1 <?xml version="1.0"?> <!-- -*- sgml -*- --> 2 <!DOCTYPE chapter PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN" 3 "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd" 4 [ <!ENTITY % vg-entities SYSTEM "../../docs/xml/vg-entities.xml"> %vg-entities; ]> 5 6 7 <chapter id="drd-manual" xreflabel="DRD: a thread error detector"> 8 <title>DRD: a thread error detector</title> 9 10 <para>To use this tool, you must specify 11 <option>--tool=drd</option> 12 on the Valgrind command line.</para> 13 14 15 <sect1 id="drd-manual.overview" xreflabel="Overview"> 16 <title>Overview</title> 17 18 <para> 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 22 primitives. 23 </para> 24 25 <sect2 id="drd-manual.mt-progr-models" xreflabel="MT-progr-models"> 26 <title>Multithreaded Programming Paradigms</title> 27 28 <para> 29 There are two possible reasons for using multithreading in a program: 30 <itemizedlist> 31 <listitem> 32 <para> 33 To model concurrent activities. Assigning one thread to each activity 34 can be a great simplification compared to multiplexing the states of 35 multiple activities in a single thread. This is why most server software 36 and embedded software is multithreaded. 37 </para> 38 </listitem> 39 <listitem> 40 <para> 41 To use multiple CPU cores simultaneously for speeding up 42 computations. This is why many High Performance Computing (HPC) 43 applications are multithreaded. 44 </para> 45 </listitem> 46 </itemizedlist> 47 </para> 48 49 <para> 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: 53 <itemizedlist> 54 <listitem> 55 <para> 56 Locking. Data that is shared over threads is protected from concurrent 57 accesses via locking. E.g. the POSIX threads library, the Qt library 58 and the Boost.Thread library support this paradigm directly. 59 </para> 60 </listitem> 61 <listitem> 62 <para> 63 Message passing. No data is shared between threads, but threads exchange 64 data by passing messages to each other. Examples of implementations of 65 the message passing paradigm are MPI and CORBA. 66 </para> 67 </listitem> 68 <listitem> 69 <para> 70 Automatic parallelization. A compiler converts a sequential program into 71 a multithreaded program. The original program may or may not contain 72 parallelization hints. One example of such parallelization hints is the 73 OpenMP standard. In this standard a set of directives are defined which 74 tell a compiler how to parallelize a C, C++ or Fortran program. OpenMP 75 is well suited for computational intensive applications. As an example, 76 an open source image processing software package is using OpenMP to 77 maximize performance on systems with multiple CPU 78 cores. GCC supports the 79 OpenMP standard from version 4.2.0 on. 80 </para> 81 </listitem> 82 <listitem> 83 <para> 84 Software Transactional Memory (STM). Any data that is shared between 85 threads is updated via transactions. After each transaction it is 86 verified whether there were any conflicting transactions. If there were 87 conflicts, the transaction is aborted, otherwise it is committed. This 88 is a so-called optimistic approach. There is a prototype of the Intel C++ 89 Compiler available that supports STM. Research about the addition of 90 STM support to GCC is ongoing. 91 </para> 92 </listitem> 93 </itemizedlist> 94 </para> 95 96 <para> 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. 102 </para> 103 104 </sect2> 105 106 107 <sect2 id="drd-manual.pthreads-model" xreflabel="Pthreads-model"> 108 <title>POSIX Threads Programming Model</title> 109 110 <para> 111 POSIX threads, also known as Pthreads, is the most widely available 112 threading library on Unix systems. 113 </para> 114 115 <para> 116 The POSIX threads programming model is based on the following abstractions: 117 <itemizedlist> 118 <listitem> 119 <para> 120 A shared address space. All threads running within the same 121 process share the same address space. All data, whether shared or 122 not, is identified by its address. 123 </para> 124 </listitem> 125 <listitem> 126 <para> 127 Regular load and store operations, which allow to read values 128 from or to write values to the memory shared by all threads 129 running in the same process. 130 </para> 131 </listitem> 132 <listitem> 133 <para> 134 Atomic store and load-modify-store operations. While these are 135 not mentioned in the POSIX threads standard, most 136 microprocessors support atomic memory operations. 137 </para> 138 </listitem> 139 <listitem> 140 <para> 141 Threads. Each thread represents a concurrent activity. 142 </para> 143 </listitem> 144 <listitem> 145 <para> 146 Synchronization objects and operations on these synchronization 147 objects. The following types of synchronization objects have been 148 defined in the POSIX threads standard: mutexes, condition variables, 149 semaphores, reader-writer synchronization objects, barriers and 150 spinlocks. 151 </para> 152 </listitem> 153 </itemizedlist> 154 </para> 155 156 <para> 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>. 163 </para> 164 165 <para> 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>. 170 </para> 171 172 </sect2> 173 174 175 <sect2 id="drd-manual.mt-problems" xreflabel="MT-Problems"> 176 <title>Multithreaded Programming Problems</title> 177 178 <para> 179 Depending on which multithreading paradigm is being used in a program, 180 one or more of the following problems can occur: 181 <itemizedlist> 182 <listitem> 183 <para> 184 Data races. One or more threads access the same memory location without 185 sufficient locking. Most but not all data races are programming errors 186 and are the cause of subtle and hard-to-find bugs. 187 </para> 188 </listitem> 189 <listitem> 190 <para> 191 Lock contention. One thread blocks the progress of one or more other 192 threads by holding a lock too long. 193 </para> 194 </listitem> 195 <listitem> 196 <para> 197 Improper use of the POSIX threads API. Most implementations of the POSIX 198 threads API have been optimized for runtime speed. Such implementations 199 will not complain on certain errors, e.g. when a mutex is being unlocked 200 by another thread than the thread that obtained a lock on the mutex. 201 </para> 202 </listitem> 203 <listitem> 204 <para> 205 Deadlock. A deadlock occurs when two or more threads wait for 206 each other indefinitely. 207 </para> 208 </listitem> 209 <listitem> 210 <para> 211 False sharing. If threads that run on different processor cores 212 access different variables located in the same cache line 213 frequently, this will slow down the involved threads a lot due 214 to frequent exchange of cache lines. 215 </para> 216 </listitem> 217 </itemizedlist> 218 </para> 219 220 <para> 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. 226 </para> 227 228 </sect2> 229 230 231 <sect2 id="drd-manual.data-race-detection" xreflabel="data-race-detection"> 232 <title>Data Race Detection</title> 233 234 <para> 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 237 determined by: 238 <orderedlist> 239 <listitem> 240 <para> 241 All memory operations performed by the same thread are performed in 242 <emphasis>program order</emphasis>, that is, the order determined by the 243 program source code and the results of previous load operations. 244 </para> 245 </listitem> 246 <listitem> 247 <para> 248 Synchronization operations determine certain ordering constraints on 249 memory operations performed by different threads. These ordering 250 constraints are called the <emphasis>synchronization order</emphasis>. 251 </para> 252 </listitem> 253 </orderedlist> 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, 258 p.234-243, May 1991. 259 </para> 260 261 <para> 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. 265 </para> 266 267 <para> 268 A multithreaded program is <emphasis>data-race free</emphasis> if all 269 conflicting memory accesses are ordered by synchronization 270 operations. 271 </para> 272 273 <para> 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. 278 </para> 279 280 <para> 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 289 contain data races. 290 </para> 291 292 <para> 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 302 positives. 303 </para> 304 305 <para> 306 DRD is based on the happens-before algorithm. 307 </para> 308 309 </sect2> 310 311 312 </sect1> 313 314 315 <sect1 id="drd-manual.using-drd" xreflabel="Using DRD"> 316 <title>Using DRD</title> 317 318 <sect2 id="drd-manual.options" xreflabel="DRD Command-line Options"> 319 <title>DRD Command-line Options</title> 320 321 <para>The following command-line options are available for controlling the 322 behavior of the DRD tool itself:</para> 323 324 <!-- start of xi:include in the manpage --> 325 <variablelist id="drd.opts.list"> 326 <varlistentry> 327 <term> 328 <option><![CDATA[--check-stack-var=<yes|no> [default: no]]]></option> 329 </term> 330 <listitem> 331 <para> 332 Controls whether DRD detects data races on stack 333 variables. Verifying stack variables is disabled by default because 334 most programs do not share stack variables over threads. 335 </para> 336 </listitem> 337 </varlistentry> 338 <varlistentry> 339 <term> 340 <option><![CDATA[--exclusive-threshold=<n> [default: off]]]></option> 341 </term> 342 <listitem> 343 <para> 344 Print an error message if any mutex or writer lock has been 345 held longer than the time specified in milliseconds. This 346 option enables the detection of lock contention. 347 </para> 348 </listitem> 349 </varlistentry> 350 <varlistentry> 351 <term> 352 <option><![CDATA[--join-list-vol=<n> [default: 10]]]></option> 353 </term> 354 <listitem> 355 <para> 356 Data races that occur between a statement at the end of one thread 357 and another thread can be missed if memory access information is 358 discarded immediately after a thread has been joined. This option 359 allows to specify for how many joined threads memory access information 360 should be retained. 361 </para> 362 </listitem> 363 </varlistentry> 364 <varlistentry> 365 <term> 366 <option> 367 <![CDATA[--first-race-only=<yes|no> [default: no]]]> 368 </option> 369 </term> 370 <listitem> 371 <para> 372 Whether to report only the first data race that has been detected on a 373 memory location or all data races that have been detected on a memory 374 location. 375 </para> 376 </listitem> 377 </varlistentry> 378 <varlistentry> 379 <term> 380 <option> 381 <![CDATA[--free-is-write=<yes|no> [default: no]]]> 382 </option> 383 </term> 384 <listitem> 385 <para> 386 Whether to report races between accessing memory and freeing 387 memory. Enabling this option may cause DRD to run slightly 388 slower. Notes: 389 <itemizedlist> 390 <listitem> 391 <para> 392 Don't enable this option when using custom memory allocators 393 that use 394 the <computeroutput>VG_USERREQ__MALLOCLIKE_BLOCK</computeroutput> 395 and <computeroutput>VG_USERREQ__FREELIKE_BLOCK</computeroutput> 396 because that would result in false positives. 397 </para> 398 </listitem> 399 <listitem> 400 <para>Don't enable this option when using reference-counted 401 objects because that will result in false positives, even when 402 that code has been annotated properly with 403 <computeroutput>ANNOTATE_HAPPENS_BEFORE</computeroutput> 404 and <computeroutput>ANNOTATE_HAPPENS_AFTER</computeroutput>. See 405 e.g. the output of the following command for an example: 406 <computeroutput>valgrind --tool=drd --free-is-write=yes 407 drd/tests/annotate_smart_pointer</computeroutput>. 408 </para> 409 </listitem> 410 </itemizedlist> 411 </para> 412 </listitem> 413 </varlistentry> 414 <varlistentry> 415 <term> 416 <option> 417 <![CDATA[--report-signal-unlocked=<yes|no> [default: yes]]]> 418 </option> 419 </term> 420 <listitem> 421 <para> 422 Whether to report calls to 423 <function>pthread_cond_signal</function> and 424 <function>pthread_cond_broadcast</function> where the mutex 425 associated with the signal through 426 <function>pthread_cond_wait</function> or 427 <function>pthread_cond_timed_wait</function>is not locked at 428 the time the signal is sent. Sending a signal without holding 429 a lock on the associated mutex is a common programming error 430 which can cause subtle race conditions and unpredictable 431 behavior. There exist some uncommon synchronization patterns 432 however where it is safe to send a signal without holding a 433 lock on the associated mutex. 434 </para> 435 </listitem> 436 </varlistentry> 437 <varlistentry> 438 <term> 439 <option><![CDATA[--segment-merging=<yes|no> [default: yes]]]></option> 440 </term> 441 <listitem> 442 <para> 443 Controls segment merging. Segment merging is an algorithm to 444 limit memory usage of the data race detection 445 algorithm. Disabling segment merging may improve the accuracy 446 of the so-called 'other segments' displayed in race reports 447 but can also trigger an out of memory error. 448 </para> 449 </listitem> 450 </varlistentry> 451 <varlistentry> 452 <term> 453 <option><![CDATA[--segment-merging-interval=<n> [default: 10]]]></option> 454 </term> 455 <listitem> 456 <para> 457 Perform segment merging only after the specified number of new 458 segments have been created. This is an advanced configuration option 459 that allows to choose whether to minimize DRD's memory usage by 460 choosing a low value or to let DRD run faster by choosing a slightly 461 higher value. The optimal value for this parameter depends on the 462 program being analyzed. The default value works well for most programs. 463 </para> 464 </listitem> 465 </varlistentry> 466 <varlistentry> 467 <term> 468 <option><![CDATA[--shared-threshold=<n> [default: off]]]></option> 469 </term> 470 <listitem> 471 <para> 472 Print an error message if a reader lock has been held longer 473 than the specified time (in milliseconds). This option enables 474 the detection of lock contention. 475 </para> 476 </listitem> 477 </varlistentry> 478 <varlistentry> 479 <term> 480 <option><![CDATA[--show-confl-seg=<yes|no> [default: yes]]]></option> 481 </term> 482 <listitem> 483 <para> 484 Show conflicting segments in race reports. Since this 485 information can help to find the cause of a data race, this 486 option is enabled by default. Disabling this option makes the 487 output of DRD more compact. 488 </para> 489 </listitem> 490 </varlistentry> 491 <varlistentry> 492 <term> 493 <option><![CDATA[--show-stack-usage=<yes|no> [default: no]]]></option> 494 </term> 495 <listitem> 496 <para> 497 Print stack usage at thread exit time. When a program creates a large 498 number of threads it becomes important to limit the amount of virtual 499 memory allocated for thread stacks. This option makes it possible to 500 observe how much stack memory has been used by each thread of the the 501 client program. Note: the DRD tool itself allocates some temporary 502 data on the client thread stack. The space necessary for this 503 temporary data must be allocated by the client program when it 504 allocates stack memory, but is not included in stack usage reported by 505 DRD. 506 </para> 507 </listitem> 508 </varlistentry> 509 </variablelist> 510 <!-- end of xi:include in the manpage --> 511 512 <!-- start of xi:include in the manpage --> 513 <para> 514 The following options are available for monitoring the behavior of the 515 client program: 516 </para> 517 518 <variablelist id="drd.debugopts.list"> 519 <varlistentry> 520 <term> 521 <option><![CDATA[--trace-addr=<address> [default: none]]]></option> 522 </term> 523 <listitem> 524 <para> 525 Trace all load and store activity for the specified 526 address. This option may be specified more than once. 527 </para> 528 </listitem> 529 </varlistentry> 530 <varlistentry> 531 <term> 532 <option><![CDATA[--ptrace-addr=<address> [default: none]]]></option> 533 </term> 534 <listitem> 535 <para> 536 Trace all load and store activity for the specified address and keep 537 doing that even after the memory at that address has been freed and 538 reallocated. 539 </para> 540 </listitem> 541 </varlistentry> 542 <varlistentry> 543 <term> 544 <option><![CDATA[--trace-alloc=<yes|no> [default: no]]]></option> 545 </term> 546 <listitem> 547 <para> 548 Trace all memory allocations and deallocations. May produce a huge 549 amount of output. 550 </para> 551 </listitem> 552 </varlistentry> 553 <varlistentry> 554 <term> 555 <option><![CDATA[--trace-barrier=<yes|no> [default: no]]]></option> 556 </term> 557 <listitem> 558 <para> 559 Trace all barrier activity. 560 </para> 561 </listitem> 562 </varlistentry> 563 <varlistentry> 564 <term> 565 <option><![CDATA[--trace-cond=<yes|no> [default: no]]]></option> 566 </term> 567 <listitem> 568 <para> 569 Trace all condition variable activity. 570 </para> 571 </listitem> 572 </varlistentry> 573 <varlistentry> 574 <term> 575 <option><![CDATA[--trace-fork-join=<yes|no> [default: no]]]></option> 576 </term> 577 <listitem> 578 <para> 579 Trace all thread creation and all thread termination events. 580 </para> 581 </listitem> 582 </varlistentry> 583 <varlistentry> 584 <term> 585 <option><![CDATA[--trace-hb=<yes|no> [default: no]]]></option> 586 </term> 587 <listitem> 588 <para> 589 Trace execution of the <literal>ANNOTATE_HAPPENS_BEFORE()</literal>, 590 <literal>ANNOTATE_HAPPENS_AFTER()</literal> and 591 <literal>ANNOTATE_HAPPENS_DONE()</literal> client requests. 592 </para> 593 </listitem> 594 </varlistentry> 595 <varlistentry> 596 <term> 597 <option><![CDATA[--trace-mutex=<yes|no> [default: no]]]></option> 598 </term> 599 <listitem> 600 <para> 601 Trace all mutex activity. 602 </para> 603 </listitem> 604 </varlistentry> 605 <varlistentry> 606 <term> 607 <option><![CDATA[--trace-rwlock=<yes|no> [default: no]]]></option> 608 </term> 609 <listitem> 610 <para> 611 Trace all reader-writer lock activity. 612 </para> 613 </listitem> 614 </varlistentry> 615 <varlistentry> 616 <term> 617 <option><![CDATA[--trace-semaphore=<yes|no> [default: no]]]></option> 618 </term> 619 <listitem> 620 <para> 621 Trace all semaphore activity. 622 </para> 623 </listitem> 624 </varlistentry> 625 </variablelist> 626 <!-- end of xi:include in the manpage --> 627 628 </sect2> 629 630 631 <sect2 id="drd-manual.data-races" xreflabel="Data Races"> 632 <title>Detected Errors: Data Races</title> 633 634 <para> 635 DRD prints a message every time it detects a data race. Please keep 636 the following in mind when interpreting DRD's output: 637 <itemizedlist> 638 <listitem> 639 <para> 640 Every thread is assigned a <emphasis>thread ID</emphasis> by the DRD 641 tool. A thread ID is a number. Thread ID's start at one and are never 642 recycled. 643 </para> 644 </listitem> 645 <listitem> 646 <para> 647 The term <emphasis>segment</emphasis> refers to a consecutive 648 sequence of load, store and synchronization operations, all 649 issued by the same thread. A segment always starts and ends at a 650 synchronization operation. Data race analysis is performed 651 between segments instead of between individual load and store 652 operations because of performance reasons. 653 </para> 654 </listitem> 655 <listitem> 656 <para> 657 There are always at least two memory accesses involved in a data 658 race. Memory accesses involved in a data race are called 659 <emphasis>conflicting memory accesses</emphasis>. DRD prints a 660 report for each memory access that conflicts with a past memory 661 access. 662 </para> 663 </listitem> 664 </itemizedlist> 665 </para> 666 667 <para> 668 Below you can find an example of a message printed by DRD when it 669 detects a data race: 670 </para> 671 <programlisting><![CDATA[ 672 $ valgrind --tool=drd --read-var-info=yes drd/tests/rwlock_race 673 ... 674 ==9466== Thread 3: 675 ==9466== Conflicting load by thread 3 at 0x006020b8 size 4 676 ==9466== at 0x400B6C: thread_func (rwlock_race.c:29) 677 ==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186) 678 ==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so) 679 ==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so) 680 ==9466== Location 0x6020b8 is 0 bytes inside local var "s_racy" 681 ==9466== declared at rwlock_race.c:18, in frame #0 of thread 3 682 ==9466== Other segment start (thread 2) 683 ==9466== at 0x4C2847D: pthread_rwlock_rdlock* (drd_pthread_intercepts.c:813) 684 ==9466== by 0x400B6B: thread_func (rwlock_race.c:28) 685 ==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186) 686 ==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so) 687 ==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so) 688 ==9466== Other segment end (thread 2) 689 ==9466== at 0x4C28B54: pthread_rwlock_unlock* (drd_pthread_intercepts.c:912) 690 ==9466== by 0x400B84: thread_func (rwlock_race.c:30) 691 ==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186) 692 ==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so) 693 ==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so) 694 ... 695 ]]></programlisting> 696 697 <para> 698 The above report has the following meaning: 699 <itemizedlist> 700 <listitem> 701 <para> 702 The number in the column on the left is the process ID of the 703 process being analyzed by DRD. 704 </para> 705 </listitem> 706 <listitem> 707 <para> 708 The first line ("Thread 3") tells you the thread ID for 709 the thread in which context the data race has been detected. 710 </para> 711 </listitem> 712 <listitem> 713 <para> 714 The next line tells which kind of operation was performed (load or 715 store) and by which thread. On the same line the start address and the 716 number of bytes involved in the conflicting access are also displayed. 717 </para> 718 </listitem> 719 <listitem> 720 <para> 721 Next, the call stack of the conflicting access is displayed. If 722 your program has been compiled with debug information 723 (<option>-g</option>), this call stack will include file names and 724 line numbers. The two 725 bottommost frames in this call stack (<function>clone</function> 726 and <function>start_thread</function>) show how the NPTL starts 727 a thread. The third frame 728 (<function>vg_thread_wrapper</function>) is added by DRD. The 729 fourth frame (<function>thread_func</function>) is the first 730 interesting line because it shows the thread entry point, that 731 is the function that has been passed as the third argument to 732 <function>pthread_create</function>. 733 </para> 734 </listitem> 735 <listitem> 736 <para> 737 Next, the allocation context for the conflicting address is 738 displayed. For dynamically allocated data the allocation call 739 stack is shown. For static variables and stack variables the 740 allocation context is only shown when the option 741 <option>--read-var-info=yes</option> has been 742 specified. Otherwise DRD will print <computeroutput>Allocation 743 context: unknown</computeroutput>. 744 </para> 745 </listitem> 746 <listitem> 747 <para> 748 A conflicting access involves at least two memory accesses. For 749 one of these accesses an exact call stack is displayed, and for 750 the other accesses an approximate call stack is displayed, 751 namely the start and the end of the segments of the other 752 accesses. This information can be interpreted as follows: 753 <orderedlist> 754 <listitem> 755 <para> 756 Start at the bottom of both call stacks, and count the 757 number stack frames with identical function name, file 758 name and line number. In the above example the three 759 bottommost frames are identical 760 (<function>clone</function>, 761 <function>start_thread</function> and 762 <function>vg_thread_wrapper</function>). 763 </para> 764 </listitem> 765 <listitem> 766 <para> 767 The next higher stack frame in both call stacks now tells 768 you between in which source code region the other memory 769 access happened. The above output tells that the other 770 memory access involved in the data race happened between 771 source code lines 28 and 30 in file 772 <computeroutput>rwlock_race.c</computeroutput>. 773 </para> 774 </listitem> 775 </orderedlist> 776 </para> 777 </listitem> 778 </itemizedlist> 779 </para> 780 781 </sect2> 782 783 784 <sect2 id="drd-manual.lock-contention" xreflabel="Lock Contention"> 785 <title>Detected Errors: Lock Contention</title> 786 787 <para> 788 Threads must be able to make progress without being blocked for too long by 789 other threads. Sometimes a thread has to wait until a mutex or reader-writer 790 synchronization object is unlocked by another thread. This is called 791 <emphasis>lock contention</emphasis>. 792 </para> 793 794 <para> 795 Lock contention causes delays. Such delays should be as short as 796 possible. The two command line options 797 <literal>--exclusive-threshold=<n></literal> and 798 <literal>--shared-threshold=<n></literal> make it possible to 799 detect excessive lock contention by making DRD report any lock that 800 has been held longer than the specified threshold. An example: 801 </para> 802 <programlisting><![CDATA[ 803 $ valgrind --tool=drd --exclusive-threshold=10 drd/tests/hold_lock -i 500 804 ... 805 ==10668== Acquired at: 806 ==10668== at 0x4C267C8: pthread_mutex_lock (drd_pthread_intercepts.c:395) 807 ==10668== by 0x400D92: main (hold_lock.c:51) 808 ==10668== Lock on mutex 0x7fefffd50 was held during 503 ms (threshold: 10 ms). 809 ==10668== at 0x4C26ADA: pthread_mutex_unlock (drd_pthread_intercepts.c:441) 810 ==10668== by 0x400DB5: main (hold_lock.c:55) 811 ... 812 ]]></programlisting> 813 814 <para> 815 The <literal>hold_lock</literal> test program holds a lock as long as 816 specified by the <literal>-i</literal> (interval) argument. The DRD 817 output reports that the lock acquired at line 51 in source file 818 <literal>hold_lock.c</literal> and released at line 55 was held during 819 503 ms, while a threshold of 10 ms was specified to DRD. 820 </para> 821 822 </sect2> 823 824 825 <sect2 id="drd-manual.api-checks" xreflabel="API Checks"> 826 <title>Detected Errors: Misuse of the POSIX threads API</title> 827 828 <para> 829 DRD is able to detect and report the following misuses of the POSIX 830 threads API: 831 <itemizedlist> 832 <listitem> 833 <para> 834 Passing the address of one type of synchronization object 835 (e.g. a mutex) to a POSIX API call that expects a pointer to 836 another type of synchronization object (e.g. a condition 837 variable). 838 </para> 839 </listitem> 840 <listitem> 841 <para> 842 Attempts to unlock a mutex that has not been locked. 843 </para> 844 </listitem> 845 <listitem> 846 <para> 847 Attempts to unlock a mutex that was locked by another thread. 848 </para> 849 </listitem> 850 <listitem> 851 <para> 852 Attempts to lock a mutex of type 853 <literal>PTHREAD_MUTEX_NORMAL</literal> or a spinlock 854 recursively. 855 </para> 856 </listitem> 857 <listitem> 858 <para> 859 Destruction or deallocation of a locked mutex. 860 </para> 861 </listitem> 862 <listitem> 863 <para> 864 Sending a signal to a condition variable while no lock is held 865 on the mutex associated with the condition variable. 866 </para> 867 </listitem> 868 <listitem> 869 <para> 870 Calling <function>pthread_cond_wait</function> on a mutex 871 that is not locked, that is locked by another thread or that 872 has been locked recursively. 873 </para> 874 </listitem> 875 <listitem> 876 <para> 877 Associating two different mutexes with a condition variable 878 through <function>pthread_cond_wait</function>. 879 </para> 880 </listitem> 881 <listitem> 882 <para> 883 Destruction or deallocation of a condition variable that is 884 being waited upon. 885 </para> 886 </listitem> 887 <listitem> 888 <para> 889 Destruction or deallocation of a locked reader-writer synchronization 890 object. 891 </para> 892 </listitem> 893 <listitem> 894 <para> 895 Attempts to unlock a reader-writer synchronization object that was not 896 locked by the calling thread. 897 </para> 898 </listitem> 899 <listitem> 900 <para> 901 Attempts to recursively lock a reader-writer synchronization object 902 exclusively. 903 </para> 904 </listitem> 905 <listitem> 906 <para> 907 Attempts to pass the address of a user-defined reader-writer 908 synchronization object to a POSIX threads function. 909 </para> 910 </listitem> 911 <listitem> 912 <para> 913 Attempts to pass the address of a POSIX reader-writer synchronization 914 object to one of the annotations for user-defined reader-writer 915 synchronization objects. 916 </para> 917 </listitem> 918 <listitem> 919 <para> 920 Reinitialization of a mutex, condition variable, reader-writer 921 lock, semaphore or barrier. 922 </para> 923 </listitem> 924 <listitem> 925 <para> 926 Destruction or deallocation of a semaphore or barrier that is 927 being waited upon. 928 </para> 929 </listitem> 930 <listitem> 931 <para> 932 Missing synchronization between barrier wait and barrier destruction. 933 </para> 934 </listitem> 935 <listitem> 936 <para> 937 Exiting a thread without first unlocking the spinlocks, mutexes or 938 reader-writer synchronization objects that were locked by that thread. 939 </para> 940 </listitem> 941 <listitem> 942 <para> 943 Passing an invalid thread ID to <function>pthread_join</function> 944 or <function>pthread_cancel</function>. 945 </para> 946 </listitem> 947 </itemizedlist> 948 </para> 949 950 </sect2> 951 952 953 <sect2 id="drd-manual.clientreqs" xreflabel="Client requests"> 954 <title>Client Requests</title> 955 956 <para> 957 Just as for other Valgrind tools it is possible to let a client program 958 interact with the DRD tool through client requests. In addition to the 959 client requests several macros have been defined that allow to use the 960 client requests in a convenient way. 961 </para> 962 963 <para> 964 The interface between client programs and the DRD tool is defined in 965 the header file <literal><valgrind/drd.h></literal>. The 966 available macros and client requests are: 967 <itemizedlist> 968 <listitem> 969 <para> 970 The macro <literal>DRD_GET_VALGRIND_THREADID</literal> and the 971 corresponding client 972 request <varname>VG_USERREQ__DRD_GET_VALGRIND_THREAD_ID</varname>. 973 Query the thread ID that has been assigned by the Valgrind core to the 974 thread executing this client request. Valgrind's thread ID's start at 975 one and are recycled in case a thread stops. 976 </para> 977 </listitem> 978 <listitem> 979 <para> 980 The macro <literal>DRD_GET_DRD_THREADID</literal> and the corresponding 981 client request <varname>VG_USERREQ__DRD_GET_DRD_THREAD_ID</varname>. 982 Query the thread ID that has been assigned by DRD to the thread 983 executing this client request. These are the thread ID's reported by DRD 984 in data race reports and in trace messages. DRD's thread ID's start at 985 one and are never recycled. 986 </para> 987 </listitem> 988 <listitem> 989 <para> 990 The macros <literal>DRD_IGNORE_VAR(x)</literal>, 991 <literal>ANNOTATE_TRACE_MEMORY(&x)</literal> and the corresponding 992 client request <varname>VG_USERREQ__DRD_START_SUPPRESSION</varname>. Some 993 applications contain intentional races. There exist e.g. applications 994 where the same value is assigned to a shared variable from two different 995 threads. It may be more convenient to suppress such races than to solve 996 these. This client request allows to suppress such races. 997 </para> 998 </listitem> 999 <listitem> 1000 <para> 1001 The macro <literal>DRD_STOP_IGNORING_VAR(x)</literal> and the 1002 corresponding client request 1003 <varname>VG_USERREQ__DRD_FINISH_SUPPRESSION</varname>. Tell DRD 1004 to no longer ignore data races for the address range that was suppressed 1005 either via the macro <literal>DRD_IGNORE_VAR(x)</literal> or via the 1006 client request <varname>VG_USERREQ__DRD_START_SUPPRESSION</varname>. 1007 </para> 1008 </listitem> 1009 <listitem> 1010 <para> 1011 The macro <literal>DRD_TRACE_VAR(x)</literal>. Trace all load and store 1012 activity for the address range starting at <literal>&x</literal> and 1013 occupying <literal>sizeof(x)</literal> bytes. When DRD reports a data 1014 race on a specified variable, and it's not immediately clear which 1015 source code statements triggered the conflicting accesses, it can be 1016 very helpful to trace all activity on the offending memory location. 1017 </para> 1018 </listitem> 1019 <listitem> 1020 <para> 1021 The macro <literal>DRD_STOP_TRACING_VAR(x)</literal>. Stop tracing load 1022 and store activity for the address range starting 1023 at <literal>&x</literal> and occupying <literal>sizeof(x)</literal> 1024 bytes. 1025 </para> 1026 </listitem> 1027 <listitem> 1028 <para> 1029 The macro <literal>ANNOTATE_TRACE_MEMORY(&x)</literal>. Trace all 1030 load and store activity that touches at least the single byte at the 1031 address <literal>&x</literal>. 1032 </para> 1033 </listitem> 1034 <listitem> 1035 <para> 1036 The client request <varname>VG_USERREQ__DRD_START_TRACE_ADDR</varname>, 1037 which allows to trace all load and store activity for the specified 1038 address range. 1039 </para> 1040 </listitem> 1041 <listitem> 1042 <para> 1043 The client 1044 request <varname>VG_USERREQ__DRD_STOP_TRACE_ADDR</varname>. Do no longer 1045 trace load and store activity for the specified address range. 1046 </para> 1047 </listitem> 1048 <listitem> 1049 <para> 1050 The macro <literal>ANNOTATE_HAPPENS_BEFORE(addr)</literal> tells DRD to 1051 insert a mark. Insert this macro just after an access to the variable at 1052 the specified address has been performed. 1053 </para> 1054 </listitem> 1055 <listitem> 1056 <para> 1057 The macro <literal>ANNOTATE_HAPPENS_AFTER(addr)</literal> tells DRD that 1058 the next access to the variable at the specified address should be 1059 considered to have happened after the access just before the latest 1060 <literal>ANNOTATE_HAPPENS_BEFORE(addr)</literal> annotation that 1061 references the same variable. The purpose of these two macros is to tell 1062 DRD about the order of inter-thread memory accesses implemented via 1063 atomic memory operations. See 1064 also <literal>drd/tests/annotate_smart_pointer.cpp</literal> for an 1065 example. 1066 </para> 1067 </listitem> 1068 <listitem> 1069 <para> 1070 The macro <literal>ANNOTATE_RWLOCK_CREATE(rwlock)</literal> tells DRD 1071 that the object at address <literal>rwlock</literal> is a 1072 reader-writer synchronization object that is not a 1073 <literal>pthread_rwlock_t</literal> synchronization object. See 1074 also <literal>drd/tests/annotate_rwlock.c</literal> for an example. 1075 </para> 1076 </listitem> 1077 <listitem> 1078 <para> 1079 The macro <literal>ANNOTATE_RWLOCK_DESTROY(rwlock)</literal> tells DRD 1080 that the reader-writer synchronization object at 1081 address <literal>rwlock</literal> has been destroyed. 1082 </para> 1083 </listitem> 1084 <listitem> 1085 <para> 1086 The macro <literal>ANNOTATE_WRITERLOCK_ACQUIRED(rwlock)</literal> tells 1087 DRD that a writer lock has been acquired on the reader-writer 1088 synchronization object at address <literal>rwlock</literal>. 1089 </para> 1090 </listitem> 1091 <listitem> 1092 <para> 1093 The macro <literal>ANNOTATE_READERLOCK_ACQUIRED(rwlock)</literal> tells 1094 DRD that a reader lock has been acquired on the reader-writer 1095 synchronization object at address <literal>rwlock</literal>. 1096 </para> 1097 </listitem> 1098 <listitem> 1099 <para> 1100 The macro <literal>ANNOTATE_RWLOCK_ACQUIRED(rwlock, is_w)</literal> 1101 tells DRD that a writer lock (when <literal>is_w != 0</literal>) or that 1102 a reader lock (when <literal>is_w == 0</literal>) has been acquired on 1103 the reader-writer synchronization object at 1104 address <literal>rwlock</literal>. 1105 </para> 1106 </listitem> 1107 <listitem> 1108 <para> 1109 The macro <literal>ANNOTATE_WRITERLOCK_RELEASED(rwlock)</literal> tells 1110 DRD that a writer lock has been released on the reader-writer 1111 synchronization object at address <literal>rwlock</literal>. 1112 </para> 1113 </listitem> 1114 <listitem> 1115 <para> 1116 The macro <literal>ANNOTATE_READERLOCK_RELEASED(rwlock)</literal> tells 1117 DRD that a reader lock has been released on the reader-writer 1118 synchronization object at address <literal>rwlock</literal>. 1119 </para> 1120 </listitem> 1121 <listitem> 1122 <para> 1123 The macro <literal>ANNOTATE_RWLOCK_RELEASED(rwlock, is_w)</literal> 1124 tells DRD that a writer lock (when <literal>is_w != 0</literal>) or that 1125 a reader lock (when <literal>is_w == 0</literal>) has been released on 1126 the reader-writer synchronization object at 1127 address <literal>rwlock</literal>. 1128 </para> 1129 </listitem> 1130 <listitem> 1131 <para> 1132 The macro <literal>ANNOTATE_BARRIER_INIT(barrier, count, 1133 reinitialization_allowed)</literal> tells DRD that a new barrier object 1134 at the address <literal>barrier</literal> has been initialized, 1135 that <literal>count</literal> threads participate in each barrier and 1136 also whether or not barrier reinitialization without intervening 1137 destruction should be reported as an error. See 1138 also <literal>drd/tests/annotate_barrier.c</literal> for an example. 1139 </para> 1140 </listitem> 1141 <listitem> 1142 <para> 1143 The macro <literal>ANNOTATE_BARRIER_DESTROY(barrier)</literal> 1144 tells DRD that a barrier object is about to be destroyed. 1145 </para> 1146 </listitem> 1147 <listitem> 1148 <para> 1149 The macro <literal>ANNOTATE_BARRIER_WAIT_BEFORE(barrier)</literal> 1150 tells DRD that waiting for a barrier will start. 1151 </para> 1152 </listitem> 1153 <listitem> 1154 <para> 1155 The macro <literal>ANNOTATE_BARRIER_WAIT_AFTER(barrier)</literal> 1156 tells DRD that waiting for a barrier has finished. 1157 </para> 1158 </listitem> 1159 <listitem> 1160 <para> 1161 The macro <literal>ANNOTATE_BENIGN_RACE_SIZED(addr, size, 1162 descr)</literal> tells DRD that any races detected on the specified 1163 address are benign and hence should not be 1164 reported. The <literal>descr</literal> argument is ignored but can be 1165 used to document why data races on <literal>addr</literal> are benign. 1166 </para> 1167 </listitem> 1168 <listitem> 1169 <para> 1170 The macro <literal>ANNOTATE_BENIGN_RACE_STATIC(var, descr)</literal> 1171 tells DRD that any races detected on the specified static variable are 1172 benign and hence should not be reported. The <literal>descr</literal> 1173 argument is ignored but can be used to document why data races 1174 on <literal>var</literal> are benign. Note: this macro can only be 1175 used in C++ programs and not in C programs. 1176 </para> 1177 </listitem> 1178 <listitem> 1179 <para> 1180 The macro <literal>ANNOTATE_IGNORE_READS_BEGIN</literal> tells 1181 DRD to ignore all memory loads performed by the current thread. 1182 </para> 1183 </listitem> 1184 <listitem> 1185 <para> 1186 The macro <literal>ANNOTATE_IGNORE_READS_END</literal> tells 1187 DRD to stop ignoring the memory loads performed by the current thread. 1188 </para> 1189 </listitem> 1190 <listitem> 1191 <para> 1192 The macro <literal>ANNOTATE_IGNORE_WRITES_BEGIN</literal> tells 1193 DRD to ignore all memory stores performed by the current thread. 1194 </para> 1195 </listitem> 1196 <listitem> 1197 <para> 1198 The macro <literal>ANNOTATE_IGNORE_WRITES_END</literal> tells 1199 DRD to stop ignoring the memory stores performed by the current thread. 1200 </para> 1201 </listitem> 1202 <listitem> 1203 <para> 1204 The macro <literal>ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN</literal> tells 1205 DRD to ignore all memory accesses performed by the current thread. 1206 </para> 1207 </listitem> 1208 <listitem> 1209 <para> 1210 The macro <literal>ANNOTATE_IGNORE_READS_AND_WRITES_END</literal> tells 1211 DRD to stop ignoring the memory accesses performed by the current thread. 1212 </para> 1213 </listitem> 1214 <listitem> 1215 <para> 1216 The macro <literal>ANNOTATE_NEW_MEMORY(addr, size)</literal> tells 1217 DRD that the specified memory range has been allocated by a custom 1218 memory allocator in the client program and that the client program 1219 will start using this memory range. 1220 </para> 1221 </listitem> 1222 <listitem> 1223 <para> 1224 The macro <literal>ANNOTATE_THREAD_NAME(name)</literal> tells DRD to 1225 associate the specified name with the current thread and to include this 1226 name in the error messages printed by DRD. 1227 </para> 1228 </listitem> 1229 <listitem> 1230 <para> 1231 The macros <literal>VALGRIND_MALLOCLIKE_BLOCK</literal> and 1232 <literal>VALGRIND_FREELIKE_BLOCK</literal> from the Valgrind core are 1233 implemented; they are described in 1234 <xref linkend="manual-core-adv.clientreq"/>. 1235 </para> 1236 </listitem> 1237 </itemizedlist> 1238 </para> 1239 1240 <para> 1241 Note: if you compiled Valgrind yourself, the header file 1242 <literal><valgrind/drd.h></literal> will have been installed in 1243 the directory <literal>/usr/include</literal> by the command 1244 <literal>make install</literal>. If you obtained Valgrind by 1245 installing it as a package however, you will probably have to install 1246 another package with a name like <literal>valgrind-devel</literal> 1247 before Valgrind's header files are available. 1248 </para> 1249 1250 </sect2> 1251 1252 1253 <sect2 id="drd-manual.gnome" xreflabel="GNOME"> 1254 <title>Debugging GNOME Programs</title> 1255 1256 <para> 1257 GNOME applications use the threading primitives provided by the 1258 <computeroutput>glib</computeroutput> and 1259 <computeroutput>gthread</computeroutput> libraries. These libraries 1260 are built on top of POSIX threads, and hence are directly supported by 1261 DRD. Please keep in mind that you have to call 1262 <function>g_thread_init</function> before creating any threads, or 1263 DRD will report several data races on glib functions. See also the 1264 <ulink 1265 url="http://library.gnome.org/devel/glib/stable/glib-Threads.html">GLib 1266 Reference Manual</ulink> for more information about 1267 <function>g_thread_init</function>. 1268 </para> 1269 1270 <para> 1271 One of the many facilities provided by the <literal>glib</literal> 1272 library is a block allocator, called <literal>g_slice</literal>. You 1273 have to disable this block allocator when using DRD by adding the 1274 following to the shell environment variables: 1275 <literal>G_SLICE=always-malloc</literal>. See also the <ulink 1276 url="http://library.gnome.org/devel/glib/stable/glib-Memory-Slices.html">GLib 1277 Reference Manual</ulink> for more information. 1278 </para> 1279 1280 </sect2> 1281 1282 1283 <sect2 id="drd-manual.boost.thread" xreflabel="Boost.Thread"> 1284 <title>Debugging Boost.Thread Programs</title> 1285 1286 <para> 1287 The Boost.Thread library is the threading library included with the 1288 cross-platform Boost Libraries. This threading library is an early 1289 implementation of the upcoming C++0x threading library. 1290 </para> 1291 1292 <para> 1293 Applications that use the Boost.Thread library should run fine under DRD. 1294 </para> 1295 1296 <para> 1297 More information about Boost.Thread can be found here: 1298 <itemizedlist> 1299 <listitem> 1300 <para> 1301 Anthony Williams, <ulink 1302 url="http://www.boost.org/doc/libs/1_37_0/doc/html/thread.html">Boost.Thread</ulink> 1303 Library Documentation, Boost website, 2007. 1304 </para> 1305 </listitem> 1306 <listitem> 1307 <para> 1308 Anthony Williams, <ulink 1309 url="http://www.ddj.com/cpp/211600441">What's New in Boost 1310 Threads?</ulink>, Recent changes to the Boost Thread library, 1311 Dr. Dobbs Magazine, October 2008. 1312 </para> 1313 </listitem> 1314 </itemizedlist> 1315 </para> 1316 1317 </sect2> 1318 1319 1320 <sect2 id="drd-manual.openmp" xreflabel="OpenMP"> 1321 <title>Debugging OpenMP Programs</title> 1322 1323 <para> 1324 OpenMP stands for <emphasis>Open Multi-Processing</emphasis>. The OpenMP 1325 standard consists of a set of compiler directives for C, C++ and Fortran 1326 programs that allows a compiler to transform a sequential program into a 1327 parallel program. OpenMP is well suited for HPC applications and allows to 1328 work at a higher level compared to direct use of the POSIX threads API. While 1329 OpenMP ensures that the POSIX API is used correctly, OpenMP programs can still 1330 contain data races. So it definitely makes sense to verify OpenMP programs 1331 with a thread checking tool. 1332 </para> 1333 1334 <para> 1335 DRD supports OpenMP shared-memory programs generated by GCC. GCC 1336 supports OpenMP since version 4.2.0. GCC's runtime support 1337 for OpenMP programs is provided by a library called 1338 <literal>libgomp</literal>. The synchronization primitives implemented 1339 in this library use Linux' futex system call directly, unless the 1340 library has been configured with the 1341 <literal>--disable-linux-futex</literal> option. DRD only supports 1342 libgomp libraries that have been configured with this option and in 1343 which symbol information is present. For most Linux distributions this 1344 means that you will have to recompile GCC. See also the script 1345 <literal>drd/scripts/download-and-build-gcc</literal> in the 1346 Valgrind source tree for an example of how to compile GCC. You will 1347 also have to make sure that the newly compiled 1348 <literal>libgomp.so</literal> library is loaded when OpenMP programs 1349 are started. This is possible by adding a line similar to the 1350 following to your shell startup script: 1351 </para> 1352 <programlisting><![CDATA[ 1353 export LD_LIBRARY_PATH=~/gcc-4.4.0/lib64:~/gcc-4.4.0/lib: 1354 ]]></programlisting> 1355 1356 <para> 1357 As an example, the test OpenMP test program 1358 <literal>drd/tests/omp_matinv</literal> triggers a data race 1359 when the option -r has been specified on the command line. The data 1360 race is triggered by the following code: 1361 </para> 1362 <programlisting><![CDATA[ 1363 #pragma omp parallel for private(j) 1364 for (j = 0; j < rows; j++) 1365 { 1366 if (i != j) 1367 { 1368 const elem_t factor = a[j * cols + i]; 1369 for (k = 0; k < cols; k++) 1370 { 1371 a[j * cols + k] -= a[i * cols + k] * factor; 1372 } 1373 } 1374 } 1375 ]]></programlisting> 1376 1377 <para> 1378 The above code is racy because the variable <literal>k</literal> has 1379 not been declared private. DRD will print the following error message 1380 for the above code: 1381 </para> 1382 <programlisting><![CDATA[ 1383 $ valgrind --tool=drd --check-stack-var=yes --read-var-info=yes drd/tests/omp_matinv 3 -t 2 -r 1384 ... 1385 Conflicting store by thread 1/1 at 0x7fefffbc4 size 4 1386 at 0x4014A0: gj.omp_fn.0 (omp_matinv.c:203) 1387 by 0x401211: gj (omp_matinv.c:159) 1388 by 0x40166A: invert_matrix (omp_matinv.c:238) 1389 by 0x4019B4: main (omp_matinv.c:316) 1390 Location 0x7fefffbc4 is 0 bytes inside local var "k" 1391 declared at omp_matinv.c:160, in frame #0 of thread 1 1392 ... 1393 ]]></programlisting> 1394 <para> 1395 In the above output the function name <function>gj.omp_fn.0</function> 1396 has been generated by GCC from the function name 1397 <function>gj</function>. The allocation context information shows that the 1398 data race has been caused by modifying the variable <literal>k</literal>. 1399 </para> 1400 1401 <para> 1402 Note: for GCC versions before 4.4.0, no allocation context information is 1403 shown. With these GCC versions the most usable information in the above output 1404 is the source file name and the line number where the data race has been 1405 detected (<literal>omp_matinv.c:203</literal>). 1406 </para> 1407 1408 <para> 1409 For more information about OpenMP, see also 1410 <ulink url="http://openmp.org/">openmp.org</ulink>. 1411 </para> 1412 1413 </sect2> 1414 1415 1416 <sect2 id="drd-manual.cust-mem-alloc" xreflabel="Custom Memory Allocators"> 1417 <title>DRD and Custom Memory Allocators</title> 1418 1419 <para> 1420 DRD tracks all memory allocation events that happen via the 1421 standard memory allocation and deallocation functions 1422 (<function>malloc</function>, <function>free</function>, 1423 <function>new</function> and <function>delete</function>), via entry 1424 and exit of stack frames or that have been annotated with Valgrind's 1425 memory pool client requests. DRD uses memory allocation and deallocation 1426 information for two purposes: 1427 <itemizedlist> 1428 <listitem> 1429 <para> 1430 To know where the scope ends of POSIX objects that have not been 1431 destroyed explicitly. It is e.g. not required by the POSIX 1432 threads standard to call 1433 <function>pthread_mutex_destroy</function> before freeing the 1434 memory in which a mutex object resides. 1435 </para> 1436 </listitem> 1437 <listitem> 1438 <para> 1439 To know where the scope of variables ends. If e.g. heap memory 1440 has been used by one thread, that thread frees that memory, and 1441 another thread allocates and starts using that memory, no data 1442 races must be reported for that memory. 1443 </para> 1444 </listitem> 1445 </itemizedlist> 1446 </para> 1447 1448 <para> 1449 It is essential for correct operation of DRD that the tool knows about 1450 memory allocation and deallocation events. When analyzing a client program 1451 with DRD that uses a custom memory allocator, either instrument the custom 1452 memory allocator with the <literal>VALGRIND_MALLOCLIKE_BLOCK</literal> 1453 and <literal>VALGRIND_FREELIKE_BLOCK</literal> macros or disable the 1454 custom memory allocator. 1455 </para> 1456 1457 <para> 1458 As an example, the GNU libstdc++ library can be configured 1459 to use standard memory allocation functions instead of memory pools by 1460 setting the environment variable 1461 <literal>GLIBCXX_FORCE_NEW</literal>. For more information, see also 1462 the <ulink 1463 url="http://gcc.gnu.org/onlinedocs/libstdc++/manual/bk01pt04ch11.html">libstdc++ 1464 manual</ulink>. 1465 </para> 1466 1467 </sect2> 1468 1469 1470 <sect2 id="drd-manual.drd-versus-memcheck" xreflabel="DRD Versus Memcheck"> 1471 <title>DRD Versus Memcheck</title> 1472 1473 <para> 1474 It is essential for correct operation of DRD that there are no memory 1475 errors such as dangling pointers in the client program. Which means that 1476 it is a good idea to make sure that your program is Memcheck-clean 1477 before you analyze it with DRD. It is possible however that some of 1478 the Memcheck reports are caused by data races. In this case it makes 1479 sense to run DRD before Memcheck. 1480 </para> 1481 1482 <para> 1483 So which tool should be run first? In case both DRD and Memcheck 1484 complain about a program, a possible approach is to run both tools 1485 alternatingly and to fix as many errors as possible after each run of 1486 each tool until none of the two tools prints any more error messages. 1487 </para> 1488 1489 </sect2> 1490 1491 1492 <sect2 id="drd-manual.resource-requirements" xreflabel="Resource Requirements"> 1493 <title>Resource Requirements</title> 1494 1495 <para> 1496 The requirements of DRD with regard to heap and stack memory and the 1497 effect on the execution time of client programs are as follows: 1498 <itemizedlist> 1499 <listitem> 1500 <para> 1501 When running a program under DRD with default DRD options, 1502 between 1.1 and 3.6 times more memory will be needed compared to 1503 a native run of the client program. More memory will be needed 1504 if loading debug information has been enabled 1505 (<literal>--read-var-info=yes</literal>). 1506 </para> 1507 </listitem> 1508 <listitem> 1509 <para> 1510 DRD allocates some of its temporary data structures on the stack 1511 of the client program threads. This amount of data is limited to 1512 1 - 2 KB. Make sure that thread stacks are sufficiently large. 1513 </para> 1514 </listitem> 1515 <listitem> 1516 <para> 1517 Most applications will run between 20 and 50 times slower under 1518 DRD than a native single-threaded run. The slowdown will be most 1519 noticeable for applications which perform frequent mutex lock / 1520 unlock operations. 1521 </para> 1522 </listitem> 1523 </itemizedlist> 1524 </para> 1525 1526 </sect2> 1527 1528 1529 <sect2 id="drd-manual.effective-use" xreflabel="Effective Use"> 1530 <title>Hints and Tips for Effective Use of DRD</title> 1531 1532 <para> 1533 The following information may be helpful when using DRD: 1534 <itemizedlist> 1535 <listitem> 1536 <para> 1537 Make sure that debug information is present in the executable 1538 being analyzed, such that DRD can print function name and line 1539 number information in stack traces. Most compilers can be told 1540 to include debug information via compiler option 1541 <option>-g</option>. 1542 </para> 1543 </listitem> 1544 <listitem> 1545 <para> 1546 Compile with option <option>-O1</option> instead of 1547 <option>-O0</option>. This will reduce the amount of generated 1548 code, may reduce the amount of debug info and will speed up 1549 DRD's processing of the client program. For more information, 1550 see also <xref linkend="manual-core.started"/>. 1551 </para> 1552 </listitem> 1553 <listitem> 1554 <para> 1555 If DRD reports any errors on libraries that are part of your 1556 Linux distribution like e.g. <literal>libc.so</literal> or 1557 <literal>libstdc++.so</literal>, installing the debug packages 1558 for these libraries will make the output of DRD a lot more 1559 detailed. 1560 </para> 1561 </listitem> 1562 <listitem> 1563 <para> 1564 When using C++, do not send output from more than one thread to 1565 <literal>std::cout</literal>. Doing so would not only 1566 generate multiple data race reports, it could also result in 1567 output from several threads getting mixed up. Either use 1568 <function>printf</function> or do the following: 1569 <orderedlist> 1570 <listitem> 1571 <para>Derive a class from <literal>std::ostreambuf</literal> 1572 and let that class send output line by line to 1573 <literal>stdout</literal>. This will avoid that individual 1574 lines of text produced by different threads get mixed 1575 up.</para> 1576 </listitem> 1577 <listitem> 1578 <para>Create one instance of <literal>std::ostream</literal> 1579 for each thread. This makes stream formatting settings 1580 thread-local. Pass a per-thread instance of the class 1581 derived from <literal>std::ostreambuf</literal> to the 1582 constructor of each instance. </para> 1583 </listitem> 1584 <listitem> 1585 <para>Let each thread send its output to its own instance of 1586 <literal>std::ostream</literal> instead of 1587 <literal>std::cout</literal>.</para> 1588 </listitem> 1589 </orderedlist> 1590 </para> 1591 </listitem> 1592 </itemizedlist> 1593 </para> 1594 1595 </sect2> 1596 1597 1598 </sect1> 1599 1600 1601 <sect1 id="drd-manual.Pthreads" xreflabel="Pthreads"> 1602 <title>Using the POSIX Threads API Effectively</title> 1603 1604 <sect2 id="drd-manual.mutex-types" xreflabel="mutex-types"> 1605 <title>Mutex types</title> 1606 1607 <para> 1608 The Single UNIX Specification version two defines the following four 1609 mutex types (see also the documentation of <ulink 1610 url="http://www.opengroup.org/onlinepubs/007908799/xsh/pthread_mutexattr_settype.html"><function>pthread_mutexattr_settype</function></ulink>): 1611 <itemizedlist> 1612 <listitem> 1613 <para> 1614 <emphasis>normal</emphasis>, which means that no error checking 1615 is performed, and that the mutex is non-recursive. 1616 </para> 1617 </listitem> 1618 <listitem> 1619 <para> 1620 <emphasis>error checking</emphasis>, which means that the mutex 1621 is non-recursive and that error checking is performed. 1622 </para> 1623 </listitem> 1624 <listitem> 1625 <para> 1626 <emphasis>recursive</emphasis>, which means that a mutex may be 1627 locked recursively. 1628 </para> 1629 </listitem> 1630 <listitem> 1631 <para> 1632 <emphasis>default</emphasis>, which means that error checking 1633 behavior is undefined, and that the behavior for recursive 1634 locking is also undefined. Or: portable code must neither 1635 trigger error conditions through the Pthreads API nor attempt to 1636 lock a mutex of default type recursively. 1637 </para> 1638 </listitem> 1639 </itemizedlist> 1640 </para> 1641 1642 <para> 1643 In complex applications it is not always clear from beforehand which 1644 mutex will be locked recursively and which mutex will not be locked 1645 recursively. Attempts lock a non-recursive mutex recursively will 1646 result in race conditions that are very hard to find without a thread 1647 checking tool. So either use the error checking mutex type and 1648 consistently check the return value of Pthread API mutex calls, or use 1649 the recursive mutex type. 1650 </para> 1651 1652 </sect2> 1653 1654 <sect2 id="drd-manual.condvar" xreflabel="condition-variables"> 1655 <title>Condition variables</title> 1656 1657 <para> 1658 A condition variable allows one thread to wake up one or more other 1659 threads. Condition variables are often used to notify one or more 1660 threads about state changes of shared data. Unfortunately it is very 1661 easy to introduce race conditions by using condition variables as the 1662 only means of state information propagation. A better approach is to 1663 let threads poll for changes of a state variable that is protected by 1664 a mutex, and to use condition variables only as a thread wakeup 1665 mechanism. See also the source file 1666 <computeroutput>drd/tests/monitor_example.cpp</computeroutput> for an 1667 example of how to implement this concept in C++. The monitor concept 1668 used in this example is a well known and very useful concept -- see 1669 also Wikipedia for more information about the <ulink 1670 url="http://en.wikipedia.org/wiki/Monitor_(synchronization)">monitor</ulink> 1671 concept. 1672 </para> 1673 1674 </sect2> 1675 1676 <sect2 id="drd-manual.pctw" xreflabel="pthread_cond_timedwait"> 1677 <title>pthread_cond_timedwait and timeouts</title> 1678 1679 <para> 1680 Historically the function 1681 <function>pthread_cond_timedwait</function> only allowed the 1682 specification of an absolute timeout, that is a timeout independent of 1683 the time when this function was called. However, almost every call to 1684 this function expresses a relative timeout. This typically happens by 1685 passing the sum of 1686 <computeroutput>clock_gettime(CLOCK_REALTIME)</computeroutput> and a 1687 relative timeout as the third argument. This approach is incorrect 1688 since forward or backward clock adjustments by e.g. ntpd will affect 1689 the timeout. A more reliable approach is as follows: 1690 <itemizedlist> 1691 <listitem> 1692 <para> 1693 When initializing a condition variable through 1694 <function>pthread_cond_init</function>, specify that the timeout of 1695 <function>pthread_cond_timedwait</function> will use the clock 1696 <literal>CLOCK_MONOTONIC</literal> instead of 1697 <literal>CLOCK_REALTIME</literal>. You can do this via 1698 <computeroutput>pthread_condattr_setclock(..., 1699 CLOCK_MONOTONIC)</computeroutput>. 1700 </para> 1701 </listitem> 1702 <listitem> 1703 <para> 1704 When calling <function>pthread_cond_timedwait</function>, pass 1705 the sum of 1706 <computeroutput>clock_gettime(CLOCK_MONOTONIC)</computeroutput> 1707 and a relative timeout as the third argument. 1708 </para> 1709 </listitem> 1710 </itemizedlist> 1711 See also 1712 <computeroutput>drd/tests/monitor_example.cpp</computeroutput> for an 1713 example. 1714 </para> 1715 1716 </sect2> 1717 1718 </sect1> 1719 1720 1721 <sect1 id="drd-manual.limitations" xreflabel="Limitations"> 1722 <title>Limitations</title> 1723 1724 <para>DRD currently has the following limitations:</para> 1725 1726 <itemizedlist> 1727 <listitem> 1728 <para> 1729 DRD, just like Memcheck, will refuse to start on Linux 1730 distributions where all symbol information has been removed from 1731 <filename>ld.so</filename>. This is e.g. the case for the PPC editions 1732 of openSUSE and Gentoo. You will have to install the glibc debuginfo 1733 package on these platforms before you can use DRD. See also openSUSE 1734 bug <ulink url="http://bugzilla.novell.com/show_bug.cgi?id=396197"> 1735 396197</ulink> and Gentoo bug <ulink 1736 url="http://bugs.gentoo.org/214065">214065</ulink>. 1737 </para> 1738 </listitem> 1739 <listitem> 1740 <para> 1741 With gcc 4.4.3 and before, DRD may report data races on the C++ 1742 class <literal>std::string</literal> in a multithreaded program. This is 1743 a know <literal>libstdc++</literal> issue -- see also GCC bug 1744 <ulink url="http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40518">40518</ulink> 1745 for more information. 1746 </para> 1747 </listitem> 1748 <listitem> 1749 <para> 1750 If you compile the DRD source code yourself, you need GCC 3.0 or 1751 later. GCC 2.95 is not supported. 1752 </para> 1753 </listitem> 1754 <listitem> 1755 <para> 1756 Of the two POSIX threads implementations for Linux, only the 1757 NPTL (Native POSIX Thread Library) is supported. The older 1758 LinuxThreads library is not supported. 1759 </para> 1760 </listitem> 1761 </itemizedlist> 1762 1763 </sect1> 1764 1765 1766 <sect1 id="drd-manual.feedback" xreflabel="Feedback"> 1767 <title>Feedback</title> 1768 1769 <para> 1770 If you have any comments, suggestions, feedback or bug reports about 1771 DRD, feel free to either post a message on the Valgrind users mailing 1772 list or to file a bug report. See also <ulink 1773 url="&vg-url;">&vg-url;</ulink> for more information. 1774 </para> 1775 1776 </sect1> 1777 1778 1779 </chapter> 1780
