• 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[ <!ENTITY % vg-entities SYSTEM "../../docs/xml/vg-entities.xml"> %vg-entities; ]>
5
6
7<chapter id="drd-manual" xreflabel="DRD: a thread error detector">
8  <title>DRD: a thread error detector</title>
9
10<para>To use this tool, you must specify
11<option>--tool=drd</option>
12on the Valgrind command line.</para>
13
14
15<sect1 id="drd-manual.overview" xreflabel="Overview">
16<title>Overview</title>
17
18<para>
19DRD is a Valgrind tool for detecting errors in multithreaded C and C++
20programs. The tool works for any program that uses the POSIX threading
21primitives or that uses threading concepts built on top of the POSIX threading
22primitives.
23</para>
24
25<sect2 id="drd-manual.mt-progr-models" xreflabel="MT-progr-models">
26<title>Multithreaded Programming Paradigms</title>
27
28<para>
29There are two possible reasons for using multithreading in a program:
30<itemizedlist>
31  <listitem>
32    <para>
33      To model concurrent activities. Assigning one thread to each activity
34      can be a great simplification compared to multiplexing the states of
35      multiple activities in a single thread. This is why most server software
36      and embedded software is multithreaded.
37    </para>
38  </listitem>
39  <listitem>
40    <para>
41      To use multiple CPU cores simultaneously for speeding up
42      computations. This is why many High Performance Computing (HPC)
43      applications are multithreaded.
44    </para>
45  </listitem>
46</itemizedlist>
47</para>
48
49<para>
50Multithreaded programs can use one or more of the following programming
51paradigms. Which paradigm is appropriate depends e.g. on the application type.
52Some examples of multithreaded programming paradigms are:
53<itemizedlist>
54  <listitem>
55    <para>
56      Locking. Data that is shared over threads is protected from concurrent
57      accesses via locking. E.g. the POSIX threads library, the Qt library
58      and the Boost.Thread library support this paradigm directly.
59    </para>
60  </listitem>
61  <listitem>
62    <para>
63      Message passing. No data is shared between threads, but threads exchange
64      data by passing messages to each other. Examples of implementations of
65      the message passing paradigm are MPI and CORBA.
66    </para>
67  </listitem>
68  <listitem>
69    <para>
70      Automatic parallelization. A compiler converts a sequential program into
71      a multithreaded program. The original program may or may not contain
72      parallelization hints. One example of such parallelization hints is the
73      OpenMP standard. In this standard a set of directives are defined which
74      tell a compiler how to parallelize a C, C++ or Fortran program. OpenMP
75      is well suited for computational intensive applications. As an example,
76      an open source image processing software package is using OpenMP to
77      maximize performance on systems with multiple CPU
78      cores. GCC supports the
79      OpenMP standard from version 4.2.0 on.
80    </para>
81  </listitem>
82  <listitem>
83    <para>
84      Software Transactional Memory (STM). Any data that is shared between
85      threads is updated via transactions. After each transaction it is
86      verified whether there were any conflicting transactions. If there were
87      conflicts, the transaction is aborted, otherwise it is committed. This
88      is a so-called optimistic approach. There is a prototype of the Intel C++
89      Compiler available that supports STM. Research about the addition of
90      STM support to GCC is ongoing.
91    </para>
92  </listitem>
93</itemizedlist>
94</para>
95
96<para>
97DRD supports any combination of multithreaded programming paradigms as
98long as the implementation of these paradigms is based on the POSIX
99threads primitives. DRD however does not support programs that use
100e.g. Linux' futexes directly. Attempts to analyze such programs with
101DRD will cause DRD to report many false positives.
102</para>
103
104</sect2>
105
106
107<sect2 id="drd-manual.pthreads-model" xreflabel="Pthreads-model">
108<title>POSIX Threads Programming Model</title>
109
110<para>
111POSIX threads, also known as Pthreads, is the most widely available
112threading library on Unix systems.
113</para>
114
115<para>
116The POSIX threads programming model is based on the following abstractions:
117<itemizedlist>
118  <listitem>
119    <para>
120      A shared address space. All threads running within the same
121      process share the same address space. All data, whether shared or
122      not, is identified by its address.
123    </para>
124  </listitem>
125  <listitem>
126    <para>
127      Regular load and store operations, which allow to read values
128      from or to write values to the memory shared by all threads
129      running in the same process.
130    </para>
131  </listitem>
132  <listitem>
133    <para>
134      Atomic store and load-modify-store operations. While these are
135      not mentioned in the POSIX threads standard, most
136      microprocessors support atomic memory operations.
137    </para>
138  </listitem>
139  <listitem>
140    <para>
141      Threads. Each thread represents a concurrent activity.
142    </para>
143  </listitem>
144  <listitem>
145    <para>
146      Synchronization objects and operations on these synchronization
147      objects. The following types of synchronization objects have been
148      defined in the POSIX threads standard: mutexes, condition variables,
149      semaphores, reader-writer synchronization objects, barriers and
150      spinlocks.
151    </para>
152  </listitem>
153</itemizedlist>
154</para>
155
156<para>
157Which source code statements generate which memory accesses depends on
158the <emphasis>memory model</emphasis> of the programming language being
159used. There is not yet a definitive memory model for the C and C++
160languages. For a draft memory model, see also the document
161<ulink url="http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2338.html">
162WG21/N2338: Concurrency memory model compiler consequences</ulink>.
163</para>
164
165<para>
166For more information about POSIX threads, see also the Single UNIX
167Specification version 3, also known as
168<ulink url="http://www.opengroup.org/onlinepubs/000095399/idx/threads.html">
169IEEE Std 1003.1</ulink>.
170</para>
171
172</sect2>
173
174
175<sect2 id="drd-manual.mt-problems" xreflabel="MT-Problems">
176<title>Multithreaded Programming Problems</title>
177
178<para>
179Depending on which multithreading paradigm is being used in a program,
180one or more of the following problems can occur:
181<itemizedlist>
182  <listitem>
183    <para>
184      Data races. One or more threads access the same memory location without
185      sufficient locking. Most but not all data races are programming errors
186      and are the cause of subtle and hard-to-find bugs.
187    </para>
188  </listitem>
189  <listitem>
190    <para>
191      Lock contention. One thread blocks the progress of one or more other
192      threads by holding a lock too long.
193    </para>
194  </listitem>
195  <listitem>
196    <para>
197      Improper use of the POSIX threads API. Most implementations of the POSIX
198      threads API have been optimized for runtime speed. Such implementations
199      will not complain on certain errors, e.g. when a mutex is being unlocked
200      by another thread than the thread that obtained a lock on the mutex.
201    </para>
202  </listitem>
203  <listitem>
204    <para>
205      Deadlock. A deadlock occurs when two or more threads wait for
206      each other indefinitely.
207    </para>
208  </listitem>
209  <listitem>
210    <para>
211      False sharing. If threads that run on different processor cores
212      access different variables located in the same cache line
213      frequently, this will slow down the involved threads a lot due
214      to frequent exchange of cache lines.
215    </para>
216  </listitem>
217</itemizedlist>
218</para>
219
220<para>
221Although the likelihood of the occurrence of data races can be reduced
222through a disciplined programming style, a tool for automatic
223detection of data races is a necessity when developing multithreaded
224software. DRD can detect these, as well as lock contention and
225improper use of the POSIX threads API.
226</para>
227
228</sect2>
229
230
231<sect2 id="drd-manual.data-race-detection" xreflabel="data-race-detection">
232<title>Data Race Detection</title>
233
234<para>
235The result of load and store operations performed by a multithreaded program
236depends on the order in which memory operations are performed. This order is
237determined by:
238<orderedlist>
239  <listitem>
240    <para>
241      All memory operations performed by the same thread are performed in
242      <emphasis>program order</emphasis>, that is, the order determined by the
243      program source code and the results of previous load operations.
244    </para>
245  </listitem>
246  <listitem>
247    <para>
248      Synchronization operations determine certain ordering constraints on
249      memory operations performed by different threads. These ordering
250      constraints are called the <emphasis>synchronization order</emphasis>.
251    </para>
252  </listitem>
253</orderedlist>
254The combination of program order and synchronization order is called the
255<emphasis>happens-before relationship</emphasis>. This concept was first
256defined by S. Adve et al in the paper <emphasis>Detecting data races on weak
257memory systems</emphasis>, ACM SIGARCH Computer Architecture News, v.19 n.3,
258p.234-243, May 1991.
259</para>
260
261<para>
262Two memory operations <emphasis>conflict</emphasis> if both operations are
263performed by different threads, refer to the same memory location and at least
264one of them is a store operation.
265</para>
266
267<para>
268A multithreaded program is <emphasis>data-race free</emphasis> if all
269conflicting memory accesses are ordered by synchronization
270operations.
271</para>
272
273<para>
274A well known way to ensure that a multithreaded program is data-race
275free is to ensure that a locking discipline is followed. It is e.g.
276possible to associate a mutex with each shared data item, and to hold
277a lock on the associated mutex while the shared data is accessed.
278</para>
279
280<para>
281All programs that follow a locking discipline are data-race free, but not all
282data-race free programs follow a locking discipline. There exist multithreaded
283programs where access to shared data is arbitrated via condition variables,
284semaphores or barriers. As an example, a certain class of HPC applications
285consists of a sequence of computation steps separated in time by barriers, and
286where these barriers are the only means of synchronization. Although there are
287many conflicting memory accesses in such applications and although such
288applications do not make use mutexes, most of these applications do not
289contain data races.
290</para>
291
292<para>
293There exist two different approaches for verifying the correctness of
294multithreaded programs at runtime. The approach of the so-called Eraser
295algorithm is to verify whether all shared memory accesses follow a consistent
296locking strategy. And the happens-before data race detectors verify directly
297whether all interthread memory accesses are ordered by synchronization
298operations. While the last approach is more complex to implement, and while it
299is more sensitive to OS scheduling, it is a general approach that works for
300all classes of multithreaded programs. An important advantage of
301happens-before data race detectors is that these do not report any false
302positives.
303</para>
304
305<para>
306DRD is based on the happens-before algorithm.
307</para>
308
309</sect2>
310
311
312</sect1>
313
314
315<sect1 id="drd-manual.using-drd" xreflabel="Using DRD">
316<title>Using DRD</title>
317
318<sect2 id="drd-manual.options" xreflabel="DRD Command-line Options">
319<title>DRD Command-line Options</title>
320
321<para>The following command-line options are available for controlling the
322behavior of the DRD tool itself:</para>
323
324<!-- start of xi:include in the manpage -->
325<variablelist id="drd.opts.list">
326  <varlistentry>
327    <term>
328      <option><![CDATA[--check-stack-var=<yes|no> [default: no]]]></option>
329    </term>
330    <listitem>
331      <para>
332        Controls whether DRD detects data races on stack
333        variables. Verifying stack variables is disabled by default because
334        most programs do not share stack variables over threads.
335      </para>
336    </listitem>
337  </varlistentry>
338  <varlistentry>
339    <term>
340      <option><![CDATA[--exclusive-threshold=<n> [default: off]]]></option>
341    </term>
342    <listitem>
343      <para>
344        Print an error message if any mutex or writer lock has been
345        held longer than the time specified in milliseconds. This
346        option enables the detection of lock contention.
347      </para>
348    </listitem>
349  </varlistentry>
350  <varlistentry>
351    <term>
352      <option><![CDATA[--join-list-vol=<n> [default: 10]]]></option>
353    </term>
354    <listitem>
355      <para>
356        Data races that occur between a statement at the end of one thread
357	and another thread can be missed if memory access information is
358	discarded immediately after a thread has been joined. This option
359	allows to specify for how many joined threads memory access information
360	should be retained.
361      </para>
362    </listitem>
363  </varlistentry>
364  <varlistentry>
365    <term>
366      <option>
367        <![CDATA[--first-race-only=<yes|no> [default: no]]]>
368      </option>
369    </term>
370    <listitem>
371      <para>
372        Whether to report only the first data race that has been detected on a
373        memory location or all data races that have been detected on a memory
374        location.
375      </para>
376    </listitem>
377  </varlistentry>
378  <varlistentry>
379    <term>
380      <option>
381        <![CDATA[--free-is-write=<yes|no> [default: no]]]>
382      </option>
383    </term>
384    <listitem>
385      <para>
386        Whether to report races between accessing memory and freeing
387        memory. Enabling this option may cause DRD to run slightly
388        slower. Notes:</para>
389      <itemizedlist>
390	<listitem>
391	  <para>
392	    Don't enable this option when using custom memory allocators
393	    that use
394	    the <computeroutput>VG_USERREQ__MALLOCLIKE_BLOCK</computeroutput>
395	    and <computeroutput>VG_USERREQ__FREELIKE_BLOCK</computeroutput>
396	    because that would result in false positives.
397	  </para>
398	</listitem>
399	<listitem>
400	  <para>Don't enable this option when using reference-counted
401	    objects because that will result in false positives, even when
402	    that code has been annotated properly with
403	    <computeroutput>ANNOTATE_HAPPENS_BEFORE</computeroutput>
404	    and <computeroutput>ANNOTATE_HAPPENS_AFTER</computeroutput>. See
405	    e.g.  the output of the following command for an example:
406	    <computeroutput>valgrind --tool=drd --free-is-write=yes
407	      drd/tests/annotate_smart_pointer</computeroutput>.
408	  </para>
409	</listitem>
410      </itemizedlist>
411    </listitem>
412  </varlistentry>
413  <varlistentry>
414    <term>
415      <option>
416        <![CDATA[--report-signal-unlocked=<yes|no> [default: yes]]]>
417      </option>
418    </term>
419    <listitem>
420      <para>
421        Whether to report calls to
422        <function>pthread_cond_signal</function> and
423        <function>pthread_cond_broadcast</function> where the mutex
424        associated with the signal through
425        <function>pthread_cond_wait</function> or
426        <function>pthread_cond_timed_wait</function>is not locked at
427        the time the signal is sent.  Sending a signal without holding
428        a lock on the associated mutex is a common programming error
429        which can cause subtle race conditions and unpredictable
430        behavior. There exist some uncommon synchronization patterns
431        however where it is safe to send a signal without holding a
432        lock on the associated mutex.
433      </para>
434    </listitem>
435  </varlistentry>
436  <varlistentry>
437    <term>
438      <option><![CDATA[--segment-merging=<yes|no> [default: yes]]]></option>
439    </term>
440    <listitem>
441      <para>
442        Controls segment merging. Segment merging is an algorithm to
443        limit memory usage of the data race detection
444        algorithm. Disabling segment merging may improve the accuracy
445        of the so-called 'other segments' displayed in race reports
446        but can also trigger an out of memory error.
447      </para>
448    </listitem>
449  </varlistentry>
450  <varlistentry>
451    <term>
452      <option><![CDATA[--segment-merging-interval=<n> [default: 10]]]></option>
453    </term>
454    <listitem>
455      <para>
456        Perform segment merging only after the specified number of new
457        segments have been created. This is an advanced configuration option
458        that allows to choose whether to minimize DRD's memory usage by
459        choosing a low value or to let DRD run faster by choosing a slightly
460        higher value. The optimal value for this parameter depends on the
461        program being analyzed. The default value works well for most programs.
462      </para>
463    </listitem>
464  </varlistentry>
465  <varlistentry>
466    <term>
467      <option><![CDATA[--shared-threshold=<n> [default: off]]]></option>
468    </term>
469    <listitem>
470      <para>
471        Print an error message if a reader lock has been held longer
472        than the specified time (in milliseconds). This option enables
473        the detection of lock contention.
474      </para>
475    </listitem>
476  </varlistentry>
477  <varlistentry>
478    <term>
479      <option><![CDATA[--show-confl-seg=<yes|no> [default: yes]]]></option>
480    </term>
481    <listitem>
482      <para>
483         Show conflicting segments in race reports. Since this
484         information can help to find the cause of a data race, this
485         option is enabled by default. Disabling this option makes the
486         output of DRD more compact.
487      </para>
488    </listitem>
489  </varlistentry>
490  <varlistentry>
491    <term>
492      <option><![CDATA[--show-stack-usage=<yes|no> [default: no]]]></option>
493    </term>
494    <listitem>
495      <para>
496        Print stack usage at thread exit time. When a program creates a large
497        number of threads it becomes important to limit the amount of virtual
498        memory allocated for thread stacks. This option makes it possible to
499        observe how much stack memory has been used by each thread of the
500        client program. Note: the DRD tool itself allocates some temporary
501        data on the client thread stack. The space necessary for this
502        temporary data must be allocated by the client program when it
503        allocates stack memory, but is not included in stack usage reported by
504        DRD.
505      </para>
506    </listitem>
507  </varlistentry>
508</variablelist>
509<!-- end of xi:include in the manpage -->
510
511<!-- start of xi:include in the manpage -->
512<para>
513The following options are available for monitoring the behavior of the
514client program:
515</para>
516
517<variablelist id="drd.debugopts.list">
518  <varlistentry>
519    <term>
520      <option><![CDATA[--trace-addr=<address> [default: none]]]></option>
521    </term>
522    <listitem>
523      <para>
524        Trace all load and store activity for the specified
525        address. This option may be specified more than once.
526      </para>
527    </listitem>
528  </varlistentry>
529  <varlistentry>
530    <term>
531      <option><![CDATA[--ptrace-addr=<address> [default: none]]]></option>
532    </term>
533    <listitem>
534      <para>
535        Trace all load and store activity for the specified address and keep
536        doing that even after the memory at that address has been freed and
537        reallocated.
538      </para>
539    </listitem>
540  </varlistentry>
541  <varlistentry>
542    <term>
543      <option><![CDATA[--trace-alloc=<yes|no> [default: no]]]></option>
544    </term>
545    <listitem>
546      <para>
547        Trace all memory allocations and deallocations. May produce a huge
548        amount of output.
549      </para>
550    </listitem>
551  </varlistentry>
552  <varlistentry>
553    <term>
554      <option><![CDATA[--trace-barrier=<yes|no> [default: no]]]></option>
555    </term>
556    <listitem>
557      <para>
558        Trace all barrier activity.
559      </para>
560    </listitem>
561  </varlistentry>
562  <varlistentry>
563    <term>
564      <option><![CDATA[--trace-cond=<yes|no> [default: no]]]></option>
565    </term>
566    <listitem>
567      <para>
568        Trace all condition variable activity.
569      </para>
570    </listitem>
571  </varlistentry>
572  <varlistentry>
573    <term>
574      <option><![CDATA[--trace-fork-join=<yes|no> [default: no]]]></option>
575    </term>
576    <listitem>
577      <para>
578        Trace all thread creation and all thread termination events.
579      </para>
580    </listitem>
581  </varlistentry>
582  <varlistentry>
583    <term>
584      <option><![CDATA[--trace-hb=<yes|no> [default: no]]]></option>
585    </term>
586    <listitem>
587      <para>
588        Trace execution of the <literal>ANNOTATE_HAPPENS_BEFORE()</literal>,
589	<literal>ANNOTATE_HAPPENS_AFTER()</literal> and
590	<literal>ANNOTATE_HAPPENS_DONE()</literal> client requests.
591      </para>
592    </listitem>
593  </varlistentry>
594  <varlistentry>
595    <term>
596      <option><![CDATA[--trace-mutex=<yes|no> [default: no]]]></option>
597    </term>
598    <listitem>
599      <para>
600        Trace all mutex activity.
601      </para>
602    </listitem>
603  </varlistentry>
604  <varlistentry>
605    <term>
606      <option><![CDATA[--trace-rwlock=<yes|no> [default: no]]]></option>
607    </term>
608    <listitem>
609      <para>
610         Trace all reader-writer lock activity.
611      </para>
612    </listitem>
613  </varlistentry>
614  <varlistentry>
615    <term>
616      <option><![CDATA[--trace-semaphore=<yes|no> [default: no]]]></option>
617    </term>
618    <listitem>
619      <para>
620        Trace all semaphore activity.
621      </para>
622    </listitem>
623  </varlistentry>
624</variablelist>
625<!-- end of xi:include in the manpage -->
626
627</sect2>
628
629
630<sect2 id="drd-manual.data-races" xreflabel="Data Races">
631<title>Detected Errors: Data Races</title>
632
633<para>
634DRD prints a message every time it detects a data race. Please keep
635the following in mind when interpreting DRD's output:
636<itemizedlist>
637  <listitem>
638    <para>
639      Every thread is assigned a <emphasis>thread ID</emphasis> by the DRD
640      tool. A thread ID is a number. Thread ID's start at one and are never
641      recycled.
642    </para>
643  </listitem>
644  <listitem>
645    <para>
646      The term <emphasis>segment</emphasis> refers to a consecutive
647      sequence of load, store and synchronization operations, all
648      issued by the same thread. A segment always starts and ends at a
649      synchronization operation. Data race analysis is performed
650      between segments instead of between individual load and store
651      operations because of performance reasons.
652    </para>
653  </listitem>
654  <listitem>
655    <para>
656      There are always at least two memory accesses involved in a data
657      race. Memory accesses involved in a data race are called
658      <emphasis>conflicting memory accesses</emphasis>. DRD prints a
659      report for each memory access that conflicts with a past memory
660      access.
661    </para>
662  </listitem>
663</itemizedlist>
664</para>
665
666<para>
667Below you can find an example of a message printed by DRD when it
668detects a data race:
669</para>
670<programlisting><![CDATA[
671$ valgrind --tool=drd --read-var-info=yes drd/tests/rwlock_race
672...
673==9466== Thread 3:
674==9466== Conflicting load by thread 3 at 0x006020b8 size 4
675==9466==    at 0x400B6C: thread_func (rwlock_race.c:29)
676==9466==    by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186)
677==9466==    by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so)
678==9466==    by 0x53250CC: clone (in /lib64/libc-2.8.so)
679==9466== Location 0x6020b8 is 0 bytes inside local var "s_racy"
680==9466== declared at rwlock_race.c:18, in frame #0 of thread 3
681==9466== Other segment start (thread 2)
682==9466==    at 0x4C2847D: pthread_rwlock_rdlock* (drd_pthread_intercepts.c:813)
683==9466==    by 0x400B6B: thread_func (rwlock_race.c:28)
684==9466==    by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186)
685==9466==    by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so)
686==9466==    by 0x53250CC: clone (in /lib64/libc-2.8.so)
687==9466== Other segment end (thread 2)
688==9466==    at 0x4C28B54: pthread_rwlock_unlock* (drd_pthread_intercepts.c:912)
689==9466==    by 0x400B84: thread_func (rwlock_race.c:30)
690==9466==    by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186)
691==9466==    by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so)
692==9466==    by 0x53250CC: clone (in /lib64/libc-2.8.so)
693...
694]]></programlisting>
695
696<para>
697The above report has the following meaning:
698<itemizedlist>
699  <listitem>
700    <para>
701      The number in the column on the left is the process ID of the
702      process being analyzed by DRD.
703    </para>
704  </listitem>
705  <listitem>
706    <para>
707      The first line ("Thread 3") tells you the thread ID for
708      the thread in which context the data race has been detected.
709    </para>
710  </listitem>
711  <listitem>
712    <para>
713      The next line tells which kind of operation was performed (load or
714      store) and by which thread. On the same line the start address and the
715      number of bytes involved in the conflicting access are also displayed.
716    </para>
717  </listitem>
718  <listitem>
719    <para>
720      Next, the call stack of the conflicting access is displayed. If
721      your program has been compiled with debug information
722      (<option>-g</option>), this call stack will include file names and
723      line numbers. The two
724      bottommost frames in this call stack (<function>clone</function>
725      and <function>start_thread</function>) show how the NPTL starts
726      a thread. The third frame
727      (<function>vg_thread_wrapper</function>) is added by DRD. The
728      fourth frame (<function>thread_func</function>) is the first
729      interesting line because it shows the thread entry point, that
730      is the function that has been passed as the third argument to
731      <function>pthread_create</function>.
732    </para>
733  </listitem>
734  <listitem>
735    <para>
736      Next, the allocation context for the conflicting address is
737      displayed. For dynamically allocated data the allocation call
738      stack is shown. For static variables and stack variables the
739      allocation context is only shown when the option
740      <option>--read-var-info=yes</option> has been
741      specified. Otherwise DRD will print <computeroutput>Allocation
742      context: unknown</computeroutput>.
743    </para>
744  </listitem>
745  <listitem>
746    <para>
747      A conflicting access involves at least two memory accesses. For
748      one of these accesses an exact call stack is displayed, and for
749      the other accesses an approximate call stack is displayed,
750      namely the start and the end of the segments of the other
751      accesses. This information can be interpreted as follows:
752      <orderedlist>
753        <listitem>
754          <para>
755            Start at the bottom of both call stacks, and count the
756            number stack frames with identical function name, file
757            name and line number. In the above example the three
758            bottommost frames are identical
759            (<function>clone</function>,
760            <function>start_thread</function> and
761            <function>vg_thread_wrapper</function>).
762          </para>
763        </listitem>
764        <listitem>
765          <para>
766            The next higher stack frame in both call stacks now tells
767            you between in which source code region the other memory
768            access happened. The above output tells that the other
769            memory access involved in the data race happened between
770            source code lines 28 and 30 in file
771            <computeroutput>rwlock_race.c</computeroutput>.
772          </para>
773        </listitem>
774      </orderedlist>
775    </para>
776  </listitem>
777</itemizedlist>
778</para>
779
780</sect2>
781
782
783<sect2 id="drd-manual.lock-contention" xreflabel="Lock Contention">
784<title>Detected Errors: Lock Contention</title>
785
786<para>
787Threads must be able to make progress without being blocked for too long by
788other threads. Sometimes a thread has to wait until a mutex or reader-writer
789synchronization object is unlocked by another thread. This is called
790<emphasis>lock contention</emphasis>.
791</para>
792
793<para>
794Lock contention causes delays. Such delays should be as short as
795possible. The two command line options
796<literal>--exclusive-threshold=&lt;n&gt;</literal> and
797<literal>--shared-threshold=&lt;n&gt;</literal> make it possible to
798detect excessive lock contention by making DRD report any lock that
799has been held longer than the specified threshold. An example:
800</para>
801<programlisting><![CDATA[
802$ valgrind --tool=drd --exclusive-threshold=10 drd/tests/hold_lock -i 500
803...
804==10668== Acquired at:
805==10668==    at 0x4C267C8: pthread_mutex_lock (drd_pthread_intercepts.c:395)
806==10668==    by 0x400D92: main (hold_lock.c:51)
807==10668== Lock on mutex 0x7fefffd50 was held during 503 ms (threshold: 10 ms).
808==10668==    at 0x4C26ADA: pthread_mutex_unlock (drd_pthread_intercepts.c:441)
809==10668==    by 0x400DB5: main (hold_lock.c:55)
810...
811]]></programlisting>
812
813<para>
814The <literal>hold_lock</literal> test program holds a lock as long as
815specified by the <literal>-i</literal> (interval) argument. The DRD
816output reports that the lock acquired at line 51 in source file
817<literal>hold_lock.c</literal> and released at line 55 was held during
818503 ms, while a threshold of 10 ms was specified to DRD.
819</para>
820
821</sect2>
822
823
824<sect2 id="drd-manual.api-checks" xreflabel="API Checks">
825<title>Detected Errors: Misuse of the POSIX threads API</title>
826
827<para>
828  DRD is able to detect and report the following misuses of the POSIX
829  threads API:
830  <itemizedlist>
831    <listitem>
832      <para>
833        Passing the address of one type of synchronization object
834        (e.g. a mutex) to a POSIX API call that expects a pointer to
835        another type of synchronization object (e.g. a condition
836        variable).
837      </para>
838    </listitem>
839    <listitem>
840      <para>
841        Attempts to unlock a mutex that has not been locked.
842      </para>
843    </listitem>
844    <listitem>
845      <para>
846        Attempts to unlock a mutex that was locked by another thread.
847      </para>
848    </listitem>
849    <listitem>
850      <para>
851        Attempts to lock a mutex of type
852        <literal>PTHREAD_MUTEX_NORMAL</literal> or a spinlock
853        recursively.
854      </para>
855    </listitem>
856    <listitem>
857      <para>
858        Destruction or deallocation of a locked mutex.
859      </para>
860    </listitem>
861    <listitem>
862      <para>
863        Sending a signal to a condition variable while no lock is held
864        on the mutex associated with the condition variable.
865      </para>
866    </listitem>
867    <listitem>
868      <para>
869        Calling <function>pthread_cond_wait</function> on a mutex
870        that is not locked, that is locked by another thread or that
871        has been locked recursively.
872      </para>
873    </listitem>
874    <listitem>
875      <para>
876        Associating two different mutexes with a condition variable
877        through <function>pthread_cond_wait</function>.
878      </para>
879    </listitem>
880    <listitem>
881      <para>
882        Destruction or deallocation of a condition variable that is
883        being waited upon.
884      </para>
885    </listitem>
886    <listitem>
887      <para>
888        Destruction or deallocation of a locked reader-writer synchronization
889        object.
890      </para>
891    </listitem>
892    <listitem>
893      <para>
894        Attempts to unlock a reader-writer synchronization object that was not
895        locked by the calling thread.
896      </para>
897    </listitem>
898    <listitem>
899      <para>
900        Attempts to recursively lock a reader-writer synchronization object
901        exclusively.
902      </para>
903    </listitem>
904    <listitem>
905      <para>
906        Attempts to pass the address of a user-defined reader-writer
907        synchronization object to a POSIX threads function.
908      </para>
909    </listitem>
910    <listitem>
911      <para>
912        Attempts to pass the address of a POSIX reader-writer synchronization
913        object to one of the annotations for user-defined reader-writer
914        synchronization objects.
915      </para>
916    </listitem>
917    <listitem>
918      <para>
919        Reinitialization of a mutex, condition variable, reader-writer
920        lock, semaphore or barrier.
921      </para>
922    </listitem>
923    <listitem>
924      <para>
925        Destruction or deallocation of a semaphore or barrier that is
926        being waited upon.
927      </para>
928    </listitem>
929    <listitem>
930      <para>
931        Missing synchronization between barrier wait and barrier destruction.
932      </para>
933    </listitem>
934    <listitem>
935      <para>
936        Exiting a thread without first unlocking the spinlocks, mutexes or
937        reader-writer synchronization objects that were locked by that thread.
938      </para>
939    </listitem>
940    <listitem>
941      <para>
942        Passing an invalid thread ID to <function>pthread_join</function>
943        or <function>pthread_cancel</function>.
944      </para>
945    </listitem>
946  </itemizedlist>
947</para>
948
949</sect2>
950
951
952<sect2 id="drd-manual.clientreqs" xreflabel="Client requests">
953<title>Client Requests</title>
954
955<para>
956Just as for other Valgrind tools it is possible to let a client program
957interact with the DRD tool through client requests. In addition to the
958client requests several macros have been defined that allow to use the
959client requests in a convenient way.
960</para>
961
962<para>
963The interface between client programs and the DRD tool is defined in
964the header file <literal>&lt;valgrind/drd.h&gt;</literal>. The
965available macros and client requests are:
966<itemizedlist>
967  <listitem>
968    <para>
969      The macro <literal>DRD_GET_VALGRIND_THREADID</literal> and the
970      corresponding client
971      request <varname>VG_USERREQ__DRD_GET_VALGRIND_THREAD_ID</varname>.
972      Query the thread ID that has been assigned by the Valgrind core to the
973      thread executing this client request. Valgrind's thread ID's start at
974      one and are recycled in case a thread stops.
975    </para>
976  </listitem>
977  <listitem>
978    <para>
979      The macro <literal>DRD_GET_DRD_THREADID</literal> and the corresponding
980      client request <varname>VG_USERREQ__DRD_GET_DRD_THREAD_ID</varname>.
981      Query the thread ID that has been assigned by DRD to the thread
982      executing this client request. These are the thread ID's reported by DRD
983      in data race reports and in trace messages. DRD's thread ID's start at
984      one and are never recycled.
985    </para>
986  </listitem>
987  <listitem>
988    <para>
989      The macros <literal>DRD_IGNORE_VAR(x)</literal>,
990      <literal>ANNOTATE_TRACE_MEMORY(&amp;x)</literal> and the corresponding
991      client request <varname>VG_USERREQ__DRD_START_SUPPRESSION</varname>. Some
992      applications contain intentional races. There exist e.g. applications
993      where the same value is assigned to a shared variable from two different
994      threads. It may be more convenient to suppress such races than to solve
995      these. This client request allows to suppress such races.
996    </para>
997  </listitem>
998  <listitem>
999    <para>
1000      The macro <literal>DRD_STOP_IGNORING_VAR(x)</literal> and the
1001      corresponding client request
1002      <varname>VG_USERREQ__DRD_FINISH_SUPPRESSION</varname>. Tell DRD
1003      to no longer ignore data races for the address range that was suppressed
1004      either via the macro <literal>DRD_IGNORE_VAR(x)</literal> or via the
1005      client request <varname>VG_USERREQ__DRD_START_SUPPRESSION</varname>.
1006    </para>
1007  </listitem>
1008  <listitem>
1009    <para>
1010      The macro <literal>DRD_TRACE_VAR(x)</literal>. Trace all load and store
1011      activity for the address range starting at <literal>&amp;x</literal> and
1012      occupying <literal>sizeof(x)</literal> bytes. When DRD reports a data
1013      race on a specified variable, and it's not immediately clear which
1014      source code statements triggered the conflicting accesses, it can be
1015      very helpful to trace all activity on the offending memory location.
1016    </para>
1017  </listitem>
1018  <listitem>
1019    <para>
1020      The macro <literal>DRD_STOP_TRACING_VAR(x)</literal>. Stop tracing load
1021      and store activity for the address range starting
1022      at <literal>&amp;x</literal> and occupying <literal>sizeof(x)</literal>
1023      bytes.
1024    </para>
1025  </listitem>
1026  <listitem>
1027    <para>
1028      The macro <literal>ANNOTATE_TRACE_MEMORY(&amp;x)</literal>. Trace all
1029      load and store activity that touches at least the single byte at the
1030      address <literal>&amp;x</literal>.
1031    </para>
1032  </listitem>
1033  <listitem>
1034    <para>
1035      The client request <varname>VG_USERREQ__DRD_START_TRACE_ADDR</varname>,
1036      which allows to trace all load and store activity for the specified
1037      address range.
1038    </para>
1039  </listitem>
1040  <listitem>
1041    <para>
1042      The client
1043      request <varname>VG_USERREQ__DRD_STOP_TRACE_ADDR</varname>. Do no longer
1044      trace load and store activity for the specified address range.
1045    </para>
1046  </listitem>
1047  <listitem>
1048    <para>
1049      The macro <literal>ANNOTATE_HAPPENS_BEFORE(addr)</literal> tells DRD to
1050      insert a mark. Insert this macro just after an access to the variable at
1051      the specified address has been performed.
1052    </para>
1053  </listitem>
1054  <listitem>
1055    <para>
1056      The macro <literal>ANNOTATE_HAPPENS_AFTER(addr)</literal> tells DRD that
1057      the next access to the variable at the specified address should be
1058      considered to have happened after the access just before the latest
1059      <literal>ANNOTATE_HAPPENS_BEFORE(addr)</literal> annotation that
1060      references the same variable. The purpose of these two macros is to tell
1061      DRD about the order of inter-thread memory accesses implemented via
1062      atomic memory operations. See
1063      also <literal>drd/tests/annotate_smart_pointer.cpp</literal> for an
1064      example.
1065    </para>
1066  </listitem>
1067  <listitem>
1068    <para>
1069      The macro <literal>ANNOTATE_RWLOCK_CREATE(rwlock)</literal> tells DRD
1070      that the object at address <literal>rwlock</literal> is a
1071      reader-writer synchronization object that is not a
1072      <literal>pthread_rwlock_t</literal> synchronization object.  See
1073      also <literal>drd/tests/annotate_rwlock.c</literal> for an example.
1074    </para>
1075  </listitem>
1076  <listitem>
1077    <para>
1078      The macro <literal>ANNOTATE_RWLOCK_DESTROY(rwlock)</literal> tells DRD
1079      that the reader-writer synchronization object at
1080      address <literal>rwlock</literal> has been destroyed.
1081    </para>
1082  </listitem>
1083  <listitem>
1084    <para>
1085      The macro <literal>ANNOTATE_WRITERLOCK_ACQUIRED(rwlock)</literal> tells
1086      DRD that a writer lock has been acquired on the reader-writer
1087      synchronization object at address <literal>rwlock</literal>.
1088    </para>
1089  </listitem>
1090  <listitem>
1091    <para>
1092      The macro <literal>ANNOTATE_READERLOCK_ACQUIRED(rwlock)</literal> tells
1093      DRD that a reader lock has been acquired on the reader-writer
1094      synchronization object at address <literal>rwlock</literal>.
1095    </para>
1096  </listitem>
1097  <listitem>
1098    <para>
1099      The macro <literal>ANNOTATE_RWLOCK_ACQUIRED(rwlock, is_w)</literal>
1100      tells DRD that a writer lock (when <literal>is_w != 0</literal>) or that
1101      a reader lock (when <literal>is_w == 0</literal>) has been acquired on
1102      the reader-writer synchronization object at
1103      address <literal>rwlock</literal>.
1104    </para>
1105  </listitem>
1106  <listitem>
1107    <para>
1108      The macro <literal>ANNOTATE_WRITERLOCK_RELEASED(rwlock)</literal> tells
1109      DRD that a writer lock has been released on the reader-writer
1110      synchronization object at address <literal>rwlock</literal>.
1111    </para>
1112  </listitem>
1113  <listitem>
1114    <para>
1115      The macro <literal>ANNOTATE_READERLOCK_RELEASED(rwlock)</literal> tells
1116      DRD that a reader lock has been released on the reader-writer
1117      synchronization object at address <literal>rwlock</literal>.
1118    </para>
1119  </listitem>
1120  <listitem>
1121    <para>
1122      The macro <literal>ANNOTATE_RWLOCK_RELEASED(rwlock, is_w)</literal>
1123      tells DRD that a writer lock (when <literal>is_w != 0</literal>) or that
1124      a reader lock (when <literal>is_w == 0</literal>) has been released on
1125      the reader-writer synchronization object at
1126      address <literal>rwlock</literal>.
1127    </para>
1128  </listitem>
1129  <listitem>
1130    <para>
1131      The macro <literal>ANNOTATE_BARRIER_INIT(barrier, count,
1132      reinitialization_allowed)</literal> tells DRD that a new barrier object
1133      at the address <literal>barrier</literal> has been initialized,
1134      that <literal>count</literal> threads participate in each barrier and
1135      also whether or not barrier reinitialization without intervening
1136      destruction should be reported as an error. See
1137      also <literal>drd/tests/annotate_barrier.c</literal> for an example.
1138    </para>
1139  </listitem>
1140  <listitem>
1141    <para>
1142      The macro <literal>ANNOTATE_BARRIER_DESTROY(barrier)</literal>
1143      tells DRD that a barrier object is about to be destroyed.
1144    </para>
1145  </listitem>
1146  <listitem>
1147    <para>
1148      The macro <literal>ANNOTATE_BARRIER_WAIT_BEFORE(barrier)</literal>
1149      tells DRD that waiting for a barrier will start.
1150    </para>
1151  </listitem>
1152  <listitem>
1153    <para>
1154      The macro <literal>ANNOTATE_BARRIER_WAIT_AFTER(barrier)</literal>
1155      tells DRD that waiting for a barrier has finished.
1156    </para>
1157  </listitem>
1158  <listitem>
1159    <para>
1160      The macro <literal>ANNOTATE_BENIGN_RACE_SIZED(addr, size,
1161      descr)</literal> tells DRD that any races detected on the specified
1162      address are benign and hence should not be
1163      reported. The <literal>descr</literal> argument is ignored but can be
1164      used to document why data races on <literal>addr</literal> are benign.
1165    </para>
1166  </listitem>
1167  <listitem>
1168    <para>
1169      The macro <literal>ANNOTATE_BENIGN_RACE_STATIC(var, descr)</literal>
1170      tells DRD that any races detected on the specified static variable are
1171      benign and hence should not be reported. The <literal>descr</literal>
1172      argument is ignored but can be used to document why data races
1173      on <literal>var</literal> are benign. Note: this macro can only be
1174      used in C++ programs and not in C programs.
1175    </para>
1176  </listitem>
1177  <listitem>
1178    <para>
1179      The macro <literal>ANNOTATE_IGNORE_READS_BEGIN</literal> tells
1180      DRD to ignore all memory loads performed by the current thread.
1181    </para>
1182  </listitem>
1183  <listitem>
1184    <para>
1185      The macro <literal>ANNOTATE_IGNORE_READS_END</literal> tells
1186      DRD to stop ignoring the memory loads performed by the current thread.
1187    </para>
1188  </listitem>
1189  <listitem>
1190    <para>
1191      The macro <literal>ANNOTATE_IGNORE_WRITES_BEGIN</literal> tells
1192      DRD to ignore all memory stores performed by the current thread.
1193    </para>
1194  </listitem>
1195  <listitem>
1196    <para>
1197      The macro <literal>ANNOTATE_IGNORE_WRITES_END</literal> tells
1198      DRD to stop ignoring the memory stores performed by the current thread.
1199    </para>
1200  </listitem>
1201  <listitem>
1202    <para>
1203      The macro <literal>ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN</literal> tells
1204      DRD to ignore all memory accesses performed by the current thread.
1205    </para>
1206  </listitem>
1207  <listitem>
1208    <para>
1209      The macro <literal>ANNOTATE_IGNORE_READS_AND_WRITES_END</literal> tells
1210      DRD to stop ignoring the memory accesses performed by the current thread.
1211    </para>
1212  </listitem>
1213  <listitem>
1214    <para>
1215      The macro <literal>ANNOTATE_NEW_MEMORY(addr, size)</literal> tells
1216      DRD that the specified memory range has been allocated by a custom
1217      memory allocator in the client program and that the client program
1218      will start using this memory range.
1219    </para>
1220  </listitem>
1221  <listitem>
1222    <para>
1223      The macro <literal>ANNOTATE_THREAD_NAME(name)</literal> tells DRD to
1224      associate the specified name with the current thread and to include this
1225      name in the error messages printed by DRD.
1226    </para>
1227  </listitem>
1228  <listitem>
1229    <para>
1230      The macros <literal>VALGRIND_MALLOCLIKE_BLOCK</literal> and
1231      <literal>VALGRIND_FREELIKE_BLOCK</literal> from the Valgrind core are
1232      implemented;  they are described in
1233      <xref linkend="manual-core-adv.clientreq"/>.
1234    </para>
1235  </listitem>
1236</itemizedlist>
1237</para>
1238
1239<para>
1240Note: if you compiled Valgrind yourself, the header file
1241<literal>&lt;valgrind/drd.h&gt;</literal> will have been installed in
1242the directory <literal>/usr/include</literal> by the command
1243<literal>make install</literal>. If you obtained Valgrind by
1244installing it as a package however, you will probably have to install
1245another package with a name like <literal>valgrind-devel</literal>
1246before Valgrind's header files are available.
1247</para>
1248
1249</sect2>
1250
1251
1252<sect2 id="drd-manual.C++11" xreflabel="C++11">
1253<title>Debugging C++11 Programs</title>
1254
1255<para>If you want to use the C++11 class std::thread you will need to do the
1256  following to annotate the std::shared_ptr&lt;&gt; objects used in the
1257  implementation of that class:
1258<itemizedlist>
1259  <listitem>
1260    <para>Add the following code at the start of a common header or at the
1261    start of each source file, before any C++ header files are included:</para>
1262    <programlisting>
1263#include &lt;valgrind/drd.h&gt;
1264#define _GLIBCXX_SYNCHRONIZATION_HAPPENS_BEFORE(addr) ANNOTATE_HAPPENS_BEFORE(addr)
1265#define _GLIBCXX_SYNCHRONIZATION_HAPPENS_AFTER(addr)  ANNOTATE_HAPPENS_AFTER(addr)
1266</programlisting>
1267  </listitem>
1268  <listitem>
1269    <para>Download the gcc source code and from source file
1270    libstdc++-v3/src/c++11/thread.cc copy the implementation of the
1271    <computeroutput>execute_native_thread_routine()</computeroutput>
1272    and <computeroutput>std::thread::_M_start_thread()</computeroutput>
1273    functions into a source file that is linked with your application. Make
1274    sure that also in this source file the
1275    _GLIBCXX_SYNCHRONIZATION_HAPPENS_*() macros are defined properly.</para>
1276  </listitem>
1277</itemizedlist>
1278</para>
1279<para>For more information, see also <emphasis>The
1280GNU C++ Library Manual, Debugging Support</emphasis>
1281(<ulink url="http://gcc.gnu.org/onlinedocs/libstdc++/manual/debug.html">http://gcc.gnu.org/onlinedocs/libstdc++/manual/debug.html</ulink>).</para>
1282
1283</sect2>
1284
1285
1286<sect2 id="drd-manual.gnome" xreflabel="GNOME">
1287<title>Debugging GNOME Programs</title>
1288
1289<para>
1290GNOME applications use the threading primitives provided by the
1291<computeroutput>glib</computeroutput> and
1292<computeroutput>gthread</computeroutput> libraries. These libraries
1293are built on top of POSIX threads, and hence are directly supported by
1294DRD. Please keep in mind that you have to call
1295<function>g_thread_init</function> before creating any threads, or
1296DRD will report several data races on glib functions. See also the
1297<ulink
1298url="http://library.gnome.org/devel/glib/stable/glib-Threads.html">GLib
1299Reference Manual</ulink> for more information about
1300<function>g_thread_init</function>.
1301</para>
1302
1303<para>
1304One of the many facilities provided by the <literal>glib</literal>
1305library is a block allocator, called <literal>g_slice</literal>. You
1306have to disable this block allocator when using DRD by adding the
1307following to the shell environment variables:
1308<literal>G_SLICE=always-malloc</literal>. See also the <ulink
1309url="http://library.gnome.org/devel/glib/stable/glib-Memory-Slices.html">GLib
1310Reference Manual</ulink> for more information.
1311</para>
1312
1313</sect2>
1314
1315
1316<sect2 id="drd-manual.boost.thread" xreflabel="Boost.Thread">
1317<title>Debugging Boost.Thread Programs</title>
1318
1319<para>
1320The Boost.Thread library is the threading library included with the
1321cross-platform Boost Libraries. This threading library is an early
1322implementation of the upcoming C++0x threading library.
1323</para>
1324
1325<para>
1326Applications that use the Boost.Thread library should run fine under DRD.
1327</para>
1328
1329<para>
1330More information about Boost.Thread can be found here:
1331<itemizedlist>
1332  <listitem>
1333    <para>
1334      Anthony Williams, <ulink
1335      url="http://www.boost.org/doc/libs/1_37_0/doc/html/thread.html">Boost.Thread</ulink>
1336      Library Documentation, Boost website, 2007.
1337    </para>
1338  </listitem>
1339  <listitem>
1340    <para>
1341      Anthony Williams, <ulink
1342      url="http://www.ddj.com/cpp/211600441">What's New in Boost
1343      Threads?</ulink>, Recent changes to the Boost Thread library,
1344      Dr. Dobbs Magazine, October 2008.
1345    </para>
1346  </listitem>
1347</itemizedlist>
1348</para>
1349
1350</sect2>
1351
1352
1353<sect2 id="drd-manual.openmp" xreflabel="OpenMP">
1354<title>Debugging OpenMP Programs</title>
1355
1356<para>
1357OpenMP stands for <emphasis>Open Multi-Processing</emphasis>. The OpenMP
1358standard consists of a set of compiler directives for C, C++ and Fortran
1359programs that allows a compiler to transform a sequential program into a
1360parallel program. OpenMP is well suited for HPC applications and allows to
1361work at a higher level compared to direct use of the POSIX threads API. While
1362OpenMP ensures that the POSIX API is used correctly, OpenMP programs can still
1363contain data races. So it definitely makes sense to verify OpenMP programs
1364with a thread checking tool.
1365</para>
1366
1367<para>
1368DRD supports OpenMP shared-memory programs generated by GCC. GCC
1369supports OpenMP since version 4.2.0.  GCC's runtime support
1370for OpenMP programs is provided by a library called
1371<literal>libgomp</literal>. The synchronization primitives implemented
1372in this library use Linux' futex system call directly, unless the
1373library has been configured with the
1374<literal>--disable-linux-futex</literal> option. DRD only supports
1375libgomp libraries that have been configured with this option and in
1376which symbol information is present. For most Linux distributions this
1377means that you will have to recompile GCC. See also the script
1378<literal>drd/scripts/download-and-build-gcc</literal> in the
1379Valgrind source tree for an example of how to compile GCC. You will
1380also have to make sure that the newly compiled
1381<literal>libgomp.so</literal> library is loaded when OpenMP programs
1382are started. This is possible by adding a line similar to the
1383following to your shell startup script:
1384</para>
1385<programlisting><![CDATA[
1386export LD_LIBRARY_PATH=~/gcc-4.4.0/lib64:~/gcc-4.4.0/lib:
1387]]></programlisting>
1388
1389<para>
1390As an example, the test OpenMP test program
1391<literal>drd/tests/omp_matinv</literal> triggers a data race
1392when the option -r has been specified on the command line. The data
1393race is triggered by the following code:
1394</para>
1395<programlisting><![CDATA[
1396#pragma omp parallel for private(j)
1397for (j = 0; j < rows; j++)
1398{
1399  if (i != j)
1400  {
1401    const elem_t factor = a[j * cols + i];
1402    for (k = 0; k < cols; k++)
1403    {
1404      a[j * cols + k] -= a[i * cols + k] * factor;
1405    }
1406  }
1407}
1408]]></programlisting>
1409
1410<para>
1411The above code is racy because the variable <literal>k</literal> has
1412not been declared private. DRD will print the following error message
1413for the above code:
1414</para>
1415<programlisting><![CDATA[
1416$ valgrind --tool=drd --check-stack-var=yes --read-var-info=yes drd/tests/omp_matinv 3 -t 2 -r
1417...
1418Conflicting store by thread 1/1 at 0x7fefffbc4 size 4
1419   at 0x4014A0: gj.omp_fn.0 (omp_matinv.c:203)
1420   by 0x401211: gj (omp_matinv.c:159)
1421   by 0x40166A: invert_matrix (omp_matinv.c:238)
1422   by 0x4019B4: main (omp_matinv.c:316)
1423Location 0x7fefffbc4 is 0 bytes inside local var "k"
1424declared at omp_matinv.c:160, in frame #0 of thread 1
1425...
1426]]></programlisting>
1427<para>
1428In the above output the function name <function>gj.omp_fn.0</function>
1429has been generated by GCC from the function name
1430<function>gj</function>. The allocation context information shows that the
1431data race has been caused by modifying the variable <literal>k</literal>.
1432</para>
1433
1434<para>
1435Note: for GCC versions before 4.4.0, no allocation context information is
1436shown. With these GCC versions the most usable information in the above output
1437is the source file name and the line number where the data race has been
1438detected (<literal>omp_matinv.c:203</literal>).
1439</para>
1440
1441<para>
1442For more information about OpenMP, see also
1443<ulink url="http://openmp.org/">openmp.org</ulink>.
1444</para>
1445
1446</sect2>
1447
1448
1449<sect2 id="drd-manual.cust-mem-alloc" xreflabel="Custom Memory Allocators">
1450<title>DRD and Custom Memory Allocators</title>
1451
1452<para>
1453DRD tracks all memory allocation events that happen via the
1454standard memory allocation and deallocation functions
1455(<function>malloc</function>, <function>free</function>,
1456<function>new</function> and <function>delete</function>), via entry
1457and exit of stack frames or that have been annotated with Valgrind's
1458memory pool client requests. DRD uses memory allocation and deallocation
1459information for two purposes:
1460<itemizedlist>
1461  <listitem>
1462    <para>
1463      To know where the scope ends of POSIX objects that have not been
1464      destroyed explicitly. It is e.g. not required by the POSIX
1465      threads standard to call
1466      <function>pthread_mutex_destroy</function> before freeing the
1467      memory in which a mutex object resides.
1468    </para>
1469  </listitem>
1470  <listitem>
1471    <para>
1472      To know where the scope of variables ends. If e.g. heap memory
1473      has been used by one thread, that thread frees that memory, and
1474      another thread allocates and starts using that memory, no data
1475      races must be reported for that memory.
1476    </para>
1477  </listitem>
1478</itemizedlist>
1479</para>
1480
1481<para>
1482It is essential for correct operation of DRD that the tool knows about
1483memory allocation and deallocation events. When analyzing a client program
1484with DRD that uses a custom memory allocator, either instrument the custom
1485memory allocator with the <literal>VALGRIND_MALLOCLIKE_BLOCK</literal>
1486and <literal>VALGRIND_FREELIKE_BLOCK</literal> macros or disable the
1487custom memory allocator.
1488</para>
1489
1490<para>
1491As an example, the GNU libstdc++ library can be configured
1492to use standard memory allocation functions instead of memory pools by
1493setting the environment variable
1494<literal>GLIBCXX_FORCE_NEW</literal>. For more information, see also
1495the <ulink
1496url="http://gcc.gnu.org/onlinedocs/libstdc++/manual/bk01pt04ch11.html">libstdc++
1497manual</ulink>.
1498</para>
1499
1500</sect2>
1501
1502
1503<sect2 id="drd-manual.drd-versus-memcheck" xreflabel="DRD Versus Memcheck">
1504<title>DRD Versus Memcheck</title>
1505
1506<para>
1507It is essential for correct operation of DRD that there are no memory
1508errors such as dangling pointers in the client program. Which means that
1509it is a good idea to make sure that your program is Memcheck-clean
1510before you analyze it with DRD. It is possible however that some of
1511the Memcheck reports are caused by data races. In this case it makes
1512sense to run DRD before Memcheck.
1513</para>
1514
1515<para>
1516So which tool should be run first? In case both DRD and Memcheck
1517complain about a program, a possible approach is to run both tools
1518alternatingly and to fix as many errors as possible after each run of
1519each tool until none of the two tools prints any more error messages.
1520</para>
1521
1522</sect2>
1523
1524
1525<sect2 id="drd-manual.resource-requirements" xreflabel="Resource Requirements">
1526<title>Resource Requirements</title>
1527
1528<para>
1529The requirements of DRD with regard to heap and stack memory and the
1530effect on the execution time of client programs are as follows:
1531<itemizedlist>
1532  <listitem>
1533    <para>
1534      When running a program under DRD with default DRD options,
1535      between 1.1 and 3.6 times more memory will be needed compared to
1536      a native run of the client program. More memory will be needed
1537      if loading debug information has been enabled
1538      (<literal>--read-var-info=yes</literal>).
1539    </para>
1540  </listitem>
1541  <listitem>
1542    <para>
1543      DRD allocates some of its temporary data structures on the stack
1544      of the client program threads. This amount of data is limited to
1545      1 - 2 KB. Make sure that thread stacks are sufficiently large.
1546    </para>
1547  </listitem>
1548  <listitem>
1549    <para>
1550      Most applications will run between 20 and 50 times slower under
1551      DRD than a native single-threaded run. The slowdown will be most
1552      noticeable for applications which perform frequent mutex lock /
1553      unlock operations.
1554    </para>
1555  </listitem>
1556</itemizedlist>
1557</para>
1558
1559</sect2>
1560
1561
1562<sect2 id="drd-manual.effective-use" xreflabel="Effective Use">
1563<title>Hints and Tips for Effective Use of DRD</title>
1564
1565<para>
1566The following information may be helpful when using DRD:
1567<itemizedlist>
1568  <listitem>
1569    <para>
1570      Make sure that debug information is present in the executable
1571      being analyzed, such that DRD can print function name and line
1572      number information in stack traces. Most compilers can be told
1573      to include debug information via compiler option
1574      <option>-g</option>.
1575    </para>
1576  </listitem>
1577  <listitem>
1578    <para>
1579      Compile with option <option>-O1</option> instead of
1580      <option>-O0</option>. This will reduce the amount of generated
1581      code, may reduce the amount of debug info and will speed up
1582      DRD's processing of the client program. For more information,
1583      see also <xref linkend="manual-core.started"/>.
1584    </para>
1585  </listitem>
1586  <listitem>
1587    <para>
1588      If DRD reports any errors on libraries that are part of your
1589      Linux distribution like e.g. <literal>libc.so</literal> or
1590      <literal>libstdc++.so</literal>, installing the debug packages
1591      for these libraries will make the output of DRD a lot more
1592      detailed.
1593    </para>
1594  </listitem>
1595  <listitem>
1596    <para>
1597      When using C++, do not send output from more than one thread to
1598      <literal>std::cout</literal>. Doing so would not only
1599      generate multiple data race reports, it could also result in
1600      output from several threads getting mixed up.  Either use
1601      <function>printf</function> or do the following:
1602      <orderedlist>
1603        <listitem>
1604          <para>Derive a class from <literal>std::ostreambuf</literal>
1605          and let that class send output line by line to
1606          <literal>stdout</literal>. This will avoid that individual
1607          lines of text produced by different threads get mixed
1608          up.</para>
1609        </listitem>
1610        <listitem>
1611          <para>Create one instance of <literal>std::ostream</literal>
1612          for each thread. This makes stream formatting settings
1613          thread-local. Pass a per-thread instance of the class
1614          derived from <literal>std::ostreambuf</literal> to the
1615          constructor of each instance. </para>
1616        </listitem>
1617        <listitem>
1618          <para>Let each thread send its output to its own instance of
1619          <literal>std::ostream</literal> instead of
1620          <literal>std::cout</literal>.</para>
1621        </listitem>
1622      </orderedlist>
1623    </para>
1624  </listitem>
1625</itemizedlist>
1626</para>
1627
1628</sect2>
1629
1630
1631</sect1>
1632
1633
1634<sect1 id="drd-manual.Pthreads" xreflabel="Pthreads">
1635<title>Using the POSIX Threads API Effectively</title>
1636
1637<sect2 id="drd-manual.mutex-types" xreflabel="mutex-types">
1638<title>Mutex types</title>
1639
1640<para>
1641The Single UNIX Specification version two defines the following four
1642mutex types (see also the documentation of <ulink
1643url="http://www.opengroup.org/onlinepubs/007908799/xsh/pthread_mutexattr_settype.html"><function>pthread_mutexattr_settype</function></ulink>):
1644<itemizedlist>
1645  <listitem>
1646    <para>
1647      <emphasis>normal</emphasis>, which means that no error checking
1648      is performed, and that the mutex is non-recursive.
1649    </para>
1650  </listitem>
1651  <listitem>
1652    <para>
1653      <emphasis>error checking</emphasis>, which means that the mutex
1654      is non-recursive and that error checking is performed.
1655    </para>
1656  </listitem>
1657  <listitem>
1658    <para>
1659      <emphasis>recursive</emphasis>, which means that a mutex may be
1660      locked recursively.
1661    </para>
1662  </listitem>
1663  <listitem>
1664    <para>
1665      <emphasis>default</emphasis>, which means that error checking
1666      behavior is undefined, and that the behavior for recursive
1667      locking is also undefined. Or: portable code must neither
1668      trigger error conditions through the Pthreads API nor attempt to
1669      lock a mutex of default type recursively.
1670    </para>
1671  </listitem>
1672</itemizedlist>
1673</para>
1674
1675<para>
1676In complex applications it is not always clear from beforehand which
1677mutex will be locked recursively and which mutex will not be locked
1678recursively. Attempts lock a non-recursive mutex recursively will
1679result in race conditions that are very hard to find without a thread
1680checking tool. So either use the error checking mutex type and
1681consistently check the return value of Pthread API mutex calls, or use
1682the recursive mutex type.
1683</para>
1684
1685</sect2>
1686
1687<sect2 id="drd-manual.condvar" xreflabel="condition-variables">
1688<title>Condition variables</title>
1689
1690<para>
1691A condition variable allows one thread to wake up one or more other
1692threads. Condition variables are often used to notify one or more
1693threads about state changes of shared data. Unfortunately it is very
1694easy to introduce race conditions by using condition variables as the
1695only means of state information propagation. A better approach is to
1696let threads poll for changes of a state variable that is protected by
1697a mutex, and to use condition variables only as a thread wakeup
1698mechanism. See also the source file
1699<computeroutput>drd/tests/monitor_example.cpp</computeroutput> for an
1700example of how to implement this concept in C++. The monitor concept
1701used in this example is a well known and very useful concept -- see
1702also Wikipedia for more information about the <ulink
1703url="http://en.wikipedia.org/wiki/Monitor_(synchronization)">monitor</ulink>
1704concept.
1705</para>
1706
1707</sect2>
1708
1709<sect2 id="drd-manual.pctw" xreflabel="pthread_cond_timedwait">
1710<title>pthread_cond_timedwait and timeouts</title>
1711
1712<para>
1713Historically the function
1714<function>pthread_cond_timedwait</function> only allowed the
1715specification of an absolute timeout, that is a timeout independent of
1716the time when this function was called. However, almost every call to
1717this function expresses a relative timeout. This typically happens by
1718passing the sum of
1719<computeroutput>clock_gettime(CLOCK_REALTIME)</computeroutput> and a
1720relative timeout as the third argument. This approach is incorrect
1721since forward or backward clock adjustments by e.g. ntpd will affect
1722the timeout. A more reliable approach is as follows:
1723<itemizedlist>
1724  <listitem>
1725    <para>
1726      When initializing a condition variable through
1727      <function>pthread_cond_init</function>, specify that the timeout of
1728      <function>pthread_cond_timedwait</function> will use the clock
1729      <literal>CLOCK_MONOTONIC</literal> instead of
1730      <literal>CLOCK_REALTIME</literal>. You can do this via
1731      <computeroutput>pthread_condattr_setclock(...,
1732      CLOCK_MONOTONIC)</computeroutput>.
1733    </para>
1734  </listitem>
1735  <listitem>
1736    <para>
1737      When calling <function>pthread_cond_timedwait</function>, pass
1738      the sum of
1739      <computeroutput>clock_gettime(CLOCK_MONOTONIC)</computeroutput>
1740      and a relative timeout as the third argument.
1741    </para>
1742  </listitem>
1743</itemizedlist>
1744See also
1745<computeroutput>drd/tests/monitor_example.cpp</computeroutput> for an
1746example.
1747</para>
1748
1749</sect2>
1750
1751</sect1>
1752
1753
1754<sect1 id="drd-manual.limitations" xreflabel="Limitations">
1755<title>Limitations</title>
1756
1757<para>DRD currently has the following limitations:</para>
1758
1759<itemizedlist>
1760  <listitem>
1761    <para>
1762      DRD, just like Memcheck, will refuse to start on Linux
1763      distributions where all symbol information has been removed from
1764      <filename>ld.so</filename>. This is e.g. the case for the PPC editions
1765      of openSUSE and Gentoo. You will have to install the glibc debuginfo
1766      package on these platforms before you can use DRD. See also openSUSE
1767      bug <ulink url="http://bugzilla.novell.com/show_bug.cgi?id=396197">
1768      396197</ulink> and Gentoo bug <ulink
1769      url="http://bugs.gentoo.org/214065">214065</ulink>.
1770    </para>
1771  </listitem>
1772  <listitem>
1773    <para>
1774      With gcc 4.4.3 and before, DRD may report data races on the C++
1775      class <literal>std::string</literal> in a multithreaded program. This is
1776      a know <literal>libstdc++</literal> issue -- see also GCC bug
1777      <ulink url="http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40518">40518</ulink>
1778      for more information.
1779    </para>
1780  </listitem>
1781  <listitem>
1782    <para>
1783      If you compile the DRD source code yourself, you need GCC 3.0 or
1784      later. GCC 2.95 is not supported.
1785    </para>
1786  </listitem>
1787  <listitem>
1788    <para>
1789      Of the two POSIX threads implementations for Linux, only the
1790      NPTL (Native POSIX Thread Library) is supported. The older
1791      LinuxThreads library is not supported.
1792    </para>
1793  </listitem>
1794</itemizedlist>
1795
1796</sect1>
1797
1798
1799<sect1 id="drd-manual.feedback" xreflabel="Feedback">
1800<title>Feedback</title>
1801
1802<para>
1803If you have any comments, suggestions, feedback or bug reports about
1804DRD, feel free to either post a message on the Valgrind users mailing
1805list or to file a bug report. See also <ulink
1806url="&vg-url;">&vg-url;</ulink> for more information.
1807</para>
1808
1809</sect1>
1810
1811
1812</chapter>
1813