linux/Documentation/cgroups/memory.txt
<<
>>
Prefs
   1Memory Resource Controller
   2
   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.
   6
   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.
  13
  14Benefits and Purpose of the memory controller
  15
  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
  19
  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).
  32
  33Current Status: linux-2.6.34-mmotm(development version of 2010/April)
  34
  35Features:
  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.
  45
  46 Kernel memory support is work in progress, and the current version provides
  47 basically functionality. (See Section 2.7)
  48
  49Brief summary of control files.
  50
  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
  73
  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
  76 memory.kmem.tcp.failcnt            # show the number of tcp buf memory usage hits limits
  77 memory.kmem.tcp.max_usage_in_bytes # show max tcp buf memory usage recorded
  78
  791. History
  80
  81The memory controller has a long history. A request for comments for the memory
  82controller was posted by Balbir Singh [1]. At the time the RFC was posted
  83there were several implementations for memory control. The goal of the
  84RFC was to build consensus and agreement for the minimal features required
  85for memory control. The first RSS controller was posted by Balbir Singh[2]
  86in Feb 2007. Pavel Emelianov [3][4][5] has since posted three versions of the
  87RSS controller. At OLS, at the resource management BoF, everyone suggested
  88that we handle both page cache and RSS together. Another request was raised
  89to allow user space handling of OOM. The current memory controller is
  90at version 6; it combines both mapped (RSS) and unmapped Page
  91Cache Control [11].
  92
  932. Memory Control
  94
  95Memory is a unique resource in the sense that it is present in a limited
  96amount. If a task requires a lot of CPU processing, the task can spread
  97its processing over a period of hours, days, months or years, but with
  98memory, the same physical memory needs to be reused to accomplish the task.
  99
 100The memory controller implementation has been divided into phases. These
 101are:
 102
 1031. Memory controller
 1042. mlock(2) controller
 1053. Kernel user memory accounting and slab control
 1064. user mappings length controller
 107
 108The memory controller is the first controller developed.
 109
 1102.1. Design
 111
 112The core of the design is a counter called the res_counter. The res_counter
 113tracks the current memory usage and limit of the group of processes associated
 114with the controller. Each cgroup has a memory controller specific data
 115structure (mem_cgroup) associated with it.
 116
 1172.2. Accounting
 118
 119                +--------------------+
 120                |  mem_cgroup     |
 121                |  (res_counter)     |
 122                +--------------------+
 123                 /            ^      \
 124                /             |       \
 125           +---------------+  |        +---------------+
 126           | mm_struct     |  |....    | mm_struct     |
 127           |               |  |        |               |
 128           +---------------+  |        +---------------+
 129                              |
 130                              + --------------+
 131                                              |
 132           +---------------+           +------+--------+
 133           | page          +---------->  page_cgroup|
 134           |               |           |               |
 135           +---------------+           +---------------+
 136
 137             (Figure 1: Hierarchy of Accounting)
 138
 139
 140Figure 1 shows the important aspects of the controller
 141
 1421. Accounting happens per cgroup
 1432. Each mm_struct knows about which cgroup it belongs to
 1443. Each page has a pointer to the page_cgroup, which in turn knows the
 145   cgroup it belongs to
 146
 147The accounting is done as follows: mem_cgroup_charge() is invoked to setup
 148the necessary data structures and check if the cgroup that is being charged
 149is over its limit. If it is then reclaim is invoked on the cgroup.
 150More details can be found in the reclaim section of this document.
 151If everything goes well, a page meta-data-structure called page_cgroup is
 152updated. page_cgroup has its own LRU on cgroup.
 153(*) page_cgroup structure is allocated at boot/memory-hotplug time.
 154
 1552.2.1 Accounting details
 156
 157All mapped anon pages (RSS) and cache pages (Page Cache) are accounted.
 158Some pages which are never reclaimable and will not be on the LRU
 159are not accounted. We just account pages under usual VM management.
 160
 161RSS pages are accounted at page_fault unless they've already been accounted
 162for earlier. A file page will be accounted for as Page Cache when it's
 163inserted into inode (radix-tree). While it's mapped into the page tables of
 164processes, duplicate accounting is carefully avoided.
 165
 166A RSS page is unaccounted when it's fully unmapped. A PageCache page is
 167unaccounted when it's removed from radix-tree. Even if RSS pages are fully
 168unmapped (by kswapd), they may exist as SwapCache in the system until they
 169are really freed. Such SwapCaches also also accounted.
 170A swapped-in page is not accounted until it's mapped.
 171
 172Note: The kernel does swapin-readahead and read multiple swaps at once.
 173This means swapped-in pages may contain pages for other tasks than a task
 174causing page fault. So, we avoid accounting at swap-in I/O.
 175
 176At page migration, accounting information is kept.
 177
 178Note: we just account pages-on-LRU because our purpose is to control amount
 179of used pages; not-on-LRU pages tend to be out-of-control from VM view.
 180
 1812.3 Shared Page Accounting
 182
 183Shared pages are accounted on the basis of the first touch approach. The
 184cgroup that first touches a page is accounted for the page. The principle
 185behind this approach is that a cgroup that aggressively uses a shared
 186page will eventually get charged for it (once it is uncharged from
 187the cgroup that brought it in -- this will happen on memory pressure).
 188
 189But see section 8.2: when moving a task to another cgroup, its pages may
 190be recharged to the new cgroup, if move_charge_at_immigrate has been chosen.
 191
 192Exception: If CONFIG_CGROUP_CGROUP_MEMCG_SWAP is not used.
 193When you do swapoff and make swapped-out pages of shmem(tmpfs) to
 194be backed into memory in force, charges for pages are accounted against the
 195caller of swapoff rather than the users of shmem.
 196
 1972.4 Swap Extension (CONFIG_MEMCG_SWAP)
 198
 199Swap Extension allows you to record charge for swap. A swapped-in page is
 200charged back to original page allocator if possible.
 201
 202When swap is accounted, following files are added.
 203 - memory.memsw.usage_in_bytes.
 204 - memory.memsw.limit_in_bytes.
 205
 206memsw means memory+swap. Usage of memory+swap is limited by
 207memsw.limit_in_bytes.
 208
 209Example: Assume a system with 4G of swap. A task which allocates 6G of memory
 210(by mistake) under 2G memory limitation will use all swap.
 211In this case, setting memsw.limit_in_bytes=3G will prevent bad use of swap.
 212By using memsw limit, you can avoid system OOM which can be caused by swap
 213shortage.
 214
 215* why 'memory+swap' rather than swap.
 216The global LRU(kswapd) can swap out arbitrary pages. Swap-out means
 217to move account from memory to swap...there is no change in usage of
 218memory+swap. In other words, when we want to limit the usage of swap without
 219affecting global LRU, memory+swap limit is better than just limiting swap from
 220OS point of view.
 221
 222* What happens when a cgroup hits memory.memsw.limit_in_bytes
 223When a cgroup hits memory.memsw.limit_in_bytes, it's useless to do swap-out
 224in this cgroup. Then, swap-out will not be done by cgroup routine and file
 225caches are dropped. But as mentioned above, global LRU can do swapout memory
 226from it for sanity of the system's memory management state. You can't forbid
 227it by cgroup.
 228
 2292.5 Reclaim
 230
 231Each cgroup maintains a per cgroup LRU which has the same structure as
 232global VM. When a cgroup goes over its limit, we first try
 233to reclaim memory from the cgroup so as to make space for the new
 234pages that the cgroup has touched. If the reclaim is unsuccessful,
 235an OOM routine is invoked to select and kill the bulkiest task in the
 236cgroup. (See 10. OOM Control below.)
 237
 238The reclaim algorithm has not been modified for cgroups, except that
 239pages that are selected for reclaiming come from the per cgroup LRU
 240list.
 241
 242NOTE: Reclaim does not work for the root cgroup, since we cannot set any
 243limits on the root cgroup.
 244
 245Note2: When panic_on_oom is set to "2", the whole system will panic.
 246
 247When oom event notifier is registered, event will be delivered.
 248(See oom_control section)
 249
 2502.6 Locking
 251
 252   lock_page_cgroup()/unlock_page_cgroup() should not be called under
 253   mapping->tree_lock.
 254
 255   Other lock order is following:
 256   PG_locked.
 257   mm->page_table_lock
 258       zone->lru_lock
 259          lock_page_cgroup.
 260  In many cases, just lock_page_cgroup() is called.
 261  per-zone-per-cgroup LRU (cgroup's private LRU) is just guarded by
 262  zone->lru_lock, it has no lock of its own.
 263
 2642.7 Kernel Memory Extension (CONFIG_MEMCG_KMEM)
 265
 266With the Kernel memory extension, the Memory Controller is able to limit
 267the amount of kernel memory used by the system. Kernel memory is fundamentally
 268different than user memory, since it can't be swapped out, which makes it
 269possible to DoS the system by consuming too much of this precious resource.
 270
 271Kernel memory limits are not imposed for the root cgroup. Usage for the root
 272cgroup may or may not be accounted.
 273
 274Currently no soft limit is implemented for kernel memory. It is future work
 275to trigger slab reclaim when those limits are reached.
 276
 2772.7.1 Current Kernel Memory resources accounted
 278
 279* sockets memory pressure: some sockets protocols have memory pressure
 280thresholds. The Memory Controller allows them to be controlled individually
 281per cgroup, instead of globally.
 282
 283* tcp memory pressure: sockets memory pressure for the tcp protocol.
 284
 2853. User Interface
 286
 2870. Configuration
 288
 289a. Enable CONFIG_CGROUPS
 290b. Enable CONFIG_RESOURCE_COUNTERS
 291c. Enable CONFIG_MEMCG
 292d. Enable CONFIG_MEMCG_SWAP (to use swap extension)
 293
 2941. Prepare the cgroups (see cgroups.txt, Why are cgroups needed?)
 295# mount -t tmpfs none /sys/fs/cgroup
 296# mkdir /sys/fs/cgroup/memory
 297# mount -t cgroup none /sys/fs/cgroup/memory -o memory
 298
 2992. Make the new group and move bash into it
 300# mkdir /sys/fs/cgroup/memory/0
 301# echo $$ > /sys/fs/cgroup/memory/0/tasks
 302
 303Since now we're in the 0 cgroup, we can alter the memory limit:
 304# echo 4M > /sys/fs/cgroup/memory/0/memory.limit_in_bytes
 305
 306NOTE: We can use a suffix (k, K, m, M, g or G) to indicate values in kilo,
 307mega or gigabytes. (Here, Kilo, Mega, Giga are Kibibytes, Mebibytes, Gibibytes.)
 308
 309NOTE: We can write "-1" to reset the *.limit_in_bytes(unlimited).
 310NOTE: We cannot set limits on the root cgroup any more.
 311
 312# cat /sys/fs/cgroup/memory/0/memory.limit_in_bytes
 3134194304
 314
 315We can check the usage:
 316# cat /sys/fs/cgroup/memory/0/memory.usage_in_bytes
 3171216512
 318
 319A successful write to this file does not guarantee a successful set of
 320this limit to the value written into the file. This can be due to a
 321number of factors, such as rounding up to page boundaries or the total
 322availability of memory on the system. The user is required to re-read
 323this file after a write to guarantee the value committed by the kernel.
 324
 325# echo 1 > memory.limit_in_bytes
 326# cat memory.limit_in_bytes
 3274096
 328
 329The memory.failcnt field gives the number of times that the cgroup limit was
 330exceeded.
 331
 332The memory.stat file gives accounting information. Now, the number of
 333caches, RSS and Active pages/Inactive pages are shown.
 334
 3354. Testing
 336
 337For testing features and implementation, see memcg_test.txt.
 338
 339Performance test is also important. To see pure memory controller's overhead,
 340testing on tmpfs will give you good numbers of small overheads.
 341Example: do kernel make on tmpfs.
 342
 343Page-fault scalability is also important. At measuring parallel
 344page fault test, multi-process test may be better than multi-thread
 345test because it has noise of shared objects/status.
 346
 347But the above two are testing extreme situations.
 348Trying usual test under memory controller is always helpful.
 349
 3504.1 Troubleshooting
 351
 352Sometimes a user might find that the application under a cgroup is
 353terminated by OOM killer. There are several causes for this:
 354
 3551. The cgroup limit is too low (just too low to do anything useful)
 3562. The user is using anonymous memory and swap is turned off or too low
 357
 358A sync followed by echo 1 > /proc/sys/vm/drop_caches will help get rid of
 359some of the pages cached in the cgroup (page cache pages).
 360
 361To know what happens, disable OOM_Kill by 10. OOM Control(see below) and
 362seeing what happens will be helpful.
 363
 3644.2 Task migration
 365
 366When a task migrates from one cgroup to another, its charge is not
 367carried forward by default. The pages allocated from the original cgroup still
 368remain charged to it, the charge is dropped when the page is freed or
 369reclaimed.
 370
 371You can move charges of a task along with task migration.
 372See 8. "Move charges at task migration"
 373
 3744.3 Removing a cgroup
 375
 376A cgroup can be removed by rmdir, but as discussed in sections 4.1 and 4.2, a
 377cgroup might have some charge associated with it, even though all
 378tasks have migrated away from it. (because we charge against pages, not
 379against tasks.)
 380
 381We move the stats to root (if use_hierarchy==0) or parent (if
 382use_hierarchy==1), and no change on the charge except uncharging
 383from the child.
 384
 385Charges recorded in swap information is not updated at removal of cgroup.
 386Recorded information is discarded and a cgroup which uses swap (swapcache)
 387will be charged as a new owner of it.
 388
 389About use_hierarchy, see Section 6.
 390
 3915. Misc. interfaces.
 392
 3935.1 force_empty
 394  memory.force_empty interface is provided to make cgroup's memory usage empty.
 395  You can use this interface only when the cgroup has no tasks.
 396  When writing anything to this
 397
 398  # echo 0 > memory.force_empty
 399
 400  Almost all pages tracked by this memory cgroup will be unmapped and freed.
 401  Some pages cannot be freed because they are locked or in-use. Such pages are
 402  moved to parent(if use_hierarchy==1) or root (if use_hierarchy==0) and this
 403  cgroup will be empty.
 404
 405  Typical use case of this interface is that calling this before rmdir().
 406  Because rmdir() moves all pages to parent, some out-of-use page caches can be
 407  moved to the parent. If you want to avoid that, force_empty will be useful.
 408
 409  About use_hierarchy, see Section 6.
 410
 4115.2 stat file
 412
 413memory.stat file includes following statistics
 414
 415# per-memory cgroup local status
 416cache           - # of bytes of page cache memory.
 417rss             - # of bytes of anonymous and swap cache memory.
 418mapped_file     - # of bytes of mapped file (includes tmpfs/shmem)
 419pgpgin          - # of charging events to the memory cgroup. The charging
 420                event happens each time a page is accounted as either mapped
 421                anon page(RSS) or cache page(Page Cache) to the cgroup.
 422pgpgout         - # of uncharging events to the memory cgroup. The uncharging
 423                event happens each time a page is unaccounted from the cgroup.
 424swap            - # of bytes of swap usage
 425inactive_anon   - # of bytes of anonymous memory and swap cache memory on
 426                LRU list.
 427active_anon     - # of bytes of anonymous and swap cache memory on active
 428                inactive LRU list.
 429inactive_file   - # of bytes of file-backed memory on inactive LRU list.
 430active_file     - # of bytes of file-backed memory on active LRU list.
 431unevictable     - # of bytes of memory that cannot be reclaimed (mlocked etc).
 432
 433# status considering hierarchy (see memory.use_hierarchy settings)
 434
 435hierarchical_memory_limit - # of bytes of memory limit with regard to hierarchy
 436                        under which the memory cgroup is
 437hierarchical_memsw_limit - # of bytes of memory+swap limit with regard to
 438                        hierarchy under which memory cgroup is.
 439
 440total_<counter>         - # hierarchical version of <counter>, which in
 441                        addition to the cgroup's own value includes the
 442                        sum of all hierarchical children's values of
 443                        <counter>, i.e. total_cache
 444
 445# The following additional stats are dependent on CONFIG_DEBUG_VM.
 446
 447recent_rotated_anon     - VM internal parameter. (see mm/vmscan.c)
 448recent_rotated_file     - VM internal parameter. (see mm/vmscan.c)
 449recent_scanned_anon     - VM internal parameter. (see mm/vmscan.c)
 450recent_scanned_file     - VM internal parameter. (see mm/vmscan.c)
 451
 452Memo:
 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.
 456
 457Note:
 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
 464         cache.)
 465
 4665.3 swappiness
 467
 468Similar to /proc/sys/vm/swappiness, but affecting a hierarchy of groups only.
 469Please note that unlike the global swappiness, memcg knob set to 0
 470really prevents from any swapping even if there is a swap storage
 471available. This might lead to memcg OOM killer if there are no file
 472pages to reclaim.
 473
 474Following cgroups' swappiness can't be changed.
 475- root cgroup (uses /proc/sys/vm/swappiness).
 476- a cgroup which uses hierarchy and it has other cgroup(s) below it.
 477- a cgroup which uses hierarchy and not the root of hierarchy.
 478
 4795.4 failcnt
 480
 481A memory cgroup provides memory.failcnt and memory.memsw.failcnt files.
 482This failcnt(== failure count) shows the number of times that a usage counter
 483hit its limit. When a memory cgroup hits a limit, failcnt increases and
 484memory under it will be reclaimed.
 485
 486You can reset failcnt by writing 0 to failcnt file.
 487# echo 0 > .../memory.failcnt
 488
 4895.5 usage_in_bytes
 490
 491For efficiency, as other kernel components, memory cgroup uses some optimization
 492to avoid unnecessary cacheline false sharing. usage_in_bytes is affected by the
 493method and doesn't show 'exact' value of memory(and swap) usage, it's an fuzz
 494value for efficient access. (Of course, when necessary, it's synchronized.)
 495If you want to know more exact memory usage, you should use RSS+CACHE(+SWAP)
 496value in memory.stat(see 5.2).
 497
 4985.6 numa_stat
 499
 500This is similar to numa_maps but operates on a per-memcg basis.  This is
 501useful for providing visibility into the numa locality information within
 502an memcg since the pages are allowed to be allocated from any physical
 503node.  One of the usecases is evaluating application performance by
 504combining this information with the application's cpu allocation.
 505
 506We export "total", "file", "anon" and "unevictable" pages per-node for
 507each memcg.  The ouput format of memory.numa_stat is:
 508
 509total=<total pages> N0=<node 0 pages> N1=<node 1 pages> ...
 510file=<total file pages> N0=<node 0 pages> N1=<node 1 pages> ...
 511anon=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
 512unevictable=<total anon pages> N0=<node 0 pages> N1=<node 1 pages> ...
 513
 514And we have total = file + anon + unevictable.
 515
 5166. Hierarchy support
 517
 518The memory controller supports a deep hierarchy and hierarchical accounting.
 519The hierarchy is created by creating the appropriate cgroups in the
 520cgroup filesystem. Consider for example, the following cgroup filesystem
 521hierarchy
 522
 523               root
 524             /  |   \
 525            /   |    \
 526           a    b     c
 527                      | \
 528                      |  \
 529                      d   e
 530
 531In the diagram above, with hierarchical accounting enabled, all memory
 532usage of e, is accounted to its ancestors up until the root (i.e, c and root),
 533that has memory.use_hierarchy enabled. If one of the ancestors goes over its
 534limit, the reclaim algorithm reclaims from the tasks in the ancestor and the
 535children of the ancestor.
 536
 5376.1 Enabling hierarchical accounting and reclaim
 538
 539A memory cgroup by default disables the hierarchy feature. Support
 540can be enabled by writing 1 to memory.use_hierarchy file of the root cgroup
 541
 542# echo 1 > memory.use_hierarchy
 543
 544The feature can be disabled by
 545
 546# echo 0 > memory.use_hierarchy
 547
 548NOTE1: Enabling/disabling will fail if either the cgroup already has other
 549       cgroups created below it, or if the parent cgroup has use_hierarchy
 550       enabled.
 551
 552NOTE2: When panic_on_oom is set to "2", the whole system will panic in
 553       case of an OOM event in any cgroup.
 554
 5557. Soft limits
 556
 557Soft limits allow for greater sharing of memory. The idea behind soft limits
 558is to allow control groups to use as much of the memory as needed, provided
 559
 560a. There is no memory contention
 561b. They do not exceed their hard limit
 562
 563When the system detects memory contention or low memory, control groups
 564are pushed back to their soft limits. If the soft limit of each control
 565group is very high, they are pushed back as much as possible to make
 566sure that one control group does not starve the others of memory.
 567
 568Please note that soft limits is a best effort feature, it comes with
 569no guarantees, but it does its best to make sure that when memory is
 570heavily contended for, memory is allocated based on the soft limit
 571hints/setup. Currently soft limit based reclaim is setup such that
 572it gets invoked from balance_pgdat (kswapd).
 573
 5747.1 Interface
 575
 576Soft limits can be setup by using the following commands (in this example we
 577assume a soft limit of 256 MiB)
 578
 579# echo 256M > memory.soft_limit_in_bytes
 580
 581If we want to change this to 1G, we can at any time use
 582
 583# echo 1G > memory.soft_limit_in_bytes
 584
 585NOTE1: Soft limits take effect over a long period of time, since they involve
 586       reclaiming memory for balancing between memory cgroups
 587NOTE2: It is recommended to set the soft limit always below the hard limit,
 588       otherwise the hard limit will take precedence.
 589
 5908. Move charges at task migration
 591
 592Users can move charges associated with a task along with task migration, that
 593is, uncharge task's pages from the old cgroup and charge them to the new cgroup.
 594This feature is not supported in !CONFIG_MMU environments because of lack of
 595page tables.
 596
 5978.1 Interface
 598
 599This feature is disabled by default. It can be enabled(and disabled again) by
 600writing to memory.move_charge_at_immigrate of the destination cgroup.
 601
 602If you want to enable it:
 603
 604# echo (some positive value) > memory.move_charge_at_immigrate
 605
 606Note: Each bits of move_charge_at_immigrate has its own meaning about what type
 607      of charges should be moved. See 8.2 for details.
 608Note: Charges are moved only when you move mm->owner, IOW, a leader of a thread
 609      group.
 610Note: If we cannot find enough space for the task in the destination cgroup, we
 611      try to make space by reclaiming memory. Task migration may fail if we
 612      cannot make enough space.
 613Note: It can take several seconds if you move charges much.
 614
 615And if you want disable it again:
 616
 617# echo 0 > memory.move_charge_at_immigrate
 618
 6198.2 Type of charges which can be move
 620
 621Each bits of move_charge_at_immigrate has its own meaning about what type of
 622charges should be moved. But in any cases, it must be noted that an account of
 623a page or a swap can be moved only when it is charged to the task's current(old)
 624memory cgroup.
 625
 626  bit | what type of charges would be moved ?
 627 -----+------------------------------------------------------------------------
 628   0  | A charge of an anonymous page(or swap of it) used by the target task.
 629      | You must enable Swap Extension(see 2.4) to enable move of swap charges.
 630 -----+------------------------------------------------------------------------
 631   1  | A charge of file pages(normal file, tmpfs file(e.g. ipc shared memory)
 632      | and swaps of tmpfs file) mmapped by the target task. Unlike the case of
 633      | anonymous pages, file pages(and swaps) in the range mmapped by the task
 634      | will be moved even if the task hasn't done page fault, i.e. they might
 635      | not be the task's "RSS", but other task's "RSS" that maps the same file.
 636      | And mapcount of the page is ignored(the page can be moved even if
 637      | page_mapcount(page) > 1). You must enable Swap Extension(see 2.4) to
 638      | enable move of swap charges.
 639
 6408.3 TODO
 641
 642- All of moving charge operations are done under cgroup_mutex. It's not good
 643  behavior to hold the mutex too long, so we may need some trick.
 644
 6459. Memory thresholds
 646
 647Memory cgroup implements memory thresholds using cgroups notification
 648API (see cgroups.txt). It allows to register multiple memory and memsw
 649thresholds and gets notifications when it crosses.
 650
 651To register a threshold application need:
 652- create an eventfd using eventfd(2);
 653- open memory.usage_in_bytes or memory.memsw.usage_in_bytes;
 654- write string like "<event_fd> <fd of memory.usage_in_bytes> <threshold>" to
 655  cgroup.event_control.
 656
 657Application will be notified through eventfd when memory usage crosses
 658threshold in any direction.
 659
 660It's applicable for root and non-root cgroup.
 661
 66210. OOM Control
 663
 664memory.oom_control file is for OOM notification and other controls.
 665
 666Memory cgroup implements OOM notifier using cgroup notification
 667API (See cgroups.txt). It allows to register multiple OOM notification
 668delivery and gets notification when OOM happens.
 669
 670To register a notifier, application need:
 671 - create an eventfd using eventfd(2)
 672 - open memory.oom_control file
 673 - write string like "<event_fd> <fd of memory.oom_control>" to
 674   cgroup.event_control
 675
 676Application will be notified through eventfd when OOM happens.
 677OOM notification doesn't work for root cgroup.
 678
 679You can disable OOM-killer by writing "1" to memory.oom_control file, as:
 680
 681        #echo 1 > memory.oom_control
 682
 683This operation is only allowed to the top cgroup of sub-hierarchy.
 684If OOM-killer is disabled, tasks under cgroup will hang/sleep
 685in memory cgroup's OOM-waitqueue when they request accountable memory.
 686
 687For running them, you have to relax the memory cgroup's OOM status by
 688        * enlarge limit or reduce usage.
 689To reduce usage,
 690        * kill some tasks.
 691        * move some tasks to other group with account migration.
 692        * remove some files (on tmpfs?)
 693
 694Then, stopped tasks will work again.
 695
 696At reading, current status of OOM is shown.
 697        oom_kill_disable 0 or 1 (if 1, oom-killer is disabled)
 698        under_oom        0 or 1 (if 1, the memory cgroup is under OOM, tasks may
 699                                 be stopped.)
 700
 70111. TODO
 702
 7031. Add support for accounting huge pages (as a separate controller)
 7042. Make per-cgroup scanner reclaim not-shared pages first
 7053. Teach controller to account for shared-pages
 7064. Start reclamation in the background when the limit is
 707   not yet hit but the usage is getting closer
 708
 709Summary
 710
 711Overall, the memory controller has been a stable controller and has been
 712commented and discussed quite extensively in the community.
 713
 714References
 715
 7161. Singh, Balbir. RFC: Memory Controller, http://lwn.net/Articles/206697/
 7172. Singh, Balbir. Memory Controller (RSS Control),
 718   http://lwn.net/Articles/222762/
 7193. Emelianov, Pavel. Resource controllers based on process cgroups
 720   http://lkml.org/lkml/2007/3/6/198
 7214. Emelianov, Pavel. RSS controller based on process cgroups (v2)
 722   http://lkml.org/lkml/2007/4/9/78
 7235. Emelianov, Pavel. RSS controller based on process cgroups (v3)
 724   http://lkml.org/lkml/2007/5/30/244
 7256. Menage, Paul. Control Groups v10, http://lwn.net/Articles/236032/
 7267. Vaidyanathan, Srinivasan, Control Groups: Pagecache accounting and control
 727   subsystem (v3), http://lwn.net/Articles/235534/
 7288. Singh, Balbir. RSS controller v2 test results (lmbench),
 729   http://lkml.org/lkml/2007/5/17/232
 7309. Singh, Balbir. RSS controller v2 AIM9 results
 731   http://lkml.org/lkml/2007/5/18/1
 73210. Singh, Balbir. Memory controller v6 test results,
 733    http://lkml.org/lkml/2007/8/19/36
 73411. Singh, Balbir. Memory controller introduction (v6),
 735    http://lkml.org/lkml/2007/8/17/69
 73612. Corbet, Jonathan, Controlling memory use in cgroups,
 737    http://lwn.net/Articles/243795/
 738
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.