]>
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 |
5bbe3547 | 24 | - compact_unevictable_allowed |
db0fb184 | 25 | - dirty_background_bytes |
1da177e4 | 26 | - dirty_background_ratio |
db0fb184 | 27 | - dirty_bytes |
1da177e4 | 28 | - dirty_expire_centisecs |
db0fb184 | 29 | - dirty_ratio |
fc1ca3d5 | 30 | - dirtytime_expire_seconds |
1da177e4 | 31 | - dirty_writeback_centisecs |
db0fb184 | 32 | - drop_caches |
5e771905 | 33 | - extfrag_threshold |
db0fb184 PM |
34 | - hugetlb_shm_group |
35 | - laptop_mode | |
36 | - legacy_va_layout | |
37 | - lowmem_reserve_ratio | |
1da177e4 | 38 | - max_map_count |
6a46079c AK |
39 | - memory_failure_early_kill |
40 | - memory_failure_recovery | |
1da177e4 | 41 | - min_free_kbytes |
0ff38490 | 42 | - min_slab_ratio |
db0fb184 PM |
43 | - min_unmapped_ratio |
44 | - mmap_min_addr | |
d07e2259 DC |
45 | - mmap_rnd_bits |
46 | - mmap_rnd_compat_bits | |
d5dbac87 | 47 | - nr_hugepages |
d1634e1a | 48 | - nr_hugepages_mempolicy |
d5dbac87 | 49 | - nr_overcommit_hugepages |
db0fb184 PM |
50 | - nr_trim_pages (only if CONFIG_MMU=n) |
51 | - numa_zonelist_order | |
52 | - oom_dump_tasks | |
53 | - oom_kill_allocating_task | |
49f0ce5f | 54 | - overcommit_kbytes |
db0fb184 PM |
55 | - overcommit_memory |
56 | - overcommit_ratio | |
57 | - page-cluster | |
58 | - panic_on_oom | |
59 | - percpu_pagelist_fraction | |
60 | - stat_interval | |
52b6f46b | 61 | - stat_refresh |
4518085e | 62 | - numa_stat |
db0fb184 | 63 | - swappiness |
c9b1d098 | 64 | - user_reserve_kbytes |
db0fb184 | 65 | - vfs_cache_pressure |
1c30844d | 66 | - watermark_boost_factor |
e6507a00 | 67 | - watermark_scale_factor |
db0fb184 PM |
68 | - zone_reclaim_mode |
69 | ||
1da177e4 LT |
70 | ============================================================== |
71 | ||
4eeab4f5 AS |
72 | admin_reserve_kbytes |
73 | ||
74 | The amount of free memory in the system that should be reserved for users | |
75 | with the capability cap_sys_admin. | |
76 | ||
77 | admin_reserve_kbytes defaults to min(3% of free pages, 8MB) | |
78 | ||
79 | That should provide enough for the admin to log in and kill a process, | |
80 | if necessary, under the default overcommit 'guess' mode. | |
81 | ||
82 | Systems running under overcommit 'never' should increase this to account | |
83 | for the full Virtual Memory Size of programs used to recover. Otherwise, | |
84 | root may not be able to log in to recover the system. | |
85 | ||
86 | How do you calculate a minimum useful reserve? | |
87 | ||
88 | sshd or login + bash (or some other shell) + top (or ps, kill, etc.) | |
89 | ||
90 | For overcommit 'guess', we can sum resident set sizes (RSS). | |
91 | On x86_64 this is about 8MB. | |
92 | ||
93 | For overcommit 'never', we can take the max of their virtual sizes (VSZ) | |
94 | and add the sum of their RSS. | |
95 | On x86_64 this is about 128MB. | |
96 | ||
97 | Changing this takes effect whenever an application requests memory. | |
98 | ||
99 | ============================================================== | |
100 | ||
db0fb184 | 101 | block_dump |
1da177e4 | 102 | |
db0fb184 PM |
103 | block_dump enables block I/O debugging when set to a nonzero value. More |
104 | information on block I/O debugging is in Documentation/laptops/laptop-mode.txt. | |
1da177e4 LT |
105 | |
106 | ============================================================== | |
107 | ||
76ab0f53 MG |
108 | compact_memory |
109 | ||
110 | Available only when CONFIG_COMPACTION is set. When 1 is written to the file, | |
111 | all zones are compacted such that free memory is available in contiguous | |
112 | blocks where possible. This can be important for example in the allocation of | |
113 | huge pages although processes will also directly compact memory as required. | |
114 | ||
115 | ============================================================== | |
116 | ||
5bbe3547 EM |
117 | compact_unevictable_allowed |
118 | ||
119 | Available only when CONFIG_COMPACTION is set. When set to 1, compaction is | |
120 | allowed to examine the unevictable lru (mlocked pages) for pages to compact. | |
121 | This should be used on systems where stalls for minor page faults are an | |
122 | acceptable trade for large contiguous free memory. Set to 0 to prevent | |
123 | compaction from moving pages that are unevictable. Default value is 1. | |
124 | ||
125 | ============================================================== | |
126 | ||
db0fb184 | 127 | dirty_background_bytes |
1da177e4 | 128 | |
6601fac8 AB |
129 | Contains the amount of dirty memory at which the background kernel |
130 | flusher threads will start writeback. | |
1da177e4 | 131 | |
abffc020 AR |
132 | Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only |
133 | one of them may be specified at a time. When one sysctl is written it is | |
134 | immediately taken into account to evaluate the dirty memory limits and the | |
135 | other appears as 0 when read. | |
1da177e4 | 136 | |
db0fb184 | 137 | ============================================================== |
1da177e4 | 138 | |
db0fb184 | 139 | dirty_background_ratio |
1da177e4 | 140 | |
715ea41e ZL |
141 | Contains, as a percentage of total available memory that contains free pages |
142 | and reclaimable pages, the number of pages at which the background kernel | |
143 | flusher threads will start writing out dirty data. | |
144 | ||
d83e2a4e | 145 | The total available memory is not equal to total system memory. |
1da177e4 | 146 | |
db0fb184 | 147 | ============================================================== |
1da177e4 | 148 | |
db0fb184 PM |
149 | dirty_bytes |
150 | ||
151 | Contains the amount of dirty memory at which a process generating disk writes | |
152 | will itself start writeback. | |
153 | ||
abffc020 AR |
154 | Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be |
155 | specified at a time. When one sysctl is written it is immediately taken into | |
156 | account to evaluate the dirty memory limits and the other appears as 0 when | |
157 | read. | |
1da177e4 | 158 | |
9e4a5bda AR |
159 | Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any |
160 | value lower than this limit will be ignored and the old configuration will be | |
161 | retained. | |
162 | ||
1da177e4 LT |
163 | ============================================================== |
164 | ||
db0fb184 | 165 | dirty_expire_centisecs |
1da177e4 | 166 | |
db0fb184 | 167 | This tunable is used to define when dirty data is old enough to be eligible |
6601fac8 AB |
168 | for writeout by the kernel flusher threads. It is expressed in 100'ths |
169 | of a second. Data which has been dirty in-memory for longer than this | |
170 | interval will be written out next time a flusher thread wakes up. | |
db0fb184 PM |
171 | |
172 | ============================================================== | |
173 | ||
174 | dirty_ratio | |
175 | ||
715ea41e ZL |
176 | Contains, as a percentage of total available memory that contains free pages |
177 | and reclaimable pages, the number of pages at which a process which is | |
178 | generating disk writes will itself start writing out dirty data. | |
179 | ||
d83e2a4e | 180 | The total available memory is not equal to total system memory. |
1da177e4 LT |
181 | |
182 | ============================================================== | |
183 | ||
fc1ca3d5 YS |
184 | dirtytime_expire_seconds |
185 | ||
186 | When a lazytime inode is constantly having its pages dirtied, the inode with | |
187 | an updated timestamp will never get chance to be written out. And, if the | |
188 | only thing that has happened on the file system is a dirtytime inode caused | |
189 | by an atime update, a worker will be scheduled to make sure that inode | |
190 | eventually gets pushed out to disk. This tunable is used to define when dirty | |
191 | inode is old enough to be eligible for writeback by the kernel flusher threads. | |
192 | And, it is also used as the interval to wakeup dirtytime_writeback thread. | |
193 | ||
194 | ============================================================== | |
195 | ||
db0fb184 | 196 | dirty_writeback_centisecs |
1da177e4 | 197 | |
6601fac8 | 198 | The kernel flusher threads will periodically wake up and write `old' data |
db0fb184 PM |
199 | out to disk. This tunable expresses the interval between those wakeups, in |
200 | 100'ths of a second. | |
1da177e4 | 201 | |
db0fb184 | 202 | Setting this to zero disables periodic writeback altogether. |
1da177e4 LT |
203 | |
204 | ============================================================== | |
205 | ||
db0fb184 | 206 | drop_caches |
1da177e4 | 207 | |
5509a5d2 DH |
208 | Writing to this will cause the kernel to drop clean caches, as well as |
209 | reclaimable slab objects like dentries and inodes. Once dropped, their | |
210 | memory becomes free. | |
1da177e4 | 211 | |
db0fb184 PM |
212 | To free pagecache: |
213 | echo 1 > /proc/sys/vm/drop_caches | |
5509a5d2 | 214 | To free reclaimable slab objects (includes dentries and inodes): |
db0fb184 | 215 | echo 2 > /proc/sys/vm/drop_caches |
5509a5d2 | 216 | To free slab objects and pagecache: |
db0fb184 | 217 | echo 3 > /proc/sys/vm/drop_caches |
1da177e4 | 218 | |
5509a5d2 DH |
219 | This is a non-destructive operation and will not free any dirty objects. |
220 | To increase the number of objects freed by this operation, the user may run | |
221 | `sync' prior to writing to /proc/sys/vm/drop_caches. This will minimize the | |
222 | number of dirty objects on the system and create more candidates to be | |
223 | dropped. | |
224 | ||
225 | This file is not a means to control the growth of the various kernel caches | |
226 | (inodes, dentries, pagecache, etc...) These objects are automatically | |
227 | reclaimed by the kernel when memory is needed elsewhere on the system. | |
228 | ||
229 | Use of this file can cause performance problems. Since it discards cached | |
230 | objects, it may cost a significant amount of I/O and CPU to recreate the | |
231 | dropped objects, especially if they were under heavy use. Because of this, | |
232 | use outside of a testing or debugging environment is not recommended. | |
233 | ||
234 | You may see informational messages in your kernel log when this file is | |
235 | used: | |
236 | ||
237 | cat (1234): drop_caches: 3 | |
238 | ||
239 | These are informational only. They do not mean that anything is wrong | |
240 | with your system. To disable them, echo 4 (bit 3) into drop_caches. | |
1da177e4 LT |
241 | |
242 | ============================================================== | |
243 | ||
5e771905 MG |
244 | extfrag_threshold |
245 | ||
246 | This parameter affects whether the kernel will compact memory or direct | |
a10726bb RV |
247 | reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in |
248 | debugfs shows what the fragmentation index for each order is in each zone in | |
249 | the system. Values tending towards 0 imply allocations would fail due to lack | |
250 | of memory, values towards 1000 imply failures are due to fragmentation and -1 | |
251 | implies that the allocation will succeed as long as watermarks are met. | |
5e771905 MG |
252 | |
253 | The kernel will not compact memory in a zone if the | |
254 | fragmentation index is <= extfrag_threshold. The default value is 500. | |
255 | ||
256 | ============================================================== | |
257 | ||
d09b6468 MH |
258 | highmem_is_dirtyable |
259 | ||
260 | Available only for systems with CONFIG_HIGHMEM enabled (32b systems). | |
261 | ||
262 | This parameter controls whether the high memory is considered for dirty | |
263 | writers throttling. This is not the case by default which means that | |
264 | only the amount of memory directly visible/usable by the kernel can | |
265 | be dirtied. As a result, on systems with a large amount of memory and | |
266 | lowmem basically depleted writers might be throttled too early and | |
267 | streaming writes can get very slow. | |
268 | ||
269 | Changing the value to non zero would allow more memory to be dirtied | |
270 | and thus allow writers to write more data which can be flushed to the | |
271 | storage more effectively. Note this also comes with a risk of pre-mature | |
272 | OOM killer because some writers (e.g. direct block device writes) can | |
273 | only use the low memory and they can fill it up with dirty data without | |
274 | any throttling. | |
275 | ||
276 | ============================================================== | |
277 | ||
db0fb184 | 278 | hugetlb_shm_group |
8ad4b1fb | 279 | |
db0fb184 PM |
280 | hugetlb_shm_group contains group id that is allowed to create SysV |
281 | shared memory segment using hugetlb page. | |
8ad4b1fb | 282 | |
db0fb184 | 283 | ============================================================== |
8ad4b1fb | 284 | |
db0fb184 | 285 | laptop_mode |
1743660b | 286 | |
db0fb184 PM |
287 | laptop_mode is a knob that controls "laptop mode". All the things that are |
288 | controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt. | |
1743660b | 289 | |
db0fb184 | 290 | ============================================================== |
1743660b | 291 | |
db0fb184 | 292 | legacy_va_layout |
1b2ffb78 | 293 | |
2174efb6 | 294 | If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel |
db0fb184 | 295 | will use the legacy (2.4) layout for all processes. |
1b2ffb78 | 296 | |
db0fb184 | 297 | ============================================================== |
1b2ffb78 | 298 | |
db0fb184 PM |
299 | lowmem_reserve_ratio |
300 | ||
301 | For some specialised workloads on highmem machines it is dangerous for | |
302 | the kernel to allow process memory to be allocated from the "lowmem" | |
303 | zone. This is because that memory could then be pinned via the mlock() | |
304 | system call, or by unavailability of swapspace. | |
305 | ||
306 | And on large highmem machines this lack of reclaimable lowmem memory | |
307 | can be fatal. | |
308 | ||
309 | So the Linux page allocator has a mechanism which prevents allocations | |
310 | which _could_ use highmem from using too much lowmem. This means that | |
311 | a certain amount of lowmem is defended from the possibility of being | |
312 | captured into pinned user memory. | |
313 | ||
314 | (The same argument applies to the old 16 megabyte ISA DMA region. This | |
315 | mechanism will also defend that region from allocations which could use | |
316 | highmem or lowmem). | |
317 | ||
318 | The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is | |
319 | in defending these lower zones. | |
320 | ||
321 | If you have a machine which uses highmem or ISA DMA and your | |
322 | applications are using mlock(), or if you are running with no swap then | |
323 | you probably should change the lowmem_reserve_ratio setting. | |
324 | ||
325 | The lowmem_reserve_ratio is an array. You can see them by reading this file. | |
326 | - | |
327 | % cat /proc/sys/vm/lowmem_reserve_ratio | |
328 | 256 256 32 | |
329 | - | |
db0fb184 PM |
330 | |
331 | But, these values are not used directly. The kernel calculates # of protection | |
332 | pages for each zones from them. These are shown as array of protection pages | |
333 | in /proc/zoneinfo like followings. (This is an example of x86-64 box). | |
334 | Each zone has an array of protection pages like this. | |
335 | ||
336 | - | |
337 | Node 0, zone DMA | |
338 | pages free 1355 | |
339 | min 3 | |
340 | low 3 | |
341 | high 4 | |
342 | : | |
343 | : | |
344 | numa_other 0 | |
345 | protection: (0, 2004, 2004, 2004) | |
346 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | |
347 | pagesets | |
348 | cpu: 0 pcp: 0 | |
349 | : | |
350 | - | |
351 | These protections are added to score to judge whether this zone should be used | |
352 | for page allocation or should be reclaimed. | |
353 | ||
354 | In this example, if normal pages (index=2) are required to this DMA zone and | |
41858966 MG |
355 | watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should |
356 | not be used because pages_free(1355) is smaller than watermark + protection[2] | |
db0fb184 PM |
357 | (4 + 2004 = 2008). If this protection value is 0, this zone would be used for |
358 | normal page requirement. If requirement is DMA zone(index=0), protection[0] | |
359 | (=0) is used. | |
360 | ||
361 | zone[i]'s protection[j] is calculated by following expression. | |
362 | ||
363 | (i < j): | |
364 | zone[i]->protection[j] | |
013110a7 | 365 | = (total sums of managed_pages from zone[i+1] to zone[j] on the node) |
db0fb184 PM |
366 | / lowmem_reserve_ratio[i]; |
367 | (i = j): | |
368 | (should not be protected. = 0; | |
369 | (i > j): | |
370 | (not necessary, but looks 0) | |
371 | ||
372 | The default values of lowmem_reserve_ratio[i] are | |
373 | 256 (if zone[i] means DMA or DMA32 zone) | |
374 | 32 (others). | |
375 | As above expression, they are reciprocal number of ratio. | |
013110a7 | 376 | 256 means 1/256. # of protection pages becomes about "0.39%" of total managed |
db0fb184 PM |
377 | pages of higher zones on the node. |
378 | ||
379 | If you would like to protect more pages, smaller values are effective. | |
d3cda233 JK |
380 | The minimum value is 1 (1/1 -> 100%). The value less than 1 completely |
381 | disables protection of the pages. | |
1b2ffb78 | 382 | |
db0fb184 | 383 | ============================================================== |
1b2ffb78 | 384 | |
db0fb184 | 385 | max_map_count: |
1743660b | 386 | |
db0fb184 PM |
387 | This file contains the maximum number of memory map areas a process |
388 | may have. Memory map areas are used as a side-effect of calling | |
def5efe0 DR |
389 | malloc, directly by mmap, mprotect, and madvise, and also when loading |
390 | shared libraries. | |
1743660b | 391 | |
db0fb184 PM |
392 | While most applications need less than a thousand maps, certain |
393 | programs, particularly malloc debuggers, may consume lots of them, | |
394 | e.g., up to one or two maps per allocation. | |
fadd8fbd | 395 | |
db0fb184 | 396 | The default value is 65536. |
9614634f | 397 | |
6a46079c AK |
398 | ============================================================= |
399 | ||
400 | memory_failure_early_kill: | |
401 | ||
402 | Control how to kill processes when uncorrected memory error (typically | |
403 | a 2bit error in a memory module) is detected in the background by hardware | |
404 | that cannot be handled by the kernel. In some cases (like the page | |
405 | still having a valid copy on disk) the kernel will handle the failure | |
406 | transparently without affecting any applications. But if there is | |
407 | no other uptodate copy of the data it will kill to prevent any data | |
408 | corruptions from propagating. | |
409 | ||
410 | 1: Kill all processes that have the corrupted and not reloadable page mapped | |
411 | as soon as the corruption is detected. Note this is not supported | |
412 | for a few types of pages, like kernel internally allocated data or | |
413 | the swap cache, but works for the majority of user pages. | |
414 | ||
415 | 0: Only unmap the corrupted page from all processes and only kill a process | |
416 | who tries to access it. | |
417 | ||
418 | The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can | |
419 | handle this if they want to. | |
420 | ||
421 | This is only active on architectures/platforms with advanced machine | |
422 | check handling and depends on the hardware capabilities. | |
423 | ||
424 | Applications can override this setting individually with the PR_MCE_KILL prctl | |
425 | ||
426 | ============================================================== | |
427 | ||
428 | memory_failure_recovery | |
429 | ||
430 | Enable memory failure recovery (when supported by the platform) | |
431 | ||
432 | 1: Attempt recovery. | |
433 | ||
434 | 0: Always panic on a memory failure. | |
435 | ||
db0fb184 | 436 | ============================================================== |
9614634f | 437 | |
db0fb184 | 438 | min_free_kbytes: |
9614634f | 439 | |
db0fb184 | 440 | This is used to force the Linux VM to keep a minimum number |
41858966 MG |
441 | of kilobytes free. The VM uses this number to compute a |
442 | watermark[WMARK_MIN] value for each lowmem zone in the system. | |
443 | Each lowmem zone gets a number of reserved free pages based | |
444 | proportionally on its size. | |
db0fb184 PM |
445 | |
446 | Some minimal amount of memory is needed to satisfy PF_MEMALLOC | |
447 | allocations; if you set this to lower than 1024KB, your system will | |
448 | become subtly broken, and prone to deadlock under high loads. | |
449 | ||
450 | Setting this too high will OOM your machine instantly. | |
9614634f CL |
451 | |
452 | ============================================================= | |
453 | ||
0ff38490 CL |
454 | min_slab_ratio: |
455 | ||
456 | This is available only on NUMA kernels. | |
457 | ||
458 | A percentage of the total pages in each zone. On Zone reclaim | |
459 | (fallback from the local zone occurs) slabs will be reclaimed if more | |
460 | than this percentage of pages in a zone are reclaimable slab pages. | |
461 | This insures that the slab growth stays under control even in NUMA | |
462 | systems that rarely perform global reclaim. | |
463 | ||
464 | The default is 5 percent. | |
465 | ||
466 | Note that slab reclaim is triggered in a per zone / node fashion. | |
467 | The process of reclaiming slab memory is currently not node specific | |
468 | and may not be fast. | |
469 | ||
470 | ============================================================= | |
471 | ||
db0fb184 | 472 | min_unmapped_ratio: |
fadd8fbd | 473 | |
db0fb184 | 474 | This is available only on NUMA kernels. |
fadd8fbd | 475 | |
90afa5de MG |
476 | This is a percentage of the total pages in each zone. Zone reclaim will |
477 | only occur if more than this percentage of pages are in a state that | |
478 | zone_reclaim_mode allows to be reclaimed. | |
479 | ||
480 | If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared | |
481 | against all file-backed unmapped pages including swapcache pages and tmpfs | |
482 | files. Otherwise, only unmapped pages backed by normal files but not tmpfs | |
483 | files and similar are considered. | |
2b744c01 | 484 | |
db0fb184 | 485 | The default is 1 percent. |
fadd8fbd | 486 | |
db0fb184 | 487 | ============================================================== |
2b744c01 | 488 | |
db0fb184 | 489 | mmap_min_addr |
ed032189 | 490 | |
db0fb184 | 491 | This file indicates the amount of address space which a user process will |
af901ca1 | 492 | be restricted from mmapping. Since kernel null dereference bugs could |
db0fb184 PM |
493 | accidentally operate based on the information in the first couple of pages |
494 | of memory userspace processes should not be allowed to write to them. By | |
495 | default this value is set to 0 and no protections will be enforced by the | |
496 | security module. Setting this value to something like 64k will allow the | |
497 | vast majority of applications to work correctly and provide defense in depth | |
498 | against future potential kernel bugs. | |
fe071d7e | 499 | |
db0fb184 | 500 | ============================================================== |
fef1bdd6 | 501 | |
d07e2259 DC |
502 | mmap_rnd_bits: |
503 | ||
504 | This value can be used to select the number of bits to use to | |
505 | determine the random offset to the base address of vma regions | |
506 | resulting from mmap allocations on architectures which support | |
507 | tuning address space randomization. This value will be bounded | |
508 | by the architecture's minimum and maximum supported values. | |
509 | ||
510 | This value can be changed after boot using the | |
511 | /proc/sys/vm/mmap_rnd_bits tunable | |
512 | ||
513 | ============================================================== | |
514 | ||
515 | mmap_rnd_compat_bits: | |
516 | ||
517 | This value can be used to select the number of bits to use to | |
518 | determine the random offset to the base address of vma regions | |
519 | resulting from mmap allocations for applications run in | |
520 | compatibility mode on architectures which support tuning address | |
521 | space randomization. This value will be bounded by the | |
522 | architecture's minimum and maximum supported values. | |
523 | ||
524 | This value can be changed after boot using the | |
525 | /proc/sys/vm/mmap_rnd_compat_bits tunable | |
526 | ||
527 | ============================================================== | |
528 | ||
db0fb184 | 529 | nr_hugepages |
fef1bdd6 | 530 | |
db0fb184 | 531 | Change the minimum size of the hugepage pool. |
fef1bdd6 | 532 | |
1ad1335d | 533 | See Documentation/admin-guide/mm/hugetlbpage.rst |
fef1bdd6 | 534 | |
db0fb184 | 535 | ============================================================== |
d1634e1a PD |
536 | |
537 | nr_hugepages_mempolicy | |
538 | ||
539 | Change the size of the hugepage pool at run-time on a specific | |
540 | set of NUMA nodes. | |
541 | ||
542 | See Documentation/admin-guide/mm/hugetlbpage.rst | |
543 | ||
544 | ============================================================== | |
fef1bdd6 | 545 | |
db0fb184 | 546 | nr_overcommit_hugepages |
fef1bdd6 | 547 | |
db0fb184 PM |
548 | Change the maximum size of the hugepage pool. The maximum is |
549 | nr_hugepages + nr_overcommit_hugepages. | |
fe071d7e | 550 | |
1ad1335d | 551 | See Documentation/admin-guide/mm/hugetlbpage.rst |
fe071d7e | 552 | |
db0fb184 | 553 | ============================================================== |
fe071d7e | 554 | |
db0fb184 | 555 | nr_trim_pages |
ed032189 | 556 | |
db0fb184 PM |
557 | This is available only on NOMMU kernels. |
558 | ||
559 | This value adjusts the excess page trimming behaviour of power-of-2 aligned | |
560 | NOMMU mmap allocations. | |
561 | ||
562 | A value of 0 disables trimming of allocations entirely, while a value of 1 | |
563 | trims excess pages aggressively. Any value >= 1 acts as the watermark where | |
564 | trimming of allocations is initiated. | |
565 | ||
566 | The default value is 1. | |
567 | ||
568 | See Documentation/nommu-mmap.txt for more information. | |
ed032189 | 569 | |
f0c0b2b8 KH |
570 | ============================================================== |
571 | ||
572 | numa_zonelist_order | |
573 | ||
c9bff3ee MH |
574 | This sysctl is only for NUMA and it is deprecated. Anything but |
575 | Node order will fail! | |
576 | ||
f0c0b2b8 KH |
577 | 'where the memory is allocated from' is controlled by zonelists. |
578 | (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation. | |
579 | you may be able to read ZONE_DMA as ZONE_DMA32...) | |
580 | ||
581 | In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following. | |
582 | ZONE_NORMAL -> ZONE_DMA | |
583 | This means that a memory allocation request for GFP_KERNEL will | |
584 | get memory from ZONE_DMA only when ZONE_NORMAL is not available. | |
585 | ||
586 | In NUMA case, you can think of following 2 types of order. | |
587 | Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL | |
588 | ||
589 | (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL | |
590 | (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA. | |
591 | ||
592 | Type(A) offers the best locality for processes on Node(0), but ZONE_DMA | |
593 | will be used before ZONE_NORMAL exhaustion. This increases possibility of | |
594 | out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small. | |
595 | ||
596 | Type(B) cannot offer the best locality but is more robust against OOM of | |
597 | the DMA zone. | |
598 | ||
599 | Type(A) is called as "Node" order. Type (B) is "Zone" order. | |
600 | ||
601 | "Node order" orders the zonelists by node, then by zone within each node. | |
5a3016a6 | 602 | Specify "[Nn]ode" for node order |
f0c0b2b8 KH |
603 | |
604 | "Zone Order" orders the zonelists by zone type, then by node within each | |
5a3016a6 | 605 | zone. Specify "[Zz]one" for zone order. |
f0c0b2b8 | 606 | |
7c88a292 XQ |
607 | Specify "[Dd]efault" to request automatic configuration. |
608 | ||
609 | On 32-bit, the Normal zone needs to be preserved for allocations accessible | |
610 | by the kernel, so "zone" order will be selected. | |
611 | ||
612 | On 64-bit, devices that require DMA32/DMA are relatively rare, so "node" | |
613 | order will be selected. | |
614 | ||
615 | Default order is recommended unless this is causing problems for your | |
616 | system/application. | |
d5dbac87 NA |
617 | |
618 | ============================================================== | |
619 | ||
db0fb184 | 620 | oom_dump_tasks |
d5dbac87 | 621 | |
dc6c9a35 KS |
622 | Enables a system-wide task dump (excluding kernel threads) to be produced |
623 | when the kernel performs an OOM-killing and includes such information as | |
af5b0f6a KS |
624 | pid, uid, tgid, vm size, rss, pgtables_bytes, swapents, oom_score_adj |
625 | score, and name. This is helpful to determine why the OOM killer was | |
626 | invoked, to identify the rogue task that caused it, and to determine why | |
627 | the OOM killer chose the task it did to kill. | |
d5dbac87 | 628 | |
db0fb184 PM |
629 | If this is set to zero, this information is suppressed. On very |
630 | large systems with thousands of tasks it may not be feasible to dump | |
631 | the memory state information for each one. Such systems should not | |
632 | be forced to incur a performance penalty in OOM conditions when the | |
633 | information may not be desired. | |
634 | ||
635 | If this is set to non-zero, this information is shown whenever the | |
636 | OOM killer actually kills a memory-hogging task. | |
637 | ||
ad915c43 | 638 | The default value is 1 (enabled). |
d5dbac87 NA |
639 | |
640 | ============================================================== | |
641 | ||
db0fb184 | 642 | oom_kill_allocating_task |
d5dbac87 | 643 | |
db0fb184 PM |
644 | This enables or disables killing the OOM-triggering task in |
645 | out-of-memory situations. | |
d5dbac87 | 646 | |
db0fb184 PM |
647 | If this is set to zero, the OOM killer will scan through the entire |
648 | tasklist and select a task based on heuristics to kill. This normally | |
649 | selects a rogue memory-hogging task that frees up a large amount of | |
650 | memory when killed. | |
651 | ||
652 | If this is set to non-zero, the OOM killer simply kills the task that | |
653 | triggered the out-of-memory condition. This avoids the expensive | |
654 | tasklist scan. | |
655 | ||
656 | If panic_on_oom is selected, it takes precedence over whatever value | |
657 | is used in oom_kill_allocating_task. | |
658 | ||
659 | The default value is 0. | |
dd8632a1 PM |
660 | |
661 | ============================================================== | |
662 | ||
49f0ce5f JM |
663 | overcommit_kbytes: |
664 | ||
665 | When overcommit_memory is set to 2, the committed address space is not | |
666 | permitted to exceed swap plus this amount of physical RAM. See below. | |
667 | ||
668 | Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one | |
669 | of them may be specified at a time. Setting one disables the other (which | |
670 | then appears as 0 when read). | |
671 | ||
672 | ============================================================== | |
673 | ||
db0fb184 | 674 | overcommit_memory: |
dd8632a1 | 675 | |
db0fb184 | 676 | This value contains a flag that enables memory overcommitment. |
dd8632a1 | 677 | |
db0fb184 PM |
678 | When this flag is 0, the kernel attempts to estimate the amount |
679 | of free memory left when userspace requests more memory. | |
dd8632a1 | 680 | |
db0fb184 PM |
681 | When this flag is 1, the kernel pretends there is always enough |
682 | memory until it actually runs out. | |
dd8632a1 | 683 | |
db0fb184 PM |
684 | When this flag is 2, the kernel uses a "never overcommit" |
685 | policy that attempts to prevent any overcommit of memory. | |
c9b1d098 | 686 | Note that user_reserve_kbytes affects this policy. |
dd8632a1 | 687 | |
db0fb184 PM |
688 | This feature can be very useful because there are a lot of |
689 | programs that malloc() huge amounts of memory "just-in-case" | |
690 | and don't use much of it. | |
691 | ||
692 | The default value is 0. | |
693 | ||
ad56b738 | 694 | See Documentation/vm/overcommit-accounting.rst and |
85f237a5 | 695 | mm/util.c::__vm_enough_memory() for more information. |
db0fb184 PM |
696 | |
697 | ============================================================== | |
698 | ||
699 | overcommit_ratio: | |
700 | ||
701 | When overcommit_memory is set to 2, the committed address | |
702 | space is not permitted to exceed swap plus this percentage | |
703 | of physical RAM. See above. | |
704 | ||
705 | ============================================================== | |
706 | ||
707 | page-cluster | |
708 | ||
df858fa8 CE |
709 | page-cluster controls the number of pages up to which consecutive pages |
710 | are read in from swap in a single attempt. This is the swap counterpart | |
711 | to page cache readahead. | |
712 | The mentioned consecutivity is not in terms of virtual/physical addresses, | |
713 | but consecutive on swap space - that means they were swapped out together. | |
db0fb184 PM |
714 | |
715 | It is a logarithmic value - setting it to zero means "1 page", setting | |
716 | it to 1 means "2 pages", setting it to 2 means "4 pages", etc. | |
df858fa8 | 717 | Zero disables swap readahead completely. |
db0fb184 PM |
718 | |
719 | The default value is three (eight pages at a time). There may be some | |
720 | small benefits in tuning this to a different value if your workload is | |
721 | swap-intensive. | |
722 | ||
df858fa8 CE |
723 | Lower values mean lower latencies for initial faults, but at the same time |
724 | extra faults and I/O delays for following faults if they would have been part of | |
725 | that consecutive pages readahead would have brought in. | |
726 | ||
db0fb184 PM |
727 | ============================================================= |
728 | ||
729 | panic_on_oom | |
730 | ||
731 | This enables or disables panic on out-of-memory feature. | |
732 | ||
733 | If this is set to 0, the kernel will kill some rogue process, | |
734 | called oom_killer. Usually, oom_killer can kill rogue processes and | |
735 | system will survive. | |
736 | ||
737 | If this is set to 1, the kernel panics when out-of-memory happens. | |
738 | However, if a process limits using nodes by mempolicy/cpusets, | |
739 | and those nodes become memory exhaustion status, one process | |
740 | may be killed by oom-killer. No panic occurs in this case. | |
741 | Because other nodes' memory may be free. This means system total status | |
742 | may be not fatal yet. | |
743 | ||
744 | If this is set to 2, the kernel panics compulsorily even on the | |
daaf1e68 KH |
745 | above-mentioned. Even oom happens under memory cgroup, the whole |
746 | system panics. | |
db0fb184 PM |
747 | |
748 | The default value is 0. | |
749 | 1 and 2 are for failover of clustering. Please select either | |
750 | according to your policy of failover. | |
daaf1e68 KH |
751 | panic_on_oom=2+kdump gives you very strong tool to investigate |
752 | why oom happens. You can get snapshot. | |
db0fb184 PM |
753 | |
754 | ============================================================= | |
755 | ||
756 | percpu_pagelist_fraction | |
757 | ||
758 | This is the fraction of pages at most (high mark pcp->high) in each zone that | |
759 | are allocated for each per cpu page list. The min value for this is 8. It | |
760 | means that we don't allow more than 1/8th of pages in each zone to be | |
761 | allocated in any single per_cpu_pagelist. This entry only changes the value | |
762 | of hot per cpu pagelists. User can specify a number like 100 to allocate | |
763 | 1/100th of each zone to each per cpu page list. | |
764 | ||
765 | The batch value of each per cpu pagelist is also updated as a result. It is | |
766 | set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8) | |
767 | ||
768 | The initial value is zero. Kernel does not use this value at boot time to set | |
7cd2b0a3 DR |
769 | the high water marks for each per cpu page list. If the user writes '0' to this |
770 | sysctl, it will revert to this default behavior. | |
db0fb184 PM |
771 | |
772 | ============================================================== | |
773 | ||
774 | stat_interval | |
775 | ||
776 | The time interval between which vm statistics are updated. The default | |
777 | is 1 second. | |
778 | ||
779 | ============================================================== | |
780 | ||
52b6f46b HD |
781 | stat_refresh |
782 | ||
783 | Any read or write (by root only) flushes all the per-cpu vm statistics | |
784 | into their global totals, for more accurate reports when testing | |
785 | e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo | |
786 | ||
787 | As a side-effect, it also checks for negative totals (elsewhere reported | |
788 | as 0) and "fails" with EINVAL if any are found, with a warning in dmesg. | |
789 | (At time of writing, a few stats are known sometimes to be found negative, | |
790 | with no ill effects: errors and warnings on these stats are suppressed.) | |
791 | ||
792 | ============================================================== | |
793 | ||
4518085e KW |
794 | numa_stat |
795 | ||
796 | This interface allows runtime configuration of numa statistics. | |
797 | ||
798 | When page allocation performance becomes a bottleneck and you can tolerate | |
799 | some possible tool breakage and decreased numa counter precision, you can | |
800 | do: | |
801 | echo 0 > /proc/sys/vm/numa_stat | |
802 | ||
803 | When page allocation performance is not a bottleneck and you want all | |
804 | tooling to work, you can do: | |
805 | echo 1 > /proc/sys/vm/numa_stat | |
806 | ||
807 | ============================================================== | |
808 | ||
db0fb184 PM |
809 | swappiness |
810 | ||
811 | This control is used to define how aggressive the kernel will swap | |
2743232c | 812 | memory pages. Higher values will increase aggressiveness, lower values |
8582cb96 AT |
813 | decrease the amount of swap. A value of 0 instructs the kernel not to |
814 | initiate swap until the amount of free and file-backed pages is less | |
815 | than the high water mark in a zone. | |
db0fb184 PM |
816 | |
817 | The default value is 60. | |
818 | ||
819 | ============================================================== | |
820 | ||
c9b1d098 AS |
821 | - user_reserve_kbytes |
822 | ||
633708a4 | 823 | When overcommit_memory is set to 2, "never overcommit" mode, reserve |
c9b1d098 AS |
824 | min(3% of current process size, user_reserve_kbytes) of free memory. |
825 | This is intended to prevent a user from starting a single memory hogging | |
826 | process, such that they cannot recover (kill the hog). | |
827 | ||
828 | user_reserve_kbytes defaults to min(3% of the current process size, 128MB). | |
829 | ||
830 | If this is reduced to zero, then the user will be allowed to allocate | |
831 | all free memory with a single process, minus admin_reserve_kbytes. | |
832 | Any subsequent attempts to execute a command will result in | |
833 | "fork: Cannot allocate memory". | |
834 | ||
835 | Changing this takes effect whenever an application requests memory. | |
836 | ||
837 | ============================================================== | |
838 | ||
db0fb184 PM |
839 | vfs_cache_pressure |
840 | ------------------ | |
841 | ||
4a0da71b DV |
842 | This percentage value controls the tendency of the kernel to reclaim |
843 | the memory which is used for caching of directory and inode objects. | |
db0fb184 PM |
844 | |
845 | At the default value of vfs_cache_pressure=100 the kernel will attempt to | |
846 | reclaim dentries and inodes at a "fair" rate with respect to pagecache and | |
847 | swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer | |
55c37a84 JK |
848 | to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will |
849 | never reclaim dentries and inodes due to memory pressure and this can easily | |
850 | lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100 | |
db0fb184 PM |
851 | causes the kernel to prefer to reclaim dentries and inodes. |
852 | ||
4a0da71b DV |
853 | Increasing vfs_cache_pressure significantly beyond 100 may have negative |
854 | performance impact. Reclaim code needs to take various locks to find freeable | |
855 | directory and inode objects. With vfs_cache_pressure=1000, it will look for | |
856 | ten times more freeable objects than there are. | |
857 | ||
795ae7a0 JW |
858 | ============================================================= |
859 | ||
1c30844d MG |
860 | watermark_boost_factor: |
861 | ||
862 | This factor controls the level of reclaim when memory is being fragmented. | |
863 | It defines the percentage of the high watermark of a zone that will be | |
864 | reclaimed if pages of different mobility are being mixed within pageblocks. | |
865 | The intent is that compaction has less work to do in the future and to | |
866 | increase the success rate of future high-order allocations such as SLUB | |
867 | allocations, THP and hugetlbfs pages. | |
868 | ||
869 | To make it sensible with respect to the watermark_scale_factor parameter, | |
870 | the unit is in fractions of 10,000. The default value of 15,000 means | |
871 | that up to 150% of the high watermark will be reclaimed in the event of | |
872 | a pageblock being mixed due to fragmentation. The level of reclaim is | |
873 | determined by the number of fragmentation events that occurred in the | |
874 | recent past. If this value is smaller than a pageblock then a pageblocks | |
875 | worth of pages will be reclaimed (e.g. 2MB on 64-bit x86). A boost factor | |
876 | of 0 will disable the feature. | |
877 | ||
878 | ============================================================= | |
879 | ||
795ae7a0 JW |
880 | watermark_scale_factor: |
881 | ||
882 | This factor controls the aggressiveness of kswapd. It defines the | |
883 | amount of memory left in a node/system before kswapd is woken up and | |
884 | how much memory needs to be free before kswapd goes back to sleep. | |
885 | ||
886 | The unit is in fractions of 10,000. The default value of 10 means the | |
887 | distances between watermarks are 0.1% of the available memory in the | |
888 | node/system. The maximum value is 1000, or 10% of memory. | |
889 | ||
890 | A high rate of threads entering direct reclaim (allocstall) or kswapd | |
891 | going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate | |
892 | that the number of free pages kswapd maintains for latency reasons is | |
893 | too small for the allocation bursts occurring in the system. This knob | |
894 | can then be used to tune kswapd aggressiveness accordingly. | |
895 | ||
db0fb184 PM |
896 | ============================================================== |
897 | ||
898 | zone_reclaim_mode: | |
899 | ||
900 | Zone_reclaim_mode allows someone to set more or less aggressive approaches to | |
901 | reclaim memory when a zone runs out of memory. If it is set to zero then no | |
902 | zone reclaim occurs. Allocations will be satisfied from other zones / nodes | |
903 | in the system. | |
904 | ||
905 | This is value ORed together of | |
906 | ||
907 | 1 = Zone reclaim on | |
908 | 2 = Zone reclaim writes dirty pages out | |
909 | 4 = Zone reclaim swaps pages | |
910 | ||
4f9b16a6 MG |
911 | zone_reclaim_mode is disabled by default. For file servers or workloads |
912 | that benefit from having their data cached, zone_reclaim_mode should be | |
913 | left disabled as the caching effect is likely to be more important than | |
db0fb184 PM |
914 | data locality. |
915 | ||
4f9b16a6 MG |
916 | zone_reclaim may be enabled if it's known that the workload is partitioned |
917 | such that each partition fits within a NUMA node and that accessing remote | |
918 | memory would cause a measurable performance reduction. The page allocator | |
919 | will then reclaim easily reusable pages (those page cache pages that are | |
920 | currently not used) before allocating off node pages. | |
921 | ||
db0fb184 PM |
922 | Allowing zone reclaim to write out pages stops processes that are |
923 | writing large amounts of data from dirtying pages on other nodes. Zone | |
924 | reclaim will write out dirty pages if a zone fills up and so effectively | |
925 | throttle the process. This may decrease the performance of a single process | |
926 | since it cannot use all of system memory to buffer the outgoing writes | |
927 | anymore but it preserve the memory on other nodes so that the performance | |
928 | of other processes running on other nodes will not be affected. | |
929 | ||
930 | Allowing regular swap effectively restricts allocations to the local | |
931 | node unless explicitly overridden by memory policies or cpuset | |
932 | configurations. | |
933 | ||
934 | ============ End of Document ================================= |