1Memory Resource Controller
   3NOTE: This document is hopelessly outdated and it asks for a complete
   4      rewrite. It still contains a useful information so we are keeping it
   5      here but make sure to check the current code if you need a deeper
   6      understanding.
   8NOTE: The Memory Resource Controller has generically been referred to as the
   9      memory controller in this document. Do not confuse memory controller
  10      used here with the memory controller that is used in hardware.
  12(For editors)
  13In this document:
  14      When we mention a cgroup (cgroupfs's directory) with memory controller,
  15      we call it "memory cgroup". When you see git-log and source code, you'll
  16      see patch's title and function names tend to use "memcg".
  17      In this document, we avoid using it.
  19Benefits and Purpose of the memory controller
  21The memory controller isolates the memory behaviour of a group of tasks
  22from the rest of the system. The article on LWN [12] mentions some probable
  23uses of the memory controller. The memory controller can be used to
  25a. Isolate an application or a group of applications
  26   Memory-hungry applications can be isolated and limited to a smaller
  27   amount of memory.
  28b. Create a cgroup with a limited amount of memory; this can be used
  29   as a good alternative to booting with mem=XXXX.
  30c. Virtualization solutions can control the amount of memory they want
  31   to assign to a virtual machine instance.
  32d. A CD/DVD burner could control the amount of memory used by the
  33   rest of the system to ensure that burning does not fail due to lack
  34   of available memory.
  35e. There are several other use cases; find one or use the controller just
  36   for fun (to learn and hack on the VM subsystem).
  38Current Status: linux-2.6.34-mmotm(development version of 2010/April)
  41 - accounting anonymous pages, file caches, swap caches usage and limiting them.
  42 - pages are linked to per-memcg LRU exclusively, and there is no global LRU.
  43 - optionally, memory+swap usage can be accounted and limited.
  44 - hierarchical accounting
  45 - soft limit
  46 - moving (recharging) account at moving a task is selectable.
  47 - usage threshold notifier
  48 - memory pressure notifier
  49 - oom-killer disable knob and oom-notifier
  50 - Root cgroup has no limit controls.
  52 Kernel memory support is a work in progress, and the current version provides
  53 basically functionality. (See Section 2.7)
  55Brief summary of control files.
  57 tasks                           # attach a task(thread) and show list of threads
  58 cgroup.procs                    # show list of processes
  59 cgroup.event_control            # an interface for event_fd()
  60 memory.usage_in_bytes           # show current usage for memory
  61                                 (See 5.5 for details)
  62 memory.memsw.usage_in_bytes     # show current usage for memory+Swap
  63                                 (See 5.5 for details)
  64 memory.limit_in_bytes           # set/show limit of memory usage
  65 memory.memsw.limit_in_bytes     # set/show limit of memory+Swap usage
  66 memory.failcnt                  # show the number of memory usage hits limits
  67 memory.memsw.failcnt            # show the number of memory+Swap hits limits
  68 memory.max_usage_in_bytes       # show max memory usage recorded
  69 memory.memsw.max_usage_in_bytes # show max memory+Swap usage recorded
  70 memory.soft_limit_in_bytes      # set/show soft limit of memory usage
  71 memory.stat                     # show various statistics
  72 memory.use_hierarchy            # set/show hierarchical account enabled
  73 memory.force_empty              # trigger forced move charge to parent
  74 memory.pressure_level           # set memory pressure notifications
  75 memory.swappiness               # set/show swappiness parameter of vmscan
  76                                 (See sysctl's vm.swappiness)
  77 memory.move_charge_at_immigrate # set/show controls of moving charges
  78 memory.oom_control              # set/show oom controls.
  79 memory.numa_stat                # show the number of memory usage per numa node
  81 memory.kmem.limit_in_bytes      # set/show hard limit for kernel memory
  82 memory.kmem.usage_in_bytes      # show current kernel memory allocation
  83 memory.kmem.failcnt             # show the number of kernel memory usage hits limits
  84 memory.kmem.max_usage_in_bytes  # show max kernel memory usage recorded
  86 memory.kmem.tcp.limit_in_bytes  # set/show hard limit for tcp buf memory
  87 memory.kmem.tcp.usage_in_bytes  # show current tcp buf memory allocation
  88 memory.kmem.tcp.failcnt            # show the number of tcp buf memory usage hits limits
  89 memory.kmem.tcp.max_usage_in_bytes # show max tcp buf memory usage recorded
  911. History
  93The memory controller has a long history. A request for comments for the memory
  94controller was posted by Balbir Singh [1]. At the time the RFC was posted
  95there were several implementations for memory control. The goal of the
  96RFC was to build consensus and agreement for the minimal features required
  97for memory control. The first RSS controller was posted by Balbir Singh[2]
  98in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the
  99RSS controller. At OLS, at the resource management BoF, everyone suggested
 100that we handle both page cache and RSS together. Another request was raised
 101to allow user space handling of OOM. The current memory controller is
 102at version 6; it combines both mapped (RSS) and unmapped Page
 103Cache Control [11].
 1052. Memory Control
 107Memory is a unique resource in the sense that it is present in a limited
 108amount. If a task requires a lot of CPU processing, the task can spread
 109its processing over a period of hours, days, months or years, but with
 110memory, the same physical memory needs to be reused to accomplish the task.
 112The memory controller implementation has been divided into phases. These
 1151. Memory controller
 1162. mlock(2) controller
 1173. Kernel user memory accounting and slab control
 1184. user mappings length controller
 120The memory controller is the first controller developed.
 1222.1. Design
 124The core of the design is a counter called the page_counter. The
 125page_counter tracks the current memory usage and limit of the group of
 126processes associated with the controller. Each cgroup has a memory controller
 127specific data structure (mem_cgroup) associated with it.
 1292.2. Accounting
 131                +--------------------+
 132                |  mem_cgroup        |
 133                |  (page_counter)    |
 134                +--------------------+
 135                 /            ^      \
 136                /             |       \
 137           +---------------+  |        +---------------+
 138           | mm_struct     |  |....    | mm_struct     |
 139           |               |  |        |               |
 140           +---------------+  |        +---------------+
 141                              |
 142                              + --------------+
 143                                              |
 144           +---------------+           +------+--------+
 145           | page          +---------->  page_cgroup|
 146           |               |           |               |
 147           +---------------+           +---------------+
 149             (Figure 1: Hierarchy of Accounting)
 152Figure 1 shows the important aspects of the controller
 1541. Accounting happens per cgroup
 1552. Each mm_struct knows about which cgroup it belongs to
 1563. Each page has a pointer to the page_cgroup, which in turn knows the
 157   cgroup it belongs to
 159The accounting is done as follows: mem_cgroup_charge_common() is invoked to
 160set up the necessary data structures and check if the cgroup that is being
 161charged is over its limit. If it is, then reclaim is invoked on the cgroup.
 162More details can be found in the reclaim section of this document.
 163If everything goes well, a page meta-data-structure called page_cgroup is
 164updated. page_cgroup has its own LRU on cgroup.
 165(*) page_cgroup structure is allocated at boot/memory-hotplug time.
 1672.2.1 Accounting details
 169All mapped anon pages (RSS) and cache pages (Page Cache) are accounted.
 170Some pages which are never reclaimable and will not be on the LRU
 171are not accounted. We just account pages under usual VM management.
 173RSS pages are accounted at page_fault unless they've already been accounted
 174for earlier. A file page will be accounted for as Page Cache when it's
 175inserted into inode (radix-tree). While it's mapped into the page tables of
 176processes, duplicate accounting is carefully avoided.
 178An RSS page is unaccounted when it's fully unmapped. A PageCache page is
 179unaccounted when it's removed from radix-tree. Even if RSS pages are fully
 180unmapped (by kswapd), they may exist as SwapCache in the system until they
 181are really freed. Such SwapCaches are also accounted.
 182A swapped-in page is not accounted until it's mapped.
 184Note: The kernel does swapin-readahead and reads multiple swaps at once.
 185This means swapped-in pages may contain pages for other tasks than a task
 186causing page fault. So, we avoid accounting at swap-in I/O.
 188At page migration, accounting information is kept.
 190Note: we just account pages-on-LRU because our purpose is to control amount
 191of used pages; not-on-LRU pages tend to be out-of-control from VM view.
 1932.3 Shared Page Accounting
 195Shared pages are accounted on the basis of the first touch approach. The
 196cgroup that first touches a page is accounted for the page. The principle
 197behind this approach is that a cgroup that aggressively uses a shared
 198page will eventually get charged for it (once it is uncharged from
 199the cgroup that brought it in -- this will happen on memory pressure).
 201But see section 8.2: when moving a task to another cgroup, its pages may
 202be recharged to the new cgroup, if move_charge_at_immigrate has been chosen.
 204Exception: If CONFIG_MEMCG_SWAP is not used.
 205When you do swapoff and make swapped-out pages of shmem(tmpfs) to
 206be backed into memory in force, charges for pages are accounted against the
 207caller of swapoff rather than the users of shmem.
 2092.4 Swap Extension (CONFIG_MEMCG_SWAP)
 211Swap Extension allows you to record charge for swap. A swapped-in page is
 212charged back to original page allocator if possible.
 214When swap is accounted, following files are added.
 215 - memory.memsw.usage_in_bytes.
 216 - memory.memsw.limit_in_bytes.
 218memsw means memory+swap. Usage of memory+swap is limited by
 221Example: Assume a system with 4G of swap. A task which allocates 6G of memory
 222(by mistake) under 2G memory limitation will use all swap.
 223In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap.
 224By using the memsw limit, you can avoid system OOM which can be caused by swap
 227* why 'memory+swap' rather than swap.
 228The global LRU(kswapd) can swap out arbitrary pages. Swap-out means
 229to move account from memory to swap...there is no change in usage of
 230memory+swap. In other words, when we want to limit the usage of swap without
 231affecting global LRU, memory+swap limit is better than just limiting swap from
 232an OS point of view.
 234* What happens when a cgroup hits memory.memsw.limit_in_bytes
 235When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out
 236in this cgroup. Then, swap-out will not be done by cgroup routine and file
 237caches are dropped. But as mentioned above, global LRU can do swapout memory
 238from it for sanity of the system's memory management state. You can't forbid
 239it by cgroup.
 2412.5 Reclaim
 243Each cgroup maintains a per cgroup LRU which has the same structure as
 244global VM. When a cgroup goes over its limit, we first try
 245to reclaim memory from the cgroup so as to make space for the new
 246pages that the cgroup has touched. If the reclaim is unsuccessful,
 247an OOM routine is invoked to select and kill the bulkiest task in the
 248cgroup. (See 10. OOM Control below.)
 250The reclaim algorithm has not been modified for cgroups, except that
 251pages that are selected for reclaiming come from the per-cgroup LRU
 254NOTE: Reclaim does not work for the root cgroup, since we cannot set any
 255limits on the root cgroup.
 257Note2: When panic_on_oom is set to "2", the whole system will panic.
 259When oom event notifier is registered, event will be delivered.
 260(See oom_control section)
 2622.6 Locking
 264   lock_page_cgroup()/unlock_page_cgroup() should not be called under
 265   mapping->tree_lock.
 267   Other lock order is following:
 268   PG_locked.
 269   mm->page_table_lock
 270       zone->lru_lock
 271          lock_page_cgroup.
 272  In many cases, just lock_page_cgroup() is called.
 273  per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by
 274  zone->lru_lock, it has no lock of its own.
 2762.7 Kernel Memory Extension (CONFIG_MEMCG_KMEM)
 278WARNING: Current implementation lacks reclaim support. That means allocation
 279         attempts will fail when close to the limit even if there are plenty of
 280         kmem available for reclaim. That makes this option unusable in real
 281         life so DO NOT SELECT IT unless for development purposes.
 283With the Kernel memory extension, the Memory Controller is able to limit
 284the amount of kernel memory used by the system. Kernel memory is fundamentally
 285different than user memory, since it can't be swapped out, which makes it
 286possible to DoS the system by consuming too much of this precious resource.
 288Kernel memory won't be accounted at all until limit on a group is set. This
 289allows for existing setups to continue working without disruption.  The limit
 290cannot be set if the cgroup have children, or if there are already tasks in the
 291cgroup. Attempting to set the limit under those conditions will return -EBUSY.
 292When use_hierarchy == 1 and a group is accounted, its children will
 293automatically be accounted regardless of their limit value.
 295After a group is first limited, it will be kept being accounted until it
 296is removed. The memory limitation itself, can of course be removed by writing
 297-1 to memory.kmem.limit_in_bytes. In this case, kmem will be accounted, but not
 300Kernel memory limits are not imposed for the root cgroup. Usage for the root
 301cgroup may or may not be accounted. The memory used is accumulated into
 302memory.kmem.usage_in_bytes, or in a separate counter when it makes sense.
 303(currently only for tcp).
 304The main "kmem" counter is fed into the main counter, so kmem charges will
 305also be visible from the user counter.
 307Currently no soft limit is implemented for kernel memory. It is future work
 308to trigger slab reclaim when those limits are reached.
 3102.7.1 Current Kernel Memory resources accounted
 312* stack pages: every process consumes some stack pages. By accounting into
 313kernel memory, we prevent new processes from being created when the kernel
 314memory usage is too high.
 316* slab pages: pages allocated by the SLAB or SLUB allocator are tracked. A copy
 317of each kmem_cache is created every time the cache is touched by the first time
 318from inside the memcg. The creation is done lazily, so some objects can still be
 319skipped while the cache is being created. All objects in a slab page should
 320belong to the same memcg. This only fails to hold when a task is migrated to a
 321different memcg during the page allocation by the cache.
 323* sockets memory pressure: some sockets protocols have memory pressure
 324thresholds. The Memory Controller allows them to be controlled individually
 325per cgroup, instead of globally.
 327* tcp memory pressure: sockets memory pressure for the tcp protocol.
 3292.7.2 Common use cases
 331Because the "kmem" counter is fed to the main user counter, kernel memory can
 332never be limited completely independently of user memory. Say "U" is the user
 333limit, and "K" the kernel limit. There are three possible ways limits can be
 336    U != 0, K = unlimited:
 337    This is the standard memcg limitation mechanism already present before kmem
 338    accounting. Kernel memory is completely ignored.
 340    U != 0, K < U:
 341    Kernel memory is a subset of the user memory. This setup is useful in
 342    deployments where the total amount of memory per-cgroup is overcommited.
 343    Overcommiting kernel memory limits is definitely not recommended, since the
 344    box can still run out of non-reclaimable memory.
 345    In this case, the admin could set up K so that the sum of all groups is
 346    never greater than the total memory, and freely set U at the cost of his
 347    QoS.
 349    U != 0, K >= U:
 350    Since kmem charges will also be fed to the user counter and reclaim will be
 351    triggered for the cgroup for both kinds of memory. This setup gives the
 352    admin a unified view of memory, and it is also useful for people who just
 353    want to track kernel memory usage.
 3553. User Interface
 3573.0. Configuration
 359a. Enable CONFIG_CGROUPS
 360b. Enable CONFIG_MEMCG
 361c. Enable CONFIG_MEMCG_SWAP (to use swap extension)
 362d. Enable CONFIG_MEMCG_KMEM (to use kmem extension)
 3643.1. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?)
 365# mount -t tmpfs none /sys/fs/cgroup
 366# mkdir /sys/fs/cgroup/memory
 367# mount -t cgroup none /sys/fs/cgroup/memory -o memory
 3693.2. Make the new group and move bash into it
 370# mkdir /sys/fs/cgroup/memory/0
 371# echo $$ > /sys/fs/cgroup/memory/0/tasks
 373Since now we're in the 0 cgroup, we can alter the memory limit:
 374# echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes
 376NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo,
 377mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.)
 379NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited).
 380NOTE: We cannot set limits on the root cgroup any more.
 382# cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes
 385We can check the usage:
 386# cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes
 389A successful write to this file does not guarantee a successful setting of
 390this limit to the value written into the file. This can be due to a
 391number of factors, such as rounding up to page boundaries or the total
 392availability of memory on the system. The user is required to re-read
 393this file after a write to guarantee the value committed by the kernel.
 395# echo 1 > memory.limit_in_bytes
 396# cat memory.limit_in_bytes
 399The memory.failcnt field gives the number of times that the cgroup limit was
 402The memory.stat file gives accounting information. Now, the number of
 403caches, RSS and Active pages/Inactive pages are shown.
 4054. Testing
 407For testing features and implementation, see memcg_test.txt.
 409Performance test is also important. To see pure memory controller's overhead,
 410testing on tmpfs will give you good numbers of small overheads.
 411Example: do kernel make on tmpfs.
 413Page-fault scalability is also important. At measuring parallel
 414page fault test, multi-process test may be better than multi-thread
 415test because it has noise of shared objects/status.
 417But the above two are testing extreme situations.
 418Trying usual test under memory controller is always helpful.
 4204.1 Troubleshooting
 422Sometimes a user might find that the application under a cgroup is
 423terminated by the OOM killer. There are several causes for this:
 4251. The cgroup limit is too low (just too low to do anything useful)
 4262. The user is using anonymous memory and swap is turned off or too low
 428A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of
 429some of the pages cached in the cgroup (page cache pages).
 431To know what happens, disabling OOM_Kill as per "10. OOM Control" (below) and
 432seeing what happens will be helpful.
 4344.2 Task migration
 436When a task migrates from one cgroup to another, its charge is not
 437carried forward by default. The pages allocated from the original cgroup still
 438remain charged to it, the charge is dropped when the page is freed or
 441You can move charges of a task along with task migration.
 442See 8. "Move charges at task migration"
 4444.3 Removing a cgroup
 446A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a
 447cgroup might have some charge associated with it, even though all
 448tasks have migrated away from it. (because we charge against pages, not
 449against tasks.)
 451We move the stats to root (if use_hierarchy==0) or parent (if
 452use_hierarchy==1), and no change on the charge except uncharging
 453from the child.
 455Charges recorded in swap information is not updated at removal of cgroup.
 456Recorded information is discarded and a cgroup which uses swap (swapcache)
 457will be charged as a new owner of it.
 459About use_hierarchy, see Section 6.
 4615. Misc. interfaces.
 4635.1 force_empty
 464  memory.force_empty interface is provided to make cgroup's memory usage empty.
 465  When writing anything to this
 467  # echo 0 > memory.force_empty
 469  the cgroup will be reclaimed and as many pages reclaimed as possible.
 471  The typical use case for this interface is before calling rmdir().
 472  Because rmdir() moves all pages to parent, some out-of-use page caches can be
 473  moved to the parent. If you want to avoid that, force_empty will be useful.
 475  Also, note that when memory.kmem.limit_in_bytes is set the charges due to
 476  kernel pages will still be seen. This is not considered a failure and the
 477  write will still return success. In this case, it is expected that
 478  memory.kmem.usage_in_bytes == memory.usage_in_bytes.
 480  About use_hierarchy, see Section 6.
 4825.2 stat file
 484memory.stat file includes following statistics
 486# per-memory cgroup local status
 487cache           - # of bytes of page cache memory.
 488rss             - # of bytes of anonymous and swap cache memory (includes
 489                transparent hugepages).
 490rss_huge        - # of bytes of anonymous transparent hugepages.
 491mapped_file     - # of bytes of mapped file (includes tmpfs/shmem)
 492pgpgin          - # of charging events to the memory cgroup. The charging
 493                event happens each time a page is accounted as either mapped
 494                anon page(RSS) or cache page(Page Cache) to the cgroup.
 495pgpgout         - # of uncharging events to the memory cgroup. The uncharging
 496                event happens each time a page is unaccounted from the cgroup.
 497swap            - # of bytes of swap usage
 498writeback       - # of bytes of file/anon cache that are queued for syncing to
 499                disk.
 500inactive_anon   - # of bytes of anonymous and swap cache memory on inactive
 501                LRU list.
 502active_anon     - # of bytes of anonymous and swap cache memory on active
 503                LRU list.
 504inactive_file   - # of bytes of file-backed memory on inactive LRU list.
 505active_file     - # of bytes of file-backed memory on active LRU list.
 506unevictable     - # of bytes of memory that cannot be reclaimed (mlocked etc).
 508# status considering hierarchy (see memory.use_hierarchy settings)
 510hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy
 511                        under which the memory cgroup is
 512hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to
 513                        hierarchy under which memory cgroup is.
 515total_<counter>         - # hierarchical version of <counter>, which in
 516                        addition to the cgroup's own value includes the
 517                        sum of all hierarchical children's values of
 518                        <counter>, i.e. total_cache
 520# The following additional stats are dependent on CONFIG_DEBUG_VM.
 522recent_rotated_anon     - VM internal parameter. (see mm/vmscan.c)
 523recent_rotated_file     - VM internal parameter. (see mm/vmscan.c)
 524recent_scanned_anon     - VM internal parameter. (see mm/vmscan.c)
 525recent_scanned_file     - VM internal parameter. (see mm/vmscan.c)
 528        recent_rotated means recent frequency of LRU rotation.
 529        recent_scanned means recent # of scans to LRU.
 530        showing for better debug please see the code for meanings.
 533        Only anonymous and swap cache memory is listed as part of 'rss' stat.
 534        This should not be confused with the true 'resident set size' or the
 535        amount of physical memory used by the cgroup.
 536        'rss + file_mapped" will give you resident set size of cgroup.
 537        (Note: file and shmem may be shared among other cgroups. In that case,
 538         file_mapped is accounted only when the memory cgroup is owner of page
 539         cache.)
 5415.3 swappiness
 543Overrides /proc/sys/vm/swappiness for the particular group. The tunable
 544in the root cgroup corresponds to the global swappiness setting.
 546Please note that unlike during the global reclaim, limit reclaim
 547enforces that 0 swappiness really prevents from any swapping even if
 548there is a swap storage available. This might lead to memcg OOM killer
 549if there are no file pages to reclaim.
 5515.4 failcnt
 553A memory cgroup provides memory.failcnt and memory.memsw.failcnt files.
 554This failcnt(== failure count) shows the number of times that a usage counter
 555hit its limit. When a memory cgroup hits a limit, failcnt increases and
 556memory under it will be reclaimed.
 558You can reset failcnt by writing 0 to failcnt file.
 559# echo 0 > .../memory.failcnt
 5615.5 usage_in_bytes
 563For efficiency, as other kernel components, memory cgroup uses some optimization
 564to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the
 565method and doesn't show 'exact' value of memory (and swap) usage, it's a fuzz
 566value for efficient access. (Of course, when necessary, it's synchronized.)
 567If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP)
 568value in memory.stat(see 5.2).
 5705.6 numa_stat
 572This is similar to numa_maps but operates on a per-memcg basis.  This is
 573useful for providing visibility into the numa locality information within
 574an memcg since the pages are allowed to be allocated from any physical
 575node.  One of the use cases is evaluating application performance by
 576combining this information with the application's CPU allocation.
 578Each memcg's numa_stat file includes "total", "file", "anon" and "unevictable"
 579per-node page counts including "hierarchical_<counter>" which sums up all
 580hierarchical children's values in addition to the memcg's own value.
 582The output format of memory.numa_stat is:
 584total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ...
 585file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ...
 586anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
 587unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
 588hierarchical_<counter>=<counter pages> N0=<node 0 pages> N1=<node 1 pages> ...
 590The "total" count is sum of file + anon + unevictable.
 5926. Hierarchy support
 594The memory controller supports a deep hierarchy and hierarchical accounting.
 595The hierarchy is created by creating the appropriate cgroups in the
 596cgroup filesystem. Consider for example, the following cgroup filesystem
 599               root
 600             /  |   \
 601            /   |    \
 602           a    b     c
 603                      | \
 604                      |  \
 605                      d   e
 607In the diagram above, with hierarchical accounting enabled, all memory
 608usage of e, is accounted to its ancestors up until the root (i.e, c and root),
 609that has memory.use_hierarchy enabled. If one of the ancestors goes over its
 610limit, the reclaim algorithm reclaims from the tasks in the ancestor and the
 611children of the ancestor.
 6136.1 Enabling hierarchical accounting and reclaim
 615A memory cgroup by default disables the hierarchy feature. Support
 616can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup
 618# echo 1 > memory.use_hierarchy
 620The feature can be disabled by
 622# echo 0 > memory.use_hierarchy
 624NOTE1: Enabling/disabling will fail if either the cgroup already has other
 625       cgroups created below it, or if the parent cgroup has use_hierarchy
 626       enabled.
 628NOTE2: When panic_on_oom is set to "2", the whole system will panic in
 629       case of an OOM event in any cgroup.
 6317. Soft limits
 633Soft limits allow for greater sharing of memory. The idea behind soft limits
 634is to allow control groups to use as much of the memory as needed, provided
 636a. There is no memory contention
 637b. They do not exceed their hard limit
 639When the system detects memory contention or low memory, control groups
 640are pushed back to their soft limits. If the soft limit of each control
 641group is very high, they are pushed back as much as possible to make
 642sure that one control group does not starve the others of memory.
 644Please note that soft limits is a best-effort feature; it comes with
 645no guarantees, but it does its best to make sure that when memory is
 646heavily contended for, memory is allocated based on the soft limit
 647hints/setup. Currently soft limit based reclaim is set up such that
 648it gets invoked from balance_pgdat (kswapd).
 6507.1 Interface
 652Soft limits can be setup by using the following commands (in this example we
 653assume a soft limit of 256 MiB)
 655# echo 256M > memory.soft_limit_in_bytes
 657If we want to change this to 1G, we can at any time use
 659# echo 1G > memory.soft_limit_in_bytes
 661NOTE1: Soft limits take effect over a long period of time, since they involve
 662       reclaiming memory for balancing between memory cgroups
 663NOTE2: It is recommended to set the soft limit always below the hard limit,
 664       otherwise the hard limit will take precedence.
 6668. Move charges at task migration
 668Users can move charges associated with a task along with task migration, that
 669is, uncharge task's pages from the old cgroup and charge them to the new cgroup.
 670This feature is not supported in !CONFIG_MMU environments because of lack of
 671page tables.
 6738.1 Interface
 675This feature is disabled by default. It can be enabled (and disabled again) by
 676writing to memory.move_charge_at_immigrate of the destination cgroup.
 678If you want to enable it:
 680# echo (some positive value) > memory.move_charge_at_immigrate
 682Note: Each bits of move_charge_at_immigrate has its own meaning about what type
 683      of charges should be moved. See 8.2 for details.
 684Note: Charges are moved only when you move mm->owner, in other words,
 685      a leader of a thread group.
 686Note: If we cannot find enough space for the task in the destination cgroup, we
 687      try to make space by reclaiming memory. Task migration may fail if we
 688      cannot make enough space.
 689Note: It can take several seconds if you move charges much.
 691And if you want disable it again:
 693# echo 0 > memory.move_charge_at_immigrate
 6958.2 Type of charges which can be moved
 697Each bit in move_charge_at_immigrate has its own meaning about what type of
 698charges should be moved. But in any case, it must be noted that an account of
 699a page or a swap can be moved only when it is charged to the task's current
 700(old) memory cgroup.
 702  bit | what type of charges would be moved ?
 703 -----+------------------------------------------------------------------------
 704   0  | A charge of an anonymous page (or swap of it) used by the target task.
 705      | You must enable Swap Extension (see 2.4) to enable move of swap charges.
 706 -----+------------------------------------------------------------------------
 707   1  | A charge of file pages (normal file, tmpfs file (e.g. ipc shared memory)
 708      | and swaps of tmpfs file) mmapped by the target task. Unlike the case of
 709      | anonymous pages, file pages (and swaps) in the range mmapped by the task
 710      | will be moved even if the task hasn't done page fault, i.e. they might
 711      | not be the task's "RSS", but other task's "RSS" that maps the same file.
 712      | And mapcount of the page is ignored (the page can be moved even if
 713      | page_mapcount(page) > 1). You must enable Swap Extension (see 2.4) to
 714      | enable move of swap charges.
 7168.3 TODO
 718- All of moving charge operations are done under cgroup_mutex. It's not good
 719  behavior to hold the mutex too long, so we may need some trick.
 7219. Memory thresholds
 723Memory cgroup implements memory thresholds using the cgroups notification
 724API (see cgroups.txt). It allows to register multiple memory and memsw
 725thresholds and gets notifications when it crosses.
 727To register a threshold, an application must:
 728- create an eventfd using eventfd(2);
 729- open memory.usage_in_bytes or memory.memsw.usage_in_bytes;
 730- write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to
 731  cgroup.event_control.
 733Application will be notified through eventfd when memory usage crosses
 734threshold in any direction.
 736It's applicable for root and non-root cgroup.
 73810. OOM Control
 740memory.oom_control file is for OOM notification and other controls.
 742Memory cgroup implements OOM notifier using the cgroup notification
 743API (See cgroups.txt). It allows to register multiple OOM notification
 744delivery and gets notification when OOM happens.
 746To register a notifier, an application must:
 747 - create an eventfd using eventfd(2)
 748 - open memory.oom_control file
 749 - write string like "<event_fd> <fd of memory.oom_control>" to
 750   cgroup.event_control
 752The application will be notified through eventfd when OOM happens.
 753OOM notification doesn't work for the root cgroup.
 755You can disable the OOM-killer by writing "1" to memory.oom_control file, as:
 757        #echo 1 > memory.oom_control
 759If OOM-killer is disabled, tasks under cgroup will hang/sleep
 760in memory cgroup's OOM-waitqueue when they request accountable memory.
 762For running them, you have to relax the memory cgroup's OOM status by
 763        * enlarge limit or reduce usage.
 764To reduce usage,
 765        * kill some tasks.
 766        * move some tasks to other group with account migration.
 767        * remove some files (on tmpfs?)
 769Then, stopped tasks will work again.
 771At reading, current status of OOM is shown.
 772        oom_kill_disable 0 or 1 (if 1, oom-killer is disabled)
 773        under_oom        0 or 1 (if 1, the memory cgroup is under OOM, tasks may
 774                                 be stopped.)
 77611. Memory Pressure
 778The pressure level notifications can be used to monitor the memory
 779allocation cost; based on the pressure, applications can implement
 780different strategies of managing their memory resources. The pressure
 781levels are defined as following:
 783The "low" level means that the system is reclaiming memory for new
 784allocations. Monitoring this reclaiming activity might be useful for
 785maintaining cache level. Upon notification, the program (typically
 786"Activity Manager") might analyze vmstat and act in advance (i.e.
 787prematurely shutdown unimportant services).
 789The "medium" level means that the system is experiencing medium memory
 790pressure, the system might be making swap, paging out active file caches,
 791etc. Upon this event applications may decide to further analyze
 792vmstat/zoneinfo/memcg or internal memory usage statistics and free any
 793resources that can be easily reconstructed or re-read from a disk.
 795The "critical" level means that the system is actively thrashing, it is
 796about to out of memory (OOM) or even the in-kernel OOM killer is on its
 797way to trigger. Applications should do whatever they can to help the
 798system. It might be too late to consult with vmstat or any other
 799statistics, so it's advisable to take an immediate action.
 801The events are propagated upward until the event is handled, i.e. the
 802events are not pass-through. Here is what this means: for example you have
 803three cgroups: A->B->C. Now you set up an event listener on cgroups A, B
 804and C, and suppose group C experiences some pressure. In this situation,
 805only group C will receive the notification, i.e. groups A and B will not
 806receive it. This is done to avoid excessive "broadcasting" of messages,
 807which disturbs the system and which is especially bad if we are low on
 808memory or thrashing. So, organize the cgroups wisely, or propagate the
 809events manually (or, ask us to implement the pass-through events,
 810explaining why would you need them.)
 812The file memory.pressure_level is only used to setup an eventfd. To
 813register a notification, an application must:
 815- create an eventfd using eventfd(2);
 816- open memory.pressure_level;
 817- write string like "<event_fd> <fd of memory.pressure_level> <level>"
 818  to cgroup.event_control.
 820Application will be notified through eventfd when memory pressure is at
 821the specific level (or higher). Read/write operations to
 822memory.pressure_level are no implemented.
 826   Here is a small script example that makes a new cgroup, sets up a
 827   memory limit, sets up a notification in the cgroup and then makes child
 828   cgroup experience a critical pressure:
 830   # cd /sys/fs/cgroup/memory/
 831   # mkdir foo
 832   # cd foo
 833   # cgroup_event_listener memory.pressure_level low &
 834   # echo 8000000 > memory.limit_in_bytes
 835   # echo 8000000 > memory.memsw.limit_in_bytes
 836   # echo $$ > tasks
 837   # dd if=/dev/zero | read x
 839   (Expect a bunch of notifications, and eventually, the oom-killer will
 840   trigger.)
 84212. TODO
 8441. Make per-cgroup scanner reclaim not-shared pages first
 8452. Teach controller to account for shared-pages
 8463. Start reclamation in the background when the limit is
 847   not yet hit but the usage is getting closer
 851Overall, the memory controller has been a stable controller and has been
 852commented and discussed quite extensively in the community.
 8561. Singh, Balbir. RFC: Memory Controller,
 8572. Singh, Balbir. Memory Controller (RSS Control),
 8593. Emelianov, Pavel. Resource controllers based on process cgroups
 8614. Emelianov, Pavel. RSS controller based on process cgroups (v2)
 8635. Emelianov, Pavel. RSS controller based on process cgroups (v3)
 8656. Menage, Paul. Control Groups v10,
 8667. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control
 867   subsystem (v3),
 8688. Singh, Balbir. RSS controller v2 test results (lmbench),
 8709. Singh, Balbir. RSS controller v2 AIM9 results
 87210. Singh, Balbir. Memory controller v6 test results,
 87411. Singh, Balbir. Memory controller introduction (v6),
 87612. Corbet, Jonathan, Controlling memory use in cgroups,
 878 kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.