1 Memory Resource Controller
3 NOTE: The Memory Resource Controller has generically been referred to as the
4 memory controller in this document. Do not confuse memory controller
5 used here with the memory controller that is used in hardware.
9 When we mention a cgroup (cgroupfs's directory) with memory controller,
10 we call it "memory cgroup". When you see git-log and source code, you'll
11 see patch's title and function names tend to use "memcg".
12 In this document, we avoid using it.
14 Benefits and Purpose of the memory controller
16 The memory controller isolates the memory behaviour of a group of tasks
17 from the rest of the system. The article on LWN [12] mentions some probable
18 uses of the memory controller. The memory controller can be used to
20 a. Isolate an application or a group of applications
21 Memory hungry applications can be isolated and limited to a smaller
23 b. Create a cgroup with limited amount of memory, this can be used
24 as a good alternative to booting with mem=XXXX.
25 c. Virtualization solutions can control the amount of memory they want
26 to assign to a virtual machine instance.
27 d. A CD/DVD burner could control the amount of memory used by the
28 rest of the system to ensure that burning does not fail due to lack
30 e. There are several other use cases, find one or use the controller just
31 for fun (to learn and hack on the VM subsystem).
33 Current Status: linux-2.6.34-mmotm(development version of 2010/April)
36 - accounting anonymous pages, file caches, swap caches usage and limiting them.
37 - pages are linked to per-memcg LRU exclusively, and there is no global LRU.
38 - optionally, memory+swap usage can be accounted and limited.
39 - hierarchical accounting
41 - moving(recharging) account at moving a task is selectable.
42 - usage threshold notifier
43 - oom-killer disable knob and oom-notifier
44 - Root cgroup has no limit controls.
46 Kernel memory support is work in progress, and the current version provides
47 basically functionality. (See Section 2.7)
49 Brief summary of control files.
51 tasks # attach a task(thread) and show list of threads
52 cgroup.procs # show list of processes
53 cgroup.event_control # an interface for event_fd()
54 memory.usage_in_bytes # show current res_counter usage for memory
56 memory.memsw.usage_in_bytes # show current res_counter usage for memory+Swap
58 memory.limit_in_bytes # set/show limit of memory usage
59 memory.memsw.limit_in_bytes # set/show limit of memory+Swap usage
60 memory.failcnt # show the number of memory usage hits limits
61 memory.memsw.failcnt # show the number of memory+Swap hits limits
62 memory.max_usage_in_bytes # show max memory usage recorded
63 memory.memsw.max_usage_in_bytes # show max memory+Swap usage recorded
64 memory.soft_limit_in_bytes # set/show soft limit of memory usage
65 memory.stat # show various statistics
66 memory.use_hierarchy # set/show hierarchical account enabled
67 memory.force_empty # trigger forced move charge to parent
68 memory.swappiness # set/show swappiness parameter of vmscan
69 (See sysctl's vm.swappiness)
70 memory.move_charge_at_immigrate # set/show controls of moving charges
71 memory.oom_control # set/show oom controls.
72 memory.numa_stat # show the number of memory usage per numa node
74 memory.kmem.tcp.limit_in_bytes # set/show hard limit for tcp buf memory
75 memory.kmem.tcp.usage_in_bytes # show current tcp buf memory allocation
79 The memory controller has a long history. A request for comments for the memory
80 controller was posted by Balbir Singh [1]. At the time the RFC was posted
81 there were several implementations for memory control. The goal of the
82 RFC was to build consensus and agreement for the minimal features required
83 for memory control. The first RSS controller was posted by Balbir Singh[2]
84 in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the
85 RSS controller. At OLS, at the resource management BoF, everyone suggested
86 that we handle both page cache and RSS together. Another request was raised
87 to allow user space handling of OOM. The current memory controller is
88 at version 6; it combines both mapped (RSS) and unmapped Page
93 Memory is a unique resource in the sense that it is present in a limited
94 amount. If a task requires a lot of CPU processing, the task can spread
95 its processing over a period of hours, days, months or years, but with
96 memory, the same physical memory needs to be reused to accomplish the task.
98 The memory controller implementation has been divided into phases. These
102 2. mlock(2) controller
103 3. Kernel user memory accounting and slab control
104 4. user mappings length controller
106 The memory controller is the first controller developed.
110 The core of the design is a counter called the res_counter. The res_counter
111 tracks the current memory usage and limit of the group of processes associated
112 with the controller. Each cgroup has a memory controller specific data
113 structure (mem_cgroup) associated with it.
117 +--------------------+
120 +--------------------+
123 +---------------+ | +---------------+
124 | mm_struct | |.... | mm_struct |
126 +---------------+ | +---------------+
130 +---------------+ +------+--------+
131 | page +----------> page_cgroup|
133 +---------------+ +---------------+
135 (Figure 1: Hierarchy of Accounting)
138 Figure 1 shows the important aspects of the controller
140 1. Accounting happens per cgroup
141 2. Each mm_struct knows about which cgroup it belongs to
142 3. Each page has a pointer to the page_cgroup, which in turn knows the
145 The accounting is done as follows: mem_cgroup_charge() is invoked to setup
146 the necessary data structures and check if the cgroup that is being charged
147 is over its limit. If it is then reclaim is invoked on the cgroup.
148 More details can be found in the reclaim section of this document.
149 If everything goes well, a page meta-data-structure called page_cgroup is
150 updated. page_cgroup has its own LRU on cgroup.
151 (*) page_cgroup structure is allocated at boot/memory-hotplug time.
153 2.2.1 Accounting details
155 All mapped anon pages (RSS) and cache pages (Page Cache) are accounted.
156 Some pages which are never reclaimable and will not be on the LRU
157 are not accounted. We just account pages under usual VM management.
159 RSS pages are accounted at page_fault unless they've already been accounted
160 for earlier. A file page will be accounted for as Page Cache when it's
161 inserted into inode (radix-tree). While it's mapped into the page tables of
162 processes, duplicate accounting is carefully avoided.
164 A RSS page is unaccounted when it's fully unmapped. A PageCache page is
165 unaccounted when it's removed from radix-tree. Even if RSS pages are fully
166 unmapped (by kswapd), they may exist as SwapCache in the system until they
167 are really freed. Such SwapCaches also also accounted.
168 A swapped-in page is not accounted until it's mapped.
170 Note: The kernel does swapin-readahead and read multiple swaps at once.
171 This means swapped-in pages may contain pages for other tasks than a task
172 causing page fault. So, we avoid accounting at swap-in I/O.
174 At page migration, accounting information is kept.
176 Note: we just account pages-on-LRU because our purpose is to control amount
177 of used pages; not-on-LRU pages tend to be out-of-control from VM view.
179 2.3 Shared Page Accounting
181 Shared pages are accounted on the basis of the first touch approach. The
182 cgroup that first touches a page is accounted for the page. The principle
183 behind this approach is that a cgroup that aggressively uses a shared
184 page will eventually get charged for it (once it is uncharged from
185 the cgroup that brought it in -- this will happen on memory pressure).
187 Exception: If CONFIG_CGROUP_CGROUP_MEM_RES_CTLR_SWAP is not used.
188 When you do swapoff and make swapped-out pages of shmem(tmpfs) to
189 be backed into memory in force, charges for pages are accounted against the
190 caller of swapoff rather than the users of shmem.
193 2.4 Swap Extension (CONFIG_CGROUP_MEM_RES_CTLR_SWAP)
195 Swap Extension allows you to record charge for swap. A swapped-in page is
196 charged back to original page allocator if possible.
198 When swap is accounted, following files are added.
199 - memory.memsw.usage_in_bytes.
200 - memory.memsw.limit_in_bytes.
202 memsw means memory+swap. Usage of memory+swap is limited by
203 memsw.limit_in_bytes.
205 Example: Assume a system with 4G of swap. A task which allocates 6G of memory
206 (by mistake) under 2G memory limitation will use all swap.
207 In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap.
208 By using memsw limit, you can avoid system OOM which can be caused by swap
211 * why 'memory+swap' rather than swap.
212 The global LRU(kswapd) can swap out arbitrary pages. Swap-out means
213 to move account from memory to swap...there is no change in usage of
214 memory+swap. In other words, when we want to limit the usage of swap without
215 affecting global LRU, memory+swap limit is better than just limiting swap from
218 * What happens when a cgroup hits memory.memsw.limit_in_bytes
219 When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out
220 in this cgroup. Then, swap-out will not be done by cgroup routine and file
221 caches are dropped. But as mentioned above, global LRU can do swapout memory
222 from it for sanity of the system's memory management state. You can't forbid
227 Each cgroup maintains a per cgroup LRU which has the same structure as
228 global VM. When a cgroup goes over its limit, we first try
229 to reclaim memory from the cgroup so as to make space for the new
230 pages that the cgroup has touched. If the reclaim is unsuccessful,
231 an OOM routine is invoked to select and kill the bulkiest task in the
232 cgroup. (See 10. OOM Control below.)
234 The reclaim algorithm has not been modified for cgroups, except that
235 pages that are selected for reclaiming come from the per cgroup LRU
238 NOTE: Reclaim does not work for the root cgroup, since we cannot set any
239 limits on the root cgroup.
241 Note2: When panic_on_oom is set to "2", the whole system will panic.
243 When oom event notifier is registered, event will be delivered.
244 (See oom_control section)
248 lock_page_cgroup()/unlock_page_cgroup() should not be called under
251 Other lock order is following:
256 In many cases, just lock_page_cgroup() is called.
257 per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by
258 zone->lru_lock, it has no lock of its own.
260 2.7 Kernel Memory Extension (CONFIG_CGROUP_MEM_RES_CTLR_KMEM)
262 With the Kernel memory extension, the Memory Controller is able to limit
263 the amount of kernel memory used by the system. Kernel memory is fundamentally
264 different than user memory, since it can't be swapped out, which makes it
265 possible to DoS the system by consuming too much of this precious resource.
267 Kernel memory limits are not imposed for the root cgroup. Usage for the root
268 cgroup may or may not be accounted.
270 Currently no soft limit is implemented for kernel memory. It is future work
271 to trigger slab reclaim when those limits are reached.
273 2.7.1 Current Kernel Memory resources accounted
275 * sockets memory pressure: some sockets protocols have memory pressure
276 thresholds. The Memory Controller allows them to be controlled individually
277 per cgroup, instead of globally.
279 * tcp memory pressure: sockets memory pressure for the tcp protocol.
285 a. Enable CONFIG_CGROUPS
286 b. Enable CONFIG_RESOURCE_COUNTERS
287 c. Enable CONFIG_CGROUP_MEM_RES_CTLR
288 d. Enable CONFIG_CGROUP_MEM_RES_CTLR_SWAP (to use swap extension)
290 1. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?)
291 # mount -t tmpfs none /sys/fs/cgroup
292 # mkdir /sys/fs/cgroup/memory
293 # mount -t cgroup none /sys/fs/cgroup/memory -o memory
295 2. Make the new group and move bash into it
296 # mkdir /sys/fs/cgroup/memory/0
297 # echo $$ > /sys/fs/cgroup/memory/0/tasks
299 Since now we're in the 0 cgroup, we can alter the memory limit:
300 # echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes
302 NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo,
303 mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.)
305 NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited).
306 NOTE: We cannot set limits on the root cgroup any more.
308 # cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes
311 We can check the usage:
312 # cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes
315 A successful write to this file does not guarantee a successful set of
316 this limit to the value written into the file. This can be due to a
317 number of factors, such as rounding up to page boundaries or the total
318 availability of memory on the system. The user is required to re-read
319 this file after a write to guarantee the value committed by the kernel.
321 # echo 1 > memory.limit_in_bytes
322 # cat memory.limit_in_bytes
325 The memory.failcnt field gives the number of times that the cgroup limit was
328 The memory.stat file gives accounting information. Now, the number of
329 caches, RSS and Active pages/Inactive pages are shown.
333 For testing features and implementation, see memcg_test.txt.
335 Performance test is also important. To see pure memory controller's overhead,
336 testing on tmpfs will give you good numbers of small overheads.
337 Example: do kernel make on tmpfs.
339 Page-fault scalability is also important. At measuring parallel
340 page fault test, multi-process test may be better than multi-thread
341 test because it has noise of shared objects/status.
343 But the above two are testing extreme situations.
344 Trying usual test under memory controller is always helpful.
348 Sometimes a user might find that the application under a cgroup is
349 terminated by OOM killer. There are several causes for this:
351 1. The cgroup limit is too low (just too low to do anything useful)
352 2. The user is using anonymous memory and swap is turned off or too low
354 A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of
355 some of the pages cached in the cgroup (page cache pages).
357 To know what happens, disable OOM_Kill by 10. OOM Control(see below) and
358 seeing what happens will be helpful.
362 When a task migrates from one cgroup to another, its charge is not
363 carried forward by default. The pages allocated from the original cgroup still
364 remain charged to it, the charge is dropped when the page is freed or
367 You can move charges of a task along with task migration.
368 See 8. "Move charges at task migration"
370 4.3 Removing a cgroup
372 A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a
373 cgroup might have some charge associated with it, even though all
374 tasks have migrated away from it. (because we charge against pages, not
377 Such charges are freed or moved to their parent. At moving, both of RSS
378 and CACHES are moved to parent.
379 rmdir() may return -EBUSY if freeing/moving fails. See 5.1 also.
381 Charges recorded in swap information is not updated at removal of cgroup.
382 Recorded information is discarded and a cgroup which uses swap (swapcache)
383 will be charged as a new owner of it.
389 memory.force_empty interface is provided to make cgroup's memory usage empty.
390 You can use this interface only when the cgroup has no tasks.
391 When writing anything to this
393 # echo 0 > memory.force_empty
395 Almost all pages tracked by this memory cgroup will be unmapped and freed.
396 Some pages cannot be freed because they are locked or in-use. Such pages are
397 moved to parent and this cgroup will be empty. This may return -EBUSY if
398 VM is too busy to free/move all pages immediately.
400 Typical use case of this interface is that calling this before rmdir().
401 Because rmdir() moves all pages to parent, some out-of-use page caches can be
402 moved to the parent. If you want to avoid that, force_empty will be useful.
406 memory.stat file includes following statistics
408 # per-memory cgroup local status
409 cache - # of bytes of page cache memory.
410 rss - # of bytes of anonymous and swap cache memory.
411 mapped_file - # of bytes of mapped file (includes tmpfs/shmem)
412 pgpgin - # of charging events to the memory cgroup. The charging
413 event happens each time a page is accounted as either mapped
414 anon page(RSS) or cache page(Page Cache) to the cgroup.
415 pgpgout - # of uncharging events to the memory cgroup. The uncharging
416 event happens each time a page is unaccounted from the cgroup.
417 swap - # of bytes of swap usage
418 inactive_anon - # of bytes of anonymous memory and swap cache memory on
420 active_anon - # of bytes of anonymous and swap cache memory on active
422 inactive_file - # of bytes of file-backed memory on inactive LRU list.
423 active_file - # of bytes of file-backed memory on active LRU list.
424 unevictable - # of bytes of memory that cannot be reclaimed (mlocked etc).
426 # status considering hierarchy (see memory.use_hierarchy settings)
428 hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy
429 under which the memory cgroup is
430 hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to
431 hierarchy under which memory cgroup is.
433 total_cache - sum of all children's "cache"
434 total_rss - sum of all children's "rss"
435 total_mapped_file - sum of all children's "cache"
436 total_pgpgin - sum of all children's "pgpgin"
437 total_pgpgout - sum of all children's "pgpgout"
438 total_swap - sum of all children's "swap"
439 total_inactive_anon - sum of all children's "inactive_anon"
440 total_active_anon - sum of all children's "active_anon"
441 total_inactive_file - sum of all children's "inactive_file"
442 total_active_file - sum of all children's "active_file"
443 total_unevictable - sum of all children's "unevictable"
445 # The following additional stats are dependent on CONFIG_DEBUG_VM.
447 recent_rotated_anon - VM internal parameter. (see mm/vmscan.c)
448 recent_rotated_file - VM internal parameter. (see mm/vmscan.c)
449 recent_scanned_anon - VM internal parameter. (see mm/vmscan.c)
450 recent_scanned_file - VM internal parameter. (see mm/vmscan.c)
453 recent_rotated means recent frequency of LRU rotation.
454 recent_scanned means recent # of scans to LRU.
455 showing for better debug please see the code for meanings.
458 Only anonymous and swap cache memory is listed as part of 'rss' stat.
459 This should not be confused with the true 'resident set size' or the
460 amount of physical memory used by the cgroup.
461 'rss + file_mapped" will give you resident set size of cgroup.
462 (Note: file and shmem may be shared among other cgroups. In that case,
463 file_mapped is accounted only when the memory cgroup is owner of page
468 Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only.
470 Following cgroups' swappiness can't be changed.
471 - root cgroup (uses /proc/sys/vm/swappiness).
472 - a cgroup which uses hierarchy and it has other cgroup(s) below it.
473 - a cgroup which uses hierarchy and not the root of hierarchy.
477 A memory cgroup provides memory.failcnt and memory.memsw.failcnt files.
478 This failcnt(== failure count) shows the number of times that a usage counter
479 hit its limit. When a memory cgroup hits a limit, failcnt increases and
480 memory under it will be reclaimed.
482 You can reset failcnt by writing 0 to failcnt file.
483 # echo 0 > .../memory.failcnt
487 For efficiency, as other kernel components, memory cgroup uses some optimization
488 to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the
489 method and doesn't show 'exact' value of memory(and swap) usage, it's an fuzz
490 value for efficient access. (Of course, when necessary, it's synchronized.)
491 If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP)
492 value in memory.stat(see 5.2).
496 This is similar to numa_maps but operates on a per-memcg basis. This is
497 useful for providing visibility into the numa locality information within
498 an memcg since the pages are allowed to be allocated from any physical
499 node. One of the usecases is evaluating application performance by
500 combining this information with the application's cpu allocation.
502 We export "total", "file", "anon" and "unevictable" pages per-node for
503 each memcg. The ouput format of memory.numa_stat is:
505 total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ...
506 file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ...
507 anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
508 unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
510 And we have total = file + anon + unevictable.
514 The memory controller supports a deep hierarchy and hierarchical accounting.
515 The hierarchy is created by creating the appropriate cgroups in the
516 cgroup filesystem. Consider for example, the following cgroup filesystem
527 In the diagram above, with hierarchical accounting enabled, all memory
528 usage of e, is accounted to its ancestors up until the root (i.e, c and root),
529 that has memory.use_hierarchy enabled. If one of the ancestors goes over its
530 limit, the reclaim algorithm reclaims from the tasks in the ancestor and the
531 children of the ancestor.
533 6.1 Enabling hierarchical accounting and reclaim
535 A memory cgroup by default disables the hierarchy feature. Support
536 can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup
538 # echo 1 > memory.use_hierarchy
540 The feature can be disabled by
542 # echo 0 > memory.use_hierarchy
544 NOTE1: Enabling/disabling will fail if either the cgroup already has other
545 cgroups created below it, or if the parent cgroup has use_hierarchy
548 NOTE2: When panic_on_oom is set to "2", the whole system will panic in
549 case of an OOM event in any cgroup.
553 Soft limits allow for greater sharing of memory. The idea behind soft limits
554 is to allow control groups to use as much of the memory as needed, provided
556 a. There is no memory contention
557 b. They do not exceed their hard limit
559 When the system detects memory contention or low memory, control groups
560 are pushed back to their soft limits. If the soft limit of each control
561 group is very high, they are pushed back as much as possible to make
562 sure that one control group does not starve the others of memory.
564 Please note that soft limits is a best effort feature, it comes with
565 no guarantees, but it does its best to make sure that when memory is
566 heavily contended for, memory is allocated based on the soft limit
567 hints/setup. Currently soft limit based reclaim is setup such that
568 it gets invoked from balance_pgdat (kswapd).
572 Soft limits can be setup by using the following commands (in this example we
573 assume a soft limit of 256 MiB)
575 # echo 256M > memory.soft_limit_in_bytes
577 If we want to change this to 1G, we can at any time use
579 # echo 1G > memory.soft_limit_in_bytes
581 NOTE1: Soft limits take effect over a long period of time, since they involve
582 reclaiming memory for balancing between memory cgroups
583 NOTE2: It is recommended to set the soft limit always below the hard limit,
584 otherwise the hard limit will take precedence.
586 8. Move charges at task migration
588 Users can move charges associated with a task along with task migration, that
589 is, uncharge task's pages from the old cgroup and charge them to the new cgroup.
590 This feature is not supported in !CONFIG_MMU environments because of lack of
595 This feature is disabled by default. It can be enabled(and disabled again) by
596 writing to memory.move_charge_at_immigrate of the destination cgroup.
598 If you want to enable it:
600 # echo (some positive value) > memory.move_charge_at_immigrate
602 Note: Each bits of move_charge_at_immigrate has its own meaning about what type
603 of charges should be moved. See 8.2 for details.
604 Note: Charges are moved only when you move mm->owner, IOW, a leader of a thread
606 Note: If we cannot find enough space for the task in the destination cgroup, we
607 try to make space by reclaiming memory. Task migration may fail if we
608 cannot make enough space.
609 Note: It can take several seconds if you move charges much.
611 And if you want disable it again:
613 # echo 0 > memory.move_charge_at_immigrate
615 8.2 Type of charges which can be move
617 Each bits of move_charge_at_immigrate has its own meaning about what type of
618 charges should be moved. But in any cases, it must be noted that an account of
619 a page or a swap can be moved only when it is charged to the task's current(old)
622 bit | what type of charges would be moved ?
623 -----+------------------------------------------------------------------------
624 0 | A charge of an anonymous page(or swap of it) used by the target task.
625 | Those pages and swaps must be used only by the target task. You must
626 | enable Swap Extension(see 2.4) to enable move of swap charges.
627 -----+------------------------------------------------------------------------
628 1 | A charge of file pages(normal file, tmpfs file(e.g. ipc shared memory)
629 | and swaps of tmpfs file) mmapped by the target task. Unlike the case of
630 | anonymous pages, file pages(and swaps) in the range mmapped by the task
631 | will be moved even if the task hasn't done page fault, i.e. they might
632 | not be the task's "RSS", but other task's "RSS" that maps the same file.
633 | And mapcount of the page is ignored(the page can be moved even if
634 | page_mapcount(page) > 1). You must enable Swap Extension(see 2.4) to
635 | enable move of swap charges.
639 - Implement madvise(2) to let users decide the vma to be moved or not to be
641 - All of moving charge operations are done under cgroup_mutex. It's not good
642 behavior to hold the mutex too long, so we may need some trick.
646 Memory cgroup implements memory thresholds using cgroups notification
647 API (see cgroups.txt). It allows to register multiple memory and memsw
648 thresholds and gets notifications when it crosses.
650 To register a threshold application need:
651 - create an eventfd using eventfd(2);
652 - open memory.usage_in_bytes or memory.memsw.usage_in_bytes;
653 - write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to
654 cgroup.event_control.
656 Application will be notified through eventfd when memory usage crosses
657 threshold in any direction.
659 It's applicable for root and non-root cgroup.
663 memory.oom_control file is for OOM notification and other controls.
665 Memory cgroup implements OOM notifier using cgroup notification
666 API (See cgroups.txt). It allows to register multiple OOM notification
667 delivery and gets notification when OOM happens.
669 To register a notifier, application need:
670 - create an eventfd using eventfd(2)
671 - open memory.oom_control file
672 - write string like "<event_fd> <fd of memory.oom_control>" to
675 Application will be notified through eventfd when OOM happens.
676 OOM notification doesn't work for root cgroup.
678 You can disable OOM-killer by writing "1" to memory.oom_control file, as:
680 #echo 1 > memory.oom_control
682 This operation is only allowed to the top cgroup of sub-hierarchy.
683 If OOM-killer is disabled, tasks under cgroup will hang/sleep
684 in memory cgroup's OOM-waitqueue when they request accountable memory.
686 For running them, you have to relax the memory cgroup's OOM status by
687 * enlarge limit or reduce usage.
690 * move some tasks to other group with account migration.
691 * remove some files (on tmpfs?)
693 Then, stopped tasks will work again.
695 At reading, current status of OOM is shown.
696 oom_kill_disable 0 or 1 (if 1, oom-killer is disabled)
697 under_oom 0 or 1 (if 1, the memory cgroup is under OOM, tasks may
702 1. Add support for accounting huge pages (as a separate controller)
703 2. Make per-cgroup scanner reclaim not-shared pages first
704 3. Teach controller to account for shared-pages
705 4. Start reclamation in the background when the limit is
706 not yet hit but the usage is getting closer
710 Overall, the memory controller has been a stable controller and has been
711 commented and discussed quite extensively in the community.
715 1. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/
716 2. Singh, Balbir. Memory Controller (RSS Control),
717 http://lwn.net/Articles/222762/
718 3. Emelianov, Pavel. Resource controllers based on process cgroups
719 http://lkml.org/lkml/2007/3/6/198
720 4. Emelianov, Pavel. RSS controller based on process cgroups (v2)
721 http://lkml.org/lkml/2007/4/9/78
722 5. Emelianov, Pavel. RSS controller based on process cgroups (v3)
723 http://lkml.org/lkml/2007/5/30/244
724 6. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/
725 7. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control
726 subsystem (v3), http://lwn.net/Articles/235534/
727 8. Singh, Balbir. RSS controller v2 test results (lmbench),
728 http://lkml.org/lkml/2007/5/17/232
729 9. Singh, Balbir. RSS controller v2 AIM9 results
730 http://lkml.org/lkml/2007/5/18/1
731 10. Singh, Balbir. Memory controller v6 test results,
732 http://lkml.org/lkml/2007/8/19/36
733 11. Singh, Balbir. Memory controller introduction (v6),
734 http://lkml.org/lkml/2007/8/17/69
735 12. Corbet, Jonathan, Controlling memory use in cgroups,
736 http://lwn.net/Articles/243795/