• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1<?xml version="1.0"?> <!-- -*- sgml -*- -->
2<!DOCTYPE chapter PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN"
3          "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd">
4
5
6<chapter id="mc-manual" xreflabel="Memcheck: a memory error detector">
7<title>Memcheck: a memory error detector</title>
8
9<para>To use this tool, you may specify <option>--tool=memcheck</option>
10on the Valgrind command line.  You don't have to, though, since Memcheck
11is the default tool.</para>
12
13
14<sect1 id="mc-manual.overview" xreflabel="Overview">
15<title>Overview</title>
16
17<para>Memcheck is a memory error detector.  It can detect the following
18problems that are common in C and C++ programs.</para>
19
20<itemizedlist>
21  <listitem>
22    <para>Accessing memory you shouldn't, e.g. overrunning and underrunning
23    heap blocks, overrunning the top of the stack, and accessing memory after
24    it has been freed.</para>
25  </listitem>
26
27  <listitem>
28    <para>Using undefined values, i.e. values that have not been initialised,
29    or that have been derived from other undefined values.</para>
30  </listitem>
31
32  <listitem>
33    <para>Incorrect freeing of heap memory, such as double-freeing heap
34    blocks, or mismatched use of
35    <function>malloc</function>/<computeroutput>new</computeroutput>/<computeroutput>new[]</computeroutput>
36    versus
37    <function>free</function>/<computeroutput>delete</computeroutput>/<computeroutput>delete[]</computeroutput></para>
38  </listitem>
39
40  <listitem>
41    <para>Overlapping <computeroutput>src</computeroutput> and
42    <computeroutput>dst</computeroutput> pointers in
43    <computeroutput>memcpy</computeroutput> and related
44    functions.</para>
45  </listitem>
46
47  <listitem>
48    <para>Memory leaks.</para>
49  </listitem>
50</itemizedlist>
51
52<para>Problems like these can be difficult to find by other means,
53often remaining undetected for long periods, then causing occasional,
54difficult-to-diagnose crashes.</para>
55
56</sect1>
57
58
59
60<sect1 id="mc-manual.errormsgs"
61       xreflabel="Explanation of error messages from Memcheck">
62<title>Explanation of error messages from Memcheck</title>
63
64<para>Memcheck issues a range of error messages.  This section presents a
65quick summary of what error messages mean.  The precise behaviour of the
66error-checking machinery is described in <xref
67linkend="mc-manual.machine"/>.</para>
68
69
70<sect2 id="mc-manual.badrw"
71       xreflabel="Illegal read / Illegal write errors">
72<title>Illegal read / Illegal write errors</title>
73
74<para>For example:</para>
75<programlisting><![CDATA[
76Invalid read of size 4
77   at 0x40F6BBCC: (within /usr/lib/libpng.so.2.1.0.9)
78   by 0x40F6B804: (within /usr/lib/libpng.so.2.1.0.9)
79   by 0x40B07FF4: read_png_image(QImageIO *) (kernel/qpngio.cpp:326)
80   by 0x40AC751B: QImageIO::read() (kernel/qimage.cpp:3621)
81 Address 0xBFFFF0E0 is not stack'd, malloc'd or free'd
82]]></programlisting>
83
84<para>This happens when your program reads or writes memory at a place
85which Memcheck reckons it shouldn't.  In this example, the program did a
864-byte read at address 0xBFFFF0E0, somewhere within the system-supplied
87library libpng.so.2.1.0.9, which was called from somewhere else in the
88same library, called from line 326 of <filename>qpngio.cpp</filename>,
89and so on.</para>
90
91<para>Memcheck tries to establish what the illegal address might relate
92to, since that's often useful.  So, if it points into a block of memory
93which has already been freed, you'll be informed of this, and also where
94the block was freed.  Likewise, if it should turn out to be just off
95the end of a heap block, a common result of off-by-one-errors in
96array subscripting, you'll be informed of this fact, and also where the
97block was allocated.  If you use the <option><xref
98linkend="opt.read-var-info"/></option> option Memcheck will run more slowly
99but may give a more detailed description of any illegal address.</para>
100
101<para>In this example, Memcheck can't identify the address.  Actually
102the address is on the stack, but, for some reason, this is not a valid
103stack address -- it is below the stack pointer and that isn't allowed.
104In this particular case it's probably caused by GCC generating invalid
105code, a known bug in some ancient versions of GCC.</para>
106
107<para>Note that Memcheck only tells you that your program is about to
108access memory at an illegal address.  It can't stop the access from
109happening.  So, if your program makes an access which normally would
110result in a segmentation fault, you program will still suffer the same
111fate -- but you will get a message from Memcheck immediately prior to
112this.  In this particular example, reading junk on the stack is
113non-fatal, and the program stays alive.</para>
114
115</sect2>
116
117
118
119<sect2 id="mc-manual.uninitvals"
120       xreflabel="Use of uninitialised values">
121<title>Use of uninitialised values</title>
122
123<para>For example:</para>
124<programlisting><![CDATA[
125Conditional jump or move depends on uninitialised value(s)
126   at 0x402DFA94: _IO_vfprintf (_itoa.h:49)
127   by 0x402E8476: _IO_printf (printf.c:36)
128   by 0x8048472: main (tests/manuel1.c:8)
129]]></programlisting>
130
131<para>An uninitialised-value use error is reported when your program
132uses a value which hasn't been initialised -- in other words, is
133undefined.  Here, the undefined value is used somewhere inside the
134<function>printf</function> machinery of the C library.  This error was
135reported when running the following small program:</para>
136<programlisting><![CDATA[
137int main()
138{
139  int x;
140  printf ("x = %d\n", x);
141}]]></programlisting>
142
143<para>It is important to understand that your program can copy around
144junk (uninitialised) data as much as it likes.  Memcheck observes this
145and keeps track of the data, but does not complain.  A complaint is
146issued only when your program attempts to make use of uninitialised
147data in a way that might affect your program's externally-visible behaviour.
148In this example, <varname>x</varname> is uninitialised.  Memcheck observes
149the value being passed to <function>_IO_printf</function> and thence to
150<function>_IO_vfprintf</function>, but makes no comment.  However,
151<function>_IO_vfprintf</function> has to examine the value of
152<varname>x</varname> so it can turn it into the corresponding ASCII string,
153and it is at this point that Memcheck complains.</para>
154
155<para>Sources of uninitialised data tend to be:</para>
156<itemizedlist>
157  <listitem>
158    <para>Local variables in procedures which have not been initialised,
159    as in the example above.</para>
160  </listitem>
161  <listitem>
162    <para>The contents of heap blocks (allocated with
163    <function>malloc</function>, <function>new</function>, or a similar
164    function) before you (or a constructor) write something there.
165    </para>
166  </listitem>
167</itemizedlist>
168
169<para>To see information on the sources of uninitialised data in your
170program, use the <option>--track-origins=yes</option> option.  This
171makes Memcheck run more slowly, but can make it much easier to track down
172the root causes of uninitialised value errors.</para>
173
174</sect2>
175
176
177
178<sect2 id="mc-manual.bad-syscall-args"
179       xreflabel="Use of uninitialised or unaddressable values in system
180       calls">
181<title>Use of uninitialised or unaddressable values in system
182       calls</title>
183
184<para>Memcheck checks all parameters to system calls:
185<itemizedlist>
186  <listitem>
187    <para>It checks all the direct parameters themselves, whether they are
188    initialised.</para>
189  </listitem>
190  <listitem>
191    <para>Also, if a system call needs to read from a buffer provided by
192    your program, Memcheck checks that the entire buffer is addressable
193    and its contents are initialised.</para>
194  </listitem>
195  <listitem>
196    <para>Also, if the system call needs to write to a user-supplied
197    buffer, Memcheck checks that the buffer is addressable.</para>
198  </listitem>
199</itemizedlist>
200</para>
201
202<para>After the system call, Memcheck updates its tracked information to
203precisely reflect any changes in memory state caused by the system
204call.</para>
205
206<para>Here's an example of two system calls with invalid parameters:</para>
207<programlisting><![CDATA[
208  #include <stdlib.h>
209  #include <unistd.h>
210  int main( void )
211  {
212    char* arr  = malloc(10);
213    int*  arr2 = malloc(sizeof(int));
214    write( 1 /* stdout */, arr, 10 );
215    exit(arr2[0]);
216  }
217]]></programlisting>
218
219<para>You get these complaints ...</para>
220<programlisting><![CDATA[
221  Syscall param write(buf) points to uninitialised byte(s)
222     at 0x25A48723: __write_nocancel (in /lib/tls/libc-2.3.3.so)
223     by 0x259AFAD3: __libc_start_main (in /lib/tls/libc-2.3.3.so)
224     by 0x8048348: (within /auto/homes/njn25/grind/head4/a.out)
225   Address 0x25AB8028 is 0 bytes inside a block of size 10 alloc'd
226     at 0x259852B0: malloc (vg_replace_malloc.c:130)
227     by 0x80483F1: main (a.c:5)
228
229  Syscall param exit(error_code) contains uninitialised byte(s)
230     at 0x25A21B44: __GI__exit (in /lib/tls/libc-2.3.3.so)
231     by 0x8048426: main (a.c:8)
232]]></programlisting>
233
234<para>... because the program has (a) written uninitialised junk
235from the heap block to the standard output, and (b) passed an
236uninitialised value to <function>exit</function>.  Note that the first
237error refers to the memory pointed to by
238<computeroutput>buf</computeroutput> (not
239<computeroutput>buf</computeroutput> itself), but the second error
240refers directly to <computeroutput>exit</computeroutput>'s argument
241<computeroutput>arr2[0]</computeroutput>.</para>
242
243</sect2>
244
245
246<sect2 id="mc-manual.badfrees" xreflabel="Illegal frees">
247<title>Illegal frees</title>
248
249<para>For example:</para>
250<programlisting><![CDATA[
251Invalid free()
252   at 0x4004FFDF: free (vg_clientmalloc.c:577)
253   by 0x80484C7: main (tests/doublefree.c:10)
254 Address 0x3807F7B4 is 0 bytes inside a block of size 177 free'd
255   at 0x4004FFDF: free (vg_clientmalloc.c:577)
256   by 0x80484C7: main (tests/doublefree.c:10)
257]]></programlisting>
258
259<para>Memcheck keeps track of the blocks allocated by your program
260with <function>malloc</function>/<computeroutput>new</computeroutput>,
261so it can know exactly whether or not the argument to
262<function>free</function>/<computeroutput>delete</computeroutput> is
263legitimate or not.  Here, this test program has freed the same block
264twice.  As with the illegal read/write errors, Memcheck attempts to
265make sense of the address freed.  If, as here, the address is one
266which has previously been freed, you wil be told that -- making
267duplicate frees of the same block easy to spot.  You will also get this
268message if you try to free a pointer that doesn't point to the start of a
269heap block.</para>
270
271</sect2>
272
273
274<sect2 id="mc-manual.rudefn"
275       xreflabel="When a heap block is freed with an inappropriate deallocation
276function">
277<title>When a heap block is freed with an inappropriate deallocation
278function</title>
279
280<para>In the following example, a block allocated with
281<function>new[]</function> has wrongly been deallocated with
282<function>free</function>:</para>
283<programlisting><![CDATA[
284Mismatched free() / delete / delete []
285   at 0x40043249: free (vg_clientfuncs.c:171)
286   by 0x4102BB4E: QGArray::~QGArray(void) (tools/qgarray.cpp:149)
287   by 0x4C261C41: PptDoc::~PptDoc(void) (include/qmemarray.h:60)
288   by 0x4C261F0E: PptXml::~PptXml(void) (pptxml.cc:44)
289 Address 0x4BB292A8 is 0 bytes inside a block of size 64 alloc'd
290   at 0x4004318C: operator new[](unsigned int) (vg_clientfuncs.c:152)
291   by 0x4C21BC15: KLaola::readSBStream(int) const (klaola.cc:314)
292   by 0x4C21C155: KLaola::stream(KLaola::OLENode const *) (klaola.cc:416)
293   by 0x4C21788F: OLEFilter::convert(QCString const &) (olefilter.cc:272)
294]]></programlisting>
295
296<para>In <literal>C++</literal> it's important to deallocate memory in a
297way compatible with how it was allocated.  The deal is:</para>
298<itemizedlist>
299  <listitem>
300    <para>If allocated with
301    <function>malloc</function>,
302    <function>calloc</function>,
303    <function>realloc</function>,
304    <function>valloc</function> or
305    <function>memalign</function>, you must
306    deallocate with <function>free</function>.</para>
307  </listitem>
308  <listitem>
309   <para>If allocated with <function>new</function>, you must deallocate
310   with <function>delete</function>.</para>
311  </listitem>
312  <listitem>
313    <para>If allocated with <function>new[]</function>, you must
314    deallocate with <function>delete[]</function>.</para>
315  </listitem>
316</itemizedlist>
317
318<para>The worst thing is that on Linux apparently it doesn't matter if
319you do mix these up, but the same program may then crash on a
320different platform, Solaris for example.  So it's best to fix it
321properly.  According to the KDE folks "it's amazing how many C++
322programmers don't know this".</para>
323
324<para>The reason behind the requirement is as follows.  In some C++
325implementations, <function>delete[]</function> must be used for
326objects allocated by <function>new[]</function> because the compiler
327stores the size of the array and the pointer-to-member to the
328destructor of the array's content just before the pointer actually
329returned.  <function>delete</function> doesn't account for this and will get
330confused, possibly corrupting the heap.</para>
331
332</sect2>
333
334
335
336<sect2 id="mc-manual.overlap"
337       xreflabel="Overlapping source and destination blocks">
338<title>Overlapping source and destination blocks</title>
339
340<para>The following C library functions copy some data from one
341memory block to another (or something similar):
342<function>memcpy</function>,
343<function>strcpy</function>,
344<function>strncpy</function>,
345<function>strcat</function>,
346<function>strncat</function>.
347The blocks pointed to by their <computeroutput>src</computeroutput> and
348<computeroutput>dst</computeroutput> pointers aren't allowed to overlap.
349The POSIX standards have wording along the lines "If copying takes place
350between objects that overlap, the behavior is undefined." Therefore,
351Memcheck checks for this.
352</para>
353
354<para>For example:</para>
355<programlisting><![CDATA[
356==27492== Source and destination overlap in memcpy(0xbffff294, 0xbffff280, 21)
357==27492==    at 0x40026CDC: memcpy (mc_replace_strmem.c:71)
358==27492==    by 0x804865A: main (overlap.c:40)
359]]></programlisting>
360
361<para>You don't want the two blocks to overlap because one of them could
362get partially overwritten by the copying.</para>
363
364<para>You might think that Memcheck is being overly pedantic reporting
365this in the case where <computeroutput>dst</computeroutput> is less than
366<computeroutput>src</computeroutput>.  For example, the obvious way to
367implement <function>memcpy</function> is by copying from the first
368byte to the last.  However, the optimisation guides of some
369architectures recommend copying from the last byte down to the first.
370Also, some implementations of <function>memcpy</function> zero
371<computeroutput>dst</computeroutput> before copying, because zeroing the
372destination's cache line(s) can improve performance.</para>
373
374<para>The moral of the story is: if you want to write truly portable
375code, don't make any assumptions about the language
376implementation.</para>
377
378</sect2>
379
380
381<sect2 id="mc-manual.leaks" xreflabel="Memory leak detection">
382<title>Memory leak detection</title>
383
384<para>Memcheck keeps track of all heap blocks issued in response to
385calls to
386<function>malloc</function>/<function>new</function> et al.
387So when the program exits, it knows which blocks have not been freed.
388</para>
389
390<para>If <option>--leak-check</option> is set appropriately, for each
391remaining block, Memcheck determines if the block is reachable from pointers
392within the root-set.  The root-set consists of (a) general purpose registers
393of all threads, and (b) initialised, aligned, pointer-sized data words in
394accessible client memory, including stacks.</para>
395
396<para>There are two ways a block can be reached.  The first is with a
397"start-pointer", i.e. a pointer to the start of the block.  The second is with
398an "interior-pointer", i.e. a pointer to the middle of the block.  There are
399three ways we know of that an interior-pointer can occur:</para>
400
401<itemizedlist>
402  <listitem>
403    <para>The pointer might have originally been a start-pointer and have been
404    moved along deliberately (or not deliberately) by the program.  In
405    particular, this can happen if your program uses tagged pointers, i.e.
406    if it uses the bottom one, two or three bits of a pointer, which are
407    normally always zero due to alignment, in order to store extra
408    information.</para>
409  </listitem>
410
411  <listitem>
412    <para>It might be a random junk value in memory, entirely unrelated, just
413    a coincidence.</para>
414  </listitem>
415
416  <listitem>
417    <para>It might be a pointer to an array of C++ objects (which possess
418    destructors) allocated with <computeroutput>new[]</computeroutput>.  In
419    this case, some compilers store a "magic cookie" containing the array
420    length at the start of the allocated block, and return a pointer to just
421    past that magic cookie, i.e. an interior-pointer.
422    See <ulink url="http://theory.uwinnipeg.ca/gnu/gcc/gxxint_14.html">this
423    page</ulink> for more information.</para>
424  </listitem>
425</itemizedlist>
426
427<para>With that in mind, consider the nine possible cases described by the
428following figure.</para>
429
430<programlisting><![CDATA[
431     Pointer chain            AAA Category    BBB Category
432     -------------            ------------    ------------
433(1)  RRR ------------> BBB                    DR
434(2)  RRR ---> AAA ---> BBB    DR              IR
435(3)  RRR               BBB                    DL
436(4)  RRR      AAA ---> BBB    DL              IL
437(5)  RRR ------?-----> BBB                    (y)DR, (n)DL
438(6)  RRR ---> AAA -?-> BBB    DR              (y)IR, (n)DL
439(7)  RRR -?-> AAA ---> BBB    (y)DR, (n)DL    (y)IR, (n)IL
440(8)  RRR -?-> AAA -?-> BBB    (y)DR, (n)DL    (y,y)IR, (n,y)IL, (_,n)DL
441(9)  RRR      AAA -?-> BBB    DL              (y)IL, (n)DL
442
443Pointer chain legend:
444- RRR: a root set node or DR block
445- AAA, BBB: heap blocks
446- --->: a start-pointer
447- -?->: an interior-pointer
448
449Category legend:
450- DR: Directly reachable
451- IR: Indirectly reachable
452- DL: Directly lost
453- IL: Indirectly lost
454- (y)XY: it's XY if the interior-pointer is a real pointer
455- (n)XY: it's XY if the interior-pointer is not a real pointer
456- (_)XY: it's XY in either case
457]]></programlisting>
458
459<para>Every possible case can be reduced to one of the above nine.  Memcheck
460merges some of these cases in its output, resulting in the following four
461categories.</para>
462
463
464<itemizedlist>
465
466  <listitem>
467    <para>"Still reachable". This covers cases 1 and 2 (for the BBB blocks)
468    above.  A start-pointer or chain of start-pointers to the block is
469    found.  Since the block is still pointed at, the programmer could, at
470    least in principle, have freed it before program exit.  Because these
471    are very common and arguably not a problem, Memcheck won't report such
472    blocks individually unless <option>--show-reachable=yes</option> is
473    specified.</para>
474  </listitem>
475
476  <listitem>
477    <para>"Definitely lost".  This covers case 3 (for the BBB blocks) above.
478    This means that no pointer to the block can be found.  The block is
479    classified as "lost", because the programmer could not possibly have
480    freed it at program exit, since no pointer to it exists.  This is likely
481    a symptom of having lost the pointer at some earlier point in the
482    program.  Such cases should be fixed by the programmer.</para>
483    </listitem>
484
485  <listitem>
486    <para>"Indirectly lost".  This covers cases 4 and 9 (for the BBB blocks)
487    above.  This means that the block is lost, not because there are no
488    pointers to it, but rather because all the blocks that point to it are
489    themselves lost.  For example, if you have a binary tree and the root
490    node is lost, all its children nodes will be indirectly lost.  Because
491    the problem will disappear if the definitely lost block that caused the
492    indirect leak is fixed, Memcheck won't report such blocks individually
493    unless <option>--show-reachable=yes</option> is specified.</para>
494  </listitem>
495
496  <listitem>
497    <para>"Possibly lost".  This covers cases 5--8 (for the BBB blocks)
498    above.  This means that a chain of one or more pointers to the block has
499    been found, but at least one of the pointers is an interior-pointer.
500    This could just be a random value in memory that happens to point into a
501    block, and so you shouldn't consider this ok unless you know you have
502    interior-pointers.</para>
503  </listitem>
504
505</itemizedlist>
506
507<para>(Note: This mapping of the nine possible cases onto four categories is
508not necessarily the best way that leaks could be reported;  in particular,
509interior-pointers are treated inconsistently.  It is possible the
510categorisation may be improved in the future.)</para>
511
512<para>Furthermore, if suppressions exists for a block, it will be reported
513as "suppressed" no matter what which of the above four categories it belongs
514to.</para>
515
516
517<para>The following is an example leak summary.</para>
518
519<programlisting><![CDATA[
520LEAK SUMMARY:
521   definitely lost: 48 bytes in 3 blocks.
522   indirectly lost: 32 bytes in 2 blocks.
523     possibly lost: 96 bytes in 6 blocks.
524   still reachable: 64 bytes in 4 blocks.
525        suppressed: 0 bytes in 0 blocks.
526]]></programlisting>
527
528<para>If <option>--leak-check=full</option> is specified,
529Memcheck will give details for each definitely lost or possibly lost block,
530including where it was allocated.  (Actually, it merges results for all
531blocks that have the same category and sufficiently similar stack traces
532into a single "loss record".  The
533<option>--leak-resolution</option> lets you control the
534meaning of "sufficiently similar".)  It cannot tell you when or how or why
535the pointer to a leaked block was lost; you have to work that out for
536yourself.  In general, you should attempt to ensure your programs do not
537have any definitely lost or possibly lost blocks at exit.</para>
538
539<para>For example:</para>
540<programlisting><![CDATA[
5418 bytes in 1 blocks are definitely lost in loss record 1 of 14
542   at 0x........: malloc (vg_replace_malloc.c:...)
543   by 0x........: mk (leak-tree.c:11)
544   by 0x........: main (leak-tree.c:39)
545
54688 (8 direct, 80 indirect) bytes in 1 blocks are definitely lost in loss record 13 of 14
547   at 0x........: malloc (vg_replace_malloc.c:...)
548   by 0x........: mk (leak-tree.c:11)
549   by 0x........: main (leak-tree.c:25)
550]]></programlisting>
551
552<para>The first message describes a simple case of a single 8 byte block
553that has been definitely lost.  The second case mentions another 8 byte
554block that has been definitely lost;  the difference is that a further 80
555bytes in other blocks are indirectly lost because of this lost block.
556The loss records are not presented in any notable order, so the loss record
557numbers aren't particularly meaningful.</para>
558
559<para>If you specify <option>--show-reachable=yes</option>,
560reachable and indirectly lost blocks will also be shown, as the following
561two examples show.</para>
562
563<programlisting><![CDATA[
56464 bytes in 4 blocks are still reachable in loss record 2 of 4
565   at 0x........: malloc (vg_replace_malloc.c:177)
566   by 0x........: mk (leak-cases.c:52)
567   by 0x........: main (leak-cases.c:74)
568
56932 bytes in 2 blocks are indirectly lost in loss record 1 of 4
570   at 0x........: malloc (vg_replace_malloc.c:177)
571   by 0x........: mk (leak-cases.c:52)
572   by 0x........: main (leak-cases.c:80)
573]]></programlisting>
574
575<para>Because there are different kinds of leaks with different severities, an
576interesting question is this: which leaks should be counted as true "errors"
577and which should not?  The answer to this question affects the numbers printed
578in the <computeroutput>ERROR SUMMARY</computeroutput> line, and also the effect
579of the <option>--error-exitcode</option> option.  Memcheck uses the following
580criteria:</para>
581
582<itemizedlist>
583  <listitem>
584    <para>First, a leak is only counted as a true "error" if
585    <option>--leak-check=full</option> is specified.  In other words, an
586    unprinted leak is not considered a true "error".  If this were not the
587    case, it would be possible to get a high error count but not have any
588    errors printed, which would be confusing.</para>
589  </listitem>
590
591  <listitem>
592    <para>After that, definitely lost and possibly lost blocks are counted as
593    true "errors".  Indirectly lost and still reachable blocks are not counted
594    as true "errors", even if <option>--show-reachable=yes</option> is
595    specified and they are printed;  this is because such blocks don't need
596    direct fixing by the programmer.
597    </para>
598  </listitem>
599</itemizedlist>
600
601</sect2>
602
603</sect1>
604
605
606
607<sect1 id="mc-manual.options"
608       xreflabel="Memcheck Command-Line Options">
609<title>Memcheck Command-Line Options</title>
610
611<!-- start of xi:include in the manpage -->
612<variablelist id="mc.opts.list">
613
614  <varlistentry id="opt.leak-check" xreflabel="--leak-check">
615    <term>
616      <option><![CDATA[--leak-check=<no|summary|yes|full> [default: summary] ]]></option>
617    </term>
618    <listitem>
619      <para>When enabled, search for memory leaks when the client
620      program finishes.  If set to <varname>summary</varname>, it says how
621      many leaks occurred.  If set to <varname>full</varname> or
622      <varname>yes</varname>, it also gives details of each individual
623      leak.</para>
624    </listitem>
625  </varlistentry>
626
627  <varlistentry id="opt.show-possibly-lost" xreflabel="--show-possibly-lost">
628    <term>
629      <option><![CDATA[--show-possibly-lost=<yes|no> [default: yes] ]]></option>
630    </term>
631    <listitem>
632      <para>When disabled, the memory leak detector will not show "possibly lost" blocks.
633      </para>
634    </listitem>
635  </varlistentry>
636
637  <varlistentry id="opt.leak-resolution" xreflabel="--leak-resolution">
638    <term>
639      <option><![CDATA[--leak-resolution=<low|med|high> [default: high] ]]></option>
640    </term>
641    <listitem>
642      <para>When doing leak checking, determines how willing
643      Memcheck is to consider different backtraces to
644      be the same for the purposes of merging multiple leaks into a single
645      leak report.  When set to <varname>low</varname>, only the first
646      two entries need match.  When <varname>med</varname>, four entries
647      have to match.  When <varname>high</varname>, all entries need to
648      match.</para>
649
650      <para>For hardcore leak debugging, you probably want to use
651      <option>--leak-resolution=high</option> together with
652      <option>--num-callers=40</option> or some such large number.
653      </para>
654
655      <para>Note that the <option>--leak-resolution</option> setting
656      does not affect Memcheck's ability to find
657      leaks.  It only changes how the results are presented.</para>
658    </listitem>
659  </varlistentry>
660
661  <varlistentry id="opt.show-reachable" xreflabel="--show-reachable">
662    <term>
663      <option><![CDATA[--show-reachable=<yes|no> [default: no] ]]></option>
664    </term>
665    <listitem>
666      <para>When disabled, the memory leak detector only shows "definitely
667      lost" and "possibly lost" blocks.  When enabled, the leak detector also
668      shows "reachable" and "indirectly lost" blocks.  (In other words, it
669      shows all blocks, except suppressed ones, so
670      <option>--show-all</option> would be a better name for
671      it.)</para>
672    </listitem>
673  </varlistentry>
674
675  <varlistentry id="opt.undef-value-errors" xreflabel="--undef-value-errors">
676    <term>
677      <option><![CDATA[--undef-value-errors=<yes|no> [default: yes] ]]></option>
678    </term>
679    <listitem>
680      <para>Controls whether Memcheck reports
681      uses of undefined value errors.  Set this to
682      <varname>no</varname> if you don't want to see undefined value
683      errors.  It also has the side effect of speeding up
684      Memcheck somewhat.
685      </para>
686    </listitem>
687  </varlistentry>
688
689  <varlistentry id="opt.track-origins" xreflabel="--track-origins">
690    <term>
691      <option><![CDATA[--track-origins=<yes|no> [default: no] ]]></option>
692    </term>
693      <listitem>
694        <para>Controls whether Memcheck tracks
695        the origin of uninitialised values.  By default, it does not,
696        which means that although it can tell you that an
697        uninitialised value is being used in a dangerous way, it
698        cannot tell you where the uninitialised value came from.  This
699        often makes it difficult to track down the root problem.
700        </para>
701        <para>When set
702        to <varname>yes</varname>, Memcheck keeps
703        track of the origins of all uninitialised values.  Then, when
704        an uninitialised value error is
705        reported, Memcheck will try to show the
706        origin of the value.  An origin can be one of the following
707        four places: a heap block, a stack allocation, a client
708        request, or miscellaneous other sources (eg, a call
709        to <varname>brk</varname>).
710        </para>
711        <para>For uninitialised values originating from a heap
712        block, Memcheck shows where the block was
713        allocated.  For uninitialised values originating from a stack
714        allocation, Memcheck can tell you which
715        function allocated the value, but no more than that -- typically
716        it shows you the source location of the opening brace of the
717        function.  So you should carefully check that all of the
718        function's local variables are initialised properly.
719        </para>
720        <para>Performance overhead: origin tracking is expensive.  It
721        halves Memcheck's speed and increases
722        memory use by a minimum of 100MB, and possibly more.
723        Nevertheless it can drastically reduce the effort required to
724        identify the root cause of uninitialised value errors, and so
725        is often a programmer productivity win, despite running
726        more slowly.
727        </para>
728        <para>Accuracy: Memcheck tracks origins
729        quite accurately.  To avoid very large space and time
730        overheads, some approximations are made.  It is possible,
731        although unlikely, that Memcheck will report an incorrect origin, or
732        not be able to identify any origin.
733        </para>
734        <para>Note that the combination
735        <option>--track-origins=yes</option>
736        and <option>--undef-value-errors=no</option> is
737        nonsensical.  Memcheck checks for and
738        rejects this combination at startup.
739        </para>
740      </listitem>
741  </varlistentry>
742
743  <varlistentry id="opt.partial-loads-ok" xreflabel="--partial-loads-ok">
744    <term>
745      <option><![CDATA[--partial-loads-ok=<yes|no> [default: no] ]]></option>
746    </term>
747    <listitem>
748      <para>Controls how Memcheck handles word-sized,
749      word-aligned loads from addresses for which some bytes are
750      addressable and others are not.  When <varname>yes</varname>, such
751      loads do not produce an address error.  Instead, loaded bytes
752      originating from illegal addresses are marked as uninitialised, and
753      those corresponding to legal addresses are handled in the normal
754      way.</para>
755
756      <para>When <varname>no</varname>, loads from partially invalid
757      addresses are treated the same as loads from completely invalid
758      addresses: an illegal-address error is issued, and the resulting
759      bytes are marked as initialised.</para>
760
761      <para>Note that code that behaves in this way is in violation of
762      the the ISO C/C++ standards, and should be considered broken.  If
763      at all possible, such code should be fixed.  This option should be
764      used only as a last resort.</para>
765    </listitem>
766  </varlistentry>
767
768  <varlistentry id="opt.freelist-vol" xreflabel="--freelist-vol">
769    <term>
770      <option><![CDATA[--freelist-vol=<number> [default: 20000000] ]]></option>
771    </term>
772    <listitem>
773      <para>When the client program releases memory using
774      <function>free</function> (in <literal>C</literal>) or
775      <computeroutput>delete</computeroutput>
776      (<literal>C++</literal>), that memory is not immediately made
777      available for re-allocation.  Instead, it is marked inaccessible
778      and placed in a queue of freed blocks.  The purpose is to defer as
779      long as possible the point at which freed-up memory comes back
780      into circulation.  This increases the chance that
781      Memcheck will be able to detect invalid
782      accesses to blocks for some significant period of time after they
783      have been freed.</para>
784
785      <para>This option specifies the maximum total size, in bytes, of the
786      blocks in the queue.  The default value is twenty million bytes.
787      Increasing this increases the total amount of memory used by
788      Memcheck but may detect invalid uses of freed
789      blocks which would otherwise go undetected.</para>
790    </listitem>
791  </varlistentry>
792
793  <varlistentry id="opt.freelist-big-blocks" xreflabel="--freelist-big-blocks">
794    <term>
795      <option><![CDATA[--freelist-big-blocks=<number> [default: 1000000] ]]></option>
796    </term>
797    <listitem>
798      <para>When making blocks from the queue of freed blocks available
799      for re-allocation, Memcheck will in priority re-circulate the blocks
800      with a size greater or equal to <option>--freelist-big-blocks</option>.
801      This ensures that freeing big blocks (in particular freeing blocks bigger than
802      <option>--freelist-vol</option>) does not immediately lead to a re-circulation
803      of all (or a lot of) the small blocks in the free list. In other words,
804      this option increases the likelihood to discover dangling pointers
805      for the "small" blocks, even when big blocks are freed.</para>
806      <para>Setting a value of 0 means that all the blocks are re-circulated
807      in a FIFO order. </para>
808    </listitem>
809  </varlistentry>
810
811  <varlistentry id="opt.workaround-gcc296-bugs" xreflabel="--workaround-gcc296-bugs">
812    <term>
813      <option><![CDATA[--workaround-gcc296-bugs=<yes|no> [default: no] ]]></option>
814    </term>
815    <listitem>
816      <para>When enabled, assume that reads and writes some small
817      distance below the stack pointer are due to bugs in GCC 2.96, and
818      does not report them.  The "small distance" is 256 bytes by
819      default.  Note that GCC 2.96 is the default compiler on some ancient
820      Linux distributions (RedHat 7.X) and so you may need to use this
821      option.  Do not use it if you do not have to, as it can cause real
822      errors to be overlooked.  A better alternative is to use a more
823      recent GCC in which this bug is fixed.</para>
824
825      <para>You may also need to use this option when working with
826      GCC 3.X or 4.X on 32-bit PowerPC Linux.  This is because
827      GCC generates code which occasionally accesses below the
828      stack pointer, particularly for floating-point to/from integer
829      conversions.  This is in violation of the 32-bit PowerPC ELF
830      specification, which makes no provision for locations below the
831      stack pointer to be accessible.</para>
832    </listitem>
833  </varlistentry>
834
835  <varlistentry id="opt.ignore-ranges" xreflabel="--ignore-ranges">
836    <term>
837      <option><![CDATA[--ignore-ranges=0xPP-0xQQ[,0xRR-0xSS] ]]></option>
838    </term>
839    <listitem>
840    <para>Any ranges listed in this option (and multiple ranges can be
841    specified, separated by commas) will be ignored by Memcheck's
842    addressability checking.</para>
843    </listitem>
844  </varlistentry>
845
846  <varlistentry id="opt.malloc-fill" xreflabel="--malloc-fill">
847    <term>
848      <option><![CDATA[--malloc-fill=<hexnumber> ]]></option>
849    </term>
850    <listitem>
851      <para>Fills blocks allocated
852      by <computeroutput>malloc</computeroutput>,
853         <computeroutput>new</computeroutput>, etc, but not
854      by <computeroutput>calloc</computeroutput>, with the specified
855      byte.  This can be useful when trying to shake out obscure
856      memory corruption problems.  The allocated area is still
857      regarded by Memcheck as undefined -- this option only affects its
858      contents.
859      </para>
860    </listitem>
861  </varlistentry>
862
863  <varlistentry id="opt.free-fill" xreflabel="--free-fill">
864    <term>
865      <option><![CDATA[--free-fill=<hexnumber> ]]></option>
866    </term>
867    <listitem>
868      <para>Fills blocks freed
869      by <computeroutput>free</computeroutput>,
870         <computeroutput>delete</computeroutput>, etc, with the
871      specified byte value.  This can be useful when trying to shake out
872      obscure memory corruption problems.  The freed area is still
873      regarded by Memcheck as not valid for access -- this option only
874      affects its contents.
875      </para>
876    </listitem>
877  </varlistentry>
878
879</variablelist>
880<!-- end of xi:include in the manpage -->
881
882</sect1>
883
884
885<sect1 id="mc-manual.suppfiles" xreflabel="Writing suppression files">
886<title>Writing suppression files</title>
887
888<para>The basic suppression format is described in
889<xref linkend="manual-core.suppress"/>.</para>
890
891<para>The suppression-type (second) line should have the form:</para>
892<programlisting><![CDATA[
893Memcheck:suppression_type]]></programlisting>
894
895<para>The Memcheck suppression types are as follows:</para>
896
897<itemizedlist>
898  <listitem>
899    <para><varname>Value1</varname>,
900    <varname>Value2</varname>,
901    <varname>Value4</varname>,
902    <varname>Value8</varname>,
903    <varname>Value16</varname>,
904    meaning an uninitialised-value error when
905    using a value of 1, 2, 4, 8 or 16 bytes.</para>
906  </listitem>
907
908  <listitem>
909    <para><varname>Cond</varname> (or its old
910    name, <varname>Value0</varname>), meaning use
911    of an uninitialised CPU condition code.</para>
912  </listitem>
913
914  <listitem>
915    <para><varname>Addr1</varname>,
916    <varname>Addr2</varname>,
917    <varname>Addr4</varname>,
918    <varname>Addr8</varname>,
919    <varname>Addr16</varname>,
920    meaning an invalid address during a
921    memory access of 1, 2, 4, 8 or 16 bytes respectively.</para>
922  </listitem>
923
924  <listitem>
925    <para><varname>Jump</varname>, meaning an
926    jump to an unaddressable location error.</para>
927  </listitem>
928
929  <listitem>
930    <para><varname>Param</varname>, meaning an
931    invalid system call parameter error.</para>
932  </listitem>
933
934  <listitem>
935    <para><varname>Free</varname>, meaning an
936    invalid or mismatching free.</para>
937  </listitem>
938
939  <listitem>
940    <para><varname>Overlap</varname>, meaning a
941    <computeroutput>src</computeroutput> /
942    <computeroutput>dst</computeroutput> overlap in
943    <function>memcpy</function> or a similar function.</para>
944  </listitem>
945
946  <listitem>
947    <para><varname>Leak</varname>, meaning
948    a memory leak.</para>
949  </listitem>
950
951</itemizedlist>
952
953<para><computeroutput>Param</computeroutput> errors have an extra
954information line at this point, which is the name of the offending
955system call parameter.  No other error kinds have this extra
956line.</para>
957
958<para>The first line of the calling context: for <varname>ValueN</varname>
959and <varname>AddrN</varname> errors, it is either the name of the function
960in which the error occurred, or, failing that, the full path of the
961<filename>.so</filename> file
962or executable containing the error location.  For <varname>Free</varname> errors, is the name
963of the function doing the freeing (eg, <function>free</function>,
964<function>__builtin_vec_delete</function>, etc).  For
965<varname>Overlap</varname> errors, is the name of the function with the
966overlapping arguments (eg.  <function>memcpy</function>,
967<function>strcpy</function>, etc).</para>
968
969<para>Lastly, there's the rest of the calling context.</para>
970
971</sect1>
972
973
974
975<sect1 id="mc-manual.machine"
976       xreflabel="Details of Memcheck's checking machinery">
977<title>Details of Memcheck's checking machinery</title>
978
979<para>Read this section if you want to know, in detail, exactly
980what and how Memcheck is checking.</para>
981
982
983<sect2 id="mc-manual.value" xreflabel="Valid-value (V) bit">
984<title>Valid-value (V) bits</title>
985
986<para>It is simplest to think of Memcheck implementing a synthetic CPU
987which is identical to a real CPU, except for one crucial detail.  Every
988bit (literally) of data processed, stored and handled by the real CPU
989has, in the synthetic CPU, an associated "valid-value" bit, which says
990whether or not the accompanying bit has a legitimate value.  In the
991discussions which follow, this bit is referred to as the V (valid-value)
992bit.</para>
993
994<para>Each byte in the system therefore has a 8 V bits which follow it
995wherever it goes.  For example, when the CPU loads a word-size item (4
996bytes) from memory, it also loads the corresponding 32 V bits from a
997bitmap which stores the V bits for the process' entire address space.
998If the CPU should later write the whole or some part of that value to
999memory at a different address, the relevant V bits will be stored back
1000in the V-bit bitmap.</para>
1001
1002<para>In short, each bit in the system has (conceptually) an associated V
1003bit, which follows it around everywhere, even inside the CPU.  Yes, all the
1004CPU's registers (integer, floating point, vector and condition registers)
1005have their own V bit vectors.  For this to work, Memcheck uses a great deal
1006of compression to represent the V bits compactly.</para>
1007
1008<para>Copying values around does not cause Memcheck to check for, or
1009report on, errors.  However, when a value is used in a way which might
1010conceivably affect your program's externally-visible behaviour,
1011the associated V bits are immediately checked.  If any of these indicate
1012that the value is undefined (even partially), an error is reported.</para>
1013
1014<para>Here's an (admittedly nonsensical) example:</para>
1015<programlisting><![CDATA[
1016int i, j;
1017int a[10], b[10];
1018for ( i = 0; i < 10; i++ ) {
1019  j = a[i];
1020  b[i] = j;
1021}]]></programlisting>
1022
1023<para>Memcheck emits no complaints about this, since it merely copies
1024uninitialised values from <varname>a[]</varname> into
1025<varname>b[]</varname>, and doesn't use them in a way which could
1026affect the behaviour of the program.  However, if
1027the loop is changed to:</para>
1028<programlisting><![CDATA[
1029for ( i = 0; i < 10; i++ ) {
1030  j += a[i];
1031}
1032if ( j == 77 )
1033  printf("hello there\n");
1034]]></programlisting>
1035
1036<para>then Memcheck will complain, at the
1037<computeroutput>if</computeroutput>, that the condition depends on
1038uninitialised values.  Note that it <command>doesn't</command> complain
1039at the <varname>j += a[i];</varname>, since at that point the
1040undefinedness is not "observable".  It's only when a decision has to be
1041made as to whether or not to do the <function>printf</function> -- an
1042observable action of your program -- that Memcheck complains.</para>
1043
1044<para>Most low level operations, such as adds, cause Memcheck to use the
1045V bits for the operands to calculate the V bits for the result.  Even if
1046the result is partially or wholly undefined, it does not
1047complain.</para>
1048
1049<para>Checks on definedness only occur in three places: when a value is
1050used to generate a memory address, when control flow decision needs to
1051be made, and when a system call is detected, Memcheck checks definedness
1052of parameters as required.</para>
1053
1054<para>If a check should detect undefinedness, an error message is
1055issued.  The resulting value is subsequently regarded as well-defined.
1056To do otherwise would give long chains of error messages.  In other
1057words, once Memcheck reports an undefined value error, it tries to
1058avoid reporting further errors derived from that same undefined
1059value.</para>
1060
1061<para>This sounds overcomplicated.  Why not just check all reads from
1062memory, and complain if an undefined value is loaded into a CPU
1063register?  Well, that doesn't work well, because perfectly legitimate C
1064programs routinely copy uninitialised values around in memory, and we
1065don't want endless complaints about that.  Here's the canonical example.
1066Consider a struct like this:</para>
1067<programlisting><![CDATA[
1068struct S { int x; char c; };
1069struct S s1, s2;
1070s1.x = 42;
1071s1.c = 'z';
1072s2 = s1;
1073]]></programlisting>
1074
1075<para>The question to ask is: how large is <varname>struct S</varname>,
1076in bytes?  An <varname>int</varname> is 4 bytes and a
1077<varname>char</varname> one byte, so perhaps a <varname>struct
1078S</varname> occupies 5 bytes?  Wrong.  All non-toy compilers we know
1079of will round the size of <varname>struct S</varname> up to a whole
1080number of words, in this case 8 bytes.  Not doing this forces compilers
1081to generate truly appalling code for accessing arrays of
1082<varname>struct S</varname>'s on some architectures.</para>
1083
1084<para>So <varname>s1</varname> occupies 8 bytes, yet only 5 of them will
1085be initialised.  For the assignment <varname>s2 = s1</varname>, GCC
1086generates code to copy all 8 bytes wholesale into <varname>s2</varname>
1087without regard for their meaning.  If Memcheck simply checked values as
1088they came out of memory, it would yelp every time a structure assignment
1089like this happened.  So the more complicated behaviour described above
1090is necessary.  This allows GCC to copy
1091<varname>s1</varname> into <varname>s2</varname> any way it likes, and a
1092warning will only be emitted if the uninitialised values are later
1093used.</para>
1094
1095</sect2>
1096
1097
1098<sect2 id="mc-manual.vaddress" xreflabel=" Valid-address (A) bits">
1099<title>Valid-address (A) bits</title>
1100
1101<para>Notice that the previous subsection describes how the validity of
1102values is established and maintained without having to say whether the
1103program does or does not have the right to access any particular memory
1104location.  We now consider the latter question.</para>
1105
1106<para>As described above, every bit in memory or in the CPU has an
1107associated valid-value (V) bit.  In addition, all bytes in memory, but
1108not in the CPU, have an associated valid-address (A) bit.  This
1109indicates whether or not the program can legitimately read or write that
1110location.  It does not give any indication of the validity of the data
1111at that location -- that's the job of the V bits -- only whether or not
1112the location may be accessed.</para>
1113
1114<para>Every time your program reads or writes memory, Memcheck checks
1115the A bits associated with the address.  If any of them indicate an
1116invalid address, an error is emitted.  Note that the reads and writes
1117themselves do not change the A bits, only consult them.</para>
1118
1119<para>So how do the A bits get set/cleared?  Like this:</para>
1120
1121<itemizedlist>
1122  <listitem>
1123    <para>When the program starts, all the global data areas are
1124    marked as accessible.</para>
1125  </listitem>
1126
1127  <listitem>
1128    <para>When the program does
1129    <function>malloc</function>/<computeroutput>new</computeroutput>,
1130    the A bits for exactly the area allocated, and not a byte more,
1131    are marked as accessible.  Upon freeing the area the A bits are
1132    changed to indicate inaccessibility.</para>
1133  </listitem>
1134
1135  <listitem>
1136    <para>When the stack pointer register (<literal>SP</literal>) moves
1137    up or down, A bits are set.  The rule is that the area from
1138    <literal>SP</literal> up to the base of the stack is marked as
1139    accessible, and below <literal>SP</literal> is inaccessible.  (If
1140    that sounds illogical, bear in mind that the stack grows down, not
1141    up, on almost all Unix systems, including GNU/Linux.)  Tracking
1142    <literal>SP</literal> like this has the useful side-effect that the
1143    section of stack used by a function for local variables etc is
1144    automatically marked accessible on function entry and inaccessible
1145    on exit.</para>
1146  </listitem>
1147
1148  <listitem>
1149    <para>When doing system calls, A bits are changed appropriately.
1150    For example, <literal>mmap</literal>
1151    magically makes files appear in the process'
1152    address space, so the A bits must be updated if <literal>mmap</literal>
1153    succeeds.</para>
1154  </listitem>
1155
1156  <listitem>
1157    <para>Optionally, your program can tell Memcheck about such changes
1158    explicitly, using the client request mechanism described
1159    above.</para>
1160  </listitem>
1161
1162</itemizedlist>
1163
1164</sect2>
1165
1166
1167<sect2 id="mc-manual.together" xreflabel="Putting it all together">
1168<title>Putting it all together</title>
1169
1170<para>Memcheck's checking machinery can be summarised as
1171follows:</para>
1172
1173<itemizedlist>
1174  <listitem>
1175    <para>Each byte in memory has 8 associated V (valid-value) bits,
1176    saying whether or not the byte has a defined value, and a single A
1177    (valid-address) bit, saying whether or not the program currently has
1178    the right to read/write that address.  As mentioned above, heavy
1179    use of compression means the overhead is typically around 25%.</para>
1180  </listitem>
1181
1182  <listitem>
1183    <para>When memory is read or written, the relevant A bits are
1184    consulted.  If they indicate an invalid address, Memcheck emits an
1185    Invalid read or Invalid write error.</para>
1186  </listitem>
1187
1188  <listitem>
1189    <para>When memory is read into the CPU's registers, the relevant V
1190    bits are fetched from memory and stored in the simulated CPU.  They
1191    are not consulted.</para>
1192  </listitem>
1193
1194  <listitem>
1195    <para>When a register is written out to memory, the V bits for that
1196    register are written back to memory too.</para>
1197  </listitem>
1198
1199  <listitem>
1200    <para>When values in CPU registers are used to generate a memory
1201    address, or to determine the outcome of a conditional branch, the V
1202    bits for those values are checked, and an error emitted if any of
1203    them are undefined.</para>
1204  </listitem>
1205
1206  <listitem>
1207    <para>When values in CPU registers are used for any other purpose,
1208    Memcheck computes the V bits for the result, but does not check
1209    them.</para>
1210  </listitem>
1211
1212  <listitem>
1213    <para>Once the V bits for a value in the CPU have been checked, they
1214    are then set to indicate validity.  This avoids long chains of
1215    errors.</para>
1216  </listitem>
1217
1218  <listitem>
1219    <para>When values are loaded from memory, Memcheck checks the A bits
1220    for that location and issues an illegal-address warning if needed.
1221    In that case, the V bits loaded are forced to indicate Valid,
1222    despite the location being invalid.</para>
1223
1224    <para>This apparently strange choice reduces the amount of confusing
1225    information presented to the user.  It avoids the unpleasant
1226    phenomenon in which memory is read from a place which is both
1227    unaddressable and contains invalid values, and, as a result, you get
1228    not only an invalid-address (read/write) error, but also a
1229    potentially large set of uninitialised-value errors, one for every
1230    time the value is used.</para>
1231
1232    <para>There is a hazy boundary case to do with multi-byte loads from
1233    addresses which are partially valid and partially invalid.  See
1234    details of the option <option>--partial-loads-ok</option> for details.
1235    </para>
1236  </listitem>
1237
1238</itemizedlist>
1239
1240
1241<para>Memcheck intercepts calls to <function>malloc</function>,
1242<function>calloc</function>, <function>realloc</function>,
1243<function>valloc</function>, <function>memalign</function>,
1244<function>free</function>, <computeroutput>new</computeroutput>,
1245<computeroutput>new[]</computeroutput>,
1246<computeroutput>delete</computeroutput> and
1247<computeroutput>delete[]</computeroutput>.  The behaviour you get
1248is:</para>
1249
1250<itemizedlist>
1251
1252  <listitem>
1253    <para><function>malloc</function>/<function>new</function>/<computeroutput>new[]</computeroutput>:
1254    the returned memory is marked as addressable but not having valid
1255    values.  This means you have to write to it before you can read
1256    it.</para>
1257  </listitem>
1258
1259  <listitem>
1260    <para><function>calloc</function>: returned memory is marked both
1261    addressable and valid, since <function>calloc</function> clears
1262    the area to zero.</para>
1263  </listitem>
1264
1265  <listitem>
1266    <para><function>realloc</function>: if the new size is larger than
1267    the old, the new section is addressable but invalid, as with
1268    <function>malloc</function>.  If the new size is smaller, the
1269    dropped-off section is marked as unaddressable.  You may only pass to
1270    <function>realloc</function> a pointer previously issued to you by
1271    <function>malloc</function>/<function>calloc</function>/<function>realloc</function>.</para>
1272  </listitem>
1273
1274  <listitem>
1275    <para><function>free</function>/<computeroutput>delete</computeroutput>/<computeroutput>delete[]</computeroutput>:
1276    you may only pass to these functions a pointer previously issued
1277    to you by the corresponding allocation function.  Otherwise,
1278    Memcheck complains.  If the pointer is indeed valid, Memcheck
1279    marks the entire area it points at as unaddressable, and places
1280    the block in the freed-blocks-queue.  The aim is to defer as long
1281    as possible reallocation of this block.  Until that happens, all
1282    attempts to access it will elicit an invalid-address error, as you
1283    would hope.</para>
1284  </listitem>
1285
1286</itemizedlist>
1287
1288</sect2>
1289</sect1>
1290
1291<sect1 id="mc-manual.monitor-commands" xreflabel="Memcheck Monitor Commands">
1292<title>Memcheck Monitor Commands</title>
1293<para>The Memcheck tool provides monitor commands handled by Valgrind's
1294built-in gdbserver (see <xref linkend="manual-core-adv.gdbserver-commandhandling"/>).
1295</para>
1296
1297<itemizedlist>
1298  <listitem>
1299    <para><varname>get_vbits &lt;addr&gt; [&lt;len&gt;]</varname>
1300    shows the definedness (V) bits for &lt;len&gt; (default 1) bytes
1301    starting at &lt;addr&gt;.  The definedness of each byte in the
1302    range is given using two hexadecimal digits.  These hexadecimal
1303    digits encode the validity of each bit of the corresponding byte,
1304    using 0 if the bit is defined and 1 if the bit is undefined.
1305    If a byte is not addressable, its validity bits are replaced
1306    by <varname>__</varname> (a double underscore).
1307    </para>
1308    <para>
1309    In the following example, <varname>string10</varname> is an array
1310    of 10 characters, in which the even numbered bytes are
1311    undefined. In the below example, the byte corresponding
1312    to <varname>string10[5]</varname> is not addressable.
1313    </para>
1314<programlisting><![CDATA[
1315(gdb) p &string10
1316$4 = (char (*)[10]) 0x8049e28
1317(gdb) monitor get_vbits 0x8049e28 10
1318ff00ff00 ff__ff00 ff00
1319(gdb)
1320]]></programlisting>
1321
1322    <para> The command get_vbits cannot be used with registers. To get
1323    the validity bits of a register, you must start Valgrind with the
1324    option <option>--vgdb-shadow-registers=yes</option>. The validity
1325    bits of a register can be obtained by printing the 'shadow 1'
1326    corresponding register.  In the below x86 example, the register
1327    eax has all its bits undefined, while the register ebx is fully
1328    defined.
1329    </para>
1330<programlisting><![CDATA[
1331(gdb) p /x $eaxs1
1332$9 = 0xffffffff
1333(gdb) p /x $ebxs1
1334$10 = 0x0
1335(gdb)
1336]]></programlisting>
1337
1338  </listitem>
1339
1340  <listitem>
1341    <para><varname>make_memory
1342    [noaccess|undefined|defined|Definedifaddressable] &lt;addr&gt;
1343    [&lt;len&gt;]</varname> marks the range of &lt;len&gt; (default 1)
1344    bytes at &lt;addr&gt; as having the given status. Parameter
1345    <varname>noaccess</varname> marks the range as non-accessible, so
1346    Memcheck will report an error on any access to it.
1347    <varname>undefined</varname> or <varname>defined</varname> mark
1348    the area as accessible, but Memcheck regards the bytes in it
1349    respectively as having undefined or defined values.
1350    <varname>Definedifaddressable</varname> marks as defined, bytes in
1351    the range which are already addressible, but makes no change to
1352    the status of bytes in the range which are not addressible. Note
1353    that the first letter of <varname>Definedifaddressable</varname>
1354    is an uppercase D to avoid confusion with <varname>defined</varname>.
1355    </para>
1356
1357    <para>
1358    In the following example, the first byte of the
1359    <varname>string10</varname> is marked as defined:
1360    </para>
1361<programlisting><![CDATA[
1362(gdb) monitor make_memory defined 0x8049e28  1
1363(gdb) monitor get_vbits 0x8049e28 10
13640000ff00 ff00ff00 ff00
1365(gdb)
1366]]></programlisting>
1367  </listitem>
1368
1369  <listitem>
1370    <para><varname>check_memory [addressable|defined] &lt;addr&gt;
1371    [&lt;len&gt;]</varname> checks that the range of &lt;len&gt;
1372    (default 1) bytes at &lt;addr&gt; has the specified accessibility.
1373    It then outputs a description of &lt;addr&gt;. In the following
1374    example, a detailed description is available because the
1375    option <option>--read-var-info=yes</option> was given Valgrind at
1376    startup:
1377    </para>
1378<programlisting><![CDATA[
1379(gdb) monitor check_memory defined 0x8049e28  1
1380Address 0x8049E28 len 1 defined
1381==14698==  Location 0x8049e28 is 0 bytes inside string10[0],
1382==14698==  declared at prog.c:10, in frame #0 of thread 1
1383(gdb)
1384]]></programlisting>
1385  </listitem>
1386
1387  <listitem>
1388    <para><varname>leak_check [full*|summary]
1389                              [reachable|possibleleak*|definiteleak]
1390                              [increased*|changed|any]
1391          </varname>
1392    performs a leak check. The <varname>*</varname> in the arguments
1393    indicates the default value. </para>
1394
1395    <para> If the first argument is <varname>summary</varname>, only a
1396    summary of the leak search is given; otherwise a full leak report
1397    is produced.  A full leak report gives detailed information for
1398    each leak: the stack trace where the leaked blocks were allocated,
1399    the number of blocks leaked and their total size.  When a full
1400    report is requested, the next two arguments further specify what
1401    kind of leaks to report.  A leak's details are shown if they match
1402    both the second and third argument.
1403    </para>
1404
1405    <para>The second argument controls what kind of blocks are shown for
1406    a <varname>full</varname> leak search.  The
1407    value <varname>definiteleak</varname> specifies that only
1408    definitely leaked blocks should be shown.  The
1409    value <varname>possibleleak</varname> will also show possibly
1410    leaked blocks (those for which only an interior pointer was
1411    found).  The value
1412    <varname>reachable</varname> will show all block categories
1413    (reachable, possibly leaked, definitely leaked).
1414    </para>
1415
1416    <para>The third argument controls what kinds of changes are shown
1417    for a <varname>full</varname> leak search. The
1418    value <varname>increased</varname> specifies that only block
1419    allocation stacks with an increased number of leaked bytes or
1420    blocks since the previous leak check should be shown.  The
1421    value <varname>changed</varname> specifies that allocation stacks
1422    with any change since the previous leak check should be shown.
1423    The value <varname>any</varname> specifies that all leak entries
1424    should be shown, regardless of any increase or decrease.  When
1425    If <varname>increased</varname> or <varname>changed</varname> are
1426    specified, the leak report entries will show the delta relative to
1427    the previous leak report.
1428    </para>
1429
1430    <para>The following example shows usage of the
1431    <varname>leak_check monitor</varname> command on
1432    the <varname>memcheck/tests/leak-cases.c</varname> regression
1433    test. The first command outputs one entry having an increase in
1434    the leaked bytes.  The second command is the same as the first
1435    command, but uses the abbreviated forms accepted by GDB and the
1436    Valgrind gdbserver. It only outputs the summary information, as
1437    there was no increase since the previous leak search.</para>
1438<programlisting><![CDATA[
1439(gdb) monitor leak_check full possibleleak increased
1440==14729== 16 (+16) bytes in 1 (+1) blocks are possibly lost in loss record 13 of 16
1441==14729==    at 0x4006E9E: malloc (vg_replace_malloc.c:236)
1442==14729==    by 0x80484D5: mk (leak-cases.c:52)
1443==14729==    by 0x804855F: f (leak-cases.c:81)
1444==14729==    by 0x80488F5: main (leak-cases.c:107)
1445==14729==
1446==14729== LEAK SUMMARY:
1447==14729==    definitely lost: 32 (+0) bytes in 2 (+0) blocks
1448==14729==    indirectly lost: 16 (+0) bytes in 1 (+0) blocks
1449==14729==      possibly lost: 32 (+16) bytes in 2 (+1) blocks
1450==14729==    still reachable: 96 (+16) bytes in 6 (+1) blocks
1451==14729==         suppressed: 0 (+0) bytes in 0 (+0) blocks
1452==14729== Reachable blocks (those to which a pointer was found) are not shown.
1453==14729== To see them, add 'reachable any' args to leak_check
1454==14729==
1455(gdb) mo l
1456==14729== LEAK SUMMARY:
1457==14729==    definitely lost: 32 (+0) bytes in 2 (+0) blocks
1458==14729==    indirectly lost: 16 (+0) bytes in 1 (+0) blocks
1459==14729==      possibly lost: 32 (+0) bytes in 2 (+0) blocks
1460==14729==    still reachable: 96 (+0) bytes in 6 (+0) blocks
1461==14729==         suppressed: 0 (+0) bytes in 0 (+0) blocks
1462==14729== Reachable blocks (those to which a pointer was found) are not shown.
1463==14729== To see them, add 'reachable any' args to leak_check
1464==14729==
1465(gdb)
1466]]></programlisting>
1467    <para>Note that when using Valgrind's gdbserver, it is not
1468    necessary to rerun
1469    with <option>--leak-check=full</option>
1470    <option>--show-reachable=yes</option> to see the reachable
1471    blocks. You can obtain the same information without rerunning by
1472    using the GDB command <computeroutput>monitor leak_check full
1473    reachable any</computeroutput> (or, using
1474    abbreviation: <computeroutput>mo l f r a</computeroutput>).
1475    </para>
1476  </listitem>
1477</itemizedlist>
1478
1479</sect1>
1480
1481<sect1 id="mc-manual.clientreqs" xreflabel="Client requests">
1482<title>Client Requests</title>
1483
1484<para>The following client requests are defined in
1485<filename>memcheck.h</filename>.
1486See <filename>memcheck.h</filename> for exact details of their
1487arguments.</para>
1488
1489<itemizedlist>
1490
1491  <listitem>
1492    <para><varname>VALGRIND_MAKE_MEM_NOACCESS</varname>,
1493    <varname>VALGRIND_MAKE_MEM_UNDEFINED</varname> and
1494    <varname>VALGRIND_MAKE_MEM_DEFINED</varname>.
1495    These mark address ranges as completely inaccessible,
1496    accessible but containing undefined data, and accessible and
1497    containing defined data, respectively.</para>
1498  </listitem>
1499
1500  <listitem>
1501    <para><varname>VALGRIND_MAKE_MEM_DEFINED_IF_ADDRESSABLE</varname>.
1502    This is just like <varname>VALGRIND_MAKE_MEM_DEFINED</varname> but only
1503    affects those bytes that are already addressable.</para>
1504  </listitem>
1505
1506  <listitem>
1507    <para><varname>VALGRIND_CHECK_MEM_IS_ADDRESSABLE</varname> and
1508    <varname>VALGRIND_CHECK_MEM_IS_DEFINED</varname>: check immediately
1509    whether or not the given address range has the relevant property,
1510    and if not, print an error message.  Also, for the convenience of
1511    the client, returns zero if the relevant property holds; otherwise,
1512    the returned value is the address of the first byte for which the
1513    property is not true.  Always returns 0 when not run on
1514    Valgrind.</para>
1515  </listitem>
1516
1517  <listitem>
1518    <para><varname>VALGRIND_CHECK_VALUE_IS_DEFINED</varname>: a quick and easy
1519    way to find out whether Valgrind thinks a particular value
1520    (lvalue, to be precise) is addressable and defined.  Prints an error
1521    message if not.  It has no return value.</para>
1522  </listitem>
1523
1524  <listitem>
1525    <para><varname>VALGRIND_DO_LEAK_CHECK</varname>: does a full memory leak
1526    check (like <option>--leak-check=full</option>) right now.
1527    This is useful for incrementally checking for leaks between arbitrary
1528    places in the program's execution.  It has no return value.</para>
1529  </listitem>
1530
1531  <listitem>
1532    <para><varname>VALGRIND_DO_ADDED_LEAK_CHECK</varname>: same as
1533   <varname> VALGRIND_DO_LEAK_CHECK</varname> but only shows the
1534    entries for which there was an increase in leaked bytes or leaked
1535    number of blocks since the previous leak search.  It has no return
1536    value.</para>
1537  </listitem>
1538
1539  <listitem>
1540    <para><varname>VALGRIND_DO_CHANGED_LEAK_CHECK</varname>: same as
1541    <varname>VALGRIND_DO_LEAK_CHECK</varname> but only shows the
1542    entries for which there was an increase or decrease in leaked
1543    bytes or leaked number of blocks since the previous leak search. It
1544    has no return value.</para>
1545  </listitem>
1546
1547  <listitem>
1548    <para><varname>VALGRIND_DO_QUICK_LEAK_CHECK</varname>: like
1549    <varname>VALGRIND_DO_LEAK_CHECK</varname>, except it produces only a leak
1550    summary (like <option>--leak-check=summary</option>).
1551    It has no return value.</para>
1552  </listitem>
1553
1554  <listitem>
1555    <para><varname>VALGRIND_COUNT_LEAKS</varname>: fills in the four
1556    arguments with the number of bytes of memory found by the previous
1557    leak check to be leaked (i.e. the sum of direct leaks and indirect leaks),
1558    dubious, reachable and suppressed.  This is useful in test harness code,
1559    after calling <varname>VALGRIND_DO_LEAK_CHECK</varname> or
1560    <varname>VALGRIND_DO_QUICK_LEAK_CHECK</varname>.</para>
1561  </listitem>
1562
1563  <listitem>
1564    <para><varname>VALGRIND_COUNT_LEAK_BLOCKS</varname>: identical to
1565    <varname>VALGRIND_COUNT_LEAKS</varname> except that it returns the
1566    number of blocks rather than the number of bytes in each
1567    category.</para>
1568  </listitem>
1569
1570  <listitem>
1571    <para><varname>VALGRIND_GET_VBITS</varname> and
1572    <varname>VALGRIND_SET_VBITS</varname>: allow you to get and set the
1573    V (validity) bits for an address range.  You should probably only
1574    set V bits that you have got with
1575    <varname>VALGRIND_GET_VBITS</varname>.  Only for those who really
1576    know what they are doing.</para>
1577  </listitem>
1578
1579  <listitem>
1580    <para><varname>VALGRIND_CREATE_BLOCK</varname> and
1581    <varname>VALGRIND_DISCARD</varname>.  <varname>VALGRIND_CREATE_BLOCK</varname>
1582    takes an address, a number of bytes and a character string.  The
1583    specified address range is then associated with that string.  When
1584    Memcheck reports an invalid access to an address in the range, it
1585    will describe it in terms of this block rather than in terms of
1586    any other block it knows about.  Note that the use of this macro
1587    does not actually change the state of memory in any way -- it
1588    merely gives a name for the range.
1589    </para>
1590
1591    <para>At some point you may want Memcheck to stop reporting errors
1592    in terms of the block named
1593    by <varname>VALGRIND_CREATE_BLOCK</varname>.  To make this
1594    possible, <varname>VALGRIND_CREATE_BLOCK</varname> returns a
1595    "block handle", which is a C <varname>int</varname> value.  You
1596    can pass this block handle to <varname>VALGRIND_DISCARD</varname>.
1597    After doing so, Valgrind will no longer relate addressing errors
1598    in the specified range to the block.  Passing invalid handles to
1599    <varname>VALGRIND_DISCARD</varname> is harmless.
1600   </para>
1601  </listitem>
1602
1603</itemizedlist>
1604
1605</sect1>
1606
1607
1608
1609
1610<sect1 id="mc-manual.mempools" xreflabel="Memory Pools">
1611<title>Memory Pools: describing and working with custom allocators</title>
1612
1613<para>Some programs use custom memory allocators, often for performance
1614reasons.  Left to itself, Memcheck is unable to understand the
1615behaviour of custom allocation schemes as well as it understands the
1616standard allocators, and so may miss errors and leaks in your program.  What
1617this section describes is a way to give Memcheck enough of a description of
1618your custom allocator that it can make at least some sense of what is
1619happening.</para>
1620
1621<para>There are many different sorts of custom allocator, so Memcheck
1622attempts to reason about them using a loose, abstract model.  We
1623use the following terminology when describing custom allocation
1624systems:</para>
1625
1626<itemizedlist>
1627  <listitem>
1628    <para>Custom allocation involves a set of independent "memory pools".
1629    </para>
1630  </listitem>
1631  <listitem>
1632    <para>Memcheck's notion of a a memory pool consists of a single "anchor
1633    address" and a set of non-overlapping "chunks" associated with the
1634    anchor address.</para>
1635  </listitem>
1636  <listitem>
1637    <para>Typically a pool's anchor address is the address of a
1638    book-keeping "header" structure.</para>
1639  </listitem>
1640  <listitem>
1641    <para>Typically the pool's chunks are drawn from a contiguous
1642    "superblock" acquired through the system
1643    <function>malloc</function> or
1644    <function>mmap</function>.</para>
1645  </listitem>
1646
1647</itemizedlist>
1648
1649<para>Keep in mind that the last two points above say "typically": the
1650Valgrind mempool client request API is intentionally vague about the
1651exact structure of a mempool. There is no specific mention made of
1652headers or superblocks. Nevertheless, the following picture may help
1653elucidate the intention of the terms in the API:</para>
1654
1655<programlisting><![CDATA[
1656   "pool"
1657   (anchor address)
1658   |
1659   v
1660   +--------+---+
1661   | header | o |
1662   +--------+-|-+
1663              |
1664              v                  superblock
1665              +------+---+--------------+---+------------------+
1666              |      |rzB|  allocation  |rzB|                  |
1667              +------+---+--------------+---+------------------+
1668                         ^              ^
1669                         |              |
1670                       "addr"     "addr"+"size"
1671]]></programlisting>
1672
1673<para>
1674Note that the header and the superblock may be contiguous or
1675discontiguous, and there may be multiple superblocks associated with a
1676single header; such variations are opaque to Memcheck. The API
1677only requires that your allocation scheme can present sensible values
1678of "pool", "addr" and "size".</para>
1679
1680<para>
1681Typically, before making client requests related to mempools, a client
1682program will have allocated such a header and superblock for their
1683mempool, and marked the superblock NOACCESS using the
1684<varname>VALGRIND_MAKE_MEM_NOACCESS</varname> client request.</para>
1685
1686<para>
1687When dealing with mempools, the goal is to maintain a particular
1688invariant condition: that Memcheck believes the unallocated portions
1689of the pool's superblock (including redzones) are NOACCESS. To
1690maintain this invariant, the client program must ensure that the
1691superblock starts out in that state; Memcheck cannot make it so, since
1692Memcheck never explicitly learns about the superblock of a pool, only
1693the allocated chunks within the pool.</para>
1694
1695<para>
1696Once the header and superblock for a pool are established and properly
1697marked, there are a number of client requests programs can use to
1698inform Memcheck about changes to the state of a mempool:</para>
1699
1700<itemizedlist>
1701
1702  <listitem>
1703    <para>
1704    <varname>VALGRIND_CREATE_MEMPOOL(pool, rzB, is_zeroed)</varname>:
1705    This request registers the address <varname>pool</varname> as the anchor
1706    address for a memory pool. It also provides a size
1707    <varname>rzB</varname>, specifying how large the redzones placed around
1708    chunks allocated from the pool should be. Finally, it provides an
1709    <varname>is_zeroed</varname> argument that specifies whether the pool's
1710    chunks are zeroed (more precisely: defined) when allocated.
1711    </para>
1712    <para>
1713    Upon completion of this request, no chunks are associated with the
1714    pool.  The request simply tells Memcheck that the pool exists, so that
1715    subsequent calls can refer to it as a pool.
1716    </para>
1717  </listitem>
1718
1719  <listitem>
1720    <para><varname>VALGRIND_DESTROY_MEMPOOL(pool)</varname>:
1721    This request tells Memcheck that a pool is being torn down. Memcheck
1722    then removes all records of chunks associated with the pool, as well
1723    as its record of the pool's existence. While destroying its records of
1724    a mempool, Memcheck resets the redzones of any live chunks in the pool
1725    to NOACCESS.
1726    </para>
1727  </listitem>
1728
1729  <listitem>
1730    <para><varname>VALGRIND_MEMPOOL_ALLOC(pool, addr, size)</varname>:
1731    This request informs Memcheck that a <varname>size</varname>-byte chunk
1732    has been allocated at <varname>addr</varname>, and associates the chunk with the
1733    specified
1734    <varname>pool</varname>. If the pool was created with nonzero
1735    <varname>rzB</varname> redzones, Memcheck will mark the
1736    <varname>rzB</varname> bytes before and after the chunk as NOACCESS. If
1737    the pool was created with the <varname>is_zeroed</varname> argument set,
1738    Memcheck will mark the chunk as DEFINED, otherwise Memcheck will mark
1739    the chunk as UNDEFINED.
1740    </para>
1741  </listitem>
1742
1743  <listitem>
1744    <para><varname>VALGRIND_MEMPOOL_FREE(pool, addr)</varname>:
1745    This request informs Memcheck that the chunk at <varname>addr</varname>
1746    should no longer be considered allocated. Memcheck will mark the chunk
1747    associated with <varname>addr</varname> as NOACCESS, and delete its
1748    record of the chunk's existence.
1749    </para>
1750  </listitem>
1751
1752  <listitem>
1753    <para><varname>VALGRIND_MEMPOOL_TRIM(pool, addr, size)</varname>:
1754    This request trims the chunks associated with <varname>pool</varname>.
1755    The request only operates on chunks associated with
1756    <varname>pool</varname>. Trimming is formally defined as:</para>
1757    <itemizedlist>
1758      <listitem>
1759        <para> All chunks entirely inside the range
1760        <varname>addr..(addr+size-1)</varname> are preserved.</para>
1761      </listitem>
1762      <listitem>
1763        <para>All chunks entirely outside the range
1764        <varname>addr..(addr+size-1)</varname> are discarded, as though
1765        <varname>VALGRIND_MEMPOOL_FREE</varname> was called on them. </para>
1766      </listitem>
1767      <listitem>
1768        <para>All other chunks must intersect with the range
1769        <varname>addr..(addr+size-1)</varname>; areas outside the
1770        intersection are marked as NOACCESS, as though they had been
1771        independently freed with
1772        <varname>VALGRIND_MEMPOOL_FREE</varname>.</para>
1773      </listitem>
1774    </itemizedlist>
1775    <para>This is a somewhat rare request, but can be useful in
1776    implementing the type of mass-free operations common in custom
1777    LIFO allocators.</para>
1778  </listitem>
1779
1780  <listitem>
1781    <para><varname>VALGRIND_MOVE_MEMPOOL(poolA, poolB)</varname>: This
1782    request informs Memcheck that the pool previously anchored at
1783    address <varname>poolA</varname> has moved to anchor address
1784    <varname>poolB</varname>.  This is a rare request, typically only needed
1785    if you <function>realloc</function> the header of a mempool.</para>
1786    <para>No memory-status bits are altered by this request.</para>
1787  </listitem>
1788
1789  <listitem>
1790    <para>
1791    <varname>VALGRIND_MEMPOOL_CHANGE(pool, addrA, addrB,
1792    size)</varname>: This request informs Memcheck that the chunk
1793    previously allocated at address <varname>addrA</varname> within
1794    <varname>pool</varname> has been moved and/or resized, and should be
1795    changed to cover the region <varname>addrB..(addrB+size-1)</varname>. This
1796    is a rare request, typically only needed if you
1797    <function>realloc</function> a superblock or wish to extend a chunk
1798    without changing its memory-status bits.
1799    </para>
1800    <para>No memory-status bits are altered by this request.
1801    </para>
1802  </listitem>
1803
1804  <listitem>
1805    <para><varname>VALGRIND_MEMPOOL_EXISTS(pool)</varname>:
1806    This request informs the caller whether or not Memcheck is currently
1807    tracking a mempool at anchor address <varname>pool</varname>. It
1808    evaluates to 1 when there is a mempool associated with that address, 0
1809    otherwise. This is a rare request, only useful in circumstances when
1810    client code might have lost track of the set of active mempools.
1811    </para>
1812  </listitem>
1813
1814</itemizedlist>
1815
1816</sect1>
1817
1818
1819
1820
1821
1822
1823
1824<sect1 id="mc-manual.mpiwrap" xreflabel="MPI Wrappers">
1825<title>Debugging MPI Parallel Programs with Valgrind</title>
1826
1827<para>Memcheck supports debugging of distributed-memory applications
1828which use the MPI message passing standard.  This support consists of a
1829library of wrapper functions for the
1830<computeroutput>PMPI_*</computeroutput> interface.  When incorporated
1831into the application's address space, either by direct linking or by
1832<computeroutput>LD_PRELOAD</computeroutput>, the wrappers intercept
1833calls to <computeroutput>PMPI_Send</computeroutput>,
1834<computeroutput>PMPI_Recv</computeroutput>, etc.  They then
1835use client requests to inform Memcheck of memory state changes caused
1836by the function being wrapped.  This reduces the number of false
1837positives that Memcheck otherwise typically reports for MPI
1838applications.</para>
1839
1840<para>The wrappers also take the opportunity to carefully check
1841size and definedness of buffers passed as arguments to MPI functions, hence
1842detecting errors such as passing undefined data to
1843<computeroutput>PMPI_Send</computeroutput>, or receiving data into a
1844buffer which is too small.</para>
1845
1846<para>Unlike most of the rest of Valgrind, the wrapper library is subject to a
1847BSD-style license, so you can link it into any code base you like.
1848See the top of <computeroutput>mpi/libmpiwrap.c</computeroutput>
1849for license details.</para>
1850
1851
1852<sect2 id="mc-manual.mpiwrap.build" xreflabel="Building MPI Wrappers">
1853<title>Building and installing the wrappers</title>
1854
1855<para> The wrapper library will be built automatically if possible.
1856Valgrind's configure script will look for a suitable
1857<computeroutput>mpicc</computeroutput> to build it with.  This must be
1858the same <computeroutput>mpicc</computeroutput> you use to build the
1859MPI application you want to debug.  By default, Valgrind tries
1860<computeroutput>mpicc</computeroutput>, but you can specify a
1861different one by using the configure-time option
1862<option>--with-mpicc</option>.  Currently the
1863wrappers are only buildable with
1864<computeroutput>mpicc</computeroutput>s which are based on GNU
1865GCC or Intel's C++ Compiler.</para>
1866
1867<para>Check that the configure script prints a line like this:</para>
1868
1869<programlisting><![CDATA[
1870checking for usable MPI2-compliant mpicc and mpi.h... yes, mpicc
1871]]></programlisting>
1872
1873<para>If it says <computeroutput>... no</computeroutput>, your
1874<computeroutput>mpicc</computeroutput> has failed to compile and link
1875a test MPI2 program.</para>
1876
1877<para>If the configure test succeeds, continue in the usual way with
1878<computeroutput>make</computeroutput> and <computeroutput>make
1879install</computeroutput>.  The final install tree should then contain
1880<computeroutput>libmpiwrap-&lt;platform&gt;.so</computeroutput>.
1881</para>
1882
1883<para>Compile up a test MPI program (eg, MPI hello-world) and try
1884this:</para>
1885
1886<programlisting><![CDATA[
1887LD_PRELOAD=$prefix/lib/valgrind/libmpiwrap-<platform>.so   \
1888           mpirun [args] $prefix/bin/valgrind ./hello
1889]]></programlisting>
1890
1891<para>You should see something similar to the following</para>
1892
1893<programlisting><![CDATA[
1894valgrind MPI wrappers 31901: Active for pid 31901
1895valgrind MPI wrappers 31901: Try MPIWRAP_DEBUG=help for possible options
1896]]></programlisting>
1897
1898<para>repeated for every process in the group.  If you do not see
1899these, there is an build/installation problem of some kind.</para>
1900
1901<para> The MPI functions to be wrapped are assumed to be in an ELF
1902shared object with soname matching
1903<computeroutput>libmpi.so*</computeroutput>.  This is known to be
1904correct at least for Open MPI and Quadrics MPI, and can easily be
1905changed if required.</para>
1906</sect2>
1907
1908
1909<sect2 id="mc-manual.mpiwrap.gettingstarted"
1910       xreflabel="Getting started with MPI Wrappers">
1911<title>Getting started</title>
1912
1913<para>Compile your MPI application as usual, taking care to link it
1914using the same <computeroutput>mpicc</computeroutput> that your
1915Valgrind build was configured with.</para>
1916
1917<para>
1918Use the following basic scheme to run your application on Valgrind with
1919the wrappers engaged:</para>
1920
1921<programlisting><![CDATA[
1922MPIWRAP_DEBUG=[wrapper-args]                                  \
1923   LD_PRELOAD=$prefix/lib/valgrind/libmpiwrap-<platform>.so   \
1924   mpirun [mpirun-args]                                       \
1925   $prefix/bin/valgrind [valgrind-args]                       \
1926   [application] [app-args]
1927]]></programlisting>
1928
1929<para>As an alternative to
1930<computeroutput>LD_PRELOAD</computeroutput>ing
1931<computeroutput>libmpiwrap-&lt;platform&gt;.so</computeroutput>, you can
1932simply link it to your application if desired.  This should not disturb
1933native behaviour of your application in any way.</para>
1934</sect2>
1935
1936
1937<sect2 id="mc-manual.mpiwrap.controlling"
1938       xreflabel="Controlling the MPI Wrappers">
1939<title>Controlling the wrapper library</title>
1940
1941<para>Environment variable
1942<computeroutput>MPIWRAP_DEBUG</computeroutput> is consulted at
1943startup.  The default behaviour is to print a starting banner</para>
1944
1945<programlisting><![CDATA[
1946valgrind MPI wrappers 16386: Active for pid 16386
1947valgrind MPI wrappers 16386: Try MPIWRAP_DEBUG=help for possible options
1948]]></programlisting>
1949
1950<para> and then be relatively quiet.</para>
1951
1952<para>You can give a list of comma-separated options in
1953<computeroutput>MPIWRAP_DEBUG</computeroutput>.  These are</para>
1954
1955<itemizedlist>
1956  <listitem>
1957    <para><computeroutput>verbose</computeroutput>:
1958    show entries/exits of all wrappers.  Also show extra
1959    debugging info, such as the status of outstanding
1960    <computeroutput>MPI_Request</computeroutput>s resulting
1961    from uncompleted <computeroutput>MPI_Irecv</computeroutput>s.</para>
1962  </listitem>
1963  <listitem>
1964    <para><computeroutput>quiet</computeroutput>:
1965    opposite of <computeroutput>verbose</computeroutput>, only print
1966    anything when the wrappers want
1967    to report a detected programming error, or in case of catastrophic
1968    failure of the wrappers.</para>
1969  </listitem>
1970  <listitem>
1971    <para><computeroutput>warn</computeroutput>:
1972    by default, functions which lack proper wrappers
1973    are not commented on, just silently
1974    ignored.  This causes a warning to be printed for each unwrapped
1975    function used, up to a maximum of three warnings per function.</para>
1976  </listitem>
1977  <listitem>
1978    <para><computeroutput>strict</computeroutput>:
1979    print an error message and abort the program if
1980    a function lacking a wrapper is used.</para>
1981  </listitem>
1982</itemizedlist>
1983
1984<para> If you want to use Valgrind's XML output facility
1985(<option>--xml=yes</option>), you should pass
1986<computeroutput>quiet</computeroutput> in
1987<computeroutput>MPIWRAP_DEBUG</computeroutput> so as to get rid of any
1988extraneous printing from the wrappers.</para>
1989
1990</sect2>
1991
1992
1993<sect2 id="mc-manual.mpiwrap.limitations.functions"
1994       xreflabel="Functions: Abilities and Limitations">
1995<title>Functions</title>
1996
1997<para>All MPI2 functions except
1998<computeroutput>MPI_Wtick</computeroutput>,
1999<computeroutput>MPI_Wtime</computeroutput> and
2000<computeroutput>MPI_Pcontrol</computeroutput> have wrappers.  The
2001first two are not wrapped because they return a
2002<computeroutput>double</computeroutput>, which Valgrind's
2003function-wrap mechanism cannot handle (but it could easily be
2004extended to do so).  <computeroutput>MPI_Pcontrol</computeroutput> cannot be
2005wrapped as it has variable arity:
2006<computeroutput>int MPI_Pcontrol(const int level, ...)</computeroutput></para>
2007
2008<para>Most functions are wrapped with a default wrapper which does
2009nothing except complain or abort if it is called, depending on
2010settings in <computeroutput>MPIWRAP_DEBUG</computeroutput> listed
2011above.  The following functions have "real", do-something-useful
2012wrappers:</para>
2013
2014<programlisting><![CDATA[
2015PMPI_Send PMPI_Bsend PMPI_Ssend PMPI_Rsend
2016
2017PMPI_Recv PMPI_Get_count
2018
2019PMPI_Isend PMPI_Ibsend PMPI_Issend PMPI_Irsend
2020
2021PMPI_Irecv
2022PMPI_Wait PMPI_Waitall
2023PMPI_Test PMPI_Testall
2024
2025PMPI_Iprobe PMPI_Probe
2026
2027PMPI_Cancel
2028
2029PMPI_Sendrecv
2030
2031PMPI_Type_commit PMPI_Type_free
2032
2033PMPI_Pack PMPI_Unpack
2034
2035PMPI_Bcast PMPI_Gather PMPI_Scatter PMPI_Alltoall
2036PMPI_Reduce PMPI_Allreduce PMPI_Op_create
2037
2038PMPI_Comm_create PMPI_Comm_dup PMPI_Comm_free PMPI_Comm_rank PMPI_Comm_size
2039
2040PMPI_Error_string
2041PMPI_Init PMPI_Initialized PMPI_Finalize
2042]]></programlisting>
2043
2044<para> A few functions such as
2045<computeroutput>PMPI_Address</computeroutput> are listed as
2046<computeroutput>HAS_NO_WRAPPER</computeroutput>.  They have no wrapper
2047at all as there is nothing worth checking, and giving a no-op wrapper
2048would reduce performance for no reason.</para>
2049
2050<para> Note that the wrapper library itself can itself generate large
2051numbers of calls to the MPI implementation, especially when walking
2052complex types.  The most common functions called are
2053<computeroutput>PMPI_Extent</computeroutput>,
2054<computeroutput>PMPI_Type_get_envelope</computeroutput>,
2055<computeroutput>PMPI_Type_get_contents</computeroutput>, and
2056<computeroutput>PMPI_Type_free</computeroutput>.  </para>
2057</sect2>
2058
2059<sect2 id="mc-manual.mpiwrap.limitations.types"
2060       xreflabel="Types: Abilities and Limitations">
2061<title>Types</title>
2062
2063<para> MPI-1.1 structured types are supported, and walked exactly.
2064The currently supported combiners are
2065<computeroutput>MPI_COMBINER_NAMED</computeroutput>,
2066<computeroutput>MPI_COMBINER_CONTIGUOUS</computeroutput>,
2067<computeroutput>MPI_COMBINER_VECTOR</computeroutput>,
2068<computeroutput>MPI_COMBINER_HVECTOR</computeroutput>
2069<computeroutput>MPI_COMBINER_INDEXED</computeroutput>,
2070<computeroutput>MPI_COMBINER_HINDEXED</computeroutput> and
2071<computeroutput>MPI_COMBINER_STRUCT</computeroutput>.  This should
2072cover all MPI-1.1 types.  The mechanism (function
2073<computeroutput>walk_type</computeroutput>) should extend easily to
2074cover MPI2 combiners.</para>
2075
2076<para>MPI defines some named structured types
2077(<computeroutput>MPI_FLOAT_INT</computeroutput>,
2078<computeroutput>MPI_DOUBLE_INT</computeroutput>,
2079<computeroutput>MPI_LONG_INT</computeroutput>,
2080<computeroutput>MPI_2INT</computeroutput>,
2081<computeroutput>MPI_SHORT_INT</computeroutput>,
2082<computeroutput>MPI_LONG_DOUBLE_INT</computeroutput>) which are pairs
2083of some basic type and a C <computeroutput>int</computeroutput>.
2084Unfortunately the MPI specification makes it impossible to look inside
2085these types and see where the fields are.  Therefore these wrappers
2086assume the types are laid out as <computeroutput>struct { float val;
2087int loc; }</computeroutput> (for
2088<computeroutput>MPI_FLOAT_INT</computeroutput>), etc, and act
2089accordingly.  This appears to be correct at least for Open MPI 1.0.2
2090and for Quadrics MPI.</para>
2091
2092<para>If <computeroutput>strict</computeroutput> is an option specified
2093in <computeroutput>MPIWRAP_DEBUG</computeroutput>, the application
2094will abort if an unhandled type is encountered.  Otherwise, the
2095application will print a warning message and continue.</para>
2096
2097<para>Some effort is made to mark/check memory ranges corresponding to
2098arrays of values in a single pass.  This is important for performance
2099since asking Valgrind to mark/check any range, no matter how small,
2100carries quite a large constant cost.  This optimisation is applied to
2101arrays of primitive types (<computeroutput>double</computeroutput>,
2102<computeroutput>float</computeroutput>,
2103<computeroutput>int</computeroutput>,
2104<computeroutput>long</computeroutput>, <computeroutput>long
2105long</computeroutput>, <computeroutput>short</computeroutput>,
2106<computeroutput>char</computeroutput>, and <computeroutput>long
2107double</computeroutput> on platforms where <computeroutput>sizeof(long
2108double) == 8</computeroutput>).  For arrays of all other types, the
2109wrappers handle each element individually and so there can be a very
2110large performance cost.</para>
2111
2112</sect2>
2113
2114
2115<sect2 id="mc-manual.mpiwrap.writingwrappers"
2116       xreflabel="Writing new MPI Wrappers">
2117<title>Writing new wrappers</title>
2118
2119<para>
2120For the most part the wrappers are straightforward.  The only
2121significant complexity arises with nonblocking receives.</para>
2122
2123<para>The issue is that <computeroutput>MPI_Irecv</computeroutput>
2124states the recv buffer and returns immediately, giving a handle
2125(<computeroutput>MPI_Request</computeroutput>) for the transaction.
2126Later the user will have to poll for completion with
2127<computeroutput>MPI_Wait</computeroutput> etc, and when the
2128transaction completes successfully, the wrappers have to paint the
2129recv buffer.  But the recv buffer details are not presented to
2130<computeroutput>MPI_Wait</computeroutput> -- only the handle is.  The
2131library therefore maintains a shadow table which associates
2132uncompleted <computeroutput>MPI_Request</computeroutput>s with the
2133corresponding buffer address/count/type.  When an operation completes,
2134the table is searched for the associated address/count/type info, and
2135memory is marked accordingly.</para>
2136
2137<para>Access to the table is guarded by a (POSIX pthreads) lock, so as
2138to make the library thread-safe.</para>
2139
2140<para>The table is allocated with
2141<computeroutput>malloc</computeroutput> and never
2142<computeroutput>free</computeroutput>d, so it will show up in leak
2143checks.</para>
2144
2145<para>Writing new wrappers should be fairly easy.  The source file is
2146<computeroutput>mpi/libmpiwrap.c</computeroutput>.  If possible,
2147find an existing wrapper for a function of similar behaviour to the
2148one you want to wrap, and use it as a starting point.  The wrappers
2149are organised in sections in the same order as the MPI 1.1 spec, to
2150aid navigation.  When adding a wrapper, remember to comment out the
2151definition of the default wrapper in the long list of defaults at the
2152bottom of the file (do not remove it, just comment it out).</para>
2153</sect2>
2154
2155<sect2 id="mc-manual.mpiwrap.whattoexpect"
2156       xreflabel="What to expect with MPI Wrappers">
2157<title>What to expect when using the wrappers</title>
2158
2159<para>The wrappers should reduce Memcheck's false-error rate on MPI
2160applications.  Because the wrapping is done at the MPI interface,
2161there will still potentially be a large number of errors reported in
2162the MPI implementation below the interface.  The best you can do is
2163try to suppress them.</para>
2164
2165<para>You may also find that the input-side (buffer
2166length/definedness) checks find errors in your MPI use, for example
2167passing too short a buffer to
2168<computeroutput>MPI_Recv</computeroutput>.</para>
2169
2170<para>Functions which are not wrapped may increase the false
2171error rate.  A possible approach is to run with
2172<computeroutput>MPI_DEBUG</computeroutput> containing
2173<computeroutput>warn</computeroutput>.  This will show you functions
2174which lack proper wrappers but which are nevertheless used.  You can
2175then write wrappers for them.
2176</para>
2177
2178<para>A known source of potential false errors are the
2179<computeroutput>PMPI_Reduce</computeroutput> family of functions, when
2180using a custom (user-defined) reduction function.  In a reduction
2181operation, each node notionally sends data to a "central point" which
2182uses the specified reduction function to merge the data items into a
2183single item.  Hence, in general, data is passed between nodes and fed
2184to the reduction function, but the wrapper library cannot mark the
2185transferred data as initialised before it is handed to the reduction
2186function, because all that happens "inside" the
2187<computeroutput>PMPI_Reduce</computeroutput> call.  As a result you
2188may see false positives reported in your reduction function.</para>
2189
2190</sect2>
2191
2192</sect1>
2193
2194
2195
2196
2197
2198</chapter>
2199