3 =============================
4 Examining Process Page Tables
5 =============================
7 pagemap is a new (as of 2.6.25) set of interfaces in the kernel that allow
8 userspace programs to examine the page tables and related information by
9 reading files in ``/proc``.
11 There are four components to pagemap:
13 * ``/proc/pid/pagemap``. This file lets a userspace process find out which
14 physical frame each virtual page is mapped to. It contains one 64-bit
15 value for each virtual page, containing the following data (from
16 ``fs/proc/task_mmu.c``, above pagemap_read):
18 * Bits 0-54 page frame number (PFN) if present
19 * Bits 0-4 swap type if swapped
20 * Bits 5-54 swap offset if swapped
21 * Bit 55 pte is soft-dirty (see
22 :ref:`Documentation/admin-guide/mm/soft-dirty.rst <soft_dirty>`)
23 * Bit 56 page exclusively mapped (since 4.2)
24 * Bit 57 pte is uffd-wp write-protected (since 5.13) (see
25 :ref:`Documentation/admin-guide/mm/userfaultfd.rst <userfaultfd>`)
27 * Bit 61 page is file-page or shared-anon (since 3.5)
31 Since Linux 4.0 only users with the CAP_SYS_ADMIN capability can get PFNs.
32 In 4.0 and 4.1 opens by unprivileged fail with -EPERM. Starting from
33 4.2 the PFN field is zeroed if the user does not have CAP_SYS_ADMIN.
34 Reason: information about PFNs helps in exploiting Rowhammer vulnerability.
36 If the page is not present but in swap, then the PFN contains an
37 encoding of the swap file number and the page's offset into the
38 swap. Unmapped pages return a null PFN. This allows determining
39 precisely which pages are mapped (or in swap) and comparing mapped
40 pages between processes.
42 Efficient users of this interface will use ``/proc/pid/maps`` to
43 determine which areas of memory are actually mapped and llseek to
44 skip over unmapped regions.
46 * ``/proc/kpagecount``. This file contains a 64-bit count of the number of
47 times each page is mapped, indexed by PFN.
49 The page-types tool in the tools/vm directory can be used to query the
50 number of times a page is mapped.
52 * ``/proc/kpageflags``. This file contains a 64-bit set of flags for each
55 The flags are (from ``fs/proc/page.c``, above kpageflags_read):
85 * ``/proc/kpagecgroup``. This file contains a 64-bit inode number of the
86 memory cgroup each page is charged to, indexed by PFN. Only available when
89 Short descriptions to the page flags
90 ====================================
93 The page is being locked for exclusive access, e.g. by undergoing read/write
96 The page is managed by the SLAB/SLOB/SLUB/SLQB kernel memory allocator.
97 When compound page is used, SLUB/SLQB will only set this flag on the head
98 page; SLOB will not flag it at all.
100 A free memory block managed by the buddy system allocator.
101 The buddy system organizes free memory in blocks of various orders.
102 An order N block has 2^N physically contiguous pages, with the BUDDY flag
103 set for and _only_ for the first page.
105 A compound page with order N consists of 2^N physically contiguous pages.
106 A compound page with order 2 takes the form of "HTTT", where H donates its
107 head page and T donates its tail page(s). The major consumers of compound
108 pages are hugeTLB pages
109 (:ref:`Documentation/admin-guide/mm/hugetlbpage.rst <hugetlbpage>`),
110 the SLUB etc. memory allocators and various device drivers.
111 However in this interface, only huge/giga pages are made visible
114 A compound page tail (see description above).
116 This is an integral part of a HugeTLB page.
118 Hardware detected memory corruption on this page: don't touch the data!
120 No page frame exists at the requested address.
122 Identical memory pages dynamically shared between one or more processes.
124 Contiguous pages which construct transparent hugepages.
126 The page is logically offline.
128 Zero page for pfn_zero or huge_zero page.
130 The page has not been accessed since it was marked idle (see
131 :ref:`Documentation/admin-guide/mm/idle_page_tracking.rst <idle_page_tracking>`).
132 Note that this flag may be stale in case the page was accessed via
133 a PTE. To make sure the flag is up-to-date one has to read
134 ``/sys/kernel/mm/page_idle/bitmap`` first.
136 The page is in use as a page table.
138 IO related page flags
139 ---------------------
144 The page has up-to-date data.
145 ie. for file backed page: (in-memory data revision >= on-disk one)
147 The page has been written to, hence contains new data.
148 i.e. for file backed page: (in-memory data revision > on-disk one)
150 The page is being synced to disk.
152 LRU related page flags
153 ----------------------
156 The page is in one of the LRU lists.
158 The page is in the active LRU list.
160 The page is in the unevictable (non-)LRU list It is somehow pinned and
161 not a candidate for LRU page reclaims, e.g. ramfs pages,
162 shmctl(SHM_LOCK) and mlock() memory segments.
164 The page has been referenced since last LRU list enqueue/requeue.
166 The page will be reclaimed soon after its pageout IO completed.
168 A memory mapped page.
170 A memory mapped page that is not part of a file.
172 The page is mapped to swap space, i.e. has an associated swap entry.
174 The page is backed by swap/RAM.
176 The page-types tool in the tools/vm directory can be used to query the
179 Using pagemap to do something useful
180 ====================================
182 The general procedure for using pagemap to find out about a process' memory
183 usage goes like this:
185 1. Read ``/proc/pid/maps`` to determine which parts of the memory space are
187 2. Select the maps you are interested in -- all of them, or a particular
188 library, or the stack or the heap, etc.
189 3. Open ``/proc/pid/pagemap`` and seek to the pages you would like to examine.
190 4. Read a u64 for each page from pagemap.
191 5. Open ``/proc/kpagecount`` and/or ``/proc/kpageflags``. For each PFN you
192 just read, seek to that entry in the file, and read the data you want.
194 For example, to find the "unique set size" (USS), which is the amount of
195 memory that a process is using that is not shared with any other process,
196 you can go through every map in the process, find the PFNs, look those up
197 in kpagecount, and tally up the number of pages that are only referenced
200 Exceptions for Shared Memory
201 ============================
203 Page table entries for shared pages are cleared when the pages are zapped or
204 swapped out. This makes swapped out pages indistinguishable from never-allocated
207 In kernel space, the swap location can still be retrieved from the page cache.
208 However, values stored only on the normal PTE get lost irretrievably when the
209 page is swapped out (i.e. SOFT_DIRTY).
211 In user space, whether the page is present, swapped or none can be deduced with
212 the help of lseek and/or mincore system calls.
214 lseek() can differentiate between accessed pages (present or swapped out) and
215 holes (none/non-allocated) by specifying the SEEK_DATA flag on the file where
216 the pages are backed. For anonymous shared pages, the file can be found in
217 ``/proc/pid/map_files/``.
219 mincore() can differentiate between pages in memory (present, including swap
220 cache) and out of memory (swapped out or none/non-allocated).
225 Reading from any of the files will return -EINVAL if you are not starting
226 the read on an 8-byte boundary (e.g., if you sought an odd number of bytes
227 into the file), or if the size of the read is not a multiple of 8 bytes.
229 Before Linux 3.11 pagemap bits 55-60 were used for "page-shift" (which is
230 always 12 at most architectures). Since Linux 3.11 their meaning changes
231 after first clear of soft-dirty bits. Since Linux 4.2 they are used for
232 flags unconditionally.