• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1This document describes some caveats about the use of Valgrind with
2Python.  Valgrind is used periodically by Python developers to try
3to ensure there are no memory leaks or invalid memory reads/writes.
4
5UPDATE: Python 3.6 now supports PYTHONMALLOC=malloc environment variable which
6can be used to force the usage of the malloc() allocator of the C library.
7
8If you don't want to read about the details of using Valgrind, there
9are still two things you must do to suppress the warnings.  First,
10you must use a suppressions file.  One is supplied in
11Misc/valgrind-python.supp.  Second, you must do one of the following:
12
13  * Uncomment Py_USING_MEMORY_DEBUGGER in Objects/obmalloc.c,
14    then rebuild Python
15  * Uncomment the lines in Misc/valgrind-python.supp that
16    suppress the warnings for PyObject_Free and PyObject_Realloc
17
18If you want to use Valgrind more effectively and catch even more
19memory leaks, you will need to configure python --without-pymalloc.
20PyMalloc allocates a few blocks in big chunks and most object
21allocations don't call malloc, they use chunks doled about by PyMalloc
22from the big blocks.  This means Valgrind can't detect
23many allocations (and frees), except for those that are forwarded
24to the system malloc.  Note: configuring python --without-pymalloc
25makes Python run much slower, especially when running under Valgrind.
26You may need to run the tests in batches under Valgrind to keep
27the memory usage down to allow the tests to complete.  It seems to take
28about 5 times longer to run --without-pymalloc.
29
30Apr 15, 2006:
31  test_ctypes causes Valgrind 3.1.1 to fail (crash).
32  test_socket_ssl should be skipped when running valgrind.
33	The reason is that it purposely uses uninitialized memory.
34	This causes many spurious warnings, so it's easier to just skip it.
35
36
37Details:
38--------
39Python uses its own small-object allocation scheme on top of malloc,
40called PyMalloc.
41
42Valgrind may show some unexpected results when PyMalloc is used.
43Starting with Python 2.3, PyMalloc is used by default.  You can disable
44PyMalloc when configuring python by adding the --without-pymalloc option.
45If you disable PyMalloc, most of the information in this document and
46the supplied suppressions file will not be useful.  As discussed above,
47disabling PyMalloc can catch more problems.
48
49If you use valgrind on a default build of Python,  you will see
50many errors like:
51
52        ==6399== Use of uninitialised value of size 4
53        ==6399== at 0x4A9BDE7E: PyObject_Free (obmalloc.c:711)
54        ==6399== by 0x4A9B8198: dictresize (dictobject.c:477)
55
56These are expected and not a problem.  Tim Peters explains
57the situation:
58
59        PyMalloc needs to know whether an arbitrary address is one
60	that's managed by it, or is managed by the system malloc.
61	The current scheme allows this to be determined in constant
62	time, regardless of how many memory areas are under pymalloc's
63	control.
64
65        The memory pymalloc manages itself is in one or more "arenas",
66	each a large contiguous memory area obtained from malloc.
67	The base address of each arena is saved by pymalloc
68	in a vector.  Each arena is carved into "pools", and a field at
69	the start of each pool contains the index of that pool's arena's
70	base address in that vector.
71
72        Given an arbitrary address, pymalloc computes the pool base
73	address corresponding to it, then looks at "the index" stored
74	near there.  If the index read up is out of bounds for the
75	vector of arena base addresses pymalloc maintains, then
76	pymalloc knows for certain that this address is not under
77	pymalloc's control.  Otherwise the index is in bounds, and
78	pymalloc compares
79
80            the arena base address stored at that index in the vector
81
82        to
83
84            the arbitrary address pymalloc is investigating
85
86        pymalloc controls this arbitrary address if and only if it lies
87        in the arena the address's pool's index claims it lies in.
88
89        It doesn't matter whether the memory pymalloc reads up ("the
90	index") is initialized.  If it's not initialized, then
91	whatever trash gets read up will lead pymalloc to conclude
92	(correctly) that the address isn't controlled by it, either
93	because the index is out of bounds, or the index is in bounds
94	but the arena it represents doesn't contain the address.
95
96        This determination has to be made on every call to one of
97	pymalloc's free/realloc entry points, so its speed is critical
98	(Python allocates and frees dynamic memory at a ferocious rate
99	-- everything in Python, from integers to "stack frames",
100	lives in the heap).
101