Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / external / valgrind / a0664b9ca67b594bd6f570a61d3301167a24750c / . / docs / html / drd-manual.html
blob: 0de09553caa85c3d984361239acb687858e784ad [file] [log] [blame]
Elliott Hughesa0664b92017-04-18 17:46:52 -0700[diff] [blame^]1<html>
2<head>
3<meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1">
4<title>8. DRD: a thread error detector</title>
5<link rel="stylesheet" type="text/css" href="vg_basic.css">
6<meta name="generator" content="DocBook XSL Stylesheets V1.78.1">
7<link rel="home" href="index.html" title="Valgrind Documentation">
8<link rel="up" href="manual.html" title="Valgrind User Manual">
9<link rel="prev" href="hg-manual.html" title="7. Helgrind: a thread error detector">
10<link rel="next" href="ms-manual.html" title="9. Massif: a heap profiler">
11</head>
12<body bgcolor="white" text="black" link="#0000FF" vlink="#840084" alink="#0000FF">
13<div><table class="nav" width="100%" cellspacing="3" cellpadding="3" border="0" summary="Navigation header"><tr>
14<td width="22px" align="center" valign="middle"><a accesskey="p" href="hg-manual.html"><img src="images/prev.png" width="18" height="21" border="0" alt="Prev"></a></td>
15<td width="25px" align="center" valign="middle"><a accesskey="u" href="manual.html"><img src="images/up.png" width="21" height="18" border="0" alt="Up"></a></td>
16<td width="31px" align="center" valign="middle"><a accesskey="h" href="index.html"><img src="images/home.png" width="27" height="20" border="0" alt="Up"></a></td>
17<th align="center" valign="middle">Valgrind User Manual</th>
18<td width="22px" align="center" valign="middle"><a accesskey="n" href="ms-manual.html"><img src="images/next.png" width="18" height="21" border="0" alt="Next"></a></td>
19</tr></table></div>
20<div class="chapter">
21<div class="titlepage"><div><div><h1 class="title">
22<a name="drd-manual"></a>8. DRD: a thread error detector</h1></div></div></div>
23<div class="toc">
24<p><b>Table of Contents</b></p>
25<dl class="toc">
26<dt><span class="sect1"><a href="drd-manual.html#drd-manual.overview">8.1. Overview</a></span></dt>
27<dd><dl>
28<dt><span class="sect2"><a href="drd-manual.html#drd-manual.mt-progr-models">8.1.1. Multithreaded Programming Paradigms</a></span></dt>
29<dt><span class="sect2"><a href="drd-manual.html#drd-manual.pthreads-model">8.1.2. POSIX Threads Programming Model</a></span></dt>
30<dt><span class="sect2"><a href="drd-manual.html#drd-manual.mt-problems">8.1.3. Multithreaded Programming Problems</a></span></dt>
31<dt><span class="sect2"><a href="drd-manual.html#drd-manual.data-race-detection">8.1.4. Data Race Detection</a></span></dt>
32</dl></dd>
33<dt><span class="sect1"><a href="drd-manual.html#drd-manual.using-drd">8.2. Using DRD</a></span></dt>
34<dd><dl>
35<dt><span class="sect2"><a href="drd-manual.html#drd-manual.options">8.2.1. DRD Command-line Options</a></span></dt>
36<dt><span class="sect2"><a href="drd-manual.html#drd-manual.data-races">8.2.2. Detected Errors: Data Races</a></span></dt>
37<dt><span class="sect2"><a href="drd-manual.html#drd-manual.lock-contention">8.2.3. Detected Errors: Lock Contention</a></span></dt>
38<dt><span class="sect2"><a href="drd-manual.html#drd-manual.api-checks">8.2.4. Detected Errors: Misuse of the POSIX threads API</a></span></dt>
39<dt><span class="sect2"><a href="drd-manual.html#drd-manual.clientreqs">8.2.5. Client Requests</a></span></dt>
40<dt><span class="sect2"><a href="drd-manual.html#drd-manual.C++11">8.2.6. Debugging C++11 Programs</a></span></dt>
41<dt><span class="sect2"><a href="drd-manual.html#drd-manual.gnome">8.2.7. Debugging GNOME Programs</a></span></dt>
42<dt><span class="sect2"><a href="drd-manual.html#drd-manual.boost.thread">8.2.8. Debugging Boost.Thread Programs</a></span></dt>
43<dt><span class="sect2"><a href="drd-manual.html#drd-manual.openmp">8.2.9. Debugging OpenMP Programs</a></span></dt>
44<dt><span class="sect2"><a href="drd-manual.html#drd-manual.cust-mem-alloc">8.2.10. DRD and Custom Memory Allocators</a></span></dt>
45<dt><span class="sect2"><a href="drd-manual.html#drd-manual.drd-versus-memcheck">8.2.11. DRD Versus Memcheck</a></span></dt>
46<dt><span class="sect2"><a href="drd-manual.html#drd-manual.resource-requirements">8.2.12. Resource Requirements</a></span></dt>
47<dt><span class="sect2"><a href="drd-manual.html#drd-manual.effective-use">8.2.13. Hints and Tips for Effective Use of DRD</a></span></dt>
48</dl></dd>
49<dt><span class="sect1"><a href="drd-manual.html#drd-manual.Pthreads">8.3. Using the POSIX Threads API Effectively</a></span></dt>
50<dd><dl>
51<dt><span class="sect2"><a href="drd-manual.html#drd-manual.mutex-types">8.3.1. Mutex types</a></span></dt>
52<dt><span class="sect2"><a href="drd-manual.html#drd-manual.condvar">8.3.2. Condition variables</a></span></dt>
53<dt><span class="sect2"><a href="drd-manual.html#drd-manual.pctw">8.3.3. pthread_cond_timedwait and timeouts</a></span></dt>
54</dl></dd>
55<dt><span class="sect1"><a href="drd-manual.html#drd-manual.limitations">8.4. Limitations</a></span></dt>
56<dt><span class="sect1"><a href="drd-manual.html#drd-manual.feedback">8.5. Feedback</a></span></dt>
57</dl>
58</div>
59<p>To use this tool, you must specify
60<code class="option">--tool=drd</code>
61on the Valgrind command line.</p>
62<div class="sect1">
63<div class="titlepage"><div><div><h2 class="title" style="clear: both">
64<a name="drd-manual.overview"></a>8.1. Overview</h2></div></div></div>
65<p>
66DRD is a Valgrind tool for detecting errors in multithreaded C and C++
67programs. The tool works for any program that uses the POSIX threading
68primitives or that uses threading concepts built on top of the POSIX threading
69primitives.
70</p>
71<div class="sect2">
72<div class="titlepage"><div><div><h3 class="title">
73<a name="drd-manual.mt-progr-models"></a>8.1.1. Multithreaded Programming Paradigms</h3></div></div></div>
74<p>
75There are two possible reasons for using multithreading in a program:
76</p>
77<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
78<li class="listitem"><p>
79 To model concurrent activities. Assigning one thread to each activity
80 can be a great simplification compared to multiplexing the states of
81 multiple activities in a single thread. This is why most server software
82 and embedded software is multithreaded.
83 </p></li>
84<li class="listitem"><p>
85 To use multiple CPU cores simultaneously for speeding up
86 computations. This is why many High Performance Computing (HPC)
87 applications are multithreaded.
88 </p></li>
89</ul></div>
90<p>
91</p>
92<p>
93Multithreaded programs can use one or more of the following programming
94paradigms. Which paradigm is appropriate depends e.g. on the application type.
95Some examples of multithreaded programming paradigms are:
96</p>
97<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
98<li class="listitem"><p>
99 Locking. Data that is shared over threads is protected from concurrent
100 accesses via locking. E.g. the POSIX threads library, the Qt library
101 and the Boost.Thread library support this paradigm directly.
102 </p></li>
103<li class="listitem"><p>
104 Message passing. No data is shared between threads, but threads exchange
105 data by passing messages to each other. Examples of implementations of
106 the message passing paradigm are MPI and CORBA.
107 </p></li>
108<li class="listitem"><p>
109 Automatic parallelization. A compiler converts a sequential program into
110 a multithreaded program. The original program may or may not contain
111 parallelization hints. One example of such parallelization hints is the
112 OpenMP standard. In this standard a set of directives are defined which
113 tell a compiler how to parallelize a C, C++ or Fortran program. OpenMP
114 is well suited for computational intensive applications. As an example,
115 an open source image processing software package is using OpenMP to
116 maximize performance on systems with multiple CPU
117 cores. GCC supports the
118 OpenMP standard from version 4.2.0 on.
119 </p></li>
120<li class="listitem"><p>
121 Software Transactional Memory (STM). Any data that is shared between
122 threads is updated via transactions. After each transaction it is
123 verified whether there were any conflicting transactions. If there were
124 conflicts, the transaction is aborted, otherwise it is committed. This
125 is a so-called optimistic approach. There is a prototype of the Intel C++
126 Compiler available that supports STM. Research about the addition of
127 STM support to GCC is ongoing.
128 </p></li>
129</ul></div>
130<p>
131</p>
132<p>
133DRD supports any combination of multithreaded programming paradigms as
134long as the implementation of these paradigms is based on the POSIX
135threads primitives. DRD however does not support programs that use
136e.g. Linux' futexes directly. Attempts to analyze such programs with
137DRD will cause DRD to report many false positives.
138</p>
139</div>
140<div class="sect2">
141<div class="titlepage"><div><div><h3 class="title">
142<a name="drd-manual.pthreads-model"></a>8.1.2. POSIX Threads Programming Model</h3></div></div></div>
143<p>
144POSIX threads, also known as Pthreads, is the most widely available
145threading library on Unix systems.
146</p>
147<p>
148The POSIX threads programming model is based on the following abstractions:
149</p>
150<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
151<li class="listitem"><p>
152 A shared address space. All threads running within the same
153 process share the same address space. All data, whether shared or
154 not, is identified by its address.
155 </p></li>
156<li class="listitem"><p>
157 Regular load and store operations, which allow to read values
158 from or to write values to the memory shared by all threads
159 running in the same process.
160 </p></li>
161<li class="listitem"><p>
162 Atomic store and load-modify-store operations. While these are
163 not mentioned in the POSIX threads standard, most
164 microprocessors support atomic memory operations.
165 </p></li>
166<li class="listitem"><p>
167 Threads. Each thread represents a concurrent activity.
168 </p></li>
169<li class="listitem"><p>
170 Synchronization objects and operations on these synchronization
171 objects. The following types of synchronization objects have been
172 defined in the POSIX threads standard: mutexes, condition variables,
173 semaphores, reader-writer synchronization objects, barriers and
174 spinlocks.
175 </p></li>
176</ul></div>
177<p>
178</p>
179<p>
180Which source code statements generate which memory accesses depends on
181the <span class="emphasis"><em>memory model</em></span> of the programming language being
182used. There is not yet a definitive memory model for the C and C++
183languages. For a draft memory model, see also the document
184<a class="ulink" href="http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2338.html" target="_top">
185WG21/N2338: Concurrency memory model compiler consequences</a>.
186</p>
187<p>
188For more information about POSIX threads, see also the Single UNIX
189Specification version 3, also known as
190<a class="ulink" href="http://www.opengroup.org/onlinepubs/000095399/idx/threads.html" target="_top">
191IEEE Std 1003.1</a>.
192</p>
193</div>
194<div class="sect2">
195<div class="titlepage"><div><div><h3 class="title">
196<a name="drd-manual.mt-problems"></a>8.1.3. Multithreaded Programming Problems</h3></div></div></div>
197<p>
198Depending on which multithreading paradigm is being used in a program,
199one or more of the following problems can occur:
200</p>
201<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
202<li class="listitem"><p>
203 Data races. One or more threads access the same memory location without
204 sufficient locking. Most but not all data races are programming errors
205 and are the cause of subtle and hard-to-find bugs.
206 </p></li>
207<li class="listitem"><p>
208 Lock contention. One thread blocks the progress of one or more other
209 threads by holding a lock too long.
210 </p></li>
211<li class="listitem"><p>
212 Improper use of the POSIX threads API. Most implementations of the POSIX
213 threads API have been optimized for runtime speed. Such implementations
214 will not complain on certain errors, e.g. when a mutex is being unlocked
215 by another thread than the thread that obtained a lock on the mutex.
216 </p></li>
217<li class="listitem"><p>
218 Deadlock. A deadlock occurs when two or more threads wait for
219 each other indefinitely.
220 </p></li>
221<li class="listitem"><p>
222 False sharing. If threads that run on different processor cores
223 access different variables located in the same cache line
224 frequently, this will slow down the involved threads a lot due
225 to frequent exchange of cache lines.
226 </p></li>
227</ul></div>
228<p>
229</p>
230<p>
231Although the likelihood of the occurrence of data races can be reduced
232through a disciplined programming style, a tool for automatic
233detection of data races is a necessity when developing multithreaded
234software. DRD can detect these, as well as lock contention and
235improper use of the POSIX threads API.
236</p>
237</div>
238<div class="sect2">
239<div class="titlepage"><div><div><h3 class="title">
240<a name="drd-manual.data-race-detection"></a>8.1.4. Data Race Detection</h3></div></div></div>
241<p>
242The result of load and store operations performed by a multithreaded program
243depends on the order in which memory operations are performed. This order is
244determined by:
245</p>
246<div class="orderedlist"><ol class="orderedlist" type="1">
247<li class="listitem"><p>
248 All memory operations performed by the same thread are performed in
249 <span class="emphasis"><em>program order</em></span>, that is, the order determined by the
250 program source code and the results of previous load operations.
251 </p></li>
252<li class="listitem"><p>
253 Synchronization operations determine certain ordering constraints on
254 memory operations performed by different threads. These ordering
255 constraints are called the <span class="emphasis"><em>synchronization order</em></span>.
256 </p></li>
257</ol></div>
258<p>
259The combination of program order and synchronization order is called the
260<span class="emphasis"><em>happens-before relationship</em></span>. This concept was first
261defined by S. Adve et al in the paper <span class="emphasis"><em>Detecting data races on weak
262memory systems</em></span>, ACM SIGARCH Computer Architecture News, v.19 n.3,
263p.234-243, May 1991.
264</p>
265<p>
266Two memory operations <span class="emphasis"><em>conflict</em></span> if both operations are
267performed by different threads, refer to the same memory location and at least
268one of them is a store operation.
269</p>
270<p>
271A multithreaded program is <span class="emphasis"><em>data-race free</em></span> if all
272conflicting memory accesses are ordered by synchronization
273operations.
274</p>
275<p>
276A well known way to ensure that a multithreaded program is data-race
277free is to ensure that a locking discipline is followed. It is e.g.
278possible to associate a mutex with each shared data item, and to hold
279a lock on the associated mutex while the shared data is accessed.
280</p>
281<p>
282All programs that follow a locking discipline are data-race free, but not all
283data-race free programs follow a locking discipline. There exist multithreaded
284programs where access to shared data is arbitrated via condition variables,
285semaphores or barriers. As an example, a certain class of HPC applications
286consists of a sequence of computation steps separated in time by barriers, and
287where these barriers are the only means of synchronization. Although there are
288many conflicting memory accesses in such applications and although such
289applications do not make use mutexes, most of these applications do not
290contain data races.
291</p>
292<p>
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</p>
304<p>
305DRD is based on the happens-before algorithm.
306</p>
307</div>
308</div>
309<div class="sect1">
310<div class="titlepage"><div><div><h2 class="title" style="clear: both">
311<a name="drd-manual.using-drd"></a>8.2. Using DRD</h2></div></div></div>
312<div class="sect2">
313<div class="titlepage"><div><div><h3 class="title">
314<a name="drd-manual.options"></a>8.2.1. DRD Command-line Options</h3></div></div></div>
315<p>The following command-line options are available for controlling the
316behavior of the DRD tool itself:</p>
317<div class="variablelist">
318<a name="drd.opts.list"></a><dl class="variablelist">
319<dt><span class="term">
320 <code class="option">--check-stack-var=&lt;yes|no&gt; [default: no]</code>
321 </span></dt>
322<dd><p>
323 Controls whether DRD detects data races on stack
324 variables. Verifying stack variables is disabled by default because
325 most programs do not share stack variables over threads.
326 </p></dd>
327<dt><span class="term">
328 <code class="option">--exclusive-threshold=&lt;n&gt; [default: off]</code>
329 </span></dt>
330<dd><p>
331 Print an error message if any mutex or writer lock has been
332 held longer than the time specified in milliseconds. This
333 option enables the detection of lock contention.
334 </p></dd>
335<dt><span class="term">
336 <code class="option">--join-list-vol=&lt;n&gt; [default: 10]</code>
337 </span></dt>
338<dd><p>
339 Data races that occur between a statement at the end of one thread
340 and another thread can be missed if memory access information is
341 discarded immediately after a thread has been joined. This option
342 allows to specify for how many joined threads memory access information
343 should be retained.
344 </p></dd>
345<dt><span class="term">
346 <code class="option">
347 --first-race-only=&lt;yes|no&gt; [default: no]
348 </code>
349 </span></dt>
350<dd><p>
351 Whether to report only the first data race that has been detected on a
352 memory location or all data races that have been detected on a memory
353 location.
354 </p></dd>
355<dt><span class="term">
356 <code class="option">
357 --free-is-write=&lt;yes|no&gt; [default: no]
358 </code>
359 </span></dt>
360<dd>
361<p>
362 Whether to report races between accessing memory and freeing
363 memory. Enabling this option may cause DRD to run slightly
364 slower. Notes:</p>
365<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
366<li class="listitem"><p>
367 Don't enable this option when using custom memory allocators
368 that use
369 the <code class="computeroutput">VG_USERREQ__MALLOCLIKE_BLOCK</code>
370 and <code class="computeroutput">VG_USERREQ__FREELIKE_BLOCK</code>
371 because that would result in false positives.
372 </p></li>
373<li class="listitem"><p>Don't enable this option when using reference-counted
374 objects because that will result in false positives, even when
375 that code has been annotated properly with
376 <code class="computeroutput">ANNOTATE_HAPPENS_BEFORE</code>
377 and <code class="computeroutput">ANNOTATE_HAPPENS_AFTER</code>. See
378 e.g. the output of the following command for an example:
379 <code class="computeroutput">valgrind --tool=drd --free-is-write=yes
380 drd/tests/annotate_smart_pointer</code>.
381 </p></li>
382</ul></div>
383</dd>
384<dt><span class="term">
385 <code class="option">
386 --report-signal-unlocked=&lt;yes|no&gt; [default: yes]
387 </code>
388 </span></dt>
389<dd><p>
390 Whether to report calls to
391 <code class="function">pthread_cond_signal</code> and
392 <code class="function">pthread_cond_broadcast</code> where the mutex
393 associated with the signal through
394 <code class="function">pthread_cond_wait</code> or
395 <code class="function">pthread_cond_timed_wait</code>is not locked at
396 the time the signal is sent. Sending a signal without holding
397 a lock on the associated mutex is a common programming error
398 which can cause subtle race conditions and unpredictable
399 behavior. There exist some uncommon synchronization patterns
400 however where it is safe to send a signal without holding a
401 lock on the associated mutex.
402 </p></dd>
403<dt><span class="term">
404 <code class="option">--segment-merging=&lt;yes|no&gt; [default: yes]</code>
405 </span></dt>
406<dd><p>
407 Controls segment merging. Segment merging is an algorithm to
408 limit memory usage of the data race detection
409 algorithm. Disabling segment merging may improve the accuracy
410 of the so-called 'other segments' displayed in race reports
411 but can also trigger an out of memory error.
412 </p></dd>
413<dt><span class="term">
414 <code class="option">--segment-merging-interval=&lt;n&gt; [default: 10]</code>
415 </span></dt>
416<dd><p>
417 Perform segment merging only after the specified number of new
418 segments have been created. This is an advanced configuration option
419 that allows to choose whether to minimize DRD's memory usage by
420 choosing a low value or to let DRD run faster by choosing a slightly
421 higher value. The optimal value for this parameter depends on the
422 program being analyzed. The default value works well for most programs.
423 </p></dd>
424<dt><span class="term">
425 <code class="option">--shared-threshold=&lt;n&gt; [default: off]</code>
426 </span></dt>
427<dd><p>
428 Print an error message if a reader lock has been held longer
429 than the specified time (in milliseconds). This option enables
430 the detection of lock contention.
431 </p></dd>
432<dt><span class="term">
433 <code class="option">--show-confl-seg=&lt;yes|no&gt; [default: yes]</code>
434 </span></dt>
435<dd><p>
436 Show conflicting segments in race reports. Since this
437 information can help to find the cause of a data race, this
438 option is enabled by default. Disabling this option makes the
439 output of DRD more compact.
440 </p></dd>
441<dt><span class="term">
442 <code class="option">--show-stack-usage=&lt;yes|no&gt; [default: no]</code>
443 </span></dt>
444<dd><p>
445 Print stack usage at thread exit time. When a program creates a large
446 number of threads it becomes important to limit the amount of virtual
447 memory allocated for thread stacks. This option makes it possible to
448 observe how much stack memory has been used by each thread of the
449 client program. Note: the DRD tool itself allocates some temporary
450 data on the client thread stack. The space necessary for this
451 temporary data must be allocated by the client program when it
452 allocates stack memory, but is not included in stack usage reported by
453 DRD.
454 </p></dd>
455<dt><span class="term">
456 <code class="option">--ignore-thread-creation=&lt;yes|no&gt; [default: no]</code>
457 </span></dt>
458<dd>
459<p>
460 Controls whether all activities during thread creation should be
461 ignored. By default enabled only on Solaris.
462 Solaris provides higher throughput, parallelism and scalability than
463 other operating systems, at the cost of more fine-grained locking
464 activity. This means for example that when a thread is created under
465 glibc, just one big lock is used for all thread setup. Solaris libc
466 uses several fine-grained locks and the creator thread resumes its
467 activities as soon as possible, leaving for example stack and TLS setup
468 sequence to the created thread.
469 This situation confuses DRD as it assumes there is some false ordering
470 in place between creator and created thread; and therefore many types
471 of race conditions in the application would not be reported. To prevent
472 such false ordering, this command line option is set to
473 <code class="computeroutput">yes</code> by default on Solaris.
474 All activity (loads, stores, client requests) is therefore ignored
475 during:</p>
476<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
477<li class="listitem"><p>
478 pthread_create() call in the creator thread
479 </p></li>
480<li class="listitem"><p>
481 thread creation phase (stack and TLS setup) in the created thread
482 </p></li>
483</ul></div>
484</dd>
485</dl>
486</div>
487<p>
488The following options are available for monitoring the behavior of the
489client program:
490</p>
491<div class="variablelist">
492<a name="drd.debugopts.list"></a><dl class="variablelist">
493<dt><span class="term">
494 <code class="option">--trace-addr=&lt;address&gt; [default: none]</code>
495 </span></dt>
496<dd><p>
497 Trace all load and store activity for the specified
498 address. This option may be specified more than once.
499 </p></dd>
500<dt><span class="term">
501 <code class="option">--ptrace-addr=&lt;address&gt; [default: none]</code>
502 </span></dt>
503<dd><p>
504 Trace all load and store activity for the specified address and keep
505 doing that even after the memory at that address has been freed and
506 reallocated.
507 </p></dd>
508<dt><span class="term">
509 <code class="option">--trace-alloc=&lt;yes|no&gt; [default: no]</code>
510 </span></dt>
511<dd><p>
512 Trace all memory allocations and deallocations. May produce a huge
513 amount of output.
514 </p></dd>
515<dt><span class="term">
516 <code class="option">--trace-barrier=&lt;yes|no&gt; [default: no]</code>
517 </span></dt>
518<dd><p>
519 Trace all barrier activity.
520 </p></dd>
521<dt><span class="term">
522 <code class="option">--trace-cond=&lt;yes|no&gt; [default: no]</code>
523 </span></dt>
524<dd><p>
525 Trace all condition variable activity.
526 </p></dd>
527<dt><span class="term">
528 <code class="option">--trace-fork-join=&lt;yes|no&gt; [default: no]</code>
529 </span></dt>
530<dd><p>
531 Trace all thread creation and all thread termination events.
532 </p></dd>
533<dt><span class="term">
534 <code class="option">--trace-hb=&lt;yes|no&gt; [default: no]</code>
535 </span></dt>
536<dd><p>
537 Trace execution of the <code class="literal">ANNOTATE_HAPPENS_BEFORE()</code>,
538 <code class="literal">ANNOTATE_HAPPENS_AFTER()</code> and
539 <code class="literal">ANNOTATE_HAPPENS_DONE()</code> client requests.
540 </p></dd>
541<dt><span class="term">
542 <code class="option">--trace-mutex=&lt;yes|no&gt; [default: no]</code>
543 </span></dt>
544<dd><p>
545 Trace all mutex activity.
546 </p></dd>
547<dt><span class="term">
548 <code class="option">--trace-rwlock=&lt;yes|no&gt; [default: no]</code>
549 </span></dt>
550<dd><p>
551 Trace all reader-writer lock activity.
552 </p></dd>
553<dt><span class="term">
554 <code class="option">--trace-semaphore=&lt;yes|no&gt; [default: no]</code>
555 </span></dt>
556<dd><p>
557 Trace all semaphore activity.
558 </p></dd>
559</dl>
560</div>
561</div>
562<div class="sect2">
563<div class="titlepage"><div><div><h3 class="title">
564<a name="drd-manual.data-races"></a>8.2.2. Detected Errors: Data Races</h3></div></div></div>
565<p>
566DRD prints a message every time it detects a data race. Please keep
567the following in mind when interpreting DRD's output:
568</p>
569<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
570<li class="listitem"><p>
571 Every thread is assigned a <span class="emphasis"><em>thread ID</em></span> by the DRD
572 tool. A thread ID is a number. Thread ID's start at one and are never
573 recycled.
574 </p></li>
575<li class="listitem"><p>
576 The term <span class="emphasis"><em>segment</em></span> refers to a consecutive
577 sequence of load, store and synchronization operations, all
578 issued by the same thread. A segment always starts and ends at a
579 synchronization operation. Data race analysis is performed
580 between segments instead of between individual load and store
581 operations because of performance reasons.
582 </p></li>
583<li class="listitem"><p>
584 There are always at least two memory accesses involved in a data
585 race. Memory accesses involved in a data race are called
586 <span class="emphasis"><em>conflicting memory accesses</em></span>. DRD prints a
587 report for each memory access that conflicts with a past memory
588 access.
589 </p></li>
590</ul></div>
591<p>
592</p>
593<p>
594Below you can find an example of a message printed by DRD when it
595detects a data race:
596</p>
597<pre class="programlisting">
598$ valgrind --tool=drd --read-var-info=yes drd/tests/rwlock_race
599...
600==9466== Thread 3:
601==9466== Conflicting load by thread 3 at 0x006020b8 size 4
602==9466== at 0x400B6C: thread_func (rwlock_race.c:29)
603==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186)
604==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so)
605==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so)
606==9466== Location 0x6020b8 is 0 bytes inside local var "s_racy"
607==9466== declared at rwlock_race.c:18, in frame #0 of thread 3
608==9466== Other segment start (thread 2)
609==9466== at 0x4C2847D: pthread_rwlock_rdlock* (drd_pthread_intercepts.c:813)
610==9466== by 0x400B6B: thread_func (rwlock_race.c:28)
611==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186)
612==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so)
613==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so)
614==9466== Other segment end (thread 2)
615==9466== at 0x4C28B54: pthread_rwlock_unlock* (drd_pthread_intercepts.c:912)
616==9466== by 0x400B84: thread_func (rwlock_race.c:30)
617==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186)
618==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so)
619==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so)
620...
621</pre>
622<p>
623The above report has the following meaning:
624</p>
625<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
626<li class="listitem"><p>
627 The number in the column on the left is the process ID of the
628 process being analyzed by DRD.
629 </p></li>
630<li class="listitem"><p>
631 The first line ("Thread 3") tells you the thread ID for
632 the thread in which context the data race has been detected.
633 </p></li>
634<li class="listitem"><p>
635 The next line tells which kind of operation was performed (load or
636 store) and by which thread. On the same line the start address and the
637 number of bytes involved in the conflicting access are also displayed.
638 </p></li>
639<li class="listitem"><p>
640 Next, the call stack of the conflicting access is displayed. If
641 your program has been compiled with debug information
642 (<code class="option">-g</code>), this call stack will include file names and
643 line numbers. The two
644 bottommost frames in this call stack (<code class="function">clone</code>
645 and <code class="function">start_thread</code>) show how the NPTL starts
646 a thread. The third frame
647 (<code class="function">vg_thread_wrapper</code>) is added by DRD. The
648 fourth frame (<code class="function">thread_func</code>) is the first
649 interesting line because it shows the thread entry point, that
650 is the function that has been passed as the third argument to
651 <code class="function">pthread_create</code>.
652 </p></li>
653<li class="listitem"><p>
654 Next, the allocation context for the conflicting address is
655 displayed. For dynamically allocated data the allocation call
656 stack is shown. For static variables and stack variables the
657 allocation context is only shown when the option
658 <code class="option">--read-var-info=yes</code> has been
659 specified. Otherwise DRD will print <code class="computeroutput">Allocation
660 context: unknown</code>.
661 </p></li>
662<li class="listitem">
663<p>
664 A conflicting access involves at least two memory accesses. For
665 one of these accesses an exact call stack is displayed, and for
666 the other accesses an approximate call stack is displayed,
667 namely the start and the end of the segments of the other
668 accesses. This information can be interpreted as follows:
669 </p>
670<div class="orderedlist"><ol class="orderedlist" type="1">
671<li class="listitem"><p>
672 Start at the bottom of both call stacks, and count the
673 number stack frames with identical function name, file
674 name and line number. In the above example the three
675 bottommost frames are identical
676 (<code class="function">clone</code>,
677 <code class="function">start_thread</code> and
678 <code class="function">vg_thread_wrapper</code>).
679 </p></li>
680<li class="listitem"><p>
681 The next higher stack frame in both call stacks now tells
682 you between in which source code region the other memory
683 access happened. The above output tells that the other
684 memory access involved in the data race happened between
685 source code lines 28 and 30 in file
686 <code class="computeroutput">rwlock_race.c</code>.
687 </p></li>
688</ol></div>
689<p>
690 </p>
691</li>
692</ul></div>
693<p>
694</p>
695</div>
696<div class="sect2">
697<div class="titlepage"><div><div><h3 class="title">
698<a name="drd-manual.lock-contention"></a>8.2.3. Detected Errors: Lock Contention</h3></div></div></div>
699<p>
700Threads must be able to make progress without being blocked for too long by
701other threads. Sometimes a thread has to wait until a mutex or reader-writer
702synchronization object is unlocked by another thread. This is called
703<span class="emphasis"><em>lock contention</em></span>.
704</p>
705<p>
706Lock contention causes delays. Such delays should be as short as
707possible. The two command line options
708<code class="literal">--exclusive-threshold=&lt;n&gt;</code> and
709<code class="literal">--shared-threshold=&lt;n&gt;</code> make it possible to
710detect excessive lock contention by making DRD report any lock that
711has been held longer than the specified threshold. An example:
712</p>
713<pre class="programlisting">
714$ valgrind --tool=drd --exclusive-threshold=10 drd/tests/hold_lock -i 500
715...
716==10668== Acquired at:
717==10668== at 0x4C267C8: pthread_mutex_lock (drd_pthread_intercepts.c:395)
718==10668== by 0x400D92: main (hold_lock.c:51)
719==10668== Lock on mutex 0x7fefffd50 was held during 503 ms (threshold: 10 ms).
720==10668== at 0x4C26ADA: pthread_mutex_unlock (drd_pthread_intercepts.c:441)
721==10668== by 0x400DB5: main (hold_lock.c:55)
722...
723</pre>
724<p>
725The <code class="literal">hold_lock</code> test program holds a lock as long as
726specified by the <code class="literal">-i</code> (interval) argument. The DRD
727output reports that the lock acquired at line 51 in source file
728<code class="literal">hold_lock.c</code> and released at line 55 was held during
729503 ms, while a threshold of 10 ms was specified to DRD.
730</p>
731</div>
732<div class="sect2">
733<div class="titlepage"><div><div><h3 class="title">
734<a name="drd-manual.api-checks"></a>8.2.4. Detected Errors: Misuse of the POSIX threads API</h3></div></div></div>
735<p>
736 DRD is able to detect and report the following misuses of the POSIX
737 threads API:
738 </p>
739<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
740<li class="listitem"><p>
741 Passing the address of one type of synchronization object
742 (e.g. a mutex) to a POSIX API call that expects a pointer to
743 another type of synchronization object (e.g. a condition
744 variable).
745 </p></li>
746<li class="listitem"><p>
747 Attempts to unlock a mutex that has not been locked.
748 </p></li>
749<li class="listitem"><p>
750 Attempts to unlock a mutex that was locked by another thread.
751 </p></li>
752<li class="listitem"><p>
753 Attempts to lock a mutex of type
754 <code class="literal">PTHREAD_MUTEX_NORMAL</code> or a spinlock
755 recursively.
756 </p></li>
757<li class="listitem"><p>
758 Destruction or deallocation of a locked mutex.
759 </p></li>
760<li class="listitem"><p>
761 Sending a signal to a condition variable while no lock is held
762 on the mutex associated with the condition variable.
763 </p></li>
764<li class="listitem"><p>
765 Calling <code class="function">pthread_cond_wait</code> on a mutex
766 that is not locked, that is locked by another thread or that
767 has been locked recursively.
768 </p></li>
769<li class="listitem"><p>
770 Associating two different mutexes with a condition variable
771 through <code class="function">pthread_cond_wait</code>.
772 </p></li>
773<li class="listitem"><p>
774 Destruction or deallocation of a condition variable that is
775 being waited upon.
776 </p></li>
777<li class="listitem"><p>
778 Destruction or deallocation of a locked reader-writer synchronization
779 object.
780 </p></li>
781<li class="listitem"><p>
782 Attempts to unlock a reader-writer synchronization object that was not
783 locked by the calling thread.
784 </p></li>
785<li class="listitem"><p>
786 Attempts to recursively lock a reader-writer synchronization object
787 exclusively.
788 </p></li>
789<li class="listitem"><p>
790 Attempts to pass the address of a user-defined reader-writer
791 synchronization object to a POSIX threads function.
792 </p></li>
793<li class="listitem"><p>
794 Attempts to pass the address of a POSIX reader-writer synchronization
795 object to one of the annotations for user-defined reader-writer
796 synchronization objects.
797 </p></li>
798<li class="listitem"><p>
799 Reinitialization of a mutex, condition variable, reader-writer
800 lock, semaphore or barrier.
801 </p></li>
802<li class="listitem"><p>
803 Destruction or deallocation of a semaphore or barrier that is
804 being waited upon.
805 </p></li>
806<li class="listitem"><p>
807 Missing synchronization between barrier wait and barrier destruction.
808 </p></li>
809<li class="listitem"><p>
810 Exiting a thread without first unlocking the spinlocks, mutexes or
811 reader-writer synchronization objects that were locked by that thread.
812 </p></li>
813<li class="listitem"><p>
814 Passing an invalid thread ID to <code class="function">pthread_join</code>
815 or <code class="function">pthread_cancel</code>.
816 </p></li>
817</ul></div>
818<p>
819</p>
820</div>
821<div class="sect2">
822<div class="titlepage"><div><div><h3 class="title">
823<a name="drd-manual.clientreqs"></a>8.2.5. Client Requests</h3></div></div></div>
824<p>
825Just as for other Valgrind tools it is possible to let a client program
826interact with the DRD tool through client requests. In addition to the
827client requests several macros have been defined that allow to use the
828client requests in a convenient way.
829</p>
830<p>
831The interface between client programs and the DRD tool is defined in
832the header file <code class="literal">&lt;valgrind/drd.h&gt;</code>. The
833available macros and client requests are:
834</p>
835<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
836<li class="listitem"><p>
837 The macro <code class="literal">DRD_GET_VALGRIND_THREADID</code> and the
838 corresponding client
839 request <code class="varname">VG_USERREQ__DRD_GET_VALGRIND_THREAD_ID</code>.
840 Query the thread ID that has been assigned by the Valgrind core to the
841 thread executing this client request. Valgrind's thread ID's start at
842 one and are recycled in case a thread stops.
843 </p></li>
844<li class="listitem"><p>
845 The macro <code class="literal">DRD_GET_DRD_THREADID</code> and the corresponding
846 client request <code class="varname">VG_USERREQ__DRD_GET_DRD_THREAD_ID</code>.
847 Query the thread ID that has been assigned by DRD to the thread
848 executing this client request. These are the thread ID's reported by DRD
849 in data race reports and in trace messages. DRD's thread ID's start at
850 one and are never recycled.
851 </p></li>
852<li class="listitem"><p>
853 The macros <code class="literal">DRD_IGNORE_VAR(x)</code>,
854 <code class="literal">ANNOTATE_TRACE_MEMORY(&amp;x)</code> and the corresponding
855 client request <code class="varname">VG_USERREQ__DRD_START_SUPPRESSION</code>. Some
856 applications contain intentional races. There exist e.g. applications
857 where the same value is assigned to a shared variable from two different
858 threads. It may be more convenient to suppress such races than to solve
859 these. This client request allows to suppress such races.
860 </p></li>
861<li class="listitem"><p>
862 The macro <code class="literal">DRD_STOP_IGNORING_VAR(x)</code> and the
863 corresponding client request
864 <code class="varname">VG_USERREQ__DRD_FINISH_SUPPRESSION</code>. Tell DRD
865 to no longer ignore data races for the address range that was suppressed
866 either via the macro <code class="literal">DRD_IGNORE_VAR(x)</code> or via the
867 client request <code class="varname">VG_USERREQ__DRD_START_SUPPRESSION</code>.
868 </p></li>
869<li class="listitem"><p>
870 The macro <code class="literal">DRD_TRACE_VAR(x)</code>. Trace all load and store
871 activity for the address range starting at <code class="literal">&amp;x</code> and
872 occupying <code class="literal">sizeof(x)</code> bytes. When DRD reports a data
873 race on a specified variable, and it's not immediately clear which
874 source code statements triggered the conflicting accesses, it can be
875 very helpful to trace all activity on the offending memory location.
876 </p></li>
877<li class="listitem"><p>
878 The macro <code class="literal">DRD_STOP_TRACING_VAR(x)</code>. Stop tracing load
879 and store activity for the address range starting
880 at <code class="literal">&amp;x</code> and occupying <code class="literal">sizeof(x)</code>
881 bytes.
882 </p></li>
883<li class="listitem"><p>
884 The macro <code class="literal">ANNOTATE_TRACE_MEMORY(&amp;x)</code>. Trace all
885 load and store activity that touches at least the single byte at the
886 address <code class="literal">&amp;x</code>.
887 </p></li>
888<li class="listitem"><p>
889 The client request <code class="varname">VG_USERREQ__DRD_START_TRACE_ADDR</code>,
890 which allows to trace all load and store activity for the specified
891 address range.
892 </p></li>
893<li class="listitem"><p>
894 The client
895 request <code class="varname">VG_USERREQ__DRD_STOP_TRACE_ADDR</code>. Do no longer
896 trace load and store activity for the specified address range.
897 </p></li>
898<li class="listitem"><p>
899 The macro <code class="literal">ANNOTATE_HAPPENS_BEFORE(addr)</code> tells DRD to
900 insert a mark. Insert this macro just after an access to the variable at
901 the specified address has been performed.
902 </p></li>
903<li class="listitem"><p>
904 The macro <code class="literal">ANNOTATE_HAPPENS_AFTER(addr)</code> tells DRD that
905 the next access to the variable at the specified address should be
906 considered to have happened after the access just before the latest
907 <code class="literal">ANNOTATE_HAPPENS_BEFORE(addr)</code> annotation that
908 references the same variable. The purpose of these two macros is to tell
909 DRD about the order of inter-thread memory accesses implemented via
910 atomic memory operations. See
911 also <code class="literal">drd/tests/annotate_smart_pointer.cpp</code> for an
912 example.
913 </p></li>
914<li class="listitem"><p>
915 The macro <code class="literal">ANNOTATE_RWLOCK_CREATE(rwlock)</code> tells DRD
916 that the object at address <code class="literal">rwlock</code> is a
917 reader-writer synchronization object that is not a
918 <code class="literal">pthread_rwlock_t</code> synchronization object. See
919 also <code class="literal">drd/tests/annotate_rwlock.c</code> for an example.
920 </p></li>
921<li class="listitem"><p>
922 The macro <code class="literal">ANNOTATE_RWLOCK_DESTROY(rwlock)</code> tells DRD
923 that the reader-writer synchronization object at
924 address <code class="literal">rwlock</code> has been destroyed.
925 </p></li>
926<li class="listitem"><p>
927 The macro <code class="literal">ANNOTATE_WRITERLOCK_ACQUIRED(rwlock)</code> tells
928 DRD that a writer lock has been acquired on the reader-writer
929 synchronization object at address <code class="literal">rwlock</code>.
930 </p></li>
931<li class="listitem"><p>
932 The macro <code class="literal">ANNOTATE_READERLOCK_ACQUIRED(rwlock)</code> tells
933 DRD that a reader lock has been acquired on the reader-writer
934 synchronization object at address <code class="literal">rwlock</code>.
935 </p></li>
936<li class="listitem"><p>
937 The macro <code class="literal">ANNOTATE_RWLOCK_ACQUIRED(rwlock, is_w)</code>
938 tells DRD that a writer lock (when <code class="literal">is_w != 0</code>) or that
939 a reader lock (when <code class="literal">is_w == 0</code>) has been acquired on
940 the reader-writer synchronization object at
941 address <code class="literal">rwlock</code>.
942 </p></li>
943<li class="listitem"><p>
944 The macro <code class="literal">ANNOTATE_WRITERLOCK_RELEASED(rwlock)</code> tells
945 DRD that a writer lock has been released on the reader-writer
946 synchronization object at address <code class="literal">rwlock</code>.
947 </p></li>
948<li class="listitem"><p>
949 The macro <code class="literal">ANNOTATE_READERLOCK_RELEASED(rwlock)</code> tells
950 DRD that a reader lock has been released on the reader-writer
951 synchronization object at address <code class="literal">rwlock</code>.
952 </p></li>
953<li class="listitem"><p>
954 The macro <code class="literal">ANNOTATE_RWLOCK_RELEASED(rwlock, is_w)</code>
955 tells DRD that a writer lock (when <code class="literal">is_w != 0</code>) or that
956 a reader lock (when <code class="literal">is_w == 0</code>) has been released on
957 the reader-writer synchronization object at
958 address <code class="literal">rwlock</code>.
959 </p></li>
960<li class="listitem"><p>
961 The macro <code class="literal">ANNOTATE_BARRIER_INIT(barrier, count,
962 reinitialization_allowed)</code> tells DRD that a new barrier object
963 at the address <code class="literal">barrier</code> has been initialized,
964 that <code class="literal">count</code> threads participate in each barrier and
965 also whether or not barrier reinitialization without intervening
966 destruction should be reported as an error. See
967 also <code class="literal">drd/tests/annotate_barrier.c</code> for an example.
968 </p></li>
969<li class="listitem"><p>
970 The macro <code class="literal">ANNOTATE_BARRIER_DESTROY(barrier)</code>
971 tells DRD that a barrier object is about to be destroyed.
972 </p></li>
973<li class="listitem"><p>
974 The macro <code class="literal">ANNOTATE_BARRIER_WAIT_BEFORE(barrier)</code>
975 tells DRD that waiting for a barrier will start.
976 </p></li>
977<li class="listitem"><p>
978 The macro <code class="literal">ANNOTATE_BARRIER_WAIT_AFTER(barrier)</code>
979 tells DRD that waiting for a barrier has finished.
980 </p></li>
981<li class="listitem"><p>
982 The macro <code class="literal">ANNOTATE_BENIGN_RACE_SIZED(addr, size,
983 descr)</code> tells DRD that any races detected on the specified
984 address are benign and hence should not be
985 reported. The <code class="literal">descr</code> argument is ignored but can be
986 used to document why data races on <code class="literal">addr</code> are benign.
987 </p></li>
988<li class="listitem"><p>
989 The macro <code class="literal">ANNOTATE_BENIGN_RACE_STATIC(var, descr)</code>
990 tells DRD that any races detected on the specified static variable are
991 benign and hence should not be reported. The <code class="literal">descr</code>
992 argument is ignored but can be used to document why data races
993 on <code class="literal">var</code> are benign. Note: this macro can only be
994 used in C++ programs and not in C programs.
995 </p></li>
996<li class="listitem"><p>
997 The macro <code class="literal">ANNOTATE_IGNORE_READS_BEGIN</code> tells
998 DRD to ignore all memory loads performed by the current thread.
999 </p></li>
1000<li class="listitem"><p>
1001 The macro <code class="literal">ANNOTATE_IGNORE_READS_END</code> tells
1002 DRD to stop ignoring the memory loads performed by the current thread.
1003 </p></li>
1004<li class="listitem"><p>
1005 The macro <code class="literal">ANNOTATE_IGNORE_WRITES_BEGIN</code> tells
1006 DRD to ignore all memory stores performed by the current thread.
1007 </p></li>
1008<li class="listitem"><p>
1009 The macro <code class="literal">ANNOTATE_IGNORE_WRITES_END</code> tells
1010 DRD to stop ignoring the memory stores performed by the current thread.
1011 </p></li>
1012<li class="listitem"><p>
1013 The macro <code class="literal">ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN</code> tells
1014 DRD to ignore all memory accesses performed by the current thread.
1015 </p></li>
1016<li class="listitem"><p>
1017 The macro <code class="literal">ANNOTATE_IGNORE_READS_AND_WRITES_END</code> tells
1018 DRD to stop ignoring the memory accesses performed by the current thread.
1019 </p></li>
1020<li class="listitem"><p>
1021 The macro <code class="literal">ANNOTATE_NEW_MEMORY(addr, size)</code> tells
1022 DRD that the specified memory range has been allocated by a custom
1023 memory allocator in the client program and that the client program
1024 will start using this memory range.
1025 </p></li>
1026<li class="listitem"><p>
1027 The macro <code class="literal">ANNOTATE_THREAD_NAME(name)</code> tells DRD to
1028 associate the specified name with the current thread and to include this
1029 name in the error messages printed by DRD.
1030 </p></li>
1031<li class="listitem"><p>
1032 The macros <code class="literal">VALGRIND_MALLOCLIKE_BLOCK</code> and
1033 <code class="literal">VALGRIND_FREELIKE_BLOCK</code> from the Valgrind core are
1034 implemented; they are described in
1035 <a class="xref" href="manual-core-adv.html#manual-core-adv.clientreq" title="3.1. The Client Request mechanism">The Client Request mechanism</a>.
1036 </p></li>
1037</ul></div>
1038<p>
1039</p>
1040<p>
1041Note: if you compiled Valgrind yourself, the header file
1042<code class="literal">&lt;valgrind/drd.h&gt;</code> will have been installed in
1043the directory <code class="literal">/usr/include</code> by the command
1044<code class="literal">make install</code>. If you obtained Valgrind by
1045installing it as a package however, you will probably have to install
1046another package with a name like <code class="literal">valgrind-devel</code>
1047before Valgrind's header files are available.
1048</p>
1049</div>
1050<div class="sect2">
1051<div class="titlepage"><div><div><h3 class="title">
1052<a name="drd-manual.C++11"></a>8.2.6. Debugging C++11 Programs</h3></div></div></div>
1053<p>If you want to use the C++11 class std::thread you will need to do the
1054 following to annotate the std::shared_ptr&lt;&gt; objects used in the
1055 implementation of that class:
1056</p>
1057<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
1058<li class="listitem">
1059<p>Add the following code at the start of a common header or at the
1060 start of each source file, before any C++ header files are included:</p>
1061<pre class="programlisting">
1062#include &lt;valgrind/drd.h&gt;
1063#define _GLIBCXX_SYNCHRONIZATION_HAPPENS_BEFORE(addr) ANNOTATE_HAPPENS_BEFORE(addr)
1064#define _GLIBCXX_SYNCHRONIZATION_HAPPENS_AFTER(addr) ANNOTATE_HAPPENS_AFTER(addr)
1065</pre>
1066</li>
1067<li class="listitem"><p>Download the gcc source code and from source file
1068 libstdc++-v3/src/c++11/thread.cc copy the implementation of the
1069 <code class="computeroutput">execute_native_thread_routine()</code>
1070 and <code class="computeroutput">std::thread::_M_start_thread()</code>
1071 functions into a source file that is linked with your application. Make
1072 sure that also in this source file the
1073 _GLIBCXX_SYNCHRONIZATION_HAPPENS_*() macros are defined properly.</p></li>
1074</ul></div>
1075<p>
1076</p>
1077<p>For more information, see also <span class="emphasis"><em>The
1078GNU C++ Library Manual, Debugging Support</em></span>
1079(<a class="ulink" href="http://gcc.gnu.org/onlinedocs/libstdc++/manual/debug.html" target="_top">http://gcc.gnu.org/onlinedocs/libstdc++/manual/debug.html</a>).</p>
1080</div>
1081<div class="sect2">
1082<div class="titlepage"><div><div><h3 class="title">
1083<a name="drd-manual.gnome"></a>8.2.7. Debugging GNOME Programs</h3></div></div></div>
1084<p>
1085GNOME applications use the threading primitives provided by the
1086<code class="computeroutput">glib</code> and
1087<code class="computeroutput">gthread</code> libraries. These libraries
1088are built on top of POSIX threads, and hence are directly supported by
1089DRD. Please keep in mind that you have to call
1090<code class="function">g_thread_init</code> before creating any threads, or
1091DRD will report several data races on glib functions. See also the
1092<a class="ulink" href="http://library.gnome.org/devel/glib/stable/glib-Threads.html" target="_top">GLib
1093Reference Manual</a> for more information about
1094<code class="function">g_thread_init</code>.
1095</p>
1096<p>
1097One of the many facilities provided by the <code class="literal">glib</code>
1098library is a block allocator, called <code class="literal">g_slice</code>. You
1099have to disable this block allocator when using DRD by adding the
1100following to the shell environment variables:
1101<code class="literal">G_SLICE=always-malloc</code>. See also the <a class="ulink" href="http://library.gnome.org/devel/glib/stable/glib-Memory-Slices.html" target="_top">GLib
1102Reference Manual</a> for more information.
1103</p>
1104</div>
1105<div class="sect2">
1106<div class="titlepage"><div><div><h3 class="title">
1107<a name="drd-manual.boost.thread"></a>8.2.8. Debugging Boost.Thread Programs</h3></div></div></div>
1108<p>
1109The Boost.Thread library is the threading library included with the
1110cross-platform Boost Libraries. This threading library is an early
1111implementation of the upcoming C++0x threading library.
1112</p>
1113<p>
1114Applications that use the Boost.Thread library should run fine under DRD.
1115</p>
1116<p>
1117More information about Boost.Thread can be found here:
1118</p>
1119<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
1120<li class="listitem"><p>
1121 Anthony Williams, <a class="ulink" href="http://www.boost.org/doc/libs/1_37_0/doc/html/thread.html" target="_top">Boost.Thread</a>
1122 Library Documentation, Boost website, 2007.
1123 </p></li>
1124<li class="listitem"><p>
1125 Anthony Williams, <a class="ulink" href="http://www.ddj.com/cpp/211600441" target="_top">What's New in Boost
1126 Threads?</a>, Recent changes to the Boost Thread library,
1127 Dr. Dobbs Magazine, October 2008.
1128 </p></li>
1129</ul></div>
1130<p>
1131</p>
1132</div>
1133<div class="sect2">
1134<div class="titlepage"><div><div><h3 class="title">
1135<a name="drd-manual.openmp"></a>8.2.9. Debugging OpenMP Programs</h3></div></div></div>
1136<p>
1137OpenMP stands for <span class="emphasis"><em>Open Multi-Processing</em></span>. The OpenMP
1138standard consists of a set of compiler directives for C, C++ and Fortran
1139programs that allows a compiler to transform a sequential program into a
1140parallel program. OpenMP is well suited for HPC applications and allows to
1141work at a higher level compared to direct use of the POSIX threads API. While
1142OpenMP ensures that the POSIX API is used correctly, OpenMP programs can still
1143contain data races. So it definitely makes sense to verify OpenMP programs
1144with a thread checking tool.
1145</p>
1146<p>
1147DRD supports OpenMP shared-memory programs generated by GCC. GCC
1148supports OpenMP since version 4.2.0. GCC's runtime support
1149for OpenMP programs is provided by a library called
1150<code class="literal">libgomp</code>. The synchronization primitives implemented
1151in this library use Linux' futex system call directly, unless the
1152library has been configured with the
1153<code class="literal">--disable-linux-futex</code> option. DRD only supports
1154libgomp libraries that have been configured with this option and in
1155which symbol information is present. For most Linux distributions this
1156means that you will have to recompile GCC. See also the script
1157<code class="literal">drd/scripts/download-and-build-gcc</code> in the
1158Valgrind source tree for an example of how to compile GCC. You will
1159also have to make sure that the newly compiled
1160<code class="literal">libgomp.so</code> library is loaded when OpenMP programs
1161are started. This is possible by adding a line similar to the
1162following to your shell startup script:
1163</p>
1164<pre class="programlisting">
1165export LD_LIBRARY_PATH=~/gcc-4.4.0/lib64:~/gcc-4.4.0/lib:
1166</pre>
1167<p>
1168As an example, the test OpenMP test program
1169<code class="literal">drd/tests/omp_matinv</code> triggers a data race
1170when the option -r has been specified on the command line. The data
1171race is triggered by the following code:
1172</p>
1173<pre class="programlisting">
1174#pragma omp parallel for private(j)
1175for (j = 0; j &lt; rows; j++)
1176{
1177 if (i != j)
1178 {
1179 const elem_t factor = a[j * cols + i];
1180 for (k = 0; k &lt; cols; k++)
1181 {
1182 a[j * cols + k] -= a[i * cols + k] * factor;
1183 }
1184 }
1185}
1186</pre>
1187<p>
1188The above code is racy because the variable <code class="literal">k</code> has
1189not been declared private. DRD will print the following error message
1190for the above code:
1191</p>
1192<pre class="programlisting">
1193$ valgrind --tool=drd --check-stack-var=yes --read-var-info=yes drd/tests/omp_matinv 3 -t 2 -r
1194...
1195Conflicting store by thread 1/1 at 0x7fefffbc4 size 4
1196 at 0x4014A0: gj.omp_fn.0 (omp_matinv.c:203)
1197 by 0x401211: gj (omp_matinv.c:159)
1198 by 0x40166A: invert_matrix (omp_matinv.c:238)
1199 by 0x4019B4: main (omp_matinv.c:316)
1200Location 0x7fefffbc4 is 0 bytes inside local var "k"
1201declared at omp_matinv.c:160, in frame #0 of thread 1
1202...
1203</pre>
1204<p>
1205In the above output the function name <code class="function">gj.omp_fn.0</code>
1206has been generated by GCC from the function name
1207<code class="function">gj</code>. The allocation context information shows that the
1208data race has been caused by modifying the variable <code class="literal">k</code>.
1209</p>
1210<p>
1211Note: for GCC versions before 4.4.0, no allocation context information is
1212shown. With these GCC versions the most usable information in the above output
1213is the source file name and the line number where the data race has been
1214detected (<code class="literal">omp_matinv.c:203</code>).
1215</p>
1216<p>
1217For more information about OpenMP, see also
1218<a class="ulink" href="http://openmp.org/" target="_top">openmp.org</a>.
1219</p>
1220</div>
1221<div class="sect2">
1222<div class="titlepage"><div><div><h3 class="title">
1223<a name="drd-manual.cust-mem-alloc"></a>8.2.10. DRD and Custom Memory Allocators</h3></div></div></div>
1224<p>
1225DRD tracks all memory allocation events that happen via the
1226standard memory allocation and deallocation functions
1227(<code class="function">malloc</code>, <code class="function">free</code>,
1228<code class="function">new</code> and <code class="function">delete</code>), via entry
1229and exit of stack frames or that have been annotated with Valgrind's
1230memory pool client requests. DRD uses memory allocation and deallocation
1231information for two purposes:
1232</p>
1233<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
1234<li class="listitem"><p>
1235 To know where the scope ends of POSIX objects that have not been
1236 destroyed explicitly. It is e.g. not required by the POSIX
1237 threads standard to call
1238 <code class="function">pthread_mutex_destroy</code> before freeing the
1239 memory in which a mutex object resides.
1240 </p></li>
1241<li class="listitem"><p>
1242 To know where the scope of variables ends. If e.g. heap memory
1243 has been used by one thread, that thread frees that memory, and
1244 another thread allocates and starts using that memory, no data
1245 races must be reported for that memory.
1246 </p></li>
1247</ul></div>
1248<p>
1249</p>
1250<p>
1251It is essential for correct operation of DRD that the tool knows about
1252memory allocation and deallocation events. When analyzing a client program
1253with DRD that uses a custom memory allocator, either instrument the custom
1254memory allocator with the <code class="literal">VALGRIND_MALLOCLIKE_BLOCK</code>
1255and <code class="literal">VALGRIND_FREELIKE_BLOCK</code> macros or disable the
1256custom memory allocator.
1257</p>
1258<p>
1259As an example, the GNU libstdc++ library can be configured
1260to use standard memory allocation functions instead of memory pools by
1261setting the environment variable
1262<code class="literal">GLIBCXX_FORCE_NEW</code>. For more information, see also
1263the <a class="ulink" href="http://gcc.gnu.org/onlinedocs/libstdc++/manual/bk01pt04ch11.html" target="_top">libstdc++
1264manual</a>.
1265</p>
1266</div>
1267<div class="sect2">
1268<div class="titlepage"><div><div><h3 class="title">
1269<a name="drd-manual.drd-versus-memcheck"></a>8.2.11. DRD Versus Memcheck</h3></div></div></div>
1270<p>
1271It is essential for correct operation of DRD that there are no memory
1272errors such as dangling pointers in the client program. Which means that
1273it is a good idea to make sure that your program is Memcheck-clean
1274before you analyze it with DRD. It is possible however that some of
1275the Memcheck reports are caused by data races. In this case it makes
1276sense to run DRD before Memcheck.
1277</p>
1278<p>
1279So which tool should be run first? In case both DRD and Memcheck
1280complain about a program, a possible approach is to run both tools
1281alternatingly and to fix as many errors as possible after each run of
1282each tool until none of the two tools prints any more error messages.
1283</p>
1284</div>
1285<div class="sect2">
1286<div class="titlepage"><div><div><h3 class="title">
1287<a name="drd-manual.resource-requirements"></a>8.2.12. Resource Requirements</h3></div></div></div>
1288<p>
1289The requirements of DRD with regard to heap and stack memory and the
1290effect on the execution time of client programs are as follows:
1291</p>
1292<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
1293<li class="listitem"><p>
1294 When running a program under DRD with default DRD options,
1295 between 1.1 and 3.6 times more memory will be needed compared to
1296 a native run of the client program. More memory will be needed
1297 if loading debug information has been enabled
1298 (<code class="literal">--read-var-info=yes</code>).
1299 </p></li>
1300<li class="listitem"><p>
1301 DRD allocates some of its temporary data structures on the stack
1302 of the client program threads. This amount of data is limited to
1303 1 - 2 KB. Make sure that thread stacks are sufficiently large.
1304 </p></li>
1305<li class="listitem"><p>
1306 Most applications will run between 20 and 50 times slower under
1307 DRD than a native single-threaded run. The slowdown will be most
1308 noticeable for applications which perform frequent mutex lock /
1309 unlock operations.
1310 </p></li>
1311</ul></div>
1312<p>
1313</p>
1314</div>
1315<div class="sect2">
1316<div class="titlepage"><div><div><h3 class="title">
1317<a name="drd-manual.effective-use"></a>8.2.13. Hints and Tips for Effective Use of DRD</h3></div></div></div>
1318<p>
1319The following information may be helpful when using DRD:
1320</p>
1321<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
1322<li class="listitem"><p>
1323 Make sure that debug information is present in the executable
1324 being analyzed, such that DRD can print function name and line
1325 number information in stack traces. Most compilers can be told
1326 to include debug information via compiler option
1327 <code class="option">-g</code>.
1328 </p></li>
1329<li class="listitem"><p>
1330 Compile with option <code class="option">-O1</code> instead of
1331 <code class="option">-O0</code>. This will reduce the amount of generated
1332 code, may reduce the amount of debug info and will speed up
1333 DRD's processing of the client program. For more information,
1334 see also <a class="xref" href="manual-core.html#manual-core.started" title="2.2. Getting started">Getting started</a>.
1335 </p></li>
1336<li class="listitem"><p>
1337 If DRD reports any errors on libraries that are part of your
1338 Linux distribution like e.g. <code class="literal">libc.so</code> or
1339 <code class="literal">libstdc++.so</code>, installing the debug packages
1340 for these libraries will make the output of DRD a lot more
1341 detailed.
1342 </p></li>
1343<li class="listitem">
1344<p>
1345 When using C++, do not send output from more than one thread to
1346 <code class="literal">std::cout</code>. Doing so would not only
1347 generate multiple data race reports, it could also result in
1348 output from several threads getting mixed up. Either use
1349 <code class="function">printf</code> or do the following:
1350 </p>
1351<div class="orderedlist"><ol class="orderedlist" type="1">
1352<li class="listitem"><p>Derive a class from <code class="literal">std::ostreambuf</code>
1353 and let that class send output line by line to
1354 <code class="literal">stdout</code>. This will avoid that individual
1355 lines of text produced by different threads get mixed
1356 up.</p></li>
1357<li class="listitem"><p>Create one instance of <code class="literal">std::ostream</code>
1358 for each thread. This makes stream formatting settings
1359 thread-local. Pass a per-thread instance of the class
1360 derived from <code class="literal">std::ostreambuf</code> to the
1361 constructor of each instance. </p></li>
1362<li class="listitem"><p>Let each thread send its output to its own instance of
1363 <code class="literal">std::ostream</code> instead of
1364 <code class="literal">std::cout</code>.</p></li>
1365</ol></div>
1366<p>
1367 </p>
1368</li>
1369</ul></div>
1370<p>
1371</p>
1372</div>
1373</div>
1374<div class="sect1">
1375<div class="titlepage"><div><div><h2 class="title" style="clear: both">
1376<a name="drd-manual.Pthreads"></a>8.3. Using the POSIX Threads API Effectively</h2></div></div></div>
1377<div class="sect2">
1378<div class="titlepage"><div><div><h3 class="title">
1379<a name="drd-manual.mutex-types"></a>8.3.1. Mutex types</h3></div></div></div>
1380<p>
1381The Single UNIX Specification version two defines the following four
1382mutex types (see also the documentation of <a class="ulink" href="http://www.opengroup.org/onlinepubs/007908799/xsh/pthread_mutexattr_settype.html" target="_top"><code class="function">pthread_mutexattr_settype</code></a>):
1383</p>
1384<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
1385<li class="listitem"><p>
1386 <span class="emphasis"><em>normal</em></span>, which means that no error checking
1387 is performed, and that the mutex is non-recursive.
1388 </p></li>
1389<li class="listitem"><p>
1390 <span class="emphasis"><em>error checking</em></span>, which means that the mutex
1391 is non-recursive and that error checking is performed.
1392 </p></li>
1393<li class="listitem"><p>
1394 <span class="emphasis"><em>recursive</em></span>, which means that a mutex may be
1395 locked recursively.
1396 </p></li>
1397<li class="listitem"><p>
1398 <span class="emphasis"><em>default</em></span>, which means that error checking
1399 behavior is undefined, and that the behavior for recursive
1400 locking is also undefined. Or: portable code must neither
1401 trigger error conditions through the Pthreads API nor attempt to
1402 lock a mutex of default type recursively.
1403 </p></li>
1404</ul></div>
1405<p>
1406</p>
1407<p>
1408In complex applications it is not always clear from beforehand which
1409mutex will be locked recursively and which mutex will not be locked
1410recursively. Attempts lock a non-recursive mutex recursively will
1411result in race conditions that are very hard to find without a thread
1412checking tool. So either use the error checking mutex type and
1413consistently check the return value of Pthread API mutex calls, or use
1414the recursive mutex type.
1415</p>
1416</div>
1417<div class="sect2">
1418<div class="titlepage"><div><div><h3 class="title">
1419<a name="drd-manual.condvar"></a>8.3.2. Condition variables</h3></div></div></div>
1420<p>
1421A condition variable allows one thread to wake up one or more other
1422threads. Condition variables are often used to notify one or more
1423threads about state changes of shared data. Unfortunately it is very
1424easy to introduce race conditions by using condition variables as the
1425only means of state information propagation. A better approach is to
1426let threads poll for changes of a state variable that is protected by
1427a mutex, and to use condition variables only as a thread wakeup
1428mechanism. See also the source file
1429<code class="computeroutput">drd/tests/monitor_example.cpp</code> for an
1430example of how to implement this concept in C++. The monitor concept
1431used in this example is a well known and very useful concept -- see
1432also Wikipedia for more information about the <a class="ulink" href="http://en.wikipedia.org/wiki/Monitor_(synchronization)" target="_top">monitor</a>
1433concept.
1434</p>
1435</div>
1436<div class="sect2">
1437<div class="titlepage"><div><div><h3 class="title">
1438<a name="drd-manual.pctw"></a>8.3.3. pthread_cond_timedwait and timeouts</h3></div></div></div>
1439<p>
1440Historically the function
1441<code class="function">pthread_cond_timedwait</code> only allowed the
1442specification of an absolute timeout, that is a timeout independent of
1443the time when this function was called. However, almost every call to
1444this function expresses a relative timeout. This typically happens by
1445passing the sum of
1446<code class="computeroutput">clock_gettime(CLOCK_REALTIME)</code> and a
1447relative timeout as the third argument. This approach is incorrect
1448since forward or backward clock adjustments by e.g. ntpd will affect
1449the timeout. A more reliable approach is as follows:
1450</p>
1451<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
1452<li class="listitem"><p>
1453 When initializing a condition variable through
1454 <code class="function">pthread_cond_init</code>, specify that the timeout of
1455 <code class="function">pthread_cond_timedwait</code> will use the clock
1456 <code class="literal">CLOCK_MONOTONIC</code> instead of
1457 <code class="literal">CLOCK_REALTIME</code>. You can do this via
1458 <code class="computeroutput">pthread_condattr_setclock(...,
1459 CLOCK_MONOTONIC)</code>.
1460 </p></li>
1461<li class="listitem"><p>
1462 When calling <code class="function">pthread_cond_timedwait</code>, pass
1463 the sum of
1464 <code class="computeroutput">clock_gettime(CLOCK_MONOTONIC)</code>
1465 and a relative timeout as the third argument.
1466 </p></li>
1467</ul></div>
1468<p>
1469See also
1470<code class="computeroutput">drd/tests/monitor_example.cpp</code> for an
1471example.
1472</p>
1473</div>
1474</div>
1475<div class="sect1">
1476<div class="titlepage"><div><div><h2 class="title" style="clear: both">
1477<a name="drd-manual.limitations"></a>8.4. Limitations</h2></div></div></div>
1478<p>DRD currently has the following limitations:</p>
1479<div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; ">
1480<li class="listitem"><p>
1481 DRD, just like Memcheck, will refuse to start on Linux
1482 distributions where all symbol information has been removed from
1483 <code class="filename">ld.so</code>. This is e.g. the case for the PPC editions
1484 of openSUSE and Gentoo. You will have to install the glibc debuginfo
1485 package on these platforms before you can use DRD. See also openSUSE
1486 bug <a class="ulink" href="http://bugzilla.novell.com/show_bug.cgi?id=396197" target="_top">
1487 396197</a> and Gentoo bug <a class="ulink" href="http://bugs.gentoo.org/214065" target="_top">214065</a>.
1488 </p></li>
1489<li class="listitem"><p>
1490 With gcc 4.4.3 and before, DRD may report data races on the C++
1491 class <code class="literal">std::string</code> in a multithreaded program. This is
1492 a know <code class="literal">libstdc++</code> issue -- see also GCC bug
1493 <a class="ulink" href="http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40518" target="_top">40518</a>
1494 for more information.
1495 </p></li>
1496<li class="listitem"><p>
1497 If you compile the DRD source code yourself, you need GCC 3.0 or
1498 later. GCC 2.95 is not supported.
1499 </p></li>
1500<li class="listitem"><p>
1501 Of the two POSIX threads implementations for Linux, only the
1502 NPTL (Native POSIX Thread Library) is supported. The older
1503 LinuxThreads library is not supported.
1504 </p></li>
1505</ul></div>
1506</div>
1507<div class="sect1">
1508<div class="titlepage"><div><div><h2 class="title" style="clear: both">
1509<a name="drd-manual.feedback"></a>8.5. Feedback</h2></div></div></div>
1510<p>
1511If you have any comments, suggestions, feedback or bug reports about
1512DRD, feel free to either post a message on the Valgrind users mailing
1513list or to file a bug report. See also <a class="ulink" href="http://www.valgrind.org/" target="_top">http://www.valgrind.org/</a> for more information.
1514</p>
1515</div>
1516</div>
1517<div>
1518<br><table class="nav" width="100%" cellspacing="3" cellpadding="2" border="0" summary="Navigation footer">
1519<tr>
1520<td rowspan="2" width="40%" align="left">
1521<a accesskey="p" href="hg-manual.html">&lt;&lt; 7. Helgrind: a thread error detector</a> </td>
1522<td width="20%" align="center"><a accesskey="u" href="manual.html">Up</a></td>
1523<td rowspan="2" width="40%" align="right"> <a accesskey="n" href="ms-manual.html">9. Massif: a heap profiler &gt;&gt;</a>
1524</td>
1525</tr>
1526<tr><td width="20%" align="center"><a accesskey="h" href="index.html">Home</a></td></tr>
1527</table>
1528</div>
1529</body>
1530</html>