1Memory Resource Controller
   3NOTE: 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.
   7(For editors)
   8In this document:
   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.
  14Benefits and Purpose of the memory controller
  16The memory controller isolates the memory behaviour of a group of tasks
  17from the rest of the system. The article on LWN [12] mentions some probable
  18uses of the memory controller. The memory controller can be used to
  20a. Isolate an application or a group of applications
  21   Memory hungry applications can be isolated and limited to a smaller
  22   amount of memory.
  23b. Create a cgroup with limited amount of memory, this can be used
  24   as a good alternative to booting with mem=XXXX.
  25c. Virtualization solutions can control the amount of memory they want
  26   to assign to a virtual machine instance.
  27d. 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
  29   of available memory.
  30e. There are several other use cases, find one or use the controller just
  31   for fun (to learn and hack on the VM subsystem).
  33Current 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
  40 - soft limit
  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)
  49Brief 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
  55                                 (See 5.5 for details)
  56 memory.memsw.usage_in_bytes     # show current res_counter usage for memory+Swap
  57                                 (See 5.5 for details)
  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
  771. History
  79The memory controller has a long history. A request for comments for the memory
  80controller was posted by Balbir Singh [1]. At the time the RFC was posted
  81there were several implementations for memory control. The goal of the
  82RFC was to build consensus and agreement for the minimal features required
  83for memory control. The first RSS controller was posted by Balbir Singh[2]
  84in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the
  85RSS controller. At OLS, at the resource management BoF, everyone suggested
  86that we handle both page cache and RSS together. Another request was raised
  87to allow user space handling of OOM. The current memory controller is
  88at version 6; it combines both mapped (RSS) and unmapped Page
  89Cache Control [11].
  912. Memory Control
  93Memory is a unique resource in the sense that it is present in a limited
  94amount. If a task requires a lot of CPU processing, the task can spread
  95its processing over a period of hours, days, months or years, but with
  96memory, the same physical memory needs to be reused to accomplish the task.
  98The memory controller implementation has been divided into phases. These
 1011. Memory controller
 1022. mlock(2) controller
 1033. Kernel user memory accounting and slab control
 1044. user mappings length controller
 106The memory controller is the first controller developed.
 1082.1. Design
 110The core of the design is a counter called the res_counter. The res_counter
 111tracks the current memory usage and limit of the group of processes associated
 112with the controller. Each cgroup has a memory controller specific data
 113structure (mem_cgroup) associated with it.
 1152.2. Accounting
 117                +--------------------+
 118                |  mem_cgroup     |
 119                |  (res_counter)     |
 120                +--------------------+
 121                 /            ^      \
 122                /             |       \
 123           +---------------+  |        +---------------+
 124           | mm_struct     |  |....    | mm_struct     |
 125           |               |  |        |               |
 126           +---------------+  |        +---------------+
 127                              |
 128                              + --------------+
 129                                              |
 130           +---------------+           +------+--------+
 131           | page          +---------->  page_cgroup|
 132           |               |           |               |
 133           +---------------+           +---------------+
 135             (Figure 1: Hierarchy of Accounting)
 138Figure 1 shows the important aspects of the controller
 1401. Accounting happens per cgroup
 1412. Each mm_struct knows about which cgroup it belongs to
 1423. Each page has a pointer to the page_cgroup, which in turn knows the
 143   cgroup it belongs to
 145The accounting is done as follows: mem_cgroup_charge() is invoked to setup
 146the necessary data structures and check if the cgroup that is being charged
 147is over its limit. If it is then reclaim is invoked on the cgroup.
 148More details can be found in the reclaim section of this document.
 149If everything goes well, a page meta-data-structure called page_cgroup is
 150updated. page_cgroup has its own LRU on cgroup.
 151(*) page_cgroup structure is allocated at boot/memory-hotplug time.
 1532.2.1 Accounting details
 155All mapped anon pages (RSS) and cache pages (Page Cache) are accounted.
 156Some pages which are never reclaimable and will not be on the LRU
 157are not accounted. We just account pages under usual VM management.
 159RSS pages are accounted at page_fault unless they've already been accounted
 160for earlier. A file page will be accounted for as Page Cache when it's
 161inserted into inode (radix-tree). While it's mapped into the page tables of
 162processes, duplicate accounting is carefully avoided.
 164A RSS page is unaccounted when it's fully unmapped. A PageCache page is
 165unaccounted when it's removed from radix-tree. Even if RSS pages are fully
 166unmapped (by kswapd), they may exist as SwapCache in the system until they
 167are really freed. Such SwapCaches also also accounted.
 168A swapped-in page is not accounted until it's mapped.
 170Note: The kernel does swapin-readahead and read multiple swaps at once.
 171This means swapped-in pages may contain pages for other tasks than a task
 172causing page fault. So, we avoid accounting at swap-in I/O.
 174At page migration, accounting information is kept.
 176Note: we just account pages-on-LRU because our purpose is to control amount
 177of used pages; not-on-LRU pages tend to be out-of-control from VM view.
 1792.3 Shared Page Accounting
 181Shared pages are accounted on the basis of the first touch approach. The
 182cgroup that first touches a page is accounted for the page. The principle
 183behind this approach is that a cgroup that aggressively uses a shared
 184page will eventually get charged for it (once it is uncharged from
 185the cgroup that brought it in -- this will happen on memory pressure).
 187But see section 8.2: when moving a task to another cgroup, its pages may
 188be recharged to the new cgroup, if move_charge_at_immigrate has been chosen.
 190Exception: If CONFIG_CGROUP_CGROUP_MEM_RES_CTLR_SWAP is not used.
 191When you do swapoff and make swapped-out pages of shmem(tmpfs) to
 192be backed into memory in force, charges for pages are accounted against the
 193caller of swapoff rather than the users of shmem.
 1952.4 Swap Extension (CONFIG_CGROUP_MEM_RES_CTLR_SWAP)
 197Swap Extension allows you to record charge for swap. A swapped-in page is
 198charged back to original page allocator if possible.
 200When swap is accounted, following files are added.
 201 - memory.memsw.usage_in_bytes.
 202 - memory.memsw.limit_in_bytes.
 204memsw means memory+swap. Usage of memory+swap is limited by
 207Example: Assume a system with 4G of swap. A task which allocates 6G of memory
 208(by mistake) under 2G memory limitation will use all swap.
 209In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap.
 210By using memsw limit, you can avoid system OOM which can be caused by swap
 213* why 'memory+swap' rather than swap.
 214The global LRU(kswapd) can swap out arbitrary pages. Swap-out means
 215to move account from memory to swap...there is no change in usage of
 216memory+swap. In other words, when we want to limit the usage of swap without
 217affecting global LRU, memory+swap limit is better than just limiting swap from
 218OS point of view.
 220* What happens when a cgroup hits memory.memsw.limit_in_bytes
 221When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out
 222in this cgroup. Then, swap-out will not be done by cgroup routine and file
 223caches are dropped. But as mentioned above, global LRU can do swapout memory
 224from it for sanity of the system's memory management state. You can't forbid
 225it by cgroup.
 2272.5 Reclaim
 229Each cgroup maintains a per cgroup LRU which has the same structure as
 230global VM. When a cgroup goes over its limit, we first try
 231to reclaim memory from the cgroup so as to make space for the new
 232pages that the cgroup has touched. If the reclaim is unsuccessful,
 233an OOM routine is invoked to select and kill the bulkiest task in the
 234cgroup. (See 10. OOM Control below.)
 236The reclaim algorithm has not been modified for cgroups, except that
 237pages that are selected for reclaiming come from the per cgroup LRU
 240NOTE: Reclaim does not work for the root cgroup, since we cannot set any
 241limits on the root cgroup.
 243Note2: When panic_on_oom is set to "2", the whole system will panic.
 245When oom event notifier is registered, event will be delivered.
 246(See oom_control section)
 2482.6 Locking
 250   lock_page_cgroup()/unlock_page_cgroup() should not be called under
 251   mapping->tree_lock.
 253   Other lock order is following:
 254   PG_locked.
 255   mm->page_table_lock
 256       zone->lru_lock
 257          lock_page_cgroup.
 258  In many cases, just lock_page_cgroup() is called.
 259  per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by
 260  zone->lru_lock, it has no lock of its own.
 2622.7 Kernel Memory Extension (CONFIG_CGROUP_MEM_RES_CTLR_KMEM)
 264With the Kernel memory extension, the Memory Controller is able to limit
 265the amount of kernel memory used by the system. Kernel memory is fundamentally
 266different than user memory, since it can't be swapped out, which makes it
 267possible to DoS the system by consuming too much of this precious resource.
 269Kernel memory limits are not imposed for the root cgroup. Usage for the root
 270cgroup may or may not be accounted.
 272Currently no soft limit is implemented for kernel memory. It is future work
 273to trigger slab reclaim when those limits are reached.
 2752.7.1 Current Kernel Memory resources accounted
 277* sockets memory pressure: some sockets protocols have memory pressure
 278thresholds. The Memory Controller allows them to be controlled individually
 279per cgroup, instead of globally.
 281* tcp memory pressure: sockets memory pressure for the tcp protocol.
 2833. User Interface
 2850. Configuration
 287a. Enable CONFIG_CGROUPS
 290d. Enable CONFIG_CGROUP_MEM_RES_CTLR_SWAP (to use swap extension)
 2921. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?)
 293# mount -t tmpfs none /sys/fs/cgroup
 294# mkdir /sys/fs/cgroup/memory
 295# mount -t cgroup none /sys/fs/cgroup/memory -o memory
 2972. Make the new group and move bash into it
 298# mkdir /sys/fs/cgroup/memory/0
 299# echo $$ > /sys/fs/cgroup/memory/0/tasks
 301Since now we're in the 0 cgroup, we can alter the memory limit:
 302# echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes
 304NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo,
 305mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.)
 307NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited).
 308NOTE: We cannot set limits on the root cgroup any more.
 310# cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes
 313We can check the usage:
 314# cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes
 317A successful write to this file does not guarantee a successful set of
 318this limit to the value written into the file. This can be due to a
 319number of factors, such as rounding up to page boundaries or the total
 320availability of memory on the system. The user is required to re-read
 321this file after a write to guarantee the value committed by the kernel.
 323# echo 1 > memory.limit_in_bytes
 324# cat memory.limit_in_bytes
 327The memory.failcnt field gives the number of times that the cgroup limit was
 330The memory.stat file gives accounting information. Now, the number of
 331caches, RSS and Active pages/Inactive pages are shown.
 3334. Testing
 335For testing features and implementation, see memcg_test.txt.
 337Performance test is also important. To see pure memory controller's overhead,
 338testing on tmpfs will give you good numbers of small overheads.
 339Example: do kernel make on tmpfs.
 341Page-fault scalability is also important. At measuring parallel
 342page fault test, multi-process test may be better than multi-thread
 343test because it has noise of shared objects/status.
 345But the above two are testing extreme situations.
 346Trying usual test under memory controller is always helpful.
 3484.1 Troubleshooting
 350Sometimes a user might find that the application under a cgroup is
 351terminated by OOM killer. There are several causes for this:
 3531. The cgroup limit is too low (just too low to do anything useful)
 3542. The user is using anonymous memory and swap is turned off or too low
 356A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of
 357some of the pages cached in the cgroup (page cache pages).
 359To know what happens, disable OOM_Kill by 10. OOM Control(see below) and
 360seeing what happens will be helpful.
 3624.2 Task migration
 364When a task migrates from one cgroup to another, its charge is not
 365carried forward by default. The pages allocated from the original cgroup still
 366remain charged to it, the charge is dropped when the page is freed or
 369You can move charges of a task along with task migration.
 370See 8. "Move charges at task migration"
 3724.3 Removing a cgroup
 374A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a
 375cgroup might have some charge associated with it, even though all
 376tasks have migrated away from it. (because we charge against pages, not
 377against tasks.)
 379We move the stats to root (if use_hierarchy==0) or parent (if
 380use_hierarchy==1), and no change on the charge except uncharging
 381from the child.
 383Charges recorded in swap information is not updated at removal of cgroup.
 384Recorded information is discarded and a cgroup which uses swap (swapcache)
 385will be charged as a new owner of it.
 387About use_hierarchy, see Section 6.
 3895. Misc. interfaces.
 3915.1 force_empty
 392  memory.force_empty interface is provided to make cgroup's memory usage empty.
 393  You can use this interface only when the cgroup has no tasks.
 394  When writing anything to this
 396  # echo 0 > memory.force_empty
 398  Almost all pages tracked by this memory cgroup will be unmapped and freed.
 399  Some pages cannot be freed because they are locked or in-use. Such pages are
 400  moved to parent(if use_hierarchy==1) or root (if use_hierarchy==0) and this
 401  cgroup will be empty.
 403  Typical use case of this interface is that calling this before rmdir().
 404  Because rmdir() moves all pages to parent, some out-of-use page caches can be
 405  moved to the parent. If you want to avoid that, force_empty will be useful.
 407  About use_hierarchy, see Section 6.
 4095.2 stat file
 411memory.stat file includes following statistics
 413# per-memory cgroup local status
 414cache           - # of bytes of page cache memory.
 415rss             - # of bytes of anonymous and swap cache memory.
 416mapped_file     - # of bytes of mapped file (includes tmpfs/shmem)
 417pgpgin          - # of charging events to the memory cgroup. The charging
 418                event happens each time a page is accounted as either mapped
 419                anon page(RSS) or cache page(Page Cache) to the cgroup.
 420pgpgout         - # of uncharging events to the memory cgroup. The uncharging
 421                event happens each time a page is unaccounted from the cgroup.
 422swap            - # of bytes of swap usage
 423inactive_anon   - # of bytes of anonymous memory and swap cache memory on
 424                LRU list.
 425active_anon     - # of bytes of anonymous and swap cache memory on active
 426                inactive LRU list.
 427inactive_file   - # of bytes of file-backed memory on inactive LRU list.
 428active_file     - # of bytes of file-backed memory on active LRU list.
 429unevictable     - # of bytes of memory that cannot be reclaimed (mlocked etc).
 431# status considering hierarchy (see memory.use_hierarchy settings)
 433hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy
 434                        under which the memory cgroup is
 435hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to
 436                        hierarchy under which memory cgroup is.
 438total_<counter>         - # hierarchical version of <counter>, which in
 439                        addition to the cgroup's own value includes the
 440                        sum of all hierarchical children's values of
 441                        <counter>, i.e. total_cache
 443# The following additional stats are dependent on CONFIG_DEBUG_VM.
 445recent_rotated_anon     - VM internal parameter. (see mm/vmscan.c)
 446recent_rotated_file     - VM internal parameter. (see mm/vmscan.c)
 447recent_scanned_anon     - VM internal parameter. (see mm/vmscan.c)
 448recent_scanned_file     - VM internal parameter. (see mm/vmscan.c)
 451        recent_rotated means recent frequency of LRU rotation.
 452        recent_scanned means recent # of scans to LRU.
 453        showing for better debug please see the code for meanings.
 456        Only anonymous and swap cache memory is listed as part of 'rss' stat.
 457        This should not be confused with the true 'resident set size' or the
 458        amount of physical memory used by the cgroup.
 459        'rss + file_mapped" will give you resident set size of cgroup.
 460        (Note: file and shmem may be shared among other cgroups. In that case,
 461         file_mapped is accounted only when the memory cgroup is owner of page
 462         cache.)
 4645.3 swappiness
 466Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only.
 468Following cgroups' swappiness can't be changed.
 469- root cgroup (uses /proc/sys/vm/swappiness).
 470- a cgroup which uses hierarchy and it has other cgroup(s) below it.
 471- a cgroup which uses hierarchy and not the root of hierarchy.
 4735.4 failcnt
 475A memory cgroup provides memory.failcnt and memory.memsw.failcnt files.
 476This failcnt(== failure count) shows the number of times that a usage counter
 477hit its limit. When a memory cgroup hits a limit, failcnt increases and
 478memory under it will be reclaimed.
 480You can reset failcnt by writing 0 to failcnt file.
 481# echo 0 > .../memory.failcnt
 4835.5 usage_in_bytes
 485For efficiency, as other kernel components, memory cgroup uses some optimization
 486to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the
 487method and doesn't show 'exact' value of memory(and swap) usage, it's an fuzz
 488value for efficient access. (Of course, when necessary, it's synchronized.)
 489If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP)
 490value in memory.stat(see 5.2).
 4925.6 numa_stat
 494This is similar to numa_maps but operates on a per-memcg basis.  This is
 495useful for providing visibility into the numa locality information within
 496an memcg since the pages are allowed to be allocated from any physical
 497node.  One of the usecases is evaluating application performance by
 498combining this information with the application's cpu allocation.
 500We export "total", "file", "anon" and "unevictable" pages per-node for
 501each memcg.  The ouput format of memory.numa_stat is:
 503total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ...
 504file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ...
 505anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
 506unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
 508And we have total = file + anon + unevictable.
 5106. Hierarchy support
 512The memory controller supports a deep hierarchy and hierarchical accounting.
 513The hierarchy is created by creating the appropriate cgroups in the
 514cgroup filesystem. Consider for example, the following cgroup filesystem
 517               root
 518             /  |   \
 519            /   |    \
 520           a    b     c
 521                      | \
 522                      |  \
 523                      d   e
 525In the diagram above, with hierarchical accounting enabled, all memory
 526usage of e, is accounted to its ancestors up until the root (i.e, c and root),
 527that has memory.use_hierarchy enabled. If one of the ancestors goes over its
 528limit, the reclaim algorithm reclaims from the tasks in the ancestor and the
 529children of the ancestor.
 5316.1 Enabling hierarchical accounting and reclaim
 533A memory cgroup by default disables the hierarchy feature. Support
 534can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup
 536# echo 1 > memory.use_hierarchy
 538The feature can be disabled by
 540# echo 0 > memory.use_hierarchy
 542NOTE1: Enabling/disabling will fail if either the cgroup already has other
 543       cgroups created below it, or if the parent cgroup has use_hierarchy
 544       enabled.
 546NOTE2: When panic_on_oom is set to "2", the whole system will panic in
 547       case of an OOM event in any cgroup.
 5497. Soft limits
 551Soft limits allow for greater sharing of memory. The idea behind soft limits
 552is to allow control groups to use as much of the memory as needed, provided
 554a. There is no memory contention
 555b. They do not exceed their hard limit
 557When the system detects memory contention or low memory, control groups
 558are pushed back to their soft limits. If the soft limit of each control
 559group is very high, they are pushed back as much as possible to make
 560sure that one control group does not starve the others of memory.
 562Please note that soft limits is a best effort feature, it comes with
 563no guarantees, but it does its best to make sure that when memory is
 564heavily contended for, memory is allocated based on the soft limit
 565hints/setup. Currently soft limit based reclaim is setup such that
 566it gets invoked from balance_pgdat (kswapd).
 5687.1 Interface
 570Soft limits can be setup by using the following commands (in this example we
 571assume a soft limit of 256 MiB)
 573# echo 256M > memory.soft_limit_in_bytes
 575If we want to change this to 1G, we can at any time use
 577# echo 1G > memory.soft_limit_in_bytes
 579NOTE1: Soft limits take effect over a long period of time, since they involve
 580       reclaiming memory for balancing between memory cgroups
 581NOTE2: It is recommended to set the soft limit always below the hard limit,
 582       otherwise the hard limit will take precedence.
 5848. Move charges at task migration
 586Users can move charges associated with a task along with task migration, that
 587is, uncharge task's pages from the old cgroup and charge them to the new cgroup.
 588This feature is not supported in !CONFIG_MMU environments because of lack of
 589page tables.
 5918.1 Interface
 593This feature is disabled by default. It can be enabled(and disabled again) by
 594writing to memory.move_charge_at_immigrate of the destination cgroup.
 596If you want to enable it:
 598# echo (some positive value) > memory.move_charge_at_immigrate
 600Note: Each bits of move_charge_at_immigrate has its own meaning about what type
 601      of charges should be moved. See 8.2 for details.
 602Note: Charges are moved only when you move mm->owner, IOW, a leader of a thread
 603      group.
 604Note: If we cannot find enough space for the task in the destination cgroup, we
 605      try to make space by reclaiming memory. Task migration may fail if we
 606      cannot make enough space.
 607Note: It can take several seconds if you move charges much.
 609And if you want disable it again:
 611# echo 0 > memory.move_charge_at_immigrate
 6138.2 Type of charges which can be move
 615Each bits of move_charge_at_immigrate has its own meaning about what type of
 616charges should be moved. But in any cases, it must be noted that an account of
 617a page or a swap can be moved only when it is charged to the task's current(old)
 618memory cgroup.
 620  bit | what type of charges would be moved ?
 621 -----+------------------------------------------------------------------------
 622   0  | A charge of an anonymous page(or swap of it) used by the target task.
 623      | You must enable Swap Extension(see 2.4) to enable move of swap charges.
 624 -----+------------------------------------------------------------------------
 625   1  | A charge of file pages(normal file, tmpfs file(e.g. ipc shared memory)
 626      | and swaps of tmpfs file) mmapped by the target task. Unlike the case of
 627      | anonymous pages, file pages(and swaps) in the range mmapped by the task
 628      | will be moved even if the task hasn't done page fault, i.e. they might
 629      | not be the task's "RSS", but other task's "RSS" that maps the same file.
 630      | And mapcount of the page is ignored(the page can be moved even if
 631      | page_mapcount(page) > 1). You must enable Swap Extension(see 2.4) to
 632      | enable move of swap charges.
 6348.3 TODO
 636- All of moving charge operations are done under cgroup_mutex. It's not good
 637  behavior to hold the mutex too long, so we may need some trick.
 6399. Memory thresholds
 641Memory cgroup implements memory thresholds using cgroups notification
 642API (see cgroups.txt). It allows to register multiple memory and memsw
 643thresholds and gets notifications when it crosses.
 645To register a threshold application need:
 646- create an eventfd using eventfd(2);
 647- open memory.usage_in_bytes or memory.memsw.usage_in_bytes;
 648- write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to
 649  cgroup.event_control.
 651Application will be notified through eventfd when memory usage crosses
 652threshold in any direction.
 654It's applicable for root and non-root cgroup.
 65610. OOM Control
 658memory.oom_control file is for OOM notification and other controls.
 660Memory cgroup implements OOM notifier using cgroup notification
 661API (See cgroups.txt). It allows to register multiple OOM notification
 662delivery and gets notification when OOM happens.
 664To register a notifier, application need:
 665 - create an eventfd using eventfd(2)
 666 - open memory.oom_control file
 667 - write string like "<event_fd> <fd of memory.oom_control>" to
 668   cgroup.event_control
 670Application will be notified through eventfd when OOM happens.
 671OOM notification doesn't work for root cgroup.
 673You can disable OOM-killer by writing "1" to memory.oom_control file, as:
 675        #echo 1 > memory.oom_control
 677This operation is only allowed to the top cgroup of sub-hierarchy.
 678If OOM-killer is disabled, tasks under cgroup will hang/sleep
 679in memory cgroup's OOM-waitqueue when they request accountable memory.
 681For running them, you have to relax the memory cgroup's OOM status by
 682        * enlarge limit or reduce usage.
 683To reduce usage,
 684        * kill some tasks.
 685        * move some tasks to other group with account migration.
 686        * remove some files (on tmpfs?)
 688Then, stopped tasks will work again.
 690At reading, current status of OOM is shown.
 691        oom_kill_disable 0 or 1 (if 1, oom-killer is disabled)
 692        under_oom        0 or 1 (if 1, the memory cgroup is under OOM, tasks may
 693                                 be stopped.)
 69511. TODO
 6971. Add support for accounting huge pages (as a separate controller)
 6982. Make per-cgroup scanner reclaim not-shared pages first
 6993. Teach controller to account for shared-pages
 7004. Start reclamation in the background when the limit is
 701   not yet hit but the usage is getting closer
 705Overall, the memory controller has been a stable controller and has been
 706commented and discussed quite extensively in the community.
 7101. Singh, Balbir. RFC: Memory Controller,
 7112. Singh, Balbir. Memory Controller (RSS Control),
 7133. Emelianov, Pavel. Resource controllers based on process cgroups
 7154. Emelianov, Pavel. RSS controller based on process cgroups (v2)
 7175. Emelianov, Pavel. RSS controller based on process cgroups (v3)
 7196. Menage, Paul. Control Groups v10,
 7207. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control
 721   subsystem (v3),
 7228. Singh, Balbir. RSS controller v2 test results (lmbench),
 7249. Singh, Balbir. RSS controller v2 AIM9 results
 72610. Singh, Balbir. Memory controller v6 test results,
 72811. Singh, Balbir. Memory controller introduction (v6),
 73012. Corbet, Jonathan, Controlling memory use in cgroups,
 732 kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.