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> 353 <![CDATA[--first-race-only=<yes|no> [default: no]]]> 354 </option> 355 </term> 356 <listitem> 357 <para> 358 Whether to report only the first data race that has been detected on a 359 memory location or all data races that have been detected on a memory 360 location. 361 </para> 362 </listitem> 363 </varlistentry> 364 <varlistentry> 365 <term> 366 <option> 367 <![CDATA[--free-is-write=<yes|no> [default: no]]]> 368 </option> 369 </term> 370 <listitem> 371 <para> 372 Whether to report accessing freed memory as a race. Helps to detect 373 memory accesses that occur after memory has been freed but might cause 374 DRD to run slightly slower. 375 </para> 376 </listitem> 377 </varlistentry> 378 <varlistentry> 379 <term> 380 <option> 381 <![CDATA[--report-signal-unlocked=<yes|no> [default: yes]]]> 382 </option> 383 </term> 384 <listitem> 385 <para> 386 Whether to report calls to 387 <function>pthread_cond_signal</function> and 388 <function>pthread_cond_broadcast</function> where the mutex 389 associated with the signal through 390 <function>pthread_cond_wait</function> or 391 <function>pthread_cond_timed_wait</function>is not locked at 392 the time the signal is sent. Sending a signal without holding 393 a lock on the associated mutex is a common programming error 394 which can cause subtle race conditions and unpredictable 395 behavior. There exist some uncommon synchronization patterns 396 however where it is safe to send a signal without holding a 397 lock on the associated mutex. 398 </para> 399 </listitem> 400 </varlistentry> 401 <varlistentry> 402 <term> 403 <option><![CDATA[--segment-merging=<yes|no> [default: yes]]]></option> 404 </term> 405 <listitem> 406 <para> 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 </para> 413 </listitem> 414 </varlistentry> 415 <varlistentry> 416 <term> 417 <option><![CDATA[--segment-merging-interval=<n> [default: 10]]]></option> 418 </term> 419 <listitem> 420 <para> 421 Perform segment merging only after the specified number of new 422 segments have been created. This is an advanced configuration option 423 that allows to choose whether to minimize DRD's memory usage by 424 choosing a low value or to let DRD run faster by choosing a slightly 425 higher value. The optimal value for this parameter depends on the 426 program being analyzed. The default value works well for most programs. 427 </para> 428 </listitem> 429 </varlistentry> 430 <varlistentry> 431 <term> 432 <option><![CDATA[--shared-threshold=<n> [default: off]]]></option> 433 </term> 434 <listitem> 435 <para> 436 Print an error message if a reader lock has been held longer 437 than the specified time (in milliseconds). This option enables 438 the detection of lock contention. 439 </para> 440 </listitem> 441 </varlistentry> 442 <varlistentry> 443 <term> 444 <option><![CDATA[--show-confl-seg=<yes|no> [default: yes]]]></option> 445 </term> 446 <listitem> 447 <para> 448 Show conflicting segments in race reports. Since this 449 information can help to find the cause of a data race, this 450 option is enabled by default. Disabling this option makes the 451 output of DRD more compact. 452 </para> 453 </listitem> 454 </varlistentry> 455 <varlistentry> 456 <term> 457 <option><![CDATA[--show-stack-usage=<yes|no> [default: no]]]></option> 458 </term> 459 <listitem> 460 <para> 461 Print stack usage at thread exit time. When a program creates a large 462 number of threads it becomes important to limit the amount of virtual 463 memory allocated for thread stacks. This option makes it possible to 464 observe how much stack memory has been used by each thread of the the 465 client program. Note: the DRD tool itself allocates some temporary 466 data on the client thread stack. The space necessary for this 467 temporary data must be allocated by the client program when it 468 allocates stack memory, but is not included in stack usage reported by 469 DRD. 470 </para> 471 </listitem> 472 </varlistentry> 473 </variablelist> 474 <!-- end of xi:include in the manpage --> 475 476 <!-- start of xi:include in the manpage --> 477 <para> 478 The following options are available for monitoring the behavior of the 479 client program: 480 </para> 481 482 <variablelist id="drd.debugopts.list"> 483 <varlistentry> 484 <term> 485 <option><![CDATA[--trace-addr=<address> [default: none]]]></option> 486 </term> 487 <listitem> 488 <para> 489 Trace all load and store activity for the specified 490 address. This option may be specified more than once. 491 </para> 492 </listitem> 493 </varlistentry> 494 <varlistentry> 495 <term> 496 <option><![CDATA[--trace-alloc=<yes|no> [default: no]]]></option> 497 </term> 498 <listitem> 499 <para> 500 Trace all memory allocations and deallocations. May produce a huge 501 amount of output. 502 </para> 503 </listitem> 504 </varlistentry> 505 <varlistentry> 506 <term> 507 <option><![CDATA[--trace-barrier=<yes|no> [default: no]]]></option> 508 </term> 509 <listitem> 510 <para> 511 Trace all barrier activity. 512 </para> 513 </listitem> 514 </varlistentry> 515 <varlistentry> 516 <term> 517 <option><![CDATA[--trace-cond=<yes|no> [default: no]]]></option> 518 </term> 519 <listitem> 520 <para> 521 Trace all condition variable activity. 522 </para> 523 </listitem> 524 </varlistentry> 525 <varlistentry> 526 <term> 527 <option><![CDATA[--trace-fork-join=<yes|no> [default: no]]]></option> 528 </term> 529 <listitem> 530 <para> 531 Trace all thread creation and all thread termination events. 532 </para> 533 </listitem> 534 </varlistentry> 535 <varlistentry> 536 <term> 537 <option><![CDATA[--trace-mutex=<yes|no> [default: no]]]></option> 538 </term> 539 <listitem> 540 <para> 541 Trace all mutex activity. 542 </para> 543 </listitem> 544 </varlistentry> 545 <varlistentry> 546 <term> 547 <option><![CDATA[--trace-rwlock=<yes|no> [default: no]]]></option> 548 </term> 549 <listitem> 550 <para> 551 Trace all reader-writer lock activity. 552 </para> 553 </listitem> 554 </varlistentry> 555 <varlistentry> 556 <term> 557 <option><![CDATA[--trace-semaphore=<yes|no> [default: no]]]></option> 558 </term> 559 <listitem> 560 <para> 561 Trace all semaphore activity. 562 </para> 563 </listitem> 564 </varlistentry> 565 </variablelist> 566 <!-- end of xi:include in the manpage --> 567 568 </sect2> 569 570 571 <sect2 id="drd-manual.data-races" xreflabel="Data Races"> 572 <title>Detected Errors: Data Races</title> 573 574 <para> 575 DRD prints a message every time it detects a data race. Please keep 576 the following in mind when interpreting DRD's output: 577 <itemizedlist> 578 <listitem> 579 <para> 580 Every thread is assigned a <emphasis>thread ID</emphasis> by the DRD 581 tool. A thread ID is a number. Thread ID's start at one and are never 582 recycled. 583 </para> 584 </listitem> 585 <listitem> 586 <para> 587 The term <emphasis>segment</emphasis> refers to a consecutive 588 sequence of load, store and synchronization operations, all 589 issued by the same thread. A segment always starts and ends at a 590 synchronization operation. Data race analysis is performed 591 between segments instead of between individual load and store 592 operations because of performance reasons. 593 </para> 594 </listitem> 595 <listitem> 596 <para> 597 There are always at least two memory accesses involved in a data 598 race. Memory accesses involved in a data race are called 599 <emphasis>conflicting memory accesses</emphasis>. DRD prints a 600 report for each memory access that conflicts with a past memory 601 access. 602 </para> 603 </listitem> 604 </itemizedlist> 605 </para> 606 607 <para> 608 Below you can find an example of a message printed by DRD when it 609 detects a data race: 610 </para> 611 <programlisting><![CDATA[ 612 $ valgrind --tool=drd --read-var-info=yes drd/tests/rwlock_race 613 ... 614 ==9466== Thread 3: 615 ==9466== Conflicting load by thread 3 at 0x006020b8 size 4 616 ==9466== at 0x400B6C: thread_func (rwlock_race.c:29) 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 ==9466== Location 0x6020b8 is 0 bytes inside local var "s_racy" 621 ==9466== declared at rwlock_race.c:18, in frame #0 of thread 3 622 ==9466== Other segment start (thread 2) 623 ==9466== at 0x4C2847D: pthread_rwlock_rdlock* (drd_pthread_intercepts.c:813) 624 ==9466== by 0x400B6B: thread_func (rwlock_race.c:28) 625 ==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186) 626 ==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so) 627 ==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so) 628 ==9466== Other segment end (thread 2) 629 ==9466== at 0x4C28B54: pthread_rwlock_unlock* (drd_pthread_intercepts.c:912) 630 ==9466== by 0x400B84: thread_func (rwlock_race.c:30) 631 ==9466== by 0x4C291DF: vg_thread_wrapper (drd_pthread_intercepts.c:186) 632 ==9466== by 0x4E3403F: start_thread (in /lib64/libpthread-2.8.so) 633 ==9466== by 0x53250CC: clone (in /lib64/libc-2.8.so) 634 ... 635 ]]></programlisting> 636 637 <para> 638 The above report has the following meaning: 639 <itemizedlist> 640 <listitem> 641 <para> 642 The number in the column on the left is the process ID of the 643 process being analyzed by DRD. 644 </para> 645 </listitem> 646 <listitem> 647 <para> 648 The first line ("Thread 3") tells you the thread ID for 649 the thread in which context the data race has been detected. 650 </para> 651 </listitem> 652 <listitem> 653 <para> 654 The next line tells which kind of operation was performed (load or 655 store) and by which thread. On the same line the start address and the 656 number of bytes involved in the conflicting access are also displayed. 657 </para> 658 </listitem> 659 <listitem> 660 <para> 661 Next, the call stack of the conflicting access is displayed. If 662 your program has been compiled with debug information 663 (<option>-g</option>), this call stack will include file names and 664 line numbers. The two 665 bottommost frames in this call stack (<function>clone</function> 666 and <function>start_thread</function>) show how the NPTL starts 667 a thread. The third frame 668 (<function>vg_thread_wrapper</function>) is added by DRD. The 669 fourth frame (<function>thread_func</function>) is the first 670 interesting line because it shows the thread entry point, that 671 is the function that has been passed as the third argument to 672 <function>pthread_create</function>. 673 </para> 674 </listitem> 675 <listitem> 676 <para> 677 Next, the allocation context for the conflicting address is 678 displayed. For dynamically allocated data the allocation call 679 stack is shown. For static variables and stack variables the 680 allocation context is only shown when the option 681 <option>--read-var-info=yes</option> has been 682 specified. Otherwise DRD will print <computeroutput>Allocation 683 context: unknown</computeroutput>. 684 </para> 685 </listitem> 686 <listitem> 687 <para> 688 A conflicting access involves at least two memory accesses. For 689 one of these accesses an exact call stack is displayed, and for 690 the other accesses an approximate call stack is displayed, 691 namely the start and the end of the segments of the other 692 accesses. This information can be interpreted as follows: 693 <orderedlist> 694 <listitem> 695 <para> 696 Start at the bottom of both call stacks, and count the 697 number stack frames with identical function name, file 698 name and line number. In the above example the three 699 bottommost frames are identical 700 (<function>clone</function>, 701 <function>start_thread</function> and 702 <function>vg_thread_wrapper</function>). 703 </para> 704 </listitem> 705 <listitem> 706 <para> 707 The next higher stack frame in both call stacks now tells 708 you between in which source code region the other memory 709 access happened. The above output tells that the other 710 memory access involved in the data race happened between 711 source code lines 28 and 30 in file 712 <computeroutput>rwlock_race.c</computeroutput>. 713 </para> 714 </listitem> 715 </orderedlist> 716 </para> 717 </listitem> 718 </itemizedlist> 719 </para> 720 721 </sect2> 722 723 724 <sect2 id="drd-manual.lock-contention" xreflabel="Lock Contention"> 725 <title>Detected Errors: Lock Contention</title> 726 727 <para> 728 Threads must be able to make progress without being blocked for too long by 729 other threads. Sometimes a thread has to wait until a mutex or reader-writer 730 synchronization object is unlocked by another thread. This is called 731 <emphasis>lock contention</emphasis>. 732 </para> 733 734 <para> 735 Lock contention causes delays. Such delays should be as short as 736 possible. The two command line options 737 <literal>--exclusive-threshold=<n></literal> and 738 <literal>--shared-threshold=<n></literal> make it possible to 739 detect excessive lock contention by making DRD report any lock that 740 has been held longer than the specified threshold. An example: 741 </para> 742 <programlisting><![CDATA[ 743 $ valgrind --tool=drd --exclusive-threshold=10 drd/tests/hold_lock -i 500 744 ... 745 ==10668== Acquired at: 746 ==10668== at 0x4C267C8: pthread_mutex_lock (drd_pthread_intercepts.c:395) 747 ==10668== by 0x400D92: main (hold_lock.c:51) 748 ==10668== Lock on mutex 0x7fefffd50 was held during 503 ms (threshold: 10 ms). 749 ==10668== at 0x4C26ADA: pthread_mutex_unlock (drd_pthread_intercepts.c:441) 750 ==10668== by 0x400DB5: main (hold_lock.c:55) 751 ... 752 ]]></programlisting> 753 754 <para> 755 The <literal>hold_lock</literal> test program holds a lock as long as 756 specified by the <literal>-i</literal> (interval) argument. The DRD 757 output reports that the lock acquired at line 51 in source file 758 <literal>hold_lock.c</literal> and released at line 55 was held during 759 503 ms, while a threshold of 10 ms was specified to DRD. 760 </para> 761 762 </sect2> 763 764 765 <sect2 id="drd-manual.api-checks" xreflabel="API Checks"> 766 <title>Detected Errors: Misuse of the POSIX threads API</title> 767 768 <para> 769 DRD is able to detect and report the following misuses of the POSIX 770 threads API: 771 <itemizedlist> 772 <listitem> 773 <para> 774 Passing the address of one type of synchronization object 775 (e.g. a mutex) to a POSIX API call that expects a pointer to 776 another type of synchronization object (e.g. a condition 777 variable). 778 </para> 779 </listitem> 780 <listitem> 781 <para> 782 Attempts to unlock a mutex that has not been locked. 783 </para> 784 </listitem> 785 <listitem> 786 <para> 787 Attempts to unlock a mutex that was locked by another thread. 788 </para> 789 </listitem> 790 <listitem> 791 <para> 792 Attempts to lock a mutex of type 793 <literal>PTHREAD_MUTEX_NORMAL</literal> or a spinlock 794 recursively. 795 </para> 796 </listitem> 797 <listitem> 798 <para> 799 Destruction or deallocation of a locked mutex. 800 </para> 801 </listitem> 802 <listitem> 803 <para> 804 Sending a signal to a condition variable while no lock is held 805 on the mutex associated with the condition variable. 806 </para> 807 </listitem> 808 <listitem> 809 <para> 810 Calling <function>pthread_cond_wait</function> on a mutex 811 that is not locked, that is locked by another thread or that 812 has been locked recursively. 813 </para> 814 </listitem> 815 <listitem> 816 <para> 817 Associating two different mutexes with a condition variable 818 through <function>pthread_cond_wait</function>. 819 </para> 820 </listitem> 821 <listitem> 822 <para> 823 Destruction or deallocation of a condition variable that is 824 being waited upon. 825 </para> 826 </listitem> 827 <listitem> 828 <para> 829 Destruction or deallocation of a locked reader-writer synchronization 830 object. 831 </para> 832 </listitem> 833 <listitem> 834 <para> 835 Attempts to unlock a reader-writer synchronization object that was not 836 locked by the calling thread. 837 </para> 838 </listitem> 839 <listitem> 840 <para> 841 Attempts to recursively lock a reader-writer synchronization object 842 exclusively. 843 </para> 844 </listitem> 845 <listitem> 846 <para> 847 Attempts to pass the address of a user-defined reader-writer 848 synchronization object to a POSIX threads function. 849 </para> 850 </listitem> 851 <listitem> 852 <para> 853 Attempts to pass the address of a POSIX reader-writer synchronization 854 object to one of the annotations for user-defined reader-writer 855 synchronization objects. 856 </para> 857 </listitem> 858 <listitem> 859 <para> 860 Reinitialization of a mutex, condition variable, reader-writer 861 lock, semaphore or barrier. 862 </para> 863 </listitem> 864 <listitem> 865 <para> 866 Destruction or deallocation of a semaphore or barrier that is 867 being waited upon. 868 </para> 869 </listitem> 870 <listitem> 871 <para> 872 Missing synchronization between barrier wait and barrier destruction. 873 </para> 874 </listitem> 875 <listitem> 876 <para> 877 Exiting a thread without first unlocking the spinlocks, mutexes or 878 reader-writer synchronization objects that were locked by that thread. 879 </para> 880 </listitem> 881 <listitem> 882 <para> 883 Passing an invalid thread ID to <function>pthread_join</function> 884 or <function>pthread_cancel</function>. 885 </para> 886 </listitem> 887 </itemizedlist> 888 </para> 889 890 </sect2> 891 892 893 <sect2 id="drd-manual.clientreqs" xreflabel="Client requests"> 894 <title>Client Requests</title> 895 896 <para> 897 Just as for other Valgrind tools it is possible to let a client program 898 interact with the DRD tool through client requests. In addition to the 899 client requests several macros have been defined that allow to use the 900 client requests in a convenient way. 901 </para> 902 903 <para> 904 The interface between client programs and the DRD tool is defined in 905 the header file <literal><valgrind/drd.h></literal>. The 906 available macros and client requests are: 907 <itemizedlist> 908 <listitem> 909 <para> 910 The macro <literal>DRD_GET_VALGRIND_THREADID</literal> and the 911 corresponding client 912 request <varname>VG_USERREQ__DRD_GET_VALGRIND_THREAD_ID</varname>. 913 Query the thread ID that has been assigned by the Valgrind core to the 914 thread executing this client request. Valgrind's thread ID's start at 915 one and are recycled in case a thread stops. 916 </para> 917 </listitem> 918 <listitem> 919 <para> 920 The macro <literal>DRD_GET_DRD_THREADID</literal> and the corresponding 921 client request <varname>VG_USERREQ__DRD_GET_DRD_THREAD_ID</varname>. 922 Query the thread ID that has been assigned by DRD to the thread 923 executing this client request. These are the thread ID's reported by DRD 924 in data race reports and in trace messages. DRD's thread ID's start at 925 one and are never recycled. 926 </para> 927 </listitem> 928 <listitem> 929 <para> 930 The macros <literal>DRD_IGNORE_VAR(x)</literal>, 931 <literal>ANNOTATE_TRACE_MEMORY(&x)</literal> and the corresponding 932 client request <varname>VG_USERREQ__DRD_START_SUPPRESSION</varname>. Some 933 applications contain intentional races. There exist e.g. applications 934 where the same value is assigned to a shared variable from two different 935 threads. It may be more convenient to suppress such races than to solve 936 these. This client request allows to suppress such races. 937 </para> 938 </listitem> 939 <listitem> 940 <para> 941 The macro <literal>DRD_STOP_IGNORING_VAR(x)</literal> and the 942 corresponding client request 943 <varname>VG_USERREQ__DRD_FINISH_SUPPRESSION</varname>. Tell DRD 944 to no longer ignore data races for the address range that was suppressed 945 either via the macro <literal>DRD_IGNORE_VAR(x)</literal> or via the 946 client request <varname>VG_USERREQ__DRD_START_SUPPRESSION</varname>. 947 </para> 948 </listitem> 949 <listitem> 950 <para> 951 The macro <literal>DRD_TRACE_VAR(x)</literal>. Trace all load and store 952 activity for the address range starting at <literal>&x</literal> and 953 occupying <literal>sizeof(x)</literal> bytes. When DRD reports a data 954 race on a specified variable, and it's not immediately clear which 955 source code statements triggered the conflicting accesses, it can be 956 very helpful to trace all activity on the offending memory location. 957 </para> 958 </listitem> 959 <listitem> 960 <para> 961 The macro <literal>ANNOTATE_TRACE_MEMORY(&x)</literal>. Trace all 962 load and store activity that touches at least the single byte at the 963 address <literal>&x</literal>. 964 </para> 965 </listitem> 966 <listitem> 967 <para> 968 The client request <varname>VG_USERREQ__DRD_START_TRACE_ADDR</varname>, 969 which allows to trace all load and store activity for the specified 970 address range. 971 </para> 972 </listitem> 973 <listitem> 974 <para> 975 The client 976 request <varname>VG_USERREQ__DRD_STOP_TRACE_ADDR</varname>. Do no longer 977 trace load and store activity for the specified address range. 978 </para> 979 </listitem> 980 <listitem> 981 <para> 982 The macro <literal>ANNOTATE_HAPPENS_BEFORE(addr)</literal> tells DRD to 983 insert a mark. Insert this macro just after an access to the variable at 984 the specified address has been performed. 985 </para> 986 </listitem> 987 <listitem> 988 <para> 989 The macro <literal>ANNOTATE_HAPPENS_AFTER(addr)</literal> tells DRD that 990 the next access to the variable at the specified address should be 991 considered to have happened after the access just before the latest 992 <literal>ANNOTATE_HAPPENS_BEFORE(addr)</literal> annotation that 993 references the same variable. The purpose of these two macros is to tell 994 DRD about the order of inter-thread memory accesses implemented via 995 atomic memory operations. See 996 also <literal>drd/tests/annotate_smart_pointer.cpp</literal> for an 997 example. 998 </para> 999 </listitem> 1000 <listitem> 1001 <para> 1002 The macro <literal>ANNOTATE_RWLOCK_CREATE(rwlock)</literal> tells DRD 1003 that the object at address <literal>rwlock</literal> is a 1004 reader-writer synchronization object that is not a 1005 <literal>pthread_rwlock_t</literal> synchronization object. See 1006 also <literal>drd/tests/annotate_rwlock.c</literal> for an example. 1007 </para> 1008 </listitem> 1009 <listitem> 1010 <para> 1011 The macro <literal>ANNOTATE_RWLOCK_DESTROY(rwlock)</literal> tells DRD 1012 that the reader-writer synchronization object at 1013 address <literal>rwlock</literal> has been destroyed. 1014 </para> 1015 </listitem> 1016 <listitem> 1017 <para> 1018 The macro <literal>ANNOTATE_WRITERLOCK_ACQUIRED(rwlock)</literal> tells 1019 DRD that a writer lock has been acquired on the reader-writer 1020 synchronization object at address <literal>rwlock</literal>. 1021 </para> 1022 </listitem> 1023 <listitem> 1024 <para> 1025 The macro <literal>ANNOTATE_READERLOCK_ACQUIRED(rwlock)</literal> tells 1026 DRD that a reader lock has been acquired on the reader-writer 1027 synchronization object at address <literal>rwlock</literal>. 1028 </para> 1029 </listitem> 1030 <listitem> 1031 <para> 1032 The macro <literal>ANNOTATE_RWLOCK_ACQUIRED(rwlock, is_w)</literal> 1033 tells DRD that a writer lock (when <literal>is_w != 0</literal>) or that 1034 a reader lock (when <literal>is_w == 0</literal>) has been acquired on 1035 the reader-writer synchronization object at 1036 address <literal>rwlock</literal>. 1037 </para> 1038 </listitem> 1039 <listitem> 1040 <para> 1041 The macro <literal>ANNOTATE_WRITERLOCK_RELEASED(rwlock)</literal> tells 1042 DRD that a writer lock has been released on the reader-writer 1043 synchronization object at address <literal>rwlock</literal>. 1044 </para> 1045 </listitem> 1046 <listitem> 1047 <para> 1048 The macro <literal>ANNOTATE_READERLOCK_RELEASED(rwlock)</literal> tells 1049 DRD that a reader lock has been released on the reader-writer 1050 synchronization object at address <literal>rwlock</literal>. 1051 </para> 1052 </listitem> 1053 <listitem> 1054 <para> 1055 The macro <literal>ANNOTATE_RWLOCK_RELEASED(rwlock, is_w)</literal> 1056 tells DRD that a writer lock (when <literal>is_w != 0</literal>) or that 1057 a reader lock (when <literal>is_w == 0</literal>) has been released on 1058 the reader-writer synchronization object at 1059 address <literal>rwlock</literal>. 1060 </para> 1061 </listitem> 1062 <listitem> 1063 <para> 1064 The macro <literal>ANNOTATE_BARRIER_INIT(barrier, count, 1065 reinitialization_allowed)</literal> tells DRD that a new barrier object 1066 at the address <literal>barrier</literal> has been initialized, 1067 that <literal>count</literal> threads participate in each barrier and 1068 also whether or not barrier reinitialization without intervening 1069 destruction should be reported as an error. See 1070 also <literal>drd/tests/annotate_barrier.c</literal> for an example. 1071 </para> 1072 </listitem> 1073 <listitem> 1074 <para> 1075 The macro <literal>ANNOTATE_BARRIER_DESTROY(barrier)</literal> 1076 tells DRD that a barrier object is about to be destroyed. 1077 </para> 1078 </listitem> 1079 <listitem> 1080 <para> 1081 The macro <literal>ANNOTATE_BARRIER_WAIT_BEFORE(barrier)</literal> 1082 tells DRD that waiting for a barrier will start. 1083 </para> 1084 </listitem> 1085 <listitem> 1086 <para> 1087 The macro <literal>ANNOTATE_BARRIER_WAIT_AFTER(barrier)</literal> 1088 tells DRD that waiting for a barrier has finished. 1089 </para> 1090 </listitem> 1091 <listitem> 1092 <para> 1093 The macro <literal>ANNOTATE_BENIGN_RACE_SIZED(addr, size, 1094 descr)</literal> tells DRD that any races detected on the specified 1095 address are benign and hence should not be 1096 reported. The <literal>descr</literal> argument is ignored but can be 1097 used to document why data races on <literal>addr</literal> are benign. 1098 </para> 1099 </listitem> 1100 <listitem> 1101 <para> 1102 The macro <literal>ANNOTATE_BENIGN_RACE_STATIC(var, descr)</literal> 1103 tells DRD that any races detected on the specified static variable are 1104 benign and hence should not be reported. The <literal>descr</literal> 1105 argument is ignored but can be used to document why data races 1106 on <literal>var</literal> are benign. Note: this macro can only be 1107 used in C++ programs and not in C programs. 1108 </para> 1109 </listitem> 1110 <listitem> 1111 <para> 1112 The macro <literal>ANNOTATE_IGNORE_READS_BEGIN</literal> tells 1113 DRD to ignore all memory loads performed by the current thread. 1114 </para> 1115 </listitem> 1116 <listitem> 1117 <para> 1118 The macro <literal>ANNOTATE_IGNORE_READS_END</literal> tells 1119 DRD to stop ignoring the memory loads performed by the current thread. 1120 </para> 1121 </listitem> 1122 <listitem> 1123 <para> 1124 The macro <literal>ANNOTATE_IGNORE_WRITES_BEGIN</literal> tells 1125 DRD to ignore all memory stores performed by the current thread. 1126 </para> 1127 </listitem> 1128 <listitem> 1129 <para> 1130 The macro <literal>ANNOTATE_IGNORE_WRITES_END</literal> tells 1131 DRD to stop ignoring the memory stores performed by the current thread. 1132 </para> 1133 </listitem> 1134 <listitem> 1135 <para> 1136 The macro <literal>ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN</literal> tells 1137 DRD to ignore all memory accesses performed by the current thread. 1138 </para> 1139 </listitem> 1140 <listitem> 1141 <para> 1142 The macro <literal>ANNOTATE_IGNORE_READS_AND_WRITES_END</literal> tells 1143 DRD to stop ignoring the memory accesses performed by the current thread. 1144 </para> 1145 </listitem> 1146 <listitem> 1147 <para> 1148 The macro <literal>ANNOTATE_NEW_MEMORY(addr, size)</literal> tells 1149 DRD that the specified memory range has been allocated by a custom 1150 memory allocator in the client program and that the client program 1151 will start using this memory range. 1152 </para> 1153 </listitem> 1154 <listitem> 1155 <para> 1156 The macro <literal>ANNOTATE_THREAD_NAME(name)</literal> tells DRD to 1157 associate the specified name with the current thread and to include this 1158 name in the error messages printed by DRD. 1159 </para> 1160 </listitem> 1161 <listitem> 1162 <para> 1163 The macros <literal>VALGRIND_MALLOCLIKE_BLOCK</literal> and 1164 <literal>VALGRIND_FREELIKE_BLOCK</literal> from the Valgrind core are 1165 implemented; they are described in 1166 <xref linkend="manual-core-adv.clientreq"/>. 1167 </para> 1168 </listitem> 1169 </itemizedlist> 1170 </para> 1171 1172 <para> 1173 Note: if you compiled Valgrind yourself, the header file 1174 <literal><valgrind/drd.h></literal> will have been installed in 1175 the directory <literal>/usr/include</literal> by the command 1176 <literal>make install</literal>. If you obtained Valgrind by 1177 installing it as a package however, you will probably have to install 1178 another package with a name like <literal>valgrind-devel</literal> 1179 before Valgrind's header files are available. 1180 </para> 1181 1182 </sect2> 1183 1184 1185 <sect2 id="drd-manual.gnome" xreflabel="GNOME"> 1186 <title>Debugging GNOME Programs</title> 1187 1188 <para> 1189 GNOME applications use the threading primitives provided by the 1190 <computeroutput>glib</computeroutput> and 1191 <computeroutput>gthread</computeroutput> libraries. These libraries 1192 are built on top of POSIX threads, and hence are directly supported by 1193 DRD. Please keep in mind that you have to call 1194 <function>g_thread_init</function> before creating any threads, or 1195 DRD will report several data races on glib functions. See also the 1196 <ulink 1197 url="http://library.gnome.org/devel/glib/stable/glib-Threads.html">GLib 1198 Reference Manual</ulink> for more information about 1199 <function>g_thread_init</function>. 1200 </para> 1201 1202 <para> 1203 One of the many facilities provided by the <literal>glib</literal> 1204 library is a block allocator, called <literal>g_slice</literal>. You 1205 have to disable this block allocator when using DRD by adding the 1206 following to the shell environment variables: 1207 <literal>G_SLICE=always-malloc</literal>. See also the <ulink 1208 url="http://library.gnome.org/devel/glib/stable/glib-Memory-Slices.html">GLib 1209 Reference Manual</ulink> for more information. 1210 </para> 1211 1212 </sect2> 1213 1214 1215 <sect2 id="drd-manual.qt" xreflabel="Qt"> 1216 <title>Debugging Qt Programs</title> 1217 1218 <para> 1219 The Qt library is the GUI library used by the KDE project. Currently 1220 there are two versions of the Qt library in use: Qt3 by KDE 3 and Qt4 1221 by KDE 4. If possible, use Qt4 instead of Qt3. Qt3 is no longer 1222 supported, and there are known problems with multithreading support in 1223 Qt3. As an example, using QString objects in more than one thread will 1224 trigger race reports (this has been confirmed by Trolltech -- see also 1225 Trolltech task <ulink 1226 url="http://trolltech.com/developer/task-tracker/index_html">#206152</ulink>). 1227 </para> 1228 1229 <para> 1230 Qt4 applications are supported by DRD, but only if the 1231 <literal>libqt4-debuginfo</literal> package has been installed. Some 1232 of the synchronization and threading primitives in Qt4 bypass the 1233 POSIX threads library, and DRD can only intercept these if symbol 1234 information for the Qt4 library is available. DRD won't tell you if it 1235 has not been able to load the Qt4 debug information, but a huge number 1236 of data races will be reported on data protected via 1237 <literal>QMutex</literal> objects. 1238 </para> 1239 1240 </sect2> 1241 1242 1243 <sect2 id="drd-manual.boost.thread" xreflabel="Boost.Thread"> 1244 <title>Debugging Boost.Thread Programs</title> 1245 1246 <para> 1247 The Boost.Thread library is the threading library included with the 1248 cross-platform Boost Libraries. This threading library is an early 1249 implementation of the upcoming C++0x threading library. 1250 </para> 1251 1252 <para> 1253 Applications that use the Boost.Thread library should run fine under DRD. 1254 </para> 1255 1256 <para> 1257 More information about Boost.Thread can be found here: 1258 <itemizedlist> 1259 <listitem> 1260 <para> 1261 Anthony Williams, <ulink 1262 url="http://www.boost.org/doc/libs/1_37_0/doc/html/thread.html">Boost.Thread</ulink> 1263 Library Documentation, Boost website, 2007. 1264 </para> 1265 </listitem> 1266 <listitem> 1267 <para> 1268 Anthony Williams, <ulink 1269 url="http://www.ddj.com/cpp/211600441">What's New in Boost 1270 Threads?</ulink>, Recent changes to the Boost Thread library, 1271 Dr. Dobbs Magazine, October 2008. 1272 </para> 1273 </listitem> 1274 </itemizedlist> 1275 </para> 1276 1277 </sect2> 1278 1279 1280 <sect2 id="drd-manual.openmp" xreflabel="OpenMP"> 1281 <title>Debugging OpenMP Programs</title> 1282 1283 <para> 1284 OpenMP stands for <emphasis>Open Multi-Processing</emphasis>. The OpenMP 1285 standard consists of a set of compiler directives for C, C++ and Fortran 1286 programs that allows a compiler to transform a sequential program into a 1287 parallel program. OpenMP is well suited for HPC applications and allows to 1288 work at a higher level compared to direct use of the POSIX threads API. While 1289 OpenMP ensures that the POSIX API is used correctly, OpenMP programs can still 1290 contain data races. So it definitely makes sense to verify OpenMP programs 1291 with a thread checking tool. 1292 </para> 1293 1294 <para> 1295 DRD supports OpenMP shared-memory programs generated by GCC. GCC 1296 supports OpenMP since version 4.2.0. GCC's runtime support 1297 for OpenMP programs is provided by a library called 1298 <literal>libgomp</literal>. The synchronization primitives implemented 1299 in this library use Linux' futex system call directly, unless the 1300 library has been configured with the 1301 <literal>--disable-linux-futex</literal> option. DRD only supports 1302 libgomp libraries that have been configured with this option and in 1303 which symbol information is present. For most Linux distributions this 1304 means that you will have to recompile GCC. See also the script 1305 <literal>drd/scripts/download-and-build-gcc</literal> in the 1306 Valgrind source tree for an example of how to compile GCC. You will 1307 also have to make sure that the newly compiled 1308 <literal>libgomp.so</literal> library is loaded when OpenMP programs 1309 are started. This is possible by adding a line similar to the 1310 following to your shell startup script: 1311 </para> 1312 <programlisting><![CDATA[ 1313 export LD_LIBRARY_PATH=~/gcc-4.4.0/lib64:~/gcc-4.4.0/lib: 1314 ]]></programlisting> 1315 1316 <para> 1317 As an example, the test OpenMP test program 1318 <literal>drd/tests/omp_matinv</literal> triggers a data race 1319 when the option -r has been specified on the command line. The data 1320 race is triggered by the following code: 1321 </para> 1322 <programlisting><![CDATA[ 1323 #pragma omp parallel for private(j) 1324 for (j = 0; j < rows; j++) 1325 { 1326 if (i != j) 1327 { 1328 const elem_t factor = a[j * cols + i]; 1329 for (k = 0; k < cols; k++) 1330 { 1331 a[j * cols + k] -= a[i * cols + k] * factor; 1332 } 1333 } 1334 } 1335 ]]></programlisting> 1336 1337 <para> 1338 The above code is racy because the variable <literal>k</literal> has 1339 not been declared private. DRD will print the following error message 1340 for the above code: 1341 </para> 1342 <programlisting><![CDATA[ 1343 $ valgrind --tool=drd --check-stack-var=yes --read-var-info=yes drd/tests/omp_matinv 3 -t 2 -r 1344 ... 1345 Conflicting store by thread 1/1 at 0x7fefffbc4 size 4 1346 at 0x4014A0: gj.omp_fn.0 (omp_matinv.c:203) 1347 by 0x401211: gj (omp_matinv.c:159) 1348 by 0x40166A: invert_matrix (omp_matinv.c:238) 1349 by 0x4019B4: main (omp_matinv.c:316) 1350 Location 0x7fefffbc4 is 0 bytes inside local var "k" 1351 declared at omp_matinv.c:160, in frame #0 of thread 1 1352 ... 1353 ]]></programlisting> 1354 <para> 1355 In the above output the function name <function>gj.omp_fn.0</function> 1356 has been generated by GCC from the function name 1357 <function>gj</function>. The allocation context information shows that the 1358 data race has been caused by modifying the variable <literal>k</literal>. 1359 </para> 1360 1361 <para> 1362 Note: for GCC versions before 4.4.0, no allocation context information is 1363 shown. With these GCC versions the most usable information in the above output 1364 is the source file name and the line number where the data race has been 1365 detected (<literal>omp_matinv.c:203</literal>). 1366 </para> 1367 1368 <para> 1369 For more information about OpenMP, see also 1370 <ulink url="http://openmp.org/">openmp.org</ulink>. 1371 </para> 1372 1373 </sect2> 1374 1375 1376 <sect2 id="drd-manual.cust-mem-alloc" xreflabel="Custom Memory Allocators"> 1377 <title>DRD and Custom Memory Allocators</title> 1378 1379 <para> 1380 DRD tracks all memory allocation events that happen via the 1381 standard memory allocation and deallocation functions 1382 (<function>malloc</function>, <function>free</function>, 1383 <function>new</function> and <function>delete</function>), via entry 1384 and exit of stack frames or that have been annotated with Valgrind's 1385 memory pool client requests. DRD uses memory allocation and deallocation 1386 information for two purposes: 1387 <itemizedlist> 1388 <listitem> 1389 <para> 1390 To know where the scope ends of POSIX objects that have not been 1391 destroyed explicitly. It is e.g. not required by the POSIX 1392 threads standard to call 1393 <function>pthread_mutex_destroy</function> before freeing the 1394 memory in which a mutex object resides. 1395 </para> 1396 </listitem> 1397 <listitem> 1398 <para> 1399 To know where the scope of variables ends. If e.g. heap memory 1400 has been used by one thread, that thread frees that memory, and 1401 another thread allocates and starts using that memory, no data 1402 races must be reported for that memory. 1403 </para> 1404 </listitem> 1405 </itemizedlist> 1406 </para> 1407 1408 <para> 1409 It is essential for correct operation of DRD that the tool knows about 1410 memory allocation and deallocation events. When analyzing a client program 1411 with DRD that uses a custom memory allocator, either instrument the custom 1412 memory allocator with the <literal>VALGRIND_MALLOCLIKE_BLOCK</literal> 1413 and <literal>VALGRIND_FREELIKE_BLOCK</literal> macros or disable the 1414 custom memory allocator. 1415 </para> 1416 1417 <para> 1418 As an example, the GNU libstdc++ library can be configured 1419 to use standard memory allocation functions instead of memory pools by 1420 setting the environment variable 1421 <literal>GLIBCXX_FORCE_NEW</literal>. For more information, see also 1422 the <ulink 1423 url="http://gcc.gnu.org/onlinedocs/libstdc++/manual/bk01pt04ch11.html">libstdc++ 1424 manual</ulink>. 1425 </para> 1426 1427 </sect2> 1428 1429 1430 <sect2 id="drd-manual.drd-versus-memcheck" xreflabel="DRD Versus Memcheck"> 1431 <title>DRD Versus Memcheck</title> 1432 1433 <para> 1434 It is essential for correct operation of DRD that there are no memory 1435 errors such as dangling pointers in the client program. Which means that 1436 it is a good idea to make sure that your program is Memcheck-clean 1437 before you analyze it with DRD. It is possible however that some of 1438 the Memcheck reports are caused by data races. In this case it makes 1439 sense to run DRD before Memcheck. 1440 </para> 1441 1442 <para> 1443 So which tool should be run first? In case both DRD and Memcheck 1444 complain about a program, a possible approach is to run both tools 1445 alternatingly and to fix as many errors as possible after each run of 1446 each tool until none of the two tools prints any more error messages. 1447 </para> 1448 1449 </sect2> 1450 1451 1452 <sect2 id="drd-manual.resource-requirements" xreflabel="Resource Requirements"> 1453 <title>Resource Requirements</title> 1454 1455 <para> 1456 The requirements of DRD with regard to heap and stack memory and the 1457 effect on the execution time of client programs are as follows: 1458 <itemizedlist> 1459 <listitem> 1460 <para> 1461 When running a program under DRD with default DRD options, 1462 between 1.1 and 3.6 times more memory will be needed compared to 1463 a native run of the client program. More memory will be needed 1464 if loading debug information has been enabled 1465 (<literal>--read-var-info=yes</literal>). 1466 </para> 1467 </listitem> 1468 <listitem> 1469 <para> 1470 DRD allocates some of its temporary data structures on the stack 1471 of the client program threads. This amount of data is limited to 1472 1 - 2 KB. Make sure that thread stacks are sufficiently large. 1473 </para> 1474 </listitem> 1475 <listitem> 1476 <para> 1477 Most applications will run between 20 and 50 times slower under 1478 DRD than a native single-threaded run. The slowdown will be most 1479 noticeable for applications which perform frequent mutex lock / 1480 unlock operations. 1481 </para> 1482 </listitem> 1483 </itemizedlist> 1484 </para> 1485 1486 </sect2> 1487 1488 1489 <sect2 id="drd-manual.effective-use" xreflabel="Effective Use"> 1490 <title>Hints and Tips for Effective Use of DRD</title> 1491 1492 <para> 1493 The following information may be helpful when using DRD: 1494 <itemizedlist> 1495 <listitem> 1496 <para> 1497 Make sure that debug information is present in the executable 1498 being analyzed, such that DRD can print function name and line 1499 number information in stack traces. Most compilers can be told 1500 to include debug information via compiler option 1501 <option>-g</option>. 1502 </para> 1503 </listitem> 1504 <listitem> 1505 <para> 1506 Compile with option <option>-O1</option> instead of 1507 <option>-O0</option>. This will reduce the amount of generated 1508 code, may reduce the amount of debug info and will speed up 1509 DRD's processing of the client program. For more information, 1510 see also <xref linkend="manual-core.started"/>. 1511 </para> 1512 </listitem> 1513 <listitem> 1514 <para> 1515 If DRD reports any errors on libraries that are part of your 1516 Linux distribution like e.g. <literal>libc.so</literal> or 1517 <literal>libstdc++.so</literal>, installing the debug packages 1518 for these libraries will make the output of DRD a lot more 1519 detailed. 1520 </para> 1521 </listitem> 1522 <listitem> 1523 <para> 1524 When using C++, do not send output from more than one thread to 1525 <literal>std::cout</literal>. Doing so would not only 1526 generate multiple data race reports, it could also result in 1527 output from several threads getting mixed up. Either use 1528 <function>printf</function> or do the following: 1529 <orderedlist> 1530 <listitem> 1531 <para>Derive a class from <literal>std::ostreambuf</literal> 1532 and let that class send output line by line to 1533 <literal>stdout</literal>. This will avoid that individual 1534 lines of text produced by different threads get mixed 1535 up.</para> 1536 </listitem> 1537 <listitem> 1538 <para>Create one instance of <literal>std::ostream</literal> 1539 for each thread. This makes stream formatting settings 1540 thread-local. Pass a per-thread instance of the class 1541 derived from <literal>std::ostreambuf</literal> to the 1542 constructor of each instance. </para> 1543 </listitem> 1544 <listitem> 1545 <para>Let each thread send its output to its own instance of 1546 <literal>std::ostream</literal> instead of 1547 <literal>std::cout</literal>.</para> 1548 </listitem> 1549 </orderedlist> 1550 </para> 1551 </listitem> 1552 </itemizedlist> 1553 </para> 1554 1555 </sect2> 1556 1557 1558 </sect1> 1559 1560 1561 <sect1 id="drd-manual.Pthreads" xreflabel="Pthreads"> 1562 <title>Using the POSIX Threads API Effectively</title> 1563 1564 <sect2 id="drd-manual.mutex-types" xreflabel="mutex-types"> 1565 <title>Mutex types</title> 1566 1567 <para> 1568 The Single UNIX Specification version two defines the following four 1569 mutex types (see also the documentation of <ulink 1570 url="http://www.opengroup.org/onlinepubs/007908799/xsh/pthread_mutexattr_settype.html"><function>pthread_mutexattr_settype</function></ulink>): 1571 <itemizedlist> 1572 <listitem> 1573 <para> 1574 <emphasis>normal</emphasis>, which means that no error checking 1575 is performed, and that the mutex is non-recursive. 1576 </para> 1577 </listitem> 1578 <listitem> 1579 <para> 1580 <emphasis>error checking</emphasis>, which means that the mutex 1581 is non-recursive and that error checking is performed. 1582 </para> 1583 </listitem> 1584 <listitem> 1585 <para> 1586 <emphasis>recursive</emphasis>, which means that a mutex may be 1587 locked recursively. 1588 </para> 1589 </listitem> 1590 <listitem> 1591 <para> 1592 <emphasis>default</emphasis>, which means that error checking 1593 behavior is undefined, and that the behavior for recursive 1594 locking is also undefined. Or: portable code must neither 1595 trigger error conditions through the Pthreads API nor attempt to 1596 lock a mutex of default type recursively. 1597 </para> 1598 </listitem> 1599 </itemizedlist> 1600 </para> 1601 1602 <para> 1603 In complex applications it is not always clear from beforehand which 1604 mutex will be locked recursively and which mutex will not be locked 1605 recursively. Attempts lock a non-recursive mutex recursively will 1606 result in race conditions that are very hard to find without a thread 1607 checking tool. So either use the error checking mutex type and 1608 consistently check the return value of Pthread API mutex calls, or use 1609 the recursive mutex type. 1610 </para> 1611 1612 </sect2> 1613 1614 <sect2 id="drd-manual.condvar" xreflabel="condition-variables"> 1615 <title>Condition variables</title> 1616 1617 <para> 1618 A condition variable allows one thread to wake up one or more other 1619 threads. Condition variables are often used to notify one or more 1620 threads about state changes of shared data. Unfortunately it is very 1621 easy to introduce race conditions by using condition variables as the 1622 only means of state information propagation. A better approach is to 1623 let threads poll for changes of a state variable that is protected by 1624 a mutex, and to use condition variables only as a thread wakeup 1625 mechanism. See also the source file 1626 <computeroutput>drd/tests/monitor_example.cpp</computeroutput> for an 1627 example of how to implement this concept in C++. The monitor concept 1628 used in this example is a well known and very useful concept -- see 1629 also Wikipedia for more information about the <ulink 1630 url="http://en.wikipedia.org/wiki/Monitor_(synchronization)">monitor</ulink> 1631 concept. 1632 </para> 1633 1634 </sect2> 1635 1636 <sect2 id="drd-manual.pctw" xreflabel="pthread_cond_timedwait"> 1637 <title>pthread_cond_timedwait and timeouts</title> 1638 1639 <para> 1640 Historically the function 1641 <function>pthread_cond_timedwait</function> only allowed the 1642 specification of an absolute timeout, that is a timeout independent of 1643 the time when this function was called. However, almost every call to 1644 this function expresses a relative timeout. This typically happens by 1645 passing the sum of 1646 <computeroutput>clock_gettime(CLOCK_REALTIME)</computeroutput> and a 1647 relative timeout as the third argument. This approach is incorrect 1648 since forward or backward clock adjustments by e.g. ntpd will affect 1649 the timeout. A more reliable approach is as follows: 1650 <itemizedlist> 1651 <listitem> 1652 <para> 1653 When initializing a condition variable through 1654 <function>pthread_cond_init</function>, specify that the timeout of 1655 <function>pthread_cond_timedwait</function> will use the clock 1656 <literal>CLOCK_MONOTONIC</literal> instead of 1657 <literal>CLOCK_REALTIME</literal>. You can do this via 1658 <computeroutput>pthread_condattr_setclock(..., 1659 CLOCK_MONOTONIC)</computeroutput>. 1660 </para> 1661 </listitem> 1662 <listitem> 1663 <para> 1664 When calling <function>pthread_cond_timedwait</function>, pass 1665 the sum of 1666 <computeroutput>clock_gettime(CLOCK_MONOTONIC)</computeroutput> 1667 and a relative timeout as the third argument. 1668 </para> 1669 </listitem> 1670 </itemizedlist> 1671 See also 1672 <computeroutput>drd/tests/monitor_example.cpp</computeroutput> for an 1673 example. 1674 </para> 1675 1676 </sect2> 1677 1678 </sect1> 1679 1680 1681 <sect1 id="drd-manual.limitations" xreflabel="Limitations"> 1682 <title>Limitations</title> 1683 1684 <para>DRD currently has the following limitations:</para> 1685 1686 <itemizedlist> 1687 <listitem> 1688 <para> 1689 DRD, just like Memcheck, will refuse to start on Linux 1690 distributions where all symbol information has been removed from 1691 <filename>ld.so</filename>. This is e.g. the case for the PPC editions 1692 of openSUSE and Gentoo. You will have to install the glibc debuginfo 1693 package on these platforms before you can use DRD. See also openSUSE 1694 bug <ulink url="http://bugzilla.novell.com/show_bug.cgi?id=396197"> 1695 396197</ulink> and Gentoo bug <ulink 1696 url="http://bugs.gentoo.org/214065">214065</ulink>. 1697 </para> 1698 </listitem> 1699 <listitem> 1700 <para> 1701 With gcc 4.4.3 and before, DRD may report data races on the C++ 1702 class <literal>std::string</literal> in a multithreaded program. This is 1703 a know <literal>libstdc++</literal> issue -- see also GCC bug 1704 <ulink url="http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40518">40518</ulink> 1705 for more information. 1706 </para> 1707 </listitem> 1708 <listitem> 1709 <para> 1710 When address tracing is enabled, no information on atomic stores 1711 will be displayed. 1712 </para> 1713 </listitem> 1714 <listitem> 1715 <para> 1716 If you compile the DRD source code yourself, you need GCC 3.0 or 1717 later. GCC 2.95 is not supported. 1718 </para> 1719 </listitem> 1720 <listitem> 1721 <para> 1722 Of the two POSIX threads implementations for Linux, only the 1723 NPTL (Native POSIX Thread Library) is supported. The older 1724 LinuxThreads library is not supported. 1725 </para> 1726 </listitem> 1727 </itemizedlist> 1728 1729 </sect1> 1730 1731 1732 <sect1 id="drd-manual.feedback" xreflabel="Feedback"> 1733 <title>Feedback</title> 1734 1735 <para> 1736 If you have any comments, suggestions, feedback or bug reports about 1737 DRD, feel free to either post a message on the Valgrind users mailing 1738 list or to file a bug report. See also <ulink 1739 url="&vg-url;">&vg-url;</ulink> for more information. 1740 </para> 1741 1742 </sect1> 1743 1744 1745 </chapter> 1746
