Documentation/vm/pagemap.txt

   1 pagemap, from the userspace perspective
   2 ---------------------------------------
   3
   4 pagemap is a new (as of 2.6.25) set of interfaces in the kernel that allow
   5 userspace programs to examine the page tables and related information by
   6 reading files in /proc.
   7
   8 There are four components to pagemap:
   9
  10  * /proc/pid/pagemap.  This file lets a userspace process find out which
  11    physical frame each virtual page is mapped to.  It contains one 64-bit
  12    value for each virtual page, containing the following data (from
  13    fs/proc/task_mmu.c, above pagemap_read):
  14
  15     * Bits 0-54  page frame number (PFN) if present
  16     * Bits 0-4   swap type if swapped
  17     * Bits 5-54  swap offset if swapped
  18     * Bit  55    pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
  19     * Bit  56    page exclusively mapped (since 4.2)
  20     * Bits 57-60 zero
  21     * Bit  61    page is file-page or shared-anon (since 3.5)
  22     * Bit  62    page swapped
  23     * Bit  63    page present
  24
  25    Since Linux 4.0 only users with the CAP_SYS_ADMIN capability can get PFNs.
  26    In 4.0 and 4.1 opens by unprivileged fail with -EPERM.  Starting from
  27    4.2 the PFN field is zeroed if the user does not have CAP_SYS_ADMIN.
  28    Reason: information about PFNs helps in exploiting Rowhammer vulnerability.
  29
  30    If the page is not present but in swap, then the PFN contains an
  31    encoding of the swap file number and the page's offset into the
  32    swap. Unmapped pages return a null PFN. This allows determining
  33    precisely which pages are mapped (or in swap) and comparing mapped
  34    pages between processes.
  35
  36    Efficient users of this interface will use /proc/pid/maps to
  37    determine which areas of memory are actually mapped and llseek to
  38    skip over unmapped regions.
  39
  40  * /proc/kpagecount.  This file contains a 64-bit count of the number of
  41    times each page is mapped, indexed by PFN.
  42
  43  * /proc/kpageflags.  This file contains a 64-bit set of flags for each
  44    page, indexed by PFN.
  45
  46    The flags are (from fs/proc/page.c, above kpageflags_read):
  47
  48      0. LOCKED
  49      1. ERROR
  50      2. REFERENCED
  51      3. UPTODATE
  52      4. DIRTY
  53      5. LRU
  54      6. ACTIVE
  55      7. SLAB
  56      8. WRITEBACK
  57      9. RECLAIM
  58     10. BUDDY
  59     11. MMAP
  60     12. ANON
  61     13. SWAPCACHE
  62     14. SWAPBACKED
  63     15. COMPOUND_HEAD
  64     16. COMPOUND_TAIL
  65     16. HUGE
  66     18. UNEVICTABLE
  67     19. HWPOISON
  68     20. NOPAGE
  69     21. KSM
  70     22. THP
  71     23. BALLOON
  72     24. ZERO_PAGE
  73     25. IDLE
  74
  75  * /proc/kpagecgroup.  This file contains a 64-bit inode number of the
  76    memory cgroup each page is charged to, indexed by PFN. Only available when
  77    CONFIG_MEMCG is set.
  78
  79 Short descriptions to the page flags:
  80
  81  0. LOCKED
  82     page is being locked for exclusive access, eg. by undergoing read/write IO
  83
  84  7. SLAB
  85     page is managed by the SLAB/SLOB/SLUB/SLQB kernel memory allocator
  86     When compound page is used, SLUB/SLQB will only set this flag on the head
  87     page; SLOB will not flag it at all.
  88
  89 10. BUDDY
  90     a free memory block managed by the buddy system allocator
  91     The buddy system organizes free memory in blocks of various orders.
  92     An order N block has 2^N physically contiguous pages, with the BUDDY flag
  93     set for and _only_ for the first page.
  94
  95 15. COMPOUND_HEAD
  96 16. COMPOUND_TAIL
  97     A compound page with order N consists of 2^N physically contiguous pages.
  98     A compound page with order 2 takes the form of "HTTT", where H donates its
  99     head page and T donates its tail page(s).  The major consumers of compound
 100     pages are hugeTLB pages (Documentation/vm/hugetlbpage.txt), the SLUB etc.
 101     memory allocators and various device drivers. However in this interface,
 102     only huge/giga pages are made visible to end users.
 103 17. HUGE
 104     this is an integral part of a HugeTLB page
 105
 106 19. HWPOISON
 107     hardware detected memory corruption on this page: don't touch the data!
 108
 109 20. NOPAGE
 110     no page frame exists at the requested address
 111
 112 21. KSM
 113     identical memory pages dynamically shared between one or more processes
 114
 115 22. THP
 116     contiguous pages which construct transparent hugepages
 117
 118 23. BALLOON
 119     balloon compaction page
 120
 121 24. ZERO_PAGE
 122     zero page for pfn_zero or huge_zero page
 123
 124 25. IDLE
 125     page has not been accessed since it was marked idle (see
 126     Documentation/vm/idle_page_tracking.txt). Note that this flag may be
 127     stale in case the page was accessed via a PTE. To make sure the flag
 128     is up-to-date one has to read /sys/kernel/mm/page_idle/bitmap first.
 129
 130     [IO related page flags]
 131  1. ERROR     IO error occurred
 132  3. UPTODATE  page has up-to-date data
 133               ie. for file backed page: (in-memory data revision >= on-disk one)
 134  4. DIRTY     page has been written to, hence contains new data
 135               ie. for file backed page: (in-memory data revision >  on-disk one)
 136  8. WRITEBACK page is being synced to disk
 137
 138     [LRU related page flags]
 139  5. LRU         page is in one of the LRU lists
 140  6. ACTIVE      page is in the active LRU list
 141 18. UNEVICTABLE page is in the unevictable (non-)LRU list
 142                 It is somehow pinned and not a candidate for LRU page reclaims,
 143                 eg. ramfs pages, shmctl(SHM_LOCK) and mlock() memory segments
 144  2. REFERENCED  page has been referenced since last LRU list enqueue/requeue
 145  9. RECLAIM     page will be reclaimed soon after its pageout IO completed
 146 11. MMAP        a memory mapped page
 147 12. ANON        a memory mapped page that is not part of a file
 148 13. SWAPCACHE   page is mapped to swap space, ie. has an associated swap entry
 149 14. SWAPBACKED  page is backed by swap/RAM
 150
 151 The page-types tool in the tools/vm directory can be used to query the
 152 above flags.
 153
 154 Using pagemap to do something useful:
 155
 156 The general procedure for using pagemap to find out about a process' memory
 157 usage goes like this:
 158
 159  1. Read /proc/pid/maps to determine which parts of the memory space are
 160     mapped to what.
 161  2. Select the maps you are interested in -- all of them, or a particular
 162     library, or the stack or the heap, etc.
 163  3. Open /proc/pid/pagemap and seek to the pages you would like to examine.
 164  4. Read a u64 for each page from pagemap.
 165  5. Open /proc/kpagecount and/or /proc/kpageflags.  For each PFN you just
 166     read, seek to that entry in the file, and read the data you want.
 167
 168 For example, to find the "unique set size" (USS), which is the amount of
 169 memory that a process is using that is not shared with any other process,
 170 you can go through every map in the process, find the PFNs, look those up
 171 in kpagecount, and tally up the number of pages that are only referenced
 172 once.
 173
 174 Other notes:
 175
 176 Reading from any of the files will return -EINVAL if you are not starting
 177 the read on an 8-byte boundary (e.g., if you sought an odd number of bytes
 178 into the file), or if the size of the read is not a multiple of 8 bytes.
 179
 180 Before Linux 3.11 pagemap bits 55-60 were used for "page-shift" (which is
 181 always 12 at most architectures). Since Linux 3.11 their meaning changes
 182 after first clear of soft-dirty bits. Since Linux 4.2 they are used for
 183 flags unconditionally.