blob: 0112da3b9ab87f731006993c76e1a7fdc4eddcaa [file] [log] [blame]
Catalin Marinas04f70332009-06-11 13:22:39 +01001Kernel Memory Leak Detector
2===========================
3
4Introduction
5------------
6
7Kmemleak provides a way of detecting possible kernel memory leaks in a
8way similar to a tracing garbage collector
9(http://en.wikipedia.org/wiki/Garbage_collection_%28computer_science%29#Tracing_garbage_collectors),
10with the difference that the orphan objects are not freed but only
11reported via /sys/kernel/debug/kmemleak. A similar method is used by the
12Valgrind tool (memcheck --leak-check) to detect the memory leaks in
13user-space applications.
14
15Usage
16-----
17
18CONFIG_DEBUG_KMEMLEAK in "Kernel hacking" has to be enabled. A kernel
19thread scans the memory every 10 minutes (by default) and prints any new
20unreferenced objects found. To trigger an intermediate scan and display
21all the possible memory leaks:
22
23 # mount -t debugfs nodev /sys/kernel/debug/
24 # cat /sys/kernel/debug/kmemleak
25
26Note that the orphan objects are listed in the order they were allocated
27and one object at the beginning of the list may cause other subsequent
28objects to be reported as orphan.
29
30Memory scanning parameters can be modified at run-time by writing to the
31/sys/kernel/debug/kmemleak file. The following parameters are supported:
32
33 off - disable kmemleak (irreversible)
34 stack=on - enable the task stacks scanning
35 stack=off - disable the tasks stacks scanning
36 scan=on - start the automatic memory scanning thread
37 scan=off - stop the automatic memory scanning thread
38 scan=<secs> - set the automatic memory scanning period in seconds (0
39 to disable it)
40
41Kmemleak can also be disabled at boot-time by passing "kmemleak=off" on
42the kernel command line.
43
44Basic Algorithm
45---------------
46
47The memory allocations via kmalloc, vmalloc, kmem_cache_alloc and
48friends are traced and the pointers, together with additional
49information like size and stack trace, are stored in a prio search tree.
50The corresponding freeing function calls are tracked and the pointers
51removed from the kmemleak data structures.
52
53An allocated block of memory is considered orphan if no pointer to its
54start address or to any location inside the block can be found by
55scanning the memory (including saved registers). This means that there
56might be no way for the kernel to pass the address of the allocated
57block to a freeing function and therefore the block is considered a
58memory leak.
59
60The scanning algorithm steps:
61
62 1. mark all objects as white (remaining white objects will later be
63 considered orphan)
64 2. scan the memory starting with the data section and stacks, checking
65 the values against the addresses stored in the prio search tree. If
66 a pointer to a white object is found, the object is added to the
67 gray list
68 3. scan the gray objects for matching addresses (some white objects
69 can become gray and added at the end of the gray list) until the
70 gray set is finished
71 4. the remaining white objects are considered orphan and reported via
72 /sys/kernel/debug/kmemleak
73
74Some allocated memory blocks have pointers stored in the kernel's
75internal data structures and they cannot be detected as orphans. To
76avoid this, kmemleak can also store the number of values pointing to an
77address inside the block address range that need to be found so that the
78block is not considered a leak. One example is __vmalloc().
79
80Kmemleak API
81------------
82
83See the include/linux/kmemleak.h header for the functions prototype.
84
85kmemleak_init - initialize kmemleak
86kmemleak_alloc - notify of a memory block allocation
87kmemleak_free - notify of a memory block freeing
88kmemleak_not_leak - mark an object as not a leak
89kmemleak_ignore - do not scan or report an object as leak
90kmemleak_scan_area - add scan areas inside a memory block
91kmemleak_no_scan - do not scan a memory block
92kmemleak_erase - erase an old value in a pointer variable
93kmemleak_alloc_recursive - as kmemleak_alloc but checks the recursiveness
94kmemleak_free_recursive - as kmemleak_free but checks the recursiveness
95
96Dealing with false positives/negatives
97--------------------------------------
98
99The false negatives are real memory leaks (orphan objects) but not
100reported by kmemleak because values found during the memory scanning
101point to such objects. To reduce the number of false negatives, kmemleak
102provides the kmemleak_ignore, kmemleak_scan_area, kmemleak_no_scan and
103kmemleak_erase functions (see above). The task stacks also increase the
104amount of false negatives and their scanning is not enabled by default.
105
106The false positives are objects wrongly reported as being memory leaks
107(orphan). For objects known not to be leaks, kmemleak provides the
108kmemleak_not_leak function. The kmemleak_ignore could also be used if
109the memory block is known not to contain other pointers and it will no
110longer be scanned.
111
112Some of the reported leaks are only transient, especially on SMP
113systems, because of pointers temporarily stored in CPU registers or
114stacks. Kmemleak defines MSECS_MIN_AGE (defaulting to 1000) representing
115the minimum age of an object to be reported as a memory leak.
116
117Limitations and Drawbacks
118-------------------------
119
120The main drawback is the reduced performance of memory allocation and
121freeing. To avoid other penalties, the memory scanning is only performed
122when the /sys/kernel/debug/kmemleak file is read. Anyway, this tool is
123intended for debugging purposes where the performance might not be the
124most important requirement.
125
126To keep the algorithm simple, kmemleak scans for values pointing to any
127address inside a block's address range. This may lead to an increased
128number of false negatives. However, it is likely that a real memory leak
129will eventually become visible.
130
131Another source of false negatives is the data stored in non-pointer
132values. In a future version, kmemleak could only scan the pointer
133members in the allocated structures. This feature would solve many of
134the false negative cases described above.
135
136The tool can report false positives. These are cases where an allocated
137block doesn't need to be freed (some cases in the init_call functions),
138the pointer is calculated by other methods than the usual container_of
139macro or the pointer is stored in a location not scanned by kmemleak.
140
141Page allocations and ioremap are not tracked. Only the ARM and x86
142architectures are currently supported.