and dirty objects are not freeable, the |
183 | user should run `sync' first. | |
1da177e4 LT |
184 | |
185 | ============================================================== | |
186 | ||
5e771905 MG |
187 | extfrag_threshold |
188 | ||
189 | This parameter affects whether the kernel will compact memory or direct | |
190 | reclaim to satisfy a high-order allocation. /proc/extfrag_index shows what | |
191 | the fragmentation index for each order is in each zone in the system. Values | |
192 | tending towards 0 imply allocations would fail due to lack of memory, | |
193 | values towards 1000 imply failures are due to fragmentation and -1 implies | |
194 | that the allocation will succeed as long as watermarks are met. | |
195 | ||
196 | The kernel will not compact memory in a zone if the | |
197 | fragmentation index is <= extfrag_threshold. The default value is 500. | |
198 | ||
199 | ============================================================== | |
200 | ||
db0fb184 | 201 | hugepages_treat_as_movable |
1da177e4 | 202 | |
86cdb465 NH |
203 | This parameter controls whether we can allocate hugepages from ZONE_MOVABLE |
204 | or not. If set to non-zero, hugepages can be allocated from ZONE_MOVABLE. | |
205 | ZONE_MOVABLE is created when kernel boot parameter kernelcore= is specified, | |
206 | so this parameter has no effect if used without kernelcore=. | |
207 | ||
208 | Hugepage migration is now available in some situations which depend on the | |
209 | architecture and/or the hugepage size. If a hugepage supports migration, | |
210 | allocation from ZONE_MOVABLE is always enabled for the hugepage regardless | |
211 | of the value of this parameter. | |
212 | IOW, this parameter affects only non-migratable hugepages. | |
213 | ||
214 | Assuming that hugepages are not migratable in your system, one usecase of | |
215 | this parameter is that users can make hugepage pool more extensible by | |
216 | enabling the allocation from ZONE_MOVABLE. This is because on ZONE_MOVABLE | |
217 | page reclaim/migration/compaction work more and you can get contiguous | |
218 | memory more likely. Note that using ZONE_MOVABLE for non-migratable | |
219 | hugepages can do harm to other features like memory hotremove (because | |
220 | memory hotremove expects that memory blocks on ZONE_MOVABLE are always | |
221 | removable,) so it's a trade-off responsible for the users. | |
24950898 | 222 | |
8ad4b1fb RS |
223 | ============================================================== |
224 | ||
db0fb184 | 225 | hugetlb_shm_group |
8ad4b1fb | 226 | |
db0fb184 PM |
227 | hugetlb_shm_group contains group id that is allowed to create SysV |
228 | shared memory segment using hugetlb page. | |
8ad4b1fb | 229 | |
db0fb184 | 230 | ============================================================== |
8ad4b1fb | 231 | |
db0fb184 | 232 | laptop_mode |
1743660b | 233 | |
db0fb184 PM |
234 | laptop_mode is a knob that controls "laptop mode". All the things that are |
235 | controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt. | |
1743660b | 236 | |
db0fb184 | 237 | ============================================================== |
1743660b | 238 | |
db0fb184 | 239 | legacy_va_layout |
1b2ffb78 | 240 | |
2174efb6 | 241 | If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel |
db0fb184 | 242 | will use the legacy (2.4) layout for all processes. |
1b2ffb78 | 243 | |
db0fb184 | 244 | ============================================================== |
1b2ffb78 | 245 | |
db0fb184 PM |
246 | lowmem_reserve_ratio |
247 | ||
248 | For some specialised workloads on highmem machines it is dangerous for | |
249 | the kernel to allow process memory to be allocated from the "lowmem" | |
250 | zone. This is because that memory could then be pinned via the mlock() | |
251 | system call, or by unavailability of swapspace. | |
252 | ||
253 | And on large highmem machines this lack of reclaimable lowmem memory | |
254 | can be fatal. | |
255 | ||
256 | So the Linux page allocator has a mechanism which prevents allocations | |
257 | which _could_ use highmem from using too much lowmem. This means that | |
258 | a certain amount of lowmem is defended from the possibility of being | |
259 | captured into pinned user memory. | |
260 | ||
261 | (The same argument applies to the old 16 megabyte ISA DMA region. This | |
262 | mechanism will also defend that region from allocations which could use | |
263 | highmem or lowmem). | |
264 | ||
265 | The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is | |
266 | in defending these lower zones. | |
267 | ||
268 | If you have a machine which uses highmem or ISA DMA and your | |
269 | applications are using mlock(), or if you are running with no swap then | |
270 | you probably should change the lowmem_reserve_ratio setting. | |
271 | ||
272 | The lowmem_reserve_ratio is an array. You can see them by reading this file. | |
273 | - | |
274 | % cat /proc/sys/vm/lowmem_reserve_ratio | |
275 | 256 256 32 | |
276 | - | |
277 | Note: # of this elements is one fewer than number of zones. Because the highest | |
278 | zone's value is not necessary for following calculation. | |
279 | ||
280 | But, these values are not used directly. The kernel calculates # of protection | |
281 | pages for each zones from them. These are shown as array of protection pages | |
282 | in /proc/zoneinfo like followings. (This is an example of x86-64 box). | |
283 | Each zone has an array of protection pages like this. | |
284 | ||
285 | - | |
286 | Node 0, zone DMA | |
287 | pages free 1355 | |
288 | min 3 | |
289 | low 3 | |
290 | high 4 | |
291 | : | |
292 | : | |
293 | numa_other 0 | |
294 | protection: (0, 2004, 2004, 2004) | |
295 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | |
296 | pagesets | |
297 | cpu: 0 pcp: 0 | |
298 | : | |
299 | - | |
300 | These protections are added to score to judge whether this zone should be used | |
301 | for page allocation or should be reclaimed. | |
302 | ||
303 | In this example, if normal pages (index=2) are required to this DMA zone and | |
41858966 MG |
304 | watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should |
305 | not be used because pages_free(1355) is smaller than watermark + protection[2] | |
db0fb184 PM |
306 | (4 + 2004 = 2008). If this protection value is 0, this zone would be used for |
307 | normal page requirement. If requirement is DMA zone(index=0), protection[0] | |
308 | (=0) is used. | |
309 | ||
310 | zone[i]'s protection[j] is calculated by following expression. | |
311 | ||
312 | (i < j): | |
313 | zone[i]->protection[j] | |
314 | = (total sums of present_pages from zone[i+1] to zone[j] on the node) | |
315 | / lowmem_reserve_ratio[i]; | |
316 | (i = j): | |
317 | (should not be protected. = 0; | |
318 | (i > j): | |
319 | (not necessary, but looks 0) | |
320 | ||
321 | The default values of lowmem_reserve_ratio[i] are | |
322 | 256 (if zone[i] means DMA or DMA32 zone) | |
323 | 32 (others). | |
324 | As above expression, they are reciprocal number of ratio. | |
325 | 256 means 1/256. # of protection pages becomes about "0.39%" of total present | |
326 | pages of higher zones on the node. | |
327 | ||
328 | If you would like to protect more pages, smaller values are effective. | |
329 | The minimum value is 1 (1/1 -> 100%). | |
1b2ffb78 | 330 | |
db0fb184 | 331 | ============================================================== |
1b2ffb78 | 332 | |
db0fb184 | 333 | max_map_count: |
1743660b | 334 | |
db0fb184 PM |
335 | This file contains the maximum number of memory map areas a process |
336 | may have. Memory map areas are used as a side-effect of calling | |
337 | malloc, directly by mmap and mprotect, and also when loading shared | |
338 | libraries. | |
1743660b | 339 | |
db0fb184 PM |
340 | While most applications need less than a thousand maps, certain |
341 | programs, particularly malloc debuggers, may consume lots of them, | |
342 | e.g., up to one or two maps per allocation. | |
fadd8fbd | 343 | |
db0fb184 | 344 | The default value is 65536. |
9614634f | 345 | |
6a46079c AK |
346 | ============================================================= |
347 | ||
348 | memory_failure_early_kill: | |
349 | ||
350 | Control how to kill processes when uncorrected memory error (typically | |
351 | a 2bit error in a memory module) is detected in the background by hardware | |
352 | that cannot be handled by the kernel. In some cases (like the page | |
353 | still having a valid copy on disk) the kernel will handle the failure | |
354 | transparently without affecting any applications. But if there is | |
355 | no other uptodate copy of the data it will kill to prevent any data | |
356 | corruptions from propagating. | |
357 | ||
358 | 1: Kill all processes that have the corrupted and not reloadable page mapped | |
359 | as soon as the corruption is detected. Note this is not supported | |
360 | for a few types of pages, like kernel internally allocated data or | |
361 | the swap cache, but works for the majority of user pages. | |
362 | ||
363 | 0: Only unmap the corrupted page from all processes and only kill a process | |
364 | who tries to access it. | |
365 | ||
366 | The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can | |
367 | handle this if they want to. | |
368 | ||
369 | This is only active on architectures/platforms with advanced machine | |
370 | check handling and depends on the hardware capabilities. | |
371 | ||
372 | Applications can override this setting individually with the PR_MCE_KILL prctl | |
373 | ||
374 | ============================================================== | |
375 | ||
376 | memory_failure_recovery | |
377 | ||
378 | Enable memory failure recovery (when supported by the platform) | |
379 | ||
380 | 1: Attempt recovery. | |
381 | ||
382 | 0: Always panic on a memory failure. | |
383 | ||
db0fb184 | 384 | ============================================================== |
9614634f | 385 | |
db0fb184 | 386 | min_free_kbytes: |
9614634f | 387 | |
db0fb184 | 388 | This is used to force the Linux VM to keep a minimum number |
41858966 MG |
389 | of kilobytes free. The VM uses this number to compute a |
390 | watermark[WMARK_MIN] value for each lowmem zone in the system. | |
391 | Each lowmem zone gets a number of reserved free pages based | |
392 | proportionally on its size. | |
db0fb184 PM |
393 | |
394 | Some minimal amount of memory is needed to satisfy PF_MEMALLOC | |
395 | allocations; if you set this to lower than 1024KB, your system will | |
396 | become subtly broken, and prone to deadlock under high loads. | |
397 | ||
398 | Setting this too high will OOM your machine instantly. | |
9614634f CL |
399 | |
400 | ============================================================= | |
401 | ||
0ff38490 CL |
402 | min_slab_ratio: |
403 | ||
404 | This is available only on NUMA kernels. | |
405 | ||
406 | A percentage of the total pages in each zone. On Zone reclaim | |
407 | (fallback from the local zone occurs) slabs will be reclaimed if more | |
408 | than this percentage of pages in a zone are reclaimable slab pages. | |
409 | This insures that the slab growth stays under control even in NUMA | |
410 | systems that rarely perform global reclaim. | |
411 | ||
412 | The default is 5 percent. | |
413 | ||
414 | Note that slab reclaim is triggered in a per zone / node fashion. | |
415 | The process of reclaiming slab memory is currently not node specific | |
416 | and may not be fast. | |
417 | ||
418 | ============================================================= | |
419 | ||
db0fb184 | 420 | min_unmapped_ratio: |
fadd8fbd | 421 | |
db0fb184 | 422 | This is available only on NUMA kernels. |
fadd8fbd | 423 | |
90afa5de MG |
424 | This is a percentage of the total pages in each zone. Zone reclaim will |
425 | only occur if more than this percentage of pages are in a state that | |
426 | zone_reclaim_mode allows to be reclaimed. | |
427 | ||
428 | If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared | |
429 | against all file-backed unmapped pages including swapcache pages and tmpfs | |
430 | files. Otherwise, only unmapped pages backed by normal files but not tmpfs | |
431 | files and similar are considered. | |
2b744c01 | 432 | |
db0fb184 | 433 | The default is 1 percent. |
fadd8fbd | 434 | |
db0fb184 | 435 | ============================================================== |
2b744c01 | 436 | |
db0fb184 | 437 | mmap_min_addr |
ed032189 | 438 | |
db0fb184 | 439 | This file indicates the amount of address space which a user process will |
af901ca1 | 440 | be restricted from mmapping. Since kernel null dereference bugs could |
db0fb184 PM |
441 | accidentally operate based on the information in the first couple of pages |
442 | of memory userspace processes should not be allowed to write to them. By | |
443 | default this value is set to 0 and no protections will be enforced by the | |
444 | security module. Setting this value to something like 64k will allow the | |
445 | vast majority of applications to work correctly and provide defense in depth | |
446 | against future potential kernel bugs. | |
fe071d7e | 447 | |
db0fb184 | 448 | ============================================================== |
fef1bdd6 | 449 | |
db0fb184 | 450 | nr_hugepages |
fef1bdd6 | 451 | |
db0fb184 | 452 | Change the minimum size of the hugepage pool. |
fef1bdd6 | 453 | |
db0fb184 | 454 | See Documentation/vm/hugetlbpage.txt |
fef1bdd6 | 455 | |
db0fb184 | 456 | ============================================================== |
fef1bdd6 | 457 | |
db0fb184 | 458 | nr_overcommit_hugepages |
fef1bdd6 | 459 | |
db0fb184 PM |
460 | Change the maximum size of the hugepage pool. The maximum is |
461 | nr_hugepages + nr_overcommit_hugepages. | |
fe071d7e | 462 | |
db0fb184 | 463 | See Documentation/vm/hugetlbpage.txt |
fe071d7e | 464 | |
db0fb184 | 465 | ============================================================== |
fe071d7e | 466 | |
db0fb184 | 467 | nr_trim_pages |
ed032189 | 468 | |
db0fb184 PM |
469 | This is available only on NOMMU kernels. |
470 | ||
471 | This value adjusts the excess page trimming behaviour of power-of-2 aligned | |
472 | NOMMU mmap allocations. | |
473 | ||
474 | A value of 0 disables trimming of allocations entirely, while a value of 1 | |
475 | trims excess pages aggressively. Any value >= 1 acts as the watermark where | |
476 | trimming of allocations is initiated. | |
477 | ||
478 | The default value is 1. | |
479 | ||
480 | See Documentation/nommu-mmap.txt for more information. | |
ed032189 | 481 | |
f0c0b2b8 KH |
482 | ============================================================== |
483 | ||
484 | numa_zonelist_order | |
485 | ||
486 | This sysctl is only for NUMA. | |
487 | 'where the memory is allocated from' is controlled by zonelists. | |
488 | (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation. | |
489 | you may be able to read ZONE_DMA as ZONE_DMA32...) | |
490 | ||
491 | In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following. | |
492 | ZONE_NORMAL -> ZONE_DMA | |
493 | This means that a memory allocation request for GFP_KERNEL will | |
494 | get memory from ZONE_DMA only when ZONE_NORMAL is not available. | |
495 | ||
496 | In NUMA case, you can think of following 2 types of order. | |
497 | Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL | |
498 | ||
499 | (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL | |
500 | (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA. | |
501 | ||
502 | Type(A) offers the best locality for processes on Node(0), but ZONE_DMA | |
503 | will be used before ZONE_NORMAL exhaustion. This increases possibility of | |
504 | out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small. | |
505 | ||
506 | Type(B) cannot offer the best locality but is more robust against OOM of | |
507 | the DMA zone. | |
508 | ||
509 | Type(A) is called as "Node" order. Type (B) is "Zone" order. | |
510 | ||
511 | "Node order" orders the zonelists by node, then by zone within each node. | |
5a3016a6 | 512 | Specify "[Nn]ode" for node order |
f0c0b2b8 KH |
513 | |
514 | "Zone Order" orders the zonelists by zone type, then by node within each | |
5a3016a6 | 515 | zone. Specify "[Zz]one" for zone order. |
f0c0b2b8 KH |
516 | |
517 | Specify "[Dd]efault" to request automatic configuration. Autoconfiguration | |
518 | will select "node" order in following case. | |
519 | (1) if the DMA zone does not exist or | |
520 | (2) if the DMA zone comprises greater than 50% of the available memory or | |
f8f191f1 | 521 | (3) if any node's DMA zone comprises greater than 70% of its local memory and |
f0c0b2b8 KH |
522 | the amount of local memory is big enough. |
523 | ||
524 | Otherwise, "zone" order will be selected. Default order is recommended unless | |
525 | this is causing problems for your system/application. | |
d5dbac87 NA |
526 | |
527 | ============================================================== | |
528 | ||
db0fb184 | 529 | oom_dump_tasks |
d5dbac87 | 530 | |
db0fb184 PM |
531 | Enables a system-wide task dump (excluding kernel threads) to be |
532 | produced when the kernel performs an OOM-killing and includes such | |
de34d965 DR |
533 | information as pid, uid, tgid, vm size, rss, nr_ptes, swapents, |
534 | oom_score_adj score, and name. This is helpful to determine why the | |
535 | OOM killer was invoked, to identify the rogue task that caused it, | |
536 | and to determine why the OOM killer chose the task it did to kill. | |
d5dbac87 | 537 | |
db0fb184 PM |
538 | If this is set to zero, this information is suppressed. On very |
539 | large systems with thousands of tasks it may not be feasible to dump | |
540 | the memory state information for each one. Such systems should not | |
541 | be forced to incur a performance penalty in OOM conditions when the | |
542 | information may not be desired. | |
543 | ||
544 | If this is set to non-zero, this information is shown whenever the | |
545 | OOM killer actually kills a memory-hogging task. | |
546 | ||
ad915c43 | 547 | The default value is 1 (enabled). |
d5dbac87 NA |
548 | |
549 | ============================================================== | |
550 | ||
db0fb184 | 551 | oom_kill_allocating_task |
d5dbac87 | 552 | |
db0fb184 PM |
553 | This enables or disables killing the OOM-triggering task in |
554 | out-of-memory situations. | |
d5dbac87 | 555 | |
db0fb184 PM |
556 | If this is set to zero, the OOM killer will scan through the entire |
557 | tasklist and select a task based on heuristics to kill. This normally | |
558 | selects a rogue memory-hogging task that frees up a large amount of | |
559 | memory when killed. | |
560 | ||
561 | If this is set to non-zero, the OOM killer simply kills the task that | |
562 | triggered the out-of-memory condition. This avoids the expensive | |
563 | tasklist scan. | |
564 | ||
565 | If panic_on_oom is selected, it takes precedence over whatever value | |
566 | is used in oom_kill_allocating_task. | |
567 | ||
568 | The default value is 0. | |
dd8632a1 PM |
569 | |
570 | ============================================================== | |
571 | ||
db0fb184 | 572 | overcommit_memory: |
dd8632a1 | 573 | |
db0fb184 | 574 | This value contains a flag that enables memory overcommitment. |
dd8632a1 | 575 | |
db0fb184 PM |
576 | When this flag is 0, the kernel attempts to estimate the amount |
577 | of free memory left when userspace requests more memory. | |
dd8632a1 | 578 | |
db0fb184 PM |
579 | When this flag is 1, the kernel pretends there is always enough |
580 | memory until it actually runs out. | |
dd8632a1 | 581 | |
db0fb184 PM |
582 | When this flag is 2, the kernel uses a "never overcommit" |
583 | policy that attempts to prevent any overcommit of memory. | |
c9b1d098 | 584 | Note that user_reserve_kbytes affects this policy. |
dd8632a1 | 585 | |
db0fb184 PM |
586 | This feature can be very useful because there are a lot of |
587 | programs that malloc() huge amounts of memory "just-in-case" | |
588 | and don't use much of it. | |
589 | ||
590 | The default value is 0. | |
591 | ||
592 | See Documentation/vm/overcommit-accounting and | |
593 | security/commoncap.c::cap_vm_enough_memory() for more information. | |
594 | ||
595 | ============================================================== | |
596 | ||
597 | overcommit_ratio: | |
598 | ||
599 | When overcommit_memory is set to 2, the committed address | |
600 | space is not permitted to exceed swap plus this percentage | |
601 | of physical RAM. See above. | |
602 | ||
603 | ============================================================== | |
604 | ||
605 | page-cluster | |
606 | ||
df858fa8 CE |
607 | page-cluster controls the number of pages up to which consecutive pages |
608 | are read in from swap in a single attempt. This is the swap counterpart | |
609 | to page cache readahead. | |
610 | The mentioned consecutivity is not in terms of virtual/physical addresses, | |
611 | but consecutive on swap space - that means they were swapped out together. | |
db0fb184 PM |
612 | |
613 | It is a logarithmic value - setting it to zero means "1 page", setting | |
614 | it to 1 means "2 pages", setting it to 2 means "4 pages", etc. | |
df858fa8 | 615 | Zero disables swap readahead completely. |
db0fb184 PM |
616 | |
617 | The default value is three (eight pages at a time). There may be some | |
618 | small benefits in tuning this to a different value if your workload is | |
619 | swap-intensive. | |
620 | ||
df858fa8 CE |
621 | Lower values mean lower latencies for initial faults, but at the same time |
622 | extra faults and I/O delays for following faults if they would have been part of | |
623 | that consecutive pages readahead would have brought in. | |
624 | ||
db0fb184 PM |
625 | ============================================================= |
626 | ||
627 | panic_on_oom | |
628 | ||
629 | This enables or disables panic on out-of-memory feature. | |
630 | ||
631 | If this is set to 0, the kernel will kill some rogue process, | |
632 | called oom_killer. Usually, oom_killer can kill rogue processes and | |
633 | system will survive. | |
634 | ||
635 | If this is set to 1, the kernel panics when out-of-memory happens. | |
636 | However, if a process limits using nodes by mempolicy/cpusets, | |
637 | and those nodes become memory exhaustion status, one process | |
638 | may be killed by oom-killer. No panic occurs in this case. | |
639 | Because other nodes' memory may be free. This means system total status | |
640 | may be not fatal yet. | |
641 | ||
642 | If this is set to 2, the kernel panics compulsorily even on the | |
daaf1e68 KH |
643 | above-mentioned. Even oom happens under memory cgroup, the whole |
644 | system panics. | |
db0fb184 PM |
645 | |
646 | The default value is 0. | |
647 | 1 and 2 are for failover of clustering. Please select either | |
648 | according to your policy of failover. | |
daaf1e68 KH |
649 | panic_on_oom=2+kdump gives you very strong tool to investigate |
650 | why oom happens. You can get snapshot. | |
db0fb184 PM |
651 | |
652 | ============================================================= | |
653 | ||
654 | percpu_pagelist_fraction | |
655 | ||
656 | This is the fraction of pages at most (high mark pcp->high) in each zone that | |
657 | are allocated for each per cpu page list. The min value for this is 8. It | |
658 | means that we don't allow more than 1/8th of pages in each zone to be | |
659 | allocated in any single per_cpu_pagelist. This entry only changes the value | |
660 | of hot per cpu pagelists. User can specify a number like 100 to allocate | |
661 | 1/100th of each zone to each per cpu page list. | |
662 | ||
663 | The batch value of each per cpu pagelist is also updated as a result. It is | |
664 | set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8) | |
665 | ||
666 | The initial value is zero. Kernel does not use this value at boot time to set | |
667 | the high water marks for each per cpu page list. | |
668 | ||
669 | ============================================================== | |
670 | ||
671 | stat_interval | |
672 | ||
673 | The time interval between which vm statistics are updated. The default | |
674 | is 1 second. | |
675 | ||
676 | ============================================================== | |
677 | ||
678 | swappiness | |
679 | ||
680 | This control is used to define how aggressive the kernel will swap | |
681 | memory pages. Higher values will increase agressiveness, lower values | |
19f59460 | 682 | decrease the amount of swap. |
db0fb184 PM |
683 | |
684 | The default value is 60. | |
685 | ||
686 | ============================================================== | |
687 | ||
c9b1d098 AS |
688 | - user_reserve_kbytes |
689 | ||
690 | When overcommit_memory is set to 2, "never overommit" mode, reserve | |
691 | min(3% of current process size, user_reserve_kbytes) of free memory. | |
692 | This is intended to prevent a user from starting a single memory hogging | |
693 | process, such that they cannot recover (kill the hog). | |
694 | ||
695 | user_reserve_kbytes defaults to min(3% of the current process size, 128MB). | |
696 | ||
697 | If this is reduced to zero, then the user will be allowed to allocate | |
698 | all free memory with a single process, minus admin_reserve_kbytes. | |
699 | Any subsequent attempts to execute a command will result in | |
700 | "fork: Cannot allocate memory". | |
701 | ||
702 | Changing this takes effect whenever an application requests memory. | |
703 | ||
704 | ============================================================== | |
705 | ||
db0fb184 PM |
706 | vfs_cache_pressure |
707 | ------------------ | |
708 | ||
709 | Controls the tendency of the kernel to reclaim the memory which is used for | |
710 | caching of directory and inode objects. | |
711 | ||
712 | At the default value of vfs_cache_pressure=100 the kernel will attempt to | |
713 | reclaim dentries and inodes at a "fair" rate with respect to pagecache and | |
714 | swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer | |
55c37a84 JK |
715 | to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will |
716 | never reclaim dentries and inodes due to memory pressure and this can easily | |
717 | lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100 | |
db0fb184 PM |
718 | causes the kernel to prefer to reclaim dentries and inodes. |
719 | ||
720 | ============================================================== | |
721 | ||
722 | zone_reclaim_mode: | |
723 | ||
724 | Zone_reclaim_mode allows someone to set more or less aggressive approaches to | |
725 | reclaim memory when a zone runs out of memory. If it is set to zero then no | |
726 | zone reclaim occurs. Allocations will be satisfied from other zones / nodes | |
727 | in the system. | |
728 | ||
729 | This is value ORed together of | |
730 | ||
731 | 1 = Zone reclaim on | |
732 | 2 = Zone reclaim writes dirty pages out | |
733 | 4 = Zone reclaim swaps pages | |
734 | ||
735 | zone_reclaim_mode is set during bootup to 1 if it is determined that pages | |
736 | from remote zones will cause a measurable performance reduction. The | |
737 | page allocator will then reclaim easily reusable pages (those page | |
738 | cache pages that are currently not used) before allocating off node pages. | |
739 | ||
740 | It may be beneficial to switch off zone reclaim if the system is | |
741 | used for a file server and all of memory should be used for caching files | |
742 | from disk. In that case the caching effect is more important than | |
743 | data locality. | |
744 | ||
745 | Allowing zone reclaim to write out pages stops processes that are | |
746 | writing large amounts of data from dirtying pages on other nodes. Zone | |
747 | reclaim will write out dirty pages if a zone fills up and so effectively | |
748 | throttle the process. This may decrease the performance of a single process | |
749 | since it cannot use all of system memory to buffer the outgoing writes | |
750 | anymore but it preserve the memory on other nodes so that the performance | |
751 | of other processes running on other nodes will not be affected. | |
752 | ||
753 | Allowing regular swap effectively restricts allocations to the local | |
754 | node unless explicitly overridden by memory policies or cpuset | |
755 | configurations. | |
756 | ||
757 | ============ End of Document ================================= |