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; ]>
7 <chapter id="drd-manual" xreflabel="DRD: a thread error detector">
8 <title>DRD: a thread error detector</title>
10 <para>To use this tool, you must specify
11 <option>--tool=drd</option>
12 on the Valgrind command line.</para>
15 <sect1 id="drd-manual.overview" xreflabel="Overview">
16 <title>Overview</title>
19 DRD is a Valgrind tool for detecting errors in multithreaded C and C++
20 programs. The tool works for any program that uses the POSIX threading
21 primitives or that uses threading concepts built on top of the POSIX threading
25 <sect2 id="drd-manual.mt-progr-models" xreflabel="MT-progr-models">
26 <title>Multithreaded Programming Paradigms</title>
29 There are two possible reasons for using multithreading in a program:
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.
41 To use multiple CPU cores simultaneously for speeding up
42 computations. This is why many High Performance Computing (HPC)
43 applications are multithreaded.
50 Multithreaded programs can use one or more of the following programming
51 paradigms. Which paradigm is appropriate depends e.g. on the application type.
52 Some examples of multithreaded programming paradigms are:
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.
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.
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.
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.
97 DRD supports any combination of multithreaded programming paradigms as
98 long as the implementation of these paradigms is based on the POSIX
99 threads primitives. DRD however does not support programs that use
100 e.g. Linux' futexes directly. Attempts to analyze such programs with
101 DRD will cause DRD to report many false positives.
107 <sect2 id="drd-manual.pthreads-model" xreflabel="Pthreads-model">
108 <title>POSIX Threads Programming Model</title>
111 POSIX threads, also known as Pthreads, is the most widely available
112 threading library on Unix systems.
116 The POSIX threads programming model is based on the following abstractions:
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.
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.
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.
141 Threads. Each thread represents a concurrent activity.
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
157 Which source code statements generate which memory accesses depends on
158 the <emphasis>memory model</emphasis> of the programming language being
159 used. There is not yet a definitive memory model for the C and C++
160 languages. 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">
162 WG21/N2338: Concurrency memory model compiler consequences</ulink>.
166 For more information about POSIX threads, see also the Single UNIX
167 Specification version 3, also known as
168 <ulink url="http://www.opengroup.org/onlinepubs/000095399/idx/threads.html">
169 IEEE Std 1003.1</ulink>.
175 <sect2 id="drd-manual.mt-problems" xreflabel="MT-Problems">
176 <title>Multithreaded Programming Problems</title>
179 Depending on which multithreading paradigm is being used in a program,
180 one or more of the following problems can occur:
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.
191 Lock contention. One thread blocks the progress of one or more other
192 threads by holding a lock too long.
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.
205 Deadlock. A deadlock occurs when two or more threads wait for
206 each other indefinitely.
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.
221 Although the likelihood of the occurrence of data races can be reduced
222 through a disciplined programming style, a tool for automatic
223 detection of data races is a necessity when developing multithreaded
224 software. DRD can detect these, as well as lock contention and
225 improper use of the POSIX threads API.
231 <sect2 id="drd-manual.data-race-detection" xreflabel="data-race-detection">
232 <title>Data Race Detection</title>
235 The result of load and store operations performed by a multithreaded program
236 depends on the order in which memory operations are performed. This order is
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.
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>.
254 The combination of program order and synchronization order is called the
255 <emphasis>happens-before relationship</emphasis>. This concept was first
256 defined by S. Adve et al in the paper <emphasis>Detecting data races on weak
257 memory systems</emphasis>, ACM SIGARCH Computer Architecture News, v.19 n.3,
262 Two memory operations <emphasis>conflict</emphasis> if both operations are
263 performed by different threads, refer to the same memory location and at least
264 one of them is a store operation.
268 A multithreaded program is <emphasis>data-race free</emphasis> if all
269 conflicting memory accesses are ordered by synchronization
274 A well known way to ensure that a multithreaded program is data-race
275 free is to ensure that a locking discipline is followed. It is e.g.
276 possible to associate a mutex with each shared data item, and to hold
277 a lock on the associated mutex while the shared data is accessed.
281 All programs that follow a locking discipline are data-race free, but not all
282 data-race free programs follow a locking discipline. There exist multithreaded
283 programs where access to shared data is arbitrated via condition variables,
284 semaphores or barriers. As an example, a certain class of HPC applications
285 consists of a sequence of computation steps separated in time by barriers, and
286 where these barriers are the only means of synchronization. Although there are
287 many conflicting memory accesses in such applications and although such
288 applications do not make use mutexes, most of these applications do not
293 There exist two different approaches for verifying the correctness of
294 multithreaded programs at runtime. The approach of the so-called Eraser
295 algorithm is to verify whether all shared memory accesses follow a consistent
296 locking strategy. And the happens-before data race detectors verify directly
297 whether all interthread memory accesses are ordered by synchronization
298 operations. While the last approach is more complex to implement, and while it
299 is more sensitive to OS scheduling, it is a general approach that works for
300 all classes of multithreaded programs. An important advantage of
301 happens-before data race detectors is that these do not report any false
306 DRD is based on the happens-before algorithm.
315 <sect1 id="drd-manual.using-drd" xreflabel="Using DRD">
316 <title>Using DRD</title>
318 <sect2 id="drd-manual.options" xreflabel="DRD Command-line Options">
319 <title>DRD Command-line Options</title>
321 <para>The following command-line options are available for controlling the
322 behavior of the DRD tool itself:</para>
324 <!-- start of xi:include in the manpage -->
325 <variablelist id="drd.opts.list">
328 <option><![CDATA[--check-stack-var=<yes|no> [default: no]]]></option>
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.
340 <option><![CDATA[--exclusive-threshold=<n> [default: off]]]></option>
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.
353 <![CDATA[--first-race-only=<yes|no> [default: no]]]>
358 Whether to report only the first data race that has been detected on a
359 memory location or all data races that have been detected on a memory
367 <![CDATA[--report-signal-unlocked=<yes|no> [default: yes]]]>
372 Whether to report calls to
373 <function>pthread_cond_signal</function> and
374 <function>pthread_cond_broadcast</function> where the mutex
375 associated with the signal through
376 <function>pthread_cond_wait</function> or
377 <function>pthread_cond_timed_wait</function>is not locked at
378 the time the signal is sent. Sending a signal without holding
379 a lock on the associated mutex is a common programming error
380 which can cause subtle race conditions and unpredictable
381 behavior. There exist some uncommon synchronization patterns
382 however where it is safe to send a signal without holding a
383 lock on the associated mutex.
389 <option><![CDATA[--segment-merging=<yes|no> [default: yes]]]></option>
393 Controls segment merging. Segment merging is an algorithm to
394 limit memory usage of the data race detection
395 algorithm. Disabling segment merging may improve the accuracy
396 of the so-called 'other segments' displayed in race reports
397 but can also trigger an out of memory error.
403 <option><![CDATA[--segment-merging-interval=<n> [default: 10]]]></option>
407 Perform segment merging only after the specified number of new
408 segments have been created. This is an advanced configuration option
409 that allows to choose whether to minimize DRD's memory usage by
410 choosing a low value or to let DRD run faster by choosing a slightly
411 higher value. The optimal value for this parameter depends on the
412 program being analyzed. The default value works well for most programs.
418 <option><![CDATA[--shared-threshold=<n> [default: off]]]></option>
422 Print an error message if a reader lock has been held longer
423 than the specified time (in milliseconds). This option enables
424 the detection of lock contention.
430 <option><![CDATA[--show-confl-seg=<yes|no> [default: yes]]]></option>
434 Show conflicting segments in race reports. Since this
435 information can help to find the cause of a data race, this
436 option is enabled by default. Disabling this option makes the
437 output of DRD more compact.
443 <option><![CDATA[--show-stack-usage=<yes|no> [default: no]]]></option>
447 Print stack usage at thread exit time. When a program creates a large
448 number of threads it becomes important to limit the amount of virtual
449 memory allocated for thread stacks. This option makes it possible to
450 observe how much stack memory has been used by each thread of the the
451 client program. Note: the DRD tool itself allocates some temporary
452 data on the client thread stack. The space necessary for this
453 temporary data must be allocated by the client program when it
454 allocates stack memory, but is not included in stack usage reported by
460 <!-- end of xi:include in the manpage -->
462 <!-- start of xi:include in the manpage -->
464 The following options are available for monitoring the behavior of the
468 <variablelist id="drd.debugopts.list">
471 <option><![CDATA[--trace-addr=<address> [default: none]]]></option>
475 Trace all load and store activity for the specified
476 address. This option may be specified more than once.
482 <option><![CDATA[--trace-barrier=<yes|no> [default: no]]]></option>
486 Trace all barrier activity.
492 <option><![CDATA[--trace-cond=<yes|no> [default: no]]]></option>
496 Trace all condition variable activity.
502 <option><![CDATA[--trace-fork-join=<yes|no> [default: no]]]></option>
506 Trace all thread creation and all thread termination events.
512 <option><![CDATA[--trace-mutex=<yes|no> [default: no]]]></option>
516 Trace all mutex activity.
522 <option><![CDATA[--trace-rwlock=<yes|no> [default: no]]]></option>
526 Trace all reader-writer lock activity.
532 <option><![CDATA[--trace-semaphore=<yes|no> [default: no]]]></option>
536 Trace all semaphore activity.
541 <!-- end of xi:include in the manpage -->
546 <sect2 id="drd-manual.data-races" xreflabel="Data Races">
547 <title>Detected Errors: Data Races</title>
550 DRD prints a message every time it detects a data race. Please keep
551 the following in mind when interpreting DRD's output:
555 Every thread is assigned a <emphasis>thread ID</emphasis> by the DRD
556 tool. A thread ID is a number. Thread ID's start at one and are never
562 The term <emphasis>segment</emphasis> refers to a consecutive
563 sequence of load, store and synchronization operations, all
564 issued by the same thread. A segment always starts and ends at a
565 synchronization operation. Data race analysis is performed
566 between segments instead of between individual load and store
567 operations because of performance reasons.
572 There are always at least two memory accesses involved in a data
573 race. Memory accesses involved in a data race are called
574 <emphasis>conflicting memory accesses</emphasis>. DRD prints a
575 report for each memory access that conflicts with a past memory
583 Below you can find an example of a message printed by DRD when it
586 <programlisting><![CDATA[
587 $ valgrind --tool=drd --read-var-info=yes drd/tests/rwlock_race
590 ==9466== Conflicting load by thread 3 at 0x006020b8 size 4
591 ==9466== at 0x400B6C: thread_func (rwlock_race.c:29)
592 ==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186)
593 ==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so)
594 ==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so)
595 ==9466== Location 0x6020b8 is 0 bytes inside local var "s_racy"
596 ==9466== declared at rwlock_race.c:18, in frame #0 of thread 3
597 ==9466== Other segment start (thread 2)
598 ==9466== at 0x4C2847D: pthread_rwlock_rdlock* (drd_pthread_intercepts.c:813)
599 ==9466== by 0x400B6B: thread_func (rwlock_race.c:28)
600 ==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186)
601 ==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so)
602 ==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so)
603 ==9466== Other segment end (thread 2)
604 ==9466== at 0x4C28B54: pthread_rwlock_unlock* (drd_pthread_intercepts.c:912)
605 ==9466== by 0x400B84: thread_func (rwlock_race.c:30)
606 ==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186)
607 ==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so)
608 ==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so)
613 The above report has the following meaning:
617 The number in the column on the left is the process ID of the
618 process being analyzed by DRD.
623 The first line ("Thread 3") tells you the thread ID for
624 the thread in which context the data race has been detected.
629 The next line tells which kind of operation was performed (load or
630 store) and by which thread. On the same line the start address and the
631 number of bytes involved in the conflicting access are also displayed.
636 Next, the call stack of the conflicting access is displayed. If
637 your program has been compiled with debug information
638 (<option>-g</option>), this call stack will include file names and
639 line numbers. The two
640 bottommost frames in this call stack (<function>clone</function>
641 and <function>start_thread</function>) show how the NPTL starts
642 a thread. The third frame
643 (<function>vg_thread_wrapper</function>) is added by DRD. The
644 fourth frame (<function>thread_func</function>) is the first
645 interesting line because it shows the thread entry point, that
646 is the function that has been passed as the third argument to
647 <function>pthread_create</function>.
652 Next, the allocation context for the conflicting address is
653 displayed. For dynamically allocated data the allocation call
654 stack is shown. For static variables and stack variables the
655 allocation context is only shown when the option
656 <option>--read-var-info=yes</option> has been
657 specified. Otherwise DRD will print <computeroutput>Allocation
658 context: unknown</computeroutput>.
663 A conflicting access involves at least two memory accesses. For
664 one of these accesses an exact call stack is displayed, and for
665 the other accesses an approximate call stack is displayed,
666 namely the start and the end of the segments of the other
667 accesses. This information can be interpreted as follows:
671 Start at the bottom of both call stacks, and count the
672 number stack frames with identical function name, file
673 name and line number. In the above example the three
674 bottommost frames are identical
675 (<function>clone</function>,
676 <function>start_thread</function> and
677 <function>vg_thread_wrapper</function>).
682 The next higher stack frame in both call stacks now tells
683 you between in which source code region the other memory
684 access happened. The above output tells that the other
685 memory access involved in the data race happened between
686 source code lines 28 and 30 in file
687 <computeroutput>rwlock_race.c</computeroutput>.
699 <sect2 id="drd-manual.lock-contention" xreflabel="Lock Contention">
700 <title>Detected Errors: Lock Contention</title>
703 Threads must be able to make progress without being blocked for too long by
704 other threads. Sometimes a thread has to wait until a mutex or reader-writer
705 synchronization object is unlocked by another thread. This is called
706 <emphasis>lock contention</emphasis>.
710 Lock contention causes delays. Such delays should be as short as
711 possible. The two command line options
712 <literal>--exclusive-threshold=<n></literal> and
713 <literal>--shared-threshold=<n></literal> make it possible to
714 detect excessive lock contention by making DRD report any lock that
715 has been held longer than the specified threshold. An example:
717 <programlisting><![CDATA[
718 $ valgrind --tool=drd --exclusive-threshold=10 drd/tests/hold_lock -i 500
720 ==10668== Acquired at:
721 ==10668== at 0x4C267C8: pthread_mutex_lock (drd_pthread_intercepts.c:395)
722 ==10668== by 0x400D92: main (hold_lock.c:51)
723 ==10668== Lock on mutex 0x7fefffd50 was held during 503 ms (threshold: 10 ms).
724 ==10668== at 0x4C26ADA: pthread_mutex_unlock (drd_pthread_intercepts.c:441)
725 ==10668== by 0x400DB5: main (hold_lock.c:55)
730 The <literal>hold_lock</literal> test program holds a lock as long as
731 specified by the <literal>-i</literal> (interval) argument. The DRD
732 output reports that the lock acquired at line 51 in source file
733 <literal>hold_lock.c</literal> and released at line 55 was held during
734 503 ms, while a threshold of 10 ms was specified to DRD.
740 <sect2 id="drd-manual.api-checks" xreflabel="API Checks">
741 <title>Detected Errors: Misuse of the POSIX threads API</title>
744 DRD is able to detect and report the following misuses of the POSIX
749 Passing the address of one type of synchronization object
750 (e.g. a mutex) to a POSIX API call that expects a pointer to
751 another type of synchronization object (e.g. a condition
757 Attempts to unlock a mutex that has not been locked.
762 Attempts to unlock a mutex that was locked by another thread.
767 Attempts to lock a mutex of type
768 <literal>PTHREAD_MUTEX_NORMAL</literal> or a spinlock
774 Destruction or deallocation of a locked mutex.
779 Sending a signal to a condition variable while no lock is held
780 on the mutex associated with the condition variable.
785 Calling <function>pthread_cond_wait</function> on a mutex
786 that is not locked, that is locked by another thread or that
787 has been locked recursively.
792 Associating two different mutexes with a condition variable
793 through <function>pthread_cond_wait</function>.
798 Destruction or deallocation of a condition variable that is
804 Destruction or deallocation of a locked reader-writer synchronization
810 Attempts to unlock a reader-writer synchronization object that was not
811 locked by the calling thread.
816 Attempts to recursively lock a reader-writer synchronization object
822 Attempts to pass the address of a user-defined reader-writer
823 synchronization object to a POSIX threads function.
828 Attempts to pass the address of a POSIX reader-writer synchronization
829 object to one of the annotations for user-defined reader-writer
830 synchronization objects.
835 Reinitialization of a mutex, condition variable, reader-writer
836 lock, semaphore or barrier.
841 Destruction or deallocation of a semaphore or barrier that is
847 Missing synchronization between barrier wait and barrier destruction.
852 Exiting a thread without first unlocking the spinlocks, mutexes or
853 reader-writer synchronization objects that were locked by that thread.
858 Passing an invalid thread ID to <function>pthread_join</function>
859 or <function>pthread_cancel</function>.
868 <sect2 id="drd-manual.clientreqs" xreflabel="Client requests">
869 <title>Client Requests</title>
872 Just as for other Valgrind tools it is possible to let a client program
873 interact with the DRD tool through client requests. In addition to the
874 client requests several macros have been defined that allow to use the
875 client requests in a convenient way.
879 The interface between client programs and the DRD tool is defined in
880 the header file <literal><valgrind/drd.h></literal>. The
881 available macros and client requests are:
885 The macro <literal>DRD_GET_VALGRIND_THREADID</literal> and the
887 request <varname>VG_USERREQ__DRD_GET_VALGRIND_THREAD_ID</varname>.
888 Query the thread ID that has been assigned by the Valgrind core to the
889 thread executing this client request. Valgrind's thread ID's start at
890 one and are recycled in case a thread stops.
895 The macro <literal>DRD_GET_DRD_THREADID</literal> and the corresponding
896 client request <varname>VG_USERREQ__DRD_GET_DRD_THREAD_ID</varname>.
897 Query the thread ID that has been assigned by DRD to the thread
898 executing this client request. These are the thread ID's reported by DRD
899 in data race reports and in trace messages. DRD's thread ID's start at
900 one and are never recycled.
905 The macros <literal>DRD_IGNORE_VAR(x)</literal>,
906 <literal>ANNOTATE_TRACE_MEMORY(&x)</literal> and the corresponding
907 client request <varname>VG_USERREQ__DRD_START_SUPPRESSION</varname>. Some
908 applications contain intentional races. There exist e.g. applications
909 where the same value is assigned to a shared variable from two different
910 threads. It may be more convenient to suppress such races than to solve
911 these. This client request allows to suppress such races.
916 The macro <literal>DRD_STOP_IGNORING_VAR(x)</literal> and the
917 corresponding client request
918 <varname>VG_USERREQ__DRD_FINISH_SUPPRESSION</varname>. Tell DRD
919 to no longer ignore data races for the address range that was suppressed
920 either via the macro <literal>DRD_IGNORE_VAR(x)</literal> or via the
921 client request <varname>VG_USERREQ__DRD_START_SUPPRESSION</varname>.
926 The macro <literal>DRD_TRACE_VAR(x)</literal>. Trace all load and store
927 activity for the address range starting at <literal>&x</literal> and
928 occupying <literal>sizeof(x)</literal> bytes. When DRD reports a data
929 race on a specified variable, and it's not immediately clear which
930 source code statements triggered the conflicting accesses, it can be
931 very helpful to trace all activity on the offending memory location.
936 The macro <literal>ANNOTATE_TRACE_MEMORY(&x)</literal>. Trace all
937 load and store activity that touches at least the single byte at the
938 address <literal>&x</literal>.
943 The client request <varname>VG_USERREQ__DRD_START_TRACE_ADDR</varname>,
944 which allows to trace all load and store activity for the specified
951 request <varname>VG_USERREQ__DRD_STOP_TRACE_ADDR</varname>. Do no longer
952 trace load and store activity for the specified address range.
957 The macro <literal>ANNOTATE_HAPPENS_BEFORE(addr)</literal> tells DRD to
958 insert a mark. Insert this macro just after an access to the variable at
959 the specified address has been performed.
964 The macro <literal>ANNOTATE_HAPPENS_AFTER(addr)</literal> tells DRD that
965 the next access to the variable at the specified address should be
966 considered to have happened after the access just before the latest
967 <literal>ANNOTATE_HAPPENS_BEFORE(addr)</literal> annotation that
968 references the same variable. The purpose of these two macros is to
969 tell DRD about the order of inter-thread memory accesses implemented via
970 atomic memory operations.
975 The macro <literal>ANNOTATE_RWLOCK_CREATE(rwlock)</literal> tells DRD
976 that the object at address <literal>rwlock</literal> is a
977 reader-writer synchronization object that is not a
978 <literal>pthread_rwlock_t</literal> synchronization object.
983 The macro <literal>ANNOTATE_RWLOCK_DESTROY(rwlock)</literal> tells DRD
984 that the reader-writer synchronization object at
985 address <literal>rwlock</literal> has been destroyed.
990 The macro <literal>ANNOTATE_WRITERLOCK_ACQUIRED(rwlock)</literal> tells
991 DRD that a writer lock has been acquired on the reader-writer
992 synchronization object at address <literal>rwlock</literal>.
997 The macro <literal>ANNOTATE_READERLOCK_ACQUIRED(rwlock)</literal> tells
998 DRD that a reader lock has been acquired on the reader-writer
999 synchronization object at address <literal>rwlock</literal>.
1004 The macro <literal>ANNOTATE_RWLOCK_ACQUIRED(rwlock, is_w)</literal>
1005 tells DRD that a writer lock (when <literal>is_w != 0</literal>) or that
1006 a reader lock (when <literal>is_w == 0</literal>) has been acquired on
1007 the reader-writer synchronization object at
1008 address <literal>rwlock</literal>.
1013 The macro <literal>ANNOTATE_WRITERLOCK_RELEASED(rwlock)</literal> tells
1014 DRD that a writer lock has been released on the reader-writer
1015 synchronization object at address <literal>rwlock</literal>.
1020 The macro <literal>ANNOTATE_READERLOCK_RELEASED(rwlock)</literal> tells
1021 DRD that a reader lock has been released on the reader-writer
1022 synchronization object at address <literal>rwlock</literal>.
1027 The macro <literal>ANNOTATE_RWLOCK_RELEASED(rwlock, is_w)</literal>
1028 tells DRD that a writer lock (when <literal>is_w != 0</literal>) or that
1029 a reader lock (when <literal>is_w == 0</literal>) has been released on
1030 the reader-writer synchronization object at
1031 address <literal>rwlock</literal>.
1036 The macro <literal>ANNOTATE_BENIGN_RACE(addr, descr)</literal> tells
1037 DRD that any races detected on the specified address are benign and
1038 hence should not be reported. The <literal>descr</literal> argument is
1039 ignored but can be used to document why data races
1040 on <literal>addr</literal> are benign.
1045 The macro <literal>ANNOTATE_IGNORE_READS_BEGIN</literal> tells
1046 DRD to ignore all memory loads performed by the current thread.
1051 The macro <literal>ANNOTATE_IGNORE_READS_END</literal> tells
1052 DRD to stop ignoring the memory loads performed by the current thread.
1057 The macro <literal>ANNOTATE_IGNORE_WRITES_BEGIN</literal> tells
1058 DRD to ignore all memory stores performed by the current thread.
1063 The macro <literal>ANNOTATE_IGNORE_WRITES_END</literal> tells
1064 DRD to stop ignoring the memory stores performed by the current thread.
1069 The macro <literal>ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN</literal> tells
1070 DRD to ignore all memory accesses performed by the current thread.
1075 The macro <literal>ANNOTATE_IGNORE_READS_AND_WRITES_END</literal> tells
1076 DRD to stop ignoring the memory accesses performed by the current thread.
1081 The macro <literal>ANNOTATE_NEW_MEMORY(addr, size)</literal> tells
1082 DRD that the specified memory range has been allocated by a custom
1083 memory allocator in the client program and that the client program
1084 will start using this memory range.
1089 The macro <literal>ANNOTATE_THREAD_NAME(name)</literal> tells DRD to
1090 associate the specified name with the current thread and to include this
1091 name in the error messages printed by DRD.
1096 The macros <literal>VALGRIND_MALLOCLIKE_BLOCK</literal> and
1097 <literal>VALGRIND_FREELIKE_BLOCK</literal> from the Valgrind core are
1098 implemented; they are described in
1099 <xref linkend="manual-core-adv.clientreq"/>.
1106 For an example of how to use the annotations for user-defined reader-writer
1107 synchronization objects, see
1108 also the source file <literal>drd/tests/annotate_rwlock.c</literal> in the
1109 Valgrind source archive. And an example of how to
1110 use the <literal>ANNOTATE_HAPPENS_BEFORE</literal> and
1111 the <literal>ANNOTATE_HAPPENS_AFTER</literal> annotations can be found
1112 in the source code of the <ulink url="http://code.google.com/chromium/">Chromium</ulink>
1117 Note: if you compiled Valgrind yourself, the header file
1118 <literal><valgrind/drd.h></literal> will have been installed in
1119 the directory <literal>/usr/include</literal> by the command
1120 <literal>make install</literal>. If you obtained Valgrind by
1121 installing it as a package however, you will probably have to install
1122 another package with a name like <literal>valgrind-devel</literal>
1123 before Valgrind's header files are available.
1129 <sect2 id="drd-manual.gnome" xreflabel="GNOME">
1130 <title>Debugging GNOME Programs</title>
1133 GNOME applications use the threading primitives provided by the
1134 <computeroutput>glib</computeroutput> and
1135 <computeroutput>gthread</computeroutput> libraries. These libraries
1136 are built on top of POSIX threads, and hence are directly supported by
1137 DRD. Please keep in mind that you have to call
1138 <function>g_thread_init</function> before creating any threads, or
1139 DRD will report several data races on glib functions. See also the
1141 url="http://library.gnome.org/devel/glib/stable/glib-Threads.html">GLib
1142 Reference Manual</ulink> for more information about
1143 <function>g_thread_init</function>.
1147 One of the many facilities provided by the <literal>glib</literal>
1148 library is a block allocator, called <literal>g_slice</literal>. You
1149 have to disable this block allocator when using DRD by adding the
1150 following to the shell environment variables:
1151 <literal>G_SLICE=always-malloc</literal>. See also the <ulink
1152 url="http://library.gnome.org/devel/glib/stable/glib-Memory-Slices.html">GLib
1153 Reference Manual</ulink> for more information.
1159 <sect2 id="drd-manual.qt" xreflabel="Qt">
1160 <title>Debugging Qt Programs</title>
1163 The Qt library is the GUI library used by the KDE project. Currently
1164 there are two versions of the Qt library in use: Qt3 by KDE 3 and Qt4
1165 by KDE 4. If possible, use Qt4 instead of Qt3. Qt3 is no longer
1166 supported, and there are known problems with multithreading support in
1167 Qt3. As an example, using QString objects in more than one thread will
1168 trigger race reports (this has been confirmed by Trolltech -- see also
1169 Trolltech task <ulink
1170 url="http://trolltech.com/developer/task-tracker/index_html">#206152</ulink>).
1174 Qt4 applications are supported by DRD, but only if the
1175 <literal>libqt4-debuginfo</literal> package has been installed. Some
1176 of the synchronization and threading primitives in Qt4 bypass the
1177 POSIX threads library, and DRD can only intercept these if symbol
1178 information for the Qt4 library is available. DRD won't tell you if it
1179 has not been able to load the Qt4 debug information, but a huge number
1180 of data races will be reported on data protected via
1181 <literal>QMutex</literal> objects.
1187 <sect2 id="drd-manual.boost.thread" xreflabel="Boost.Thread">
1188 <title>Debugging Boost.Thread Programs</title>
1191 The Boost.Thread library is the threading library included with the
1192 cross-platform Boost Libraries. This threading library is an early
1193 implementation of the upcoming C++0x threading library.
1197 Applications that use the Boost.Thread library should run fine under DRD.
1201 More information about Boost.Thread can be found here:
1205 Anthony Williams, <ulink
1206 url="http://www.boost.org/doc/libs/1_37_0/doc/html/thread.html">Boost.Thread</ulink>
1207 Library Documentation, Boost website, 2007.
1212 Anthony Williams, <ulink
1213 url="http://www.ddj.com/cpp/211600441">What's New in Boost
1214 Threads?</ulink>, Recent changes to the Boost Thread library,
1215 Dr. Dobbs Magazine, October 2008.
1224 <sect2 id="drd-manual.openmp" xreflabel="OpenMP">
1225 <title>Debugging OpenMP Programs</title>
1228 OpenMP stands for <emphasis>Open Multi-Processing</emphasis>. The OpenMP
1229 standard consists of a set of compiler directives for C, C++ and Fortran
1230 programs that allows a compiler to transform a sequential program into a
1231 parallel program. OpenMP is well suited for HPC applications and allows to
1232 work at a higher level compared to direct use of the POSIX threads API. While
1233 OpenMP ensures that the POSIX API is used correctly, OpenMP programs can still
1234 contain data races. So it definitely makes sense to verify OpenMP programs
1235 with a thread checking tool.
1239 DRD supports OpenMP shared-memory programs generated by GCC. GCC
1240 supports OpenMP since version 4.2.0. GCC's runtime support
1241 for OpenMP programs is provided by a library called
1242 <literal>libgomp</literal>. The synchronization primitives implemented
1243 in this library use Linux' futex system call directly, unless the
1244 library has been configured with the
1245 <literal>--disable-linux-futex</literal> option. DRD only supports
1246 libgomp libraries that have been configured with this option and in
1247 which symbol information is present. For most Linux distributions this
1248 means that you will have to recompile GCC. See also the script
1249 <literal>drd/scripts/download-and-build-gcc</literal> in the
1250 Valgrind source tree for an example of how to compile GCC. You will
1251 also have to make sure that the newly compiled
1252 <literal>libgomp.so</literal> library is loaded when OpenMP programs
1253 are started. This is possible by adding a line similar to the
1254 following to your shell startup script:
1256 <programlisting><![CDATA[
1257 export LD_LIBRARY_PATH=~/gcc-4.4.0/lib64:~/gcc-4.4.0/lib:
1258 ]]></programlisting>
1261 As an example, the test OpenMP test program
1262 <literal>drd/tests/omp_matinv</literal> triggers a data race
1263 when the option -r has been specified on the command line. The data
1264 race is triggered by the following code:
1266 <programlisting><![CDATA[
1267 #pragma omp parallel for private(j)
1268 for (j = 0; j < rows; j++)
1272 const elem_t factor = a[j * cols + i];
1273 for (k = 0; k < cols; k++)
1275 a[j * cols + k] -= a[i * cols + k] * factor;
1279 ]]></programlisting>
1282 The above code is racy because the variable <literal>k</literal> has
1283 not been declared private. DRD will print the following error message
1286 <programlisting><![CDATA[
1287 $ valgrind --tool=drd --check-stack-var=yes --read-var-info=yes drd/tests/omp_matinv 3 -t 2 -r
1289 Conflicting store by thread 1/1 at 0x7fefffbc4 size 4
1290 at 0x4014A0: gj.omp_fn.0 (omp_matinv.c:203)
1291 by 0x401211: gj (omp_matinv.c:159)
1292 by 0x40166A: invert_matrix (omp_matinv.c:238)
1293 by 0x4019B4: main (omp_matinv.c:316)
1294 Location 0x7fefffbc4 is 0 bytes inside local var "k"
1295 declared at omp_matinv.c:160, in frame #0 of thread 1
1297 ]]></programlisting>
1299 In the above output the function name <function>gj.omp_fn.0</function>
1300 has been generated by GCC from the function name
1301 <function>gj</function>. The allocation context information shows that the
1302 data race has been caused by modifying the variable <literal>k</literal>.
1306 Note: for GCC versions before 4.4.0, no allocation context information is
1307 shown. With these GCC versions the most usable information in the above output
1308 is the source file name and the line number where the data race has been
1309 detected (<literal>omp_matinv.c:203</literal>).
1313 For more information about OpenMP, see also
1314 <ulink url="http://openmp.org/">openmp.org</ulink>.
1320 <sect2 id="drd-manual.cust-mem-alloc" xreflabel="Custom Memory Allocators">
1321 <title>DRD and Custom Memory Allocators</title>
1324 DRD tracks all memory allocation events that happen via the
1325 standard memory allocation and deallocation functions
1326 (<function>malloc</function>, <function>free</function>,
1327 <function>new</function> and <function>delete</function>), via entry
1328 and exit of stack frames or that have been annotated with Valgrind's
1329 memory pool client requests. DRD uses memory allocation and deallocation
1330 information for two purposes:
1334 To know where the scope ends of POSIX objects that have not been
1335 destroyed explicitly. It is e.g. not required by the POSIX
1336 threads standard to call
1337 <function>pthread_mutex_destroy</function> before freeing the
1338 memory in which a mutex object resides.
1343 To know where the scope of variables ends. If e.g. heap memory
1344 has been used by one thread, that thread frees that memory, and
1345 another thread allocates and starts using that memory, no data
1346 races must be reported for that memory.
1353 It is essential for correct operation of DRD that the tool knows about
1354 memory allocation and deallocation events. When analyzing a client program
1355 with DRD that uses a custom memory allocator, either instrument the custom
1356 memory allocator with the <literal>VALGRIND_MALLOCLIKE_BLOCK</literal>
1357 and <literal>VALGRIND_FREELIKE_BLOCK</literal> macros or disable the
1358 custom memory allocator.
1362 As an example, the GNU libstdc++ library can be configured
1363 to use standard memory allocation functions instead of memory pools by
1364 setting the environment variable
1365 <literal>GLIBCXX_FORCE_NEW</literal>. For more information, see also
1367 url="http://gcc.gnu.org/onlinedocs/libstdc++/manual/bk01pt04ch11.html">libstdc++
1374 <sect2 id="drd-manual.drd-versus-memcheck" xreflabel="DRD Versus Memcheck">
1375 <title>DRD Versus Memcheck</title>
1378 It is essential for correct operation of DRD that there are no memory
1379 errors such as dangling pointers in the client program. Which means that
1380 it is a good idea to make sure that your program is Memcheck-clean
1381 before you analyze it with DRD. It is possible however that some of
1382 the Memcheck reports are caused by data races. In this case it makes
1383 sense to run DRD before Memcheck.
1387 So which tool should be run first? In case both DRD and Memcheck
1388 complain about a program, a possible approach is to run both tools
1389 alternatingly and to fix as many errors as possible after each run of
1390 each tool until none of the two tools prints any more error messages.
1396 <sect2 id="drd-manual.resource-requirements" xreflabel="Resource Requirements">
1397 <title>Resource Requirements</title>
1400 The requirements of DRD with regard to heap and stack memory and the
1401 effect on the execution time of client programs are as follows:
1405 When running a program under DRD with default DRD options,
1406 between 1.1 and 3.6 times more memory will be needed compared to
1407 a native run of the client program. More memory will be needed
1408 if loading debug information has been enabled
1409 (<literal>--read-var-info=yes</literal>).
1414 DRD allocates some of its temporary data structures on the stack
1415 of the client program threads. This amount of data is limited to
1416 1 - 2 KB. Make sure that thread stacks are sufficiently large.
1421 Most applications will run between 20 and 50 times slower under
1422 DRD than a native single-threaded run. The slowdown will be most
1423 noticeable for applications which perform frequent mutex lock /
1433 <sect2 id="drd-manual.effective-use" xreflabel="Effective Use">
1434 <title>Hints and Tips for Effective Use of DRD</title>
1437 The following information may be helpful when using DRD:
1441 Make sure that debug information is present in the executable
1442 being analyzed, such that DRD can print function name and line
1443 number information in stack traces. Most compilers can be told
1444 to include debug information via compiler option
1445 <option>-g</option>.
1450 Compile with option <option>-O1</option> instead of
1451 <option>-O0</option>. This will reduce the amount of generated
1452 code, may reduce the amount of debug info and will speed up
1453 DRD's processing of the client program. For more information,
1454 see also <xref linkend="manual-core.started"/>.
1459 If DRD reports any errors on libraries that are part of your
1460 Linux distribution like e.g. <literal>libc.so</literal> or
1461 <literal>libstdc++.so</literal>, installing the debug packages
1462 for these libraries will make the output of DRD a lot more
1468 When using C++, do not send output from more than one thread to
1469 <literal>std::cout</literal>. Doing so would not only
1470 generate multiple data race reports, it could also result in
1471 output from several threads getting mixed up. Either use
1472 <function>printf</function> or do the following:
1475 <para>Derive a class from <literal>std::ostreambuf</literal>
1476 and let that class send output line by line to
1477 <literal>stdout</literal>. This will avoid that individual
1478 lines of text produced by different threads get mixed
1482 <para>Create one instance of <literal>std::ostream</literal>
1483 for each thread. This makes stream formatting settings
1484 thread-local. Pass a per-thread instance of the class
1485 derived from <literal>std::ostreambuf</literal> to the
1486 constructor of each instance. </para>
1489 <para>Let each thread send its output to its own instance of
1490 <literal>std::ostream</literal> instead of
1491 <literal>std::cout</literal>.</para>
1505 <sect1 id="drd-manual.Pthreads" xreflabel="Pthreads">
1506 <title>Using the POSIX Threads API Effectively</title>
1508 <sect2 id="drd-manual.mutex-types" xreflabel="mutex-types">
1509 <title>Mutex types</title>
1512 The Single UNIX Specification version two defines the following four
1513 mutex types (see also the documentation of <ulink
1514 url="http://www.opengroup.org/onlinepubs/007908799/xsh/pthread_mutexattr_settype.html"><function>pthread_mutexattr_settype</function></ulink>):
1518 <emphasis>normal</emphasis>, which means that no error checking
1519 is performed, and that the mutex is non-recursive.
1524 <emphasis>error checking</emphasis>, which means that the mutex
1525 is non-recursive and that error checking is performed.
1530 <emphasis>recursive</emphasis>, which means that a mutex may be
1536 <emphasis>default</emphasis>, which means that error checking
1537 behavior is undefined, and that the behavior for recursive
1538 locking is also undefined. Or: portable code must neither
1539 trigger error conditions through the Pthreads API nor attempt to
1540 lock a mutex of default type recursively.
1547 In complex applications it is not always clear from beforehand which
1548 mutex will be locked recursively and which mutex will not be locked
1549 recursively. Attempts lock a non-recursive mutex recursively will
1550 result in race conditions that are very hard to find without a thread
1551 checking tool. So either use the error checking mutex type and
1552 consistently check the return value of Pthread API mutex calls, or use
1553 the recursive mutex type.
1558 <sect2 id="drd-manual.condvar" xreflabel="condition-variables">
1559 <title>Condition variables</title>
1562 A condition variable allows one thread to wake up one or more other
1563 threads. Condition variables are often used to notify one or more
1564 threads about state changes of shared data. Unfortunately it is very
1565 easy to introduce race conditions by using condition variables as the
1566 only means of state information propagation. A better approach is to
1567 let threads poll for changes of a state variable that is protected by
1568 a mutex, and to use condition variables only as a thread wakeup
1569 mechanism. See also the source file
1570 <computeroutput>drd/tests/monitor_example.cpp</computeroutput> for an
1571 example of how to implement this concept in C++. The monitor concept
1572 used in this example is a well known and very useful concept -- see
1573 also Wikipedia for more information about the <ulink
1574 url="http://en.wikipedia.org/wiki/Monitor_(synchronization)">monitor</ulink>
1580 <sect2 id="drd-manual.pctw" xreflabel="pthread_cond_timedwait">
1581 <title><function>pthread_cond_timedwait</function> and timeouts</title>
1584 Historically the function
1585 <function>pthread_cond_timedwait</function> only allowed the
1586 specification of an absolute timeout, that is a timeout independent of
1587 the time when this function was called. However, almost every call to
1588 this function expresses a relative timeout. This typically happens by
1590 <computeroutput>clock_gettime(CLOCK_REALTIME)</computeroutput> and a
1591 relative timeout as the third argument. This approach is incorrect
1592 since forward or backward clock adjustments by e.g. ntpd will affect
1593 the timeout. A more reliable approach is as follows:
1597 When initializing a condition variable through
1598 <function>pthread_cond_init</function>, specify that the timeout of
1599 <function>pthread_cond_timedwait</function> will use the clock
1600 <literal>CLOCK_MONOTONIC</literal> instead of
1601 <literal>CLOCK_REALTIME</literal>. You can do this via
1602 <computeroutput>pthread_condattr_setclock(...,
1603 CLOCK_MONOTONIC)</computeroutput>.
1608 When calling <function>pthread_cond_timedwait</function>, pass
1610 <computeroutput>clock_gettime(CLOCK_MONOTONIC)</computeroutput>
1611 and a relative timeout as the third argument.
1616 <computeroutput>drd/tests/monitor_example.cpp</computeroutput> for an
1622 <sect2 id="drd-manual.naming-threads" xreflabel="naming threads">
1623 <title>Assigning names to threads</title>
1626 Many applications log information about changes in internal or
1627 external state to a file. When analyzing log files of a multithreaded
1628 application it can be very convenient to know which thread logged
1629 which information. One possible approach is to identify threads in
1630 logging output by including the result of
1631 <function>pthread_self</function> in every log line. However, this approach
1632 has two disadvantages: there is no direct relationship between these
1633 values and the source code and these values can be different in each
1634 run. A better approach is to assign a brief name to each thread and to
1635 include the assigned thread name in each log line. One possible
1636 approach for managing thread names is as follows:
1640 Allocate a key for the pointer to the thread name through
1641 <function>pthread_key_create</function>.
1646 Just after thread creation, set the thread name through
1647 <function>pthread_setspecific</function>.
1652 In the code that generates the logging information, query the thread
1653 name by calling <function>pthread_getspecific</function>.
1665 <sect1 id="drd-manual.limitations" xreflabel="Limitations">
1666 <title>Limitations</title>
1668 <para>DRD currently has the following limitations:</para>
1673 DRD has only been tested on Linux and Mac OS X.
1678 Of the two POSIX threads implementations for Linux, only the
1679 NPTL (Native POSIX Thread Library) is supported. The older
1680 LinuxThreads library is not supported.
1685 DRD, just like Memcheck, will refuse to start on Linux
1686 distributions where all symbol information has been removed from
1687 <filename>ld.so</filename>. This is e.g. the case for the PPC editions
1688 of openSUSE and Gentoo. You will have to install the glibc debuginfo
1689 package on these platforms before you can use DRD. See also openSUSE
1690 bug <ulink url="http://bugzilla.novell.com/show_bug.cgi?id=396197">
1691 396197</ulink> and Gentoo bug <ulink
1692 url="http://bugs.gentoo.org/214065">214065</ulink>.
1697 When address tracing is enabled, no information on atomic stores
1698 will be displayed. This functionality is easy to add
1699 however. Please contact the Valgrind authors if you would like
1700 to see this functionality enabled.
1705 If you compile the DRD source code yourself, you need GCC 3.0 or
1706 later. GCC 2.95 is not supported.
1714 <sect1 id="drd-manual.feedback" xreflabel="Feedback">
1715 <title>Feedback</title>
1718 If you have any comments, suggestions, feedback or bug reports about
1719 DRD, feel free to either post a message on the Valgrind users mailing
1720 list or to file a bug report. See also <ulink
1721 url="&vg-url;">&vg-url;</ulink> for more information.