1.. _page_owner:
2
3==================================================
4page owner: Tracking about who allocated each page
5==================================================
6
7Introduction
8============
9
10page owner is for the tracking about who allocated each page.
11It can be used to debug memory leak or to find a memory hogger.
12When allocation happens, information about allocation such as call stack
13and order of pages is stored into certain storage for each page.
14When we need to know about status of all pages, we can get and analyze
15this information.
16
17Although we already have tracepoint for tracing page allocation/free,
18using it for analyzing who allocate each page is rather complex. We need
19to enlarge the trace buffer for preventing overlapping until userspace
20program launched. And, launched program continually dump out the trace
21buffer for later analysis and it would change system behaviour with more
22possibility rather than just keeping it in memory, so bad for debugging.
23
24page owner can also be used for various purposes. For example, accurate
25fragmentation statistics can be obtained through gfp flag information of
26each page. It is already implemented and activated if page owner is
27enabled. Other usages are more than welcome.
28
29page owner is disabled in default. So, if you'd like to use it, you need
30to add "page_owner=on" into your boot cmdline. If the kernel is built
31with page owner and page owner is disabled in runtime due to no enabling
32boot option, runtime overhead is marginal. If disabled in runtime, it
33doesn't require memory to store owner information, so there is no runtime
34memory overhead. And, page owner inserts just two unlikely branches into
35the page allocator hotpath and if not enabled, then allocation is done
36like as the kernel without page owner. These two unlikely branches should
37not affect to allocation performance, especially if the static keys jump
38label patching functionality is available. Following is the kernel's code
39size change due to this facility.
40
41- Without page owner::
42
43   text    data     bss     dec     hex filename
44   48392   2333     644   51369    c8a9 mm/page_alloc.o
45
46- With page owner::
47
48   text    data     bss     dec     hex filename
49   48800   2445     644   51889    cab1 mm/page_alloc.o
50   6662     108      29    6799    1a8f mm/page_owner.o
51   1025       8       8    1041     411 mm/page_ext.o
52
53Although, roughly, 8 KB code is added in total, page_alloc.o increase by
54520 bytes and less than half of it is in hotpath. Building the kernel with
55page owner and turning it on if needed would be great option to debug
56kernel memory problem.
57
58There is one notice that is caused by implementation detail. page owner
59stores information into the memory from struct page extension. This memory
60is initialized some time later than that page allocator starts in sparse
61memory system, so, until initialization, many pages can be allocated and
62they would have no owner information. To fix it up, these early allocated
63pages are investigated and marked as allocated in initialization phase.
64Although it doesn't mean that they have the right owner information,
65at least, we can tell whether the page is allocated or not,
66more accurately. On 2GB memory x86-64 VM box, 13343 early allocated pages
67are catched and marked, although they are mostly allocated from struct
68page extension feature. Anyway, after that, no page is left in
69un-tracking state.
70
71Usage
72=====
73
741) Build user-space helper::
75
76	cd tools/vm
77	make page_owner_sort
78
792) Enable page owner: add "page_owner=on" to boot cmdline.
80
813) Do the job what you want to debug
82
834) Analyze information from page owner::
84
85	cat /sys/kernel/debug/page_owner > page_owner_full.txt
86	./page_owner_sort page_owner_full.txt sorted_page_owner.txt
87
88   The general output of ``page_owner_full.txt`` is as follows:
89
90	Page allocated via order XXX, ...
91	PFN XXX ...
92	 // Detailed stack
93
94	Page allocated via order XXX, ...
95	PFN XXX ...
96	 // Detailed stack
97
98   The ``page_owner_sort`` tool ignores ``PFN`` rows, puts the remaining rows
99   in buf, uses regexp to extract the page order value, counts the times
100   and pages of buf, and finally sorts them according to the times.
101
102   See the result about who allocated each page
103   in the ``sorted_page_owner.txt``. General output:
104
105	XXX times, XXX pages:
106	Page allocated via order XXX, ...
107	 // Detailed stack
108
109   By default, ``page_owner_sort`` is sorted according to the times of buf.
110   If you want to sort by the pages nums of buf, use the ``-m`` parameter.
111