Documentation/vm/idle_page_tracking.txt

   1 .. _idle_page_tracking:
   2
   3 ==================
   4 Idle Page Tracking
   5 ==================
   6
   7 Motivation
   8 ==========
   9
  10 The idle page tracking feature allows to track which memory pages are being
  11 accessed by a workload and which are idle. This information can be useful for
  12 estimating the workload's working set size, which, in turn, can be taken into
  13 account when configuring the workload parameters, setting memory cgroup limits,
  14 or deciding where to place the workload within a compute cluster.
  15
  16 It is enabled by CONFIG_IDLE_PAGE_TRACKING=y.
  17
  18 .. _user_api:
  19
  20 User API
  21 ========
  22
  23 The idle page tracking API is located at ``/sys/kernel/mm/page_idle``.
  24 Currently, it consists of the only read-write file,
  25 ``/sys/kernel/mm/page_idle/bitmap``.
  26
  27 The file implements a bitmap where each bit corresponds to a memory page. The
  28 bitmap is represented by an array of 8-byte integers, and the page at PFN #i is
  29 mapped to bit #i%64 of array element #i/64, byte order is native. When a bit is
  30 set, the corresponding page is idle.
  31
  32 A page is considered idle if it has not been accessed since it was marked idle
  33 (for more details on what "accessed" actually means see the :ref:`Implementation
  34 Details <impl_details>` section).
  35 To mark a page idle one has to set the bit corresponding to
  36 the page by writing to the file. A value written to the file is OR-ed with the
  37 current bitmap value.
  38
  39 Only accesses to user memory pages are tracked. These are pages mapped to a
  40 process address space, page cache and buffer pages, swap cache pages. For other
  41 page types (e.g. SLAB pages) an attempt to mark a page idle is silently ignored,
  42 and hence such pages are never reported idle.
  43
  44 For huge pages the idle flag is set only on the head page, so one has to read
  45 ``/proc/kpageflags`` in order to correctly count idle huge pages.
  46
  47 Reading from or writing to ``/sys/kernel/mm/page_idle/bitmap`` will return
  48 -EINVAL if you are not starting the read/write on an 8-byte boundary, or
  49 if the size of the read/write is not a multiple of 8 bytes. Writing to
  50 this file beyond max PFN will return -ENXIO.
  51
  52 That said, in order to estimate the amount of pages that are not used by a
  53 workload one should:
  54
  55  1. Mark all the workload's pages as idle by setting corresponding bits in
  56     ``/sys/kernel/mm/page_idle/bitmap``. The pages can be found by reading
  57     ``/proc/pid/pagemap`` if the workload is represented by a process, or by
  58     filtering out alien pages using ``/proc/kpagecgroup`` in case the workload
  59     is placed in a memory cgroup.
  60
  61  2. Wait until the workload accesses its working set.
  62
  63  3. Read ``/sys/kernel/mm/page_idle/bitmap`` and count the number of bits set.
  64     If one wants to ignore certain types of pages, e.g. mlocked pages since they
  65     are not reclaimable, he or she can filter them out using
  66     ``/proc/kpageflags``.
  67
  68 See Documentation/vm/pagemap.txt for more information about
  69 ``/proc/pid/pagemap``, ``/proc/kpageflags``, and ``/proc/kpagecgroup``.
  70
  71 .. _impl_details:
  72
  73 Implementation Details
  74 ======================
  75
  76 The kernel internally keeps track of accesses to user memory pages in order to
  77 reclaim unreferenced pages first on memory shortage conditions. A page is
  78 considered referenced if it has been recently accessed via a process address
  79 space, in which case one or more PTEs it is mapped to will have the Accessed bit
  80 set, or marked accessed explicitly by the kernel (see mark_page_accessed()). The
  81 latter happens when:
  82
  83  - a userspace process reads or writes a page using a system call (e.g. read(2)
  84    or write(2))
  85
  86  - a page that is used for storing filesystem buffers is read or written,
  87    because a process needs filesystem metadata stored in it (e.g. lists a
  88    directory tree)
  89
  90  - a page is accessed by a device driver using get_user_pages()
  91
  92 When a dirty page is written to swap or disk as a result of memory reclaim or
  93 exceeding the dirty memory limit, it is not marked referenced.
  94
  95 The idle memory tracking feature adds a new page flag, the Idle flag. This flag
  96 is set manually, by writing to ``/sys/kernel/mm/page_idle/bitmap`` (see the
  97 :ref:`User API <user_api>`
  98 section), and cleared automatically whenever a page is referenced as defined
  99 above.
 100
 101 When a page is marked idle, the Accessed bit must be cleared in all PTEs it is
 102 mapped to, otherwise we will not be able to detect accesses to the page coming
 103 from a process address space. To avoid interference with the reclaimer, which,
 104 as noted above, uses the Accessed bit to promote actively referenced pages, one
 105 more page flag is introduced, the Young flag. When the PTE Accessed bit is
 106 cleared as a result of setting or updating a page's Idle flag, the Young flag
 107 is set on the page. The reclaimer treats the Young flag as an extra PTE
 108 Accessed bit and therefore will consider such a page as referenced.
 109
 110 Since the idle memory tracking feature is based on the memory reclaimer logic,
 111 it only works with pages that are on an LRU list, other pages are silently
 112 ignored. That means it will ignore a user memory page if it is isolated, but
 113 since there are usually not many of them, it should not affect the overall
 114 result noticeably. In order not to stall scanning of the idle page bitmap,
 115 locked pages may be skipped too.