1VALGRIND(1) valgrind VALGRIND(1)
2
6 valgrind - a suite of tools for debugging and profiling programs
7
9 valgrind [valgrind-options] [your-program] [your-program-options]
10
12 Valgrind is a flexible program for debugging and profiling Linux
13 executables. It consists of a core, which provides a synthetic CPU in
14 software, and a series of debugging and profiling tools. The
15 architecture is modular, so that new tools can be created easily and
16 without disturbing the existing structure.
17 Some of the options described below work with all Valgrind tools, and
19 some only work with a few or one. The section MEMCHECK OPTIONS and
20 those below it describe tool-specific options.
21 This manual page covers only basic usage and options. For more
23 comprehensive information, please see the HTML documentation on your
24 system: $INSTALL/share/doc/valgrind/html/index.html, or online:
25 http://www.valgrind.org/docs/manual/index.html.
26
28 The single most important option.
29 --tool=<toolname> [default: memcheck]
31 Run the Valgrind tool called toolname, e.g. memcheck, cachegrind,
32 callgrind, helgrind, drd, massif, dhat, lackey, none, exp-bbv, etc.
33
35 These options work with all tools.
36 -h --help
38 Show help for all options, both for the core and for the selected
39 tool. If the option is repeated it is equivalent to giving
40 --help-debug.
41 --help-debug
43 Same as --help, but also lists debugging options which usually are
44 only of use to Valgrind's developers.
45 --version
47 Show the version number of the Valgrind core. Tools can have their
48 own version numbers. There is a scheme in place to ensure that
49 tools only execute when the core version is one they are known to
50 work with. This was done to minimise the chances of strange
51 problems arising from tool-vs-core version incompatibilities.
52 -q, --quiet
54 Run silently, and only print error messages. Useful if you are
55 running regression tests or have some other automated test
56 machinery.
57 -v, --verbose
59 Be more verbose. Gives extra information on various aspects of your
60 program, such as: the shared objects loaded, the suppressions used,
61 the progress of the instrumentation and execution engines, and
62 warnings about unusual behaviour. Repeating the option increases
63 the verbosity level.
64 --trace-children=<yes|no> [default: no]
66 When enabled, Valgrind will trace into sub-processes initiated via
67 the exec system call. This is necessary for multi-process programs.
68 Note that Valgrind does trace into the child of a fork (it would be
70 difficult not to, since fork makes an identical copy of a process),
71 so this option is arguably badly named. However, most children of
72 fork calls immediately call exec anyway.
73 --trace-children-skip=patt1,patt2,...
75 This option only has an effect when --trace-children=yes is
76 specified. It allows for some children to be skipped. The option
77 takes a comma separated list of patterns for the names of child
78 executables that Valgrind should not trace into. Patterns may
79 include the metacharacters ? and *, which have the usual meaning.
80 This can be useful for pruning uninteresting branches from a tree
82 of processes being run on Valgrind. But you should be careful when
83 using it. When Valgrind skips tracing into an executable, it
84 doesn't just skip tracing that executable, it also skips tracing
85 any of that executable's child processes. In other words, the flag
86 doesn't merely cause tracing to stop at the specified executables
87 -- it skips tracing of entire process subtrees rooted at any of the
88 specified executables.
89 --trace-children-skip-by-arg=patt1,patt2,...
91 This is the same as --trace-children-skip, with one difference: the
92 decision as to whether to trace into a child process is made by
93 examining the arguments to the child process, rather than the name
94 of its executable.
95 --child-silent-after-fork=<yes|no> [default: no]
97 When enabled, Valgrind will not show any debugging or logging
98 output for the child process resulting from a fork call. This can
99 make the output less confusing (although more misleading) when
100 dealing with processes that create children. It is particularly
101 useful in conjunction with --trace-children=. Use of this option is
102 also strongly recommended if you are requesting XML output
103 (--xml=yes), since otherwise the XML from child and parent may
104 become mixed up, which usually makes it useless.
105 --vgdb=<no|yes|full> [default: yes]
107 Valgrind will provide "gdbserver" functionality when --vgdb=yes or
108 --vgdb=full is specified. This allows an external GNU GDB debugger
109 to control and debug your program when it runs on Valgrind.
110 --vgdb=full incurs significant performance overheads, but provides
111 more precise breakpoints and watchpoints. See Debugging your
112 program using Valgrind's gdbserver and GDB for a detailed
113 description.
114 If the embedded gdbserver is enabled but no gdb is currently being
116 used, the vgdb command line utility can send "monitor commands" to
117 Valgrind from a shell. The Valgrind core provides a set of Valgrind
118 monitor commands. A tool can optionally provide tool specific
119 monitor commands, which are documented in the tool specific
120 chapter.
121 --vgdb-error=<number> [default: 999999999]
123 Use this option when the Valgrind gdbserver is enabled with
124 --vgdb=yes or --vgdb=full. Tools that report errors will wait for
125 "number" errors to be reported before freezing the program and
126 waiting for you to connect with GDB. It follows that a value of
127 zero will cause the gdbserver to be started before your program is
128 executed. This is typically used to insert GDB breakpoints before
129 execution, and also works with tools that do not report errors,
130 such as Massif.
131 --vgdb-stop-at=<set> [default: none]
133 Use this option when the Valgrind gdbserver is enabled with
134 --vgdb=yes or --vgdb=full. The Valgrind gdbserver will be invoked
135 for each error after --vgdb-error have been reported. You can
136 additionally ask the Valgrind gdbserver to be invoked for other
137 events, specified in one of the following ways:
138 • a comma separated list of one or more of startup exit abexit
140 valgrindabexit.
141 The values startup exit valgrindabexit respectively indicate to
143 invoke gdbserver before your program is executed, after the
144 last instruction of your program, on Valgrind abnormal exit
145 (e.g. internal error, out of memory, ...).
146 The option abexit is similar to exit but tells to invoke
148 gdbserver only when your application exits abnormally (i.e.
149 with an exit code different of 0).
150 Note: startup and --vgdb-error=0 will both cause Valgrind
152 gdbserver to be invoked before your program is executed. The
153 --vgdb-error=0 will in addition cause your program to stop on
154 all subsequent errors.
155 • all to specify the complete set. It is equivalent to
157 --vgdb-stop-at=startup,exit,abexit,valgrindabexit.
158 • none for the empty set.
160 --track-fds=<yes|no|all> [default: no]
162 When enabled, Valgrind will print out a list of open file
163 descriptors on exit or on request, via the gdbserver monitor
164 command v.info open_fds. Along with each file descriptor is printed
165 a stack backtrace of where the file was opened and any details
166 relating to the file descriptor such as the file name or socket
167 details. Use all to include reporting on stdin, stdout and stderr.
168 --time-stamp=<yes|no> [default: no]
170 When enabled, each message is preceded with an indication of the
171 elapsed wallclock time since startup, expressed as days, hours,
172 minutes, seconds and milliseconds.
173 --log-fd=<number> [default: 2, stderr]
175 Specifies that Valgrind should send all of its messages to the
176 specified file descriptor. The default, 2, is the standard error
177 channel (stderr). Note that this may interfere with the client's
178 own use of stderr, as Valgrind's output will be interleaved with
179 any output that the client sends to stderr.
180 --log-file=<filename>
182 Specifies that Valgrind should send all of its messages to the
183 specified file. If the file name is empty, it causes an abort.
184 There are three special format specifiers that can be used in the
185 file name.
186 %p is replaced with the current process ID. This is very useful for
188 program that invoke multiple processes. WARNING: If you use
189 --trace-children=yes and your program invokes multiple processes OR
190 your program forks without calling exec afterwards, and you don't
191 use this specifier (or the %q specifier below), the Valgrind output
192 from all those processes will go into one file, possibly jumbled
193 up, and possibly incomplete. Note: If the program forks and calls
194 exec afterwards, Valgrind output of the child from the period
195 between fork and exec will be lost. Fortunately this gap is really
196 tiny for most programs; and modern programs use posix_spawn anyway.
197 %n is replaced with a file sequence number unique for this process.
199 This is useful for processes that produces several files from the
200 same filename template.
201 %q{FOO} is replaced with the contents of the environment variable
203 FOO. If the {FOO} part is malformed, it causes an abort. This
204 specifier is rarely needed, but very useful in certain
205 circumstances (eg. when running MPI programs). The idea is that you
206 specify a variable which will be set differently for each process
207 in the job, for example BPROC_RANK or whatever is applicable in
208 your MPI setup. If the named environment variable is not set, it
209 causes an abort. Note that in some shells, the { and } characters
210 may need to be escaped with a backslash.
211 %% is replaced with %.
213 If an % is followed by any other character, it causes an abort.
215 If the file name specifies a relative file name, it is put in the
217 program's initial working directory: this is the current directory
218 when the program started its execution after the fork or after the
219 exec. If it specifies an absolute file name (ie. starts with '/')
220 then it is put there.
221 --log-socket=<ip-address:port-number>
223 Specifies that Valgrind should send all of its messages to the
224 specified port at the specified IP address. The port may be
225 omitted, in which case port 1500 is used. If a connection cannot be
226 made to the specified socket, Valgrind falls back to writing output
227 to the standard error (stderr). This option is intended to be used
228 in conjunction with the valgrind-listener program. For further
229 details, see the commentary in the manual.
230 --enable-debuginfod=<no|yes> [default: yes]
232 When enabled Valgrind will attempt to download missing debuginfo
233 from debuginfod servers if space-separated server URLs are present
234 in the $DEBUGINFOD_URLS environment variable. This option is
235 supported on Linux only.
236
238 These options are used by all tools that can report errors, e.g.
239 Memcheck, but not Cachegrind.
240 --xml=<yes|no> [default: no]
242 When enabled, the important parts of the output (e.g. tool error
243 messages) will be in XML format rather than plain text.
244 Furthermore, the XML output will be sent to a different output
245 channel than the plain text output. Therefore, you also must use
246 one of --xml-fd, --xml-file or --xml-socket to specify where the
247 XML is to be sent.
248 Less important messages will still be printed in plain text, but
250 because the XML output and plain text output are sent to different
251 output channels (the destination of the plain text output is still
252 controlled by --log-fd, --log-file and --log-socket) this should
253 not cause problems.
254 This option is aimed at making life easier for tools that consume
256 Valgrind's output as input, such as GUI front ends. Currently this
257 option works with Memcheck, Helgrind and DRD. The output format is
258 specified in the file docs/internals/xml-output-protocol4.txt in
259 the source tree for Valgrind 3.5.0 or later.
260 The recommended options for a GUI to pass, when requesting XML
262 output, are: --xml=yes to enable XML output, --xml-file to send the
263 XML output to a (presumably GUI-selected) file, --log-file to send
264 the plain text output to a second GUI-selected file,
265 --child-silent-after-fork=yes, and -q to restrict the plain text
266 output to critical error messages created by Valgrind itself. For
267 example, failure to read a specified suppressions file counts as a
268 critical error message. In this way, for a successful run the text
269 output file will be empty. But if it isn't empty, then it will
270 contain important information which the GUI user should be made
271 aware of.
272 --xml-fd=<number> [default: -1, disabled]
274 Specifies that Valgrind should send its XML output to the specified
275 file descriptor. It must be used in conjunction with --xml=yes.
276 --xml-file=<filename>
278 Specifies that Valgrind should send its XML output to the specified
279 file. It must be used in conjunction with --xml=yes. Any %p or %q
280 sequences appearing in the filename are expanded in exactly the
281 same way as they are for --log-file. See the description of --log-
282 file for details.
283 --xml-socket=<ip-address:port-number>
285 Specifies that Valgrind should send its XML output the specified
286 port at the specified IP address. It must be used in conjunction
287 with --xml=yes. The form of the argument is the same as that used
288 by --log-socket. See the description of --log-socket for further
289 details.
290 --xml-user-comment=<string>
292 Embeds an extra user comment string at the start of the XML output.
293 Only works when --xml=yes is specified; ignored otherwise.
294 --demangle=<yes|no> [default: yes]
296 Enable/disable automatic demangling (decoding) of C++ names.
297 Enabled by default. When enabled, Valgrind will attempt to
298 translate encoded C++ names back to something approaching the
299 original. The demangler handles symbols mangled by g++ versions
300 2.X, 3.X and 4.X.
301 An important fact about demangling is that function names mentioned
303 in suppressions files should be in their mangled form. Valgrind
304 does not demangle function names when searching for applicable
305 suppressions, because to do otherwise would make suppression file
306 contents dependent on the state of Valgrind's demangling machinery,
307 and also slow down suppression matching.
308 --num-callers=<number> [default: 12]
310 Specifies the maximum number of entries shown in stack traces that
311 identify program locations. Note that errors are commoned up using
312 only the top four function locations (the place in the current
313 function, and that of its three immediate callers). So this doesn't
314 affect the total number of errors reported.
315 The maximum value for this is 500. Note that higher settings will
317 make Valgrind run a bit more slowly and take a bit more memory, but
318 can be useful when working with programs with deeply-nested call
319 chains.
320 --unw-stack-scan-thresh=<number> [default: 0] ,
322 --unw-stack-scan-frames=<number> [default: 5]
323 Stack-scanning support is available only on ARM targets.
324 These flags enable and control stack unwinding by stack scanning.
326 When the normal stack unwinding mechanisms -- usage of Dwarf CFI
327 records, and frame-pointer following -- fail, stack scanning may be
328 able to recover a stack trace.
329 Note that stack scanning is an imprecise, heuristic mechanism that
331 may give very misleading results, or none at all. It should be used
332 only in emergencies, when normal unwinding fails, and it is
333 important to nevertheless have stack traces.
334 Stack scanning is a simple technique: the unwinder reads words from
336 the stack, and tries to guess which of them might be return
337 addresses, by checking to see if they point just after ARM or Thumb
338 call instructions. If so, the word is added to the backtrace.
339 The main danger occurs when a function call returns, leaving its
341 return address exposed, and a new function is called, but the new
342 function does not overwrite the old address. The result of this is
343 that the backtrace may contain entries for functions which have
344 already returned, and so be very confusing.
345 A second limitation of this implementation is that it will scan
347 only the page (4KB, normally) containing the starting stack
348 pointer. If the stack frames are large, this may result in only a
349 few (or not even any) being present in the trace. Also, if you are
350 unlucky and have an initial stack pointer near the end of its
351 containing page, the scan may miss all interesting frames.
352 By default stack scanning is disabled. The normal use case is to
354 ask for it when a stack trace would otherwise be very short. So, to
355 enable it, use --unw-stack-scan-thresh=number. This requests
356 Valgrind to try using stack scanning to "extend" stack traces which
357 contain fewer than number frames.
358 If stack scanning does take place, it will only generate at most
360 the number of frames specified by --unw-stack-scan-frames.
361 Typically, stack scanning generates so many garbage entries that
362 this value is set to a low value (5) by default. In no case will a
363 stack trace larger than the value specified by --num-callers be
364 created.
365 --error-limit=<yes|no> [default: yes]
367 When enabled, Valgrind stops reporting errors after 10,000,000 in
368 total, or 1,000 different ones, have been seen. This is to stop the
369 error tracking machinery from becoming a huge performance overhead
370 in programs with many errors.
371 --error-exitcode=<number> [default: 0]
373 Specifies an alternative exit code to return if Valgrind reported
374 any errors in the run. When set to the default value (zero), the
375 return value from Valgrind will always be the return value of the
376 process being simulated. When set to a nonzero value, that value is
377 returned instead, if Valgrind detects any errors. This is useful
378 for using Valgrind as part of an automated test suite, since it
379 makes it easy to detect test cases for which Valgrind has reported
380 errors, just by inspecting return codes. When set to a nonzero
381 value and Valgrind detects no error, the return value of Valgrind
382 will be the return value of the program being simulated.
383 --exit-on-first-error=<yes|no> [default: no]
385 If this option is enabled, Valgrind exits on the first error. A
386 nonzero exit value must be defined using --error-exitcode option.
387 Useful if you are running regression tests or have some other
388 automated test machinery.
389 --error-markers=<begin>,<end> [default: none]
391 When errors are output as plain text (i.e. XML not used),
392 --error-markers instructs to output a line containing the begin
393 (end) string before (after) each error.
394 Such marker lines facilitate searching for errors and/or extracting
396 errors in an output file that contain valgrind errors mixed with
397 the program output.
398 Note that empty markers are accepted. So, only using a begin (or an
400 end) marker is possible.
401 --show-error-list=no|yes [default: no]
403 If this option is enabled, for tools that report errors, valgrind
404 will show the list of detected errors and the list of used
405 suppressions at exit.
406 Note that at verbosity 2 and above, valgrind automatically shows
408 the list of detected errors and the list of used suppressions at
409 exit, unless --show-error-list=no is selected.
410 -s
412 Specifying -s is equivalent to --show-error-list=yes.
413 --sigill-diagnostics=<yes|no> [default: yes]
415 Enable/disable printing of illegal instruction diagnostics. Enabled
416 by default, but defaults to disabled when --quiet is given. The
417 default can always be explicitly overridden by giving this option.
418 When enabled, a warning message will be printed, along with some
420 diagnostics, whenever an instruction is encountered that Valgrind
421 cannot decode or translate, before the program is given a SIGILL
422 signal. Often an illegal instruction indicates a bug in the program
423 or missing support for the particular instruction in Valgrind. But
424 some programs do deliberately try to execute an instruction that
425 might be missing and trap the SIGILL signal to detect processor
426 features. Using this flag makes it possible to avoid the diagnostic
427 output that you would otherwise get in such cases.
428 --keep-debuginfo=<yes|no> [default: no]
430 When enabled, keep ("archive") symbols and all other debuginfo for
431 unloaded code. This allows saved stack traces to include file/line
432 info for code that has been dlclose'd (or similar). Be careful with
433 this, since it can lead to unbounded memory use for programs which
434 repeatedly load and unload shared objects.
435 Some tools and some functionalities have only limited support for
437 archived debug info. Memcheck fully supports it. Generally, tools
438 that report errors can use archived debug info to show the error
439 stack traces. The known limitations are: Helgrind's past access
440 stack trace of a race condition is does not use archived debug
441 info. Massif (and more generally the xtree Massif output format)
442 does not make use of archived debug info. Only Memcheck has been
443 (somewhat) tested with --keep-debuginfo=yes, so other tools may
444 have unknown limitations.
445 --show-below-main=<yes|no> [default: no]
447 By default, stack traces for errors do not show any functions that
448 appear beneath main because most of the time it's uninteresting C
449 library stuff and/or gobbledygook. Alternatively, if main is not
450 present in the stack trace, stack traces will not show any
451 functions below main-like functions such as glibc's
452 __libc_start_main. Furthermore, if main-like functions are present
453 in the trace, they are normalised as (below main), in order to make
454 the output more deterministic.
455 If this option is enabled, all stack trace entries will be shown
457 and main-like functions will not be normalised.
458 --fullpath-after=<string> [default: don't show source paths]
460 By default Valgrind only shows the filenames in stack traces, but
461 not full paths to source files. When using Valgrind in large
462 projects where the sources reside in multiple different
463 directories, this can be inconvenient. --fullpath-after provides a
464 flexible solution to this problem. When this option is present, the
465 path to each source file is shown, with the following all-important
466 caveat: if string is found in the path, then the path up to and
467 including string is omitted, else the path is shown unmodified.
468 Note that string is not required to be a prefix of the path.
469 For example, consider a file named
471 /home/janedoe/blah/src/foo/bar/xyzzy.c. Specifying
472 --fullpath-after=/home/janedoe/blah/src/ will cause Valgrind to
473 show the name as foo/bar/xyzzy.c.
474 Because the string is not required to be a prefix,
476 --fullpath-after=src/ will produce the same output. This is useful
477 when the path contains arbitrary machine-generated characters. For
478 example, the path /my/build/dir/C32A1B47/blah/src/foo/xyzzy can be
479 pruned to foo/xyzzy using --fullpath-after=/blah/src/.
480 If you simply want to see the full path, just specify an empty
482 string: --fullpath-after=. This isn't a special case, merely a
483 logical consequence of the above rules.
484 Finally, you can use --fullpath-after multiple times. Any
486 appearance of it causes Valgrind to switch to producing full paths
487 and applying the above filtering rule. Each produced path is
488 compared against all the --fullpath-after-specified strings, in the
489 order specified. The first string to match causes the path to be
490 truncated as described above. If none match, the full path is
491 shown. This facilitates chopping off prefixes when the sources are
492 drawn from a number of unrelated directories.
493 --extra-debuginfo-path=<path> [default: undefined and unused]
495 By default Valgrind searches in several well-known paths for debug
496 objects, such as /usr/lib/debug/.
497 However, there may be scenarios where you may wish to put debug
499 objects at an arbitrary location, such as external storage when
500 running Valgrind on a mobile device with limited local storage.
501 Another example might be a situation where you do not have
502 permission to install debug object packages on the system where you
503 are running Valgrind.
504 In these scenarios, you may provide an absolute path as an extra,
506 final place for Valgrind to search for debug objects by specifying
507 --extra-debuginfo-path=/path/to/debug/objects. The given path will
508 be prepended to the absolute path name of the searched-for object.
509 For example, if Valgrind is looking for the debuginfo for
510 /w/x/y/zz.so and --extra-debuginfo-path=/a/b/c is specified, it
511 will look for a debug object at /a/b/c/w/x/y/zz.so.
512 This flag should only be specified once. If it is specified
514 multiple times, only the last instance is honoured.
515 --debuginfo-server=ipaddr:port [default: undefined and unused]
517 This is a new, experimental, feature introduced in version 3.9.0.
518 In some scenarios it may be convenient to read debuginfo from
520 objects stored on a different machine. With this flag, Valgrind
521 will query a debuginfo server running on ipaddr and listening on
522 port port, if it cannot find the debuginfo object in the local
523 filesystem.
524 The debuginfo server must accept TCP connections on port port. The
526 debuginfo server is contained in the source file
527 auxprogs/valgrind-di-server.c. It will only serve from the
528 directory it is started in. port defaults to 1500 in both client
529 and server if not specified.
530 If Valgrind looks for the debuginfo for /w/x/y/zz.so by using the
532 debuginfo server, it will strip the pathname components and merely
533 request zz.so on the server. That in turn will look only in its
534 current working directory for a matching debuginfo object.
535 The debuginfo data is transmitted in small fragments (8 KB) as
537 requested by Valgrind. Each block is compressed using LZO to reduce
538 transmission time. The implementation has been tuned for best
539 performance over a single-stage 802.11g (WiFi) network link.
540 Note that checks for matching primary vs debug objects, using GNU
542 debuglink CRC scheme, are performed even when using the debuginfo
543 server. To disable such checking, you need to also specify
544 --allow-mismatched-debuginfo=yes.
545 By default the Valgrind build system will build valgrind-di-server
547 for the target platform, which is almost certainly not what you
548 want. So far we have been unable to find out how to get
549 automake/autoconf to build it for the build platform. If you want
550 to use it, you will have to recompile it by hand using the command
551 shown at the top of auxprogs/valgrind-di-server.c.
552 Valgrind can also download debuginfo via debuginfod. See the
554 DEBUGINFOD section for more information.
555 --allow-mismatched-debuginfo=no|yes [no]
557 When reading debuginfo from separate debuginfo objects, Valgrind
558 will by default check that the main and debuginfo objects match,
559 using the GNU debuglink mechanism. This guarantees that it does not
560 read debuginfo from out of date debuginfo objects, and also ensures
561 that Valgrind can't crash as a result of mismatches.
562 This check can be overridden using
564 --allow-mismatched-debuginfo=yes. This may be useful when the
565 debuginfo and main objects have not been split in the proper way.
566 Be careful when using this, though: it disables all consistency
567 checking, and Valgrind has been observed to crash when the main and
568 debuginfo objects don't match.
569 --suppressions=<filename> [default: $PREFIX/lib/valgrind/default.supp]
571 Specifies an extra file from which to read descriptions of errors
572 to suppress. You may use up to 100 extra suppression files.
573 --gen-suppressions=<yes|no|all> [default: no]
575 When set to yes, Valgrind will pause after every error shown and
576 print the line:
577 ---- Print suppression ? --- [Return/N/n/Y/y/C/c] ----
579 Pressing Ret, or N Ret or n Ret, causes Valgrind continue execution
581 without printing a suppression for this error.
582 Pressing Y Ret or y Ret causes Valgrind to write a suppression for
584 this error. You can then cut and paste it into a suppression file
585 if you don't want to hear about the error in the future.
586 When set to all, Valgrind will print a suppression for every
588 reported error, without querying the user.
589 This option is particularly useful with C++ programs, as it prints
591 out the suppressions with mangled names, as required.
592 Note that the suppressions printed are as specific as possible. You
594 may want to common up similar ones, by adding wildcards to function
595 names, and by using frame-level wildcards. The wildcarding
596 facilities are powerful yet flexible, and with a bit of careful
597 editing, you may be able to suppress a whole family of related
598 errors with only a few suppressions.
599 Sometimes two different errors are suppressed by the same
601 suppression, in which case Valgrind will output the suppression
602 more than once, but you only need to have one copy in your
603 suppression file (but having more than one won't cause problems).
604 Also, the suppression name is given as <insert a suppression name
605 here>; the name doesn't really matter, it's only used with the -v
606 option which prints out all used suppression records.
607 --input-fd=<number> [default: 0, stdin]
609 When using --gen-suppressions=yes, Valgrind will stop so as to read
610 keyboard input from you when each error occurs. By default it reads
611 from the standard input (stdin), which is problematic for programs
612 which close stdin. This option allows you to specify an alternative
613 file descriptor from which to read input.
614 --dsymutil=no|yes [yes]
616 This option is only relevant when running Valgrind on Mac OS X.
617 Mac OS X uses a deferred debug information (debuginfo) linking
619 scheme. When object files containing debuginfo are linked into a
620 .dylib or an executable, the debuginfo is not copied into the final
621 file. Instead, the debuginfo must be linked manually by running
622 dsymutil, a system-provided utility, on the executable or .dylib.
623 The resulting combined debuginfo is placed in a directory alongside
624 the executable or .dylib, but with the extension .dSYM.
625 With --dsymutil=no, Valgrind will detect cases where the .dSYM
627 directory is either missing, or is present but does not appear to
628 match the associated executable or .dylib, most likely because it
629 is out of date. In these cases, Valgrind will print a warning
630 message but take no further action.
631 With --dsymutil=yes, Valgrind will, in such cases, automatically
633 run dsymutil as necessary to bring the debuginfo up to date. For
634 all practical purposes, if you always use --dsymutil=yes, then
635 there is never any need to run dsymutil manually or as part of your
636 applications's build system, since Valgrind will run it as
637 necessary.
638 Valgrind will not attempt to run dsymutil on any executable or
640 library in /usr/, /bin/, /sbin/, /opt/, /sw/, /System/, /Library/
641 or /Applications/ since dsymutil will always fail in such
642 situations. It fails both because the debuginfo for such
643 pre-installed system components is not available anywhere, and also
644 because it would require write privileges in those directories.
645 Be careful when using --dsymutil=yes, since it will cause
647 pre-existing .dSYM directories to be silently deleted and
648 re-created. Also note that dsymutil is quite slow, sometimes
649 excessively so.
650 --max-stackframe=<number> [default: 2000000]
652 The maximum size of a stack frame. If the stack pointer moves by
653 more than this amount then Valgrind will assume that the program is
654 switching to a different stack.
655 You may need to use this option if your program has large
657 stack-allocated arrays. Valgrind keeps track of your program's
658 stack pointer. If it changes by more than the threshold amount,
659 Valgrind assumes your program is switching to a different stack,
660 and Memcheck behaves differently than it would for a stack pointer
661 change smaller than the threshold. Usually this heuristic works
662 well. However, if your program allocates large structures on the
663 stack, this heuristic will be fooled, and Memcheck will
664 subsequently report large numbers of invalid stack accesses. This
665 option allows you to change the threshold to a different value.
666 You should only consider use of this option if Valgrind's debug
668 output directs you to do so. In that case it will tell you the new
669 threshold you should specify.
670 In general, allocating large structures on the stack is a bad idea,
672 because you can easily run out of stack space, especially on
673 systems with limited memory or which expect to support large
674 numbers of threads each with a small stack, and also because the
675 error checking performed by Memcheck is more effective for
676 heap-allocated data than for stack-allocated data. If you have to
677 use this option, you may wish to consider rewriting your code to
678 allocate on the heap rather than on the stack.
679 --main-stacksize=<number> [default: use current 'ulimit' value]
681 Specifies the size of the main thread's stack.
682 To simplify its memory management, Valgrind reserves all required
684 space for the main thread's stack at startup. That means it needs
685 to know the required stack size at startup.
686 By default, Valgrind uses the current "ulimit" value for the stack
688 size, or 16 MB, whichever is lower. In many cases this gives a
689 stack size in the range 8 to 16 MB, which almost never overflows
690 for most applications.
691 If you need a larger total stack size, use --main-stacksize to
693 specify it. Only set it as high as you need, since reserving far
694 more space than you need (that is, hundreds of megabytes more than
695 you need) constrains Valgrind's memory allocators and may reduce
696 the total amount of memory that Valgrind can use. This is only
697 really of significance on 32-bit machines.
698 On Linux, you may request a stack of size up to 2GB. Valgrind will
700 stop with a diagnostic message if the stack cannot be allocated.
701 --main-stacksize only affects the stack size for the program's
703 initial thread. It has no bearing on the size of thread stacks, as
704 Valgrind does not allocate those.
705 You may need to use both --main-stacksize and --max-stackframe
707 together. It is important to understand that --main-stacksize sets
708 the maximum total stack size, whilst --max-stackframe specifies the
709 largest size of any one stack frame. You will have to work out the
710 --main-stacksize value for yourself (usually, if your applications
711 segfaults). But Valgrind will tell you the needed --max-stackframe
712 size, if necessary.
713 As discussed further in the description of --max-stackframe, a
715 requirement for a large stack is a sign of potential portability
716 problems. You are best advised to place all large data in
717 heap-allocated memory.
718 --max-threads=<number> [default: 500]
720 By default, Valgrind can handle to up to 500 threads. Occasionally,
721 that number is too small. Use this option to provide a different
722 limit. E.g. --max-threads=3000.
723
725 For tools that use their own version of malloc (e.g. Memcheck, Massif,
726 Helgrind, DRD), the following options apply.
727 --alignment=<number> [default: 8 or 16, depending on the platform]
729 By default Valgrind's malloc, realloc, etc, return a block whose
730 starting address is 8-byte aligned or 16-byte aligned (the value
731 depends on the platform and matches the platform default). This
732 option allows you to specify a different alignment. The supplied
733 value must be greater than or equal to the default, less than or
734 equal to 4096, and must be a power of two.
735 --redzone-size=<number> [default: depends on the tool]
737 Valgrind's malloc, realloc, etc, add padding blocks before and
738 after each heap block allocated by the program being run. Such
739 padding blocks are called redzones. The default value for the
740 redzone size depends on the tool. For example, Memcheck adds and
741 protects a minimum of 16 bytes before and after each block
742 allocated by the client. This allows it to detect block underruns
743 or overruns of up to 16 bytes.
744 Increasing the redzone size makes it possible to detect overruns of
746 larger distances, but increases the amount of memory used by
747 Valgrind. Decreasing the redzone size will reduce the memory needed
748 by Valgrind but also reduces the chances of detecting
749 over/underruns, so is not recommended.
750 --xtree-memory=none|allocs|full [none]
752 Tools replacing Valgrind's malloc, realloc, etc, can optionally
753 produce an execution tree detailing which piece of code is
754 responsible for heap memory usage. See Execution Trees for a
755 detailed explanation about execution trees.
756 When set to none, no memory execution tree is produced.
758 When set to allocs, the memory execution tree gives the current
760 number of allocated bytes and the current number of allocated
761 blocks.
762 When set to full, the memory execution tree gives 6 different
764 measurements : the current number of allocated bytes and blocks
765 (same values as for allocs), the total number of allocated bytes
766 and blocks, the total number of freed bytes and blocks.
767 Note that the overhead in cpu and memory to produce an xtree
769 depends on the tool. The overhead in cpu is small for the value
770 allocs, as the information needed to produce this report is
771 maintained in any case by the tool. For massif and helgrind,
772 specifying full implies to capture a stack trace for each free
773 operation, while normally these tools only capture an allocation
774 stack trace. For Memcheck, the cpu overhead for the value full is
775 small, as this can only be used in combination with
776 --keep-stacktraces=alloc-and-free or
777 --keep-stacktraces=alloc-then-free, which already records a stack
778 trace for each free operation. The memory overhead varies between 5
779 and 10 words per unique stacktrace in the xtree, plus the memory
780 needed to record the stack trace for the free operations, if needed
781 specifically for the xtree.
782 --xtree-memory-file=<filename> [default: xtmemory.kcg.%p]
784 Specifies that Valgrind should produce the xtree memory report in
785 the specified file. Any %p or %q sequences appearing in the
786 filename are expanded in exactly the same way as they are for
787 --log-file. See the description of --log-file for details.
788 If the filename contains the extension .ms, then the produced file
790 format will be a massif output file format. If the filename
791 contains the extension .kcg or no extension is provided or
792 recognised, then the produced file format will be a callgrind
793 output format.
794 See Execution Trees for a detailed explanation about execution
796 trees formats.
797
799 These options apply to all tools, as they affect certain obscure
800 workings of the Valgrind core. Most people won't need to use them.
801 --smc-check=<none|stack|all|all-non-file> [default: all-non-file for
803 x86/amd64/s390x, stack for other archs]
804 This option controls Valgrind's detection of self-modifying code.
805 If no checking is done, when a program executes some code, then
806 overwrites it with new code, and executes the new code, Valgrind
807 will continue to execute the translations it made for the old code.
808 This will likely lead to incorrect behaviour and/or crashes.
809 For "modern" architectures -- anything that's not x86, amd64 or
811 s390x -- the default is stack. This is because a correct program
812 must take explicit action to reestablish D-I cache coherence
813 following code modification. Valgrind observes and honours such
814 actions, with the result that self-modifying code is transparently
815 handled with zero extra cost.
816 For x86, amd64 and s390x, the program is not required to notify the
818 hardware of required D-I coherence syncing. Hence the default is
819 all-non-file, which covers the normal case of generating code into
820 an anonymous (non-file-backed) mmap'd area.
821 The meanings of the four available settings are as follows. No
823 detection (none), detect self-modifying code on the stack (which is
824 used by GCC to implement nested functions) (stack), detect
825 self-modifying code everywhere (all), and detect self-modifying
826 code everywhere except in file-backed mappings (all-non-file).
827 Running with all will slow Valgrind down noticeably. Running with
829 none will rarely speed things up, since very little code gets
830 dynamically generated in most programs. The
831 VALGRIND_DISCARD_TRANSLATIONS client request is an alternative to
832 --smc-check=all and --smc-check=all-non-file that requires more
833 programmer effort but allows Valgrind to run your program faster,
834 by telling it precisely when translations need to be re-made.
835 --smc-check=all-non-file provides a cheaper but more limited
837 version of --smc-check=all. It adds checks to any translations that
838 do not originate from file-backed memory mappings. Typical
839 applications that generate code, for example JITs in web browsers,
840 generate code into anonymous mmaped areas, whereas the "fixed" code
841 of the browser always lives in file-backed mappings.
842 --smc-check=all-non-file takes advantage of this observation,
843 limiting the overhead of checking to code which is likely to be JIT
844 generated.
845 --read-inline-info=<yes|no> [default: see below]
847 When enabled, Valgrind will read information about inlined function
848 calls from DWARF3 debug info. This slows Valgrind startup and makes
849 it use more memory (typically for each inlined piece of code, 6
850 words and space for the function name), but it results in more
851 descriptive stacktraces. Currently, this functionality is enabled
852 by default only for Linux, Android and Solaris targets and only for
853 the tools Memcheck, Massif, Helgrind and DRD. Here is an example of
854 some stacktraces with --read-inline-info=no:
855 ==15380== Conditional jump or move depends on uninitialised value(s)
857 ==15380== at 0x80484EA: main (inlinfo.c:6)
858 ==15380==
859 ==15380== Conditional jump or move depends on uninitialised value(s)
860 ==15380== at 0x8048550: fun_noninline (inlinfo.c:6)
861 ==15380== by 0x804850E: main (inlinfo.c:34)
862 ==15380==
863 ==15380== Conditional jump or move depends on uninitialised value(s)
864 ==15380== at 0x8048520: main (inlinfo.c:6)
865 And here are the same errors with --read-inline-info=yes:
867 ==15377== Conditional jump or move depends on uninitialised value(s)
869 ==15377== at 0x80484EA: fun_d (inlinfo.c:6)
870 ==15377== by 0x80484EA: fun_c (inlinfo.c:14)
871 ==15377== by 0x80484EA: fun_b (inlinfo.c:20)
872 ==15377== by 0x80484EA: fun_a (inlinfo.c:26)
873 ==15377== by 0x80484EA: main (inlinfo.c:33)
874 ==15377==
875 ==15377== Conditional jump or move depends on uninitialised value(s)
876 ==15377== at 0x8048550: fun_d (inlinfo.c:6)
877 ==15377== by 0x8048550: fun_noninline (inlinfo.c:41)
878 ==15377== by 0x804850E: main (inlinfo.c:34)
879 ==15377==
880 ==15377== Conditional jump or move depends on uninitialised value(s)
881 ==15377== at 0x8048520: fun_d (inlinfo.c:6)
882 ==15377== by 0x8048520: main (inlinfo.c:35)
883 --read-var-info=<yes|no> [default: no]
885 When enabled, Valgrind will read information about variable types
886 and locations from DWARF3 debug info. This slows Valgrind startup
887 significantly and makes it use significantly more memory, but for
888 the tools that can take advantage of it (Memcheck, Helgrind, DRD)
889 it can result in more precise error messages. For example, here are
890 some standard errors issued by Memcheck:
891 ==15363== Uninitialised byte(s) found during client check request
893 ==15363== at 0x80484A9: croak (varinfo1.c:28)
894 ==15363== by 0x8048544: main (varinfo1.c:55)
895 ==15363== Address 0x80497f7 is 7 bytes inside data symbol "global_i2"
896 ==15363==
897 ==15363== Uninitialised byte(s) found during client check request
898 ==15363== at 0x80484A9: croak (varinfo1.c:28)
899 ==15363== by 0x8048550: main (varinfo1.c:56)
900 ==15363== Address 0xbea0d0cc is on thread 1's stack
901 ==15363== in frame #1, created by main (varinfo1.c:45)
902 And here are the same errors with --read-var-info=yes:
904 ==15370== Uninitialised byte(s) found during client check request
906 ==15370== at 0x80484A9: croak (varinfo1.c:28)
907 ==15370== by 0x8048544: main (varinfo1.c:55)
908 ==15370== Location 0x80497f7 is 0 bytes inside global_i2[7],
909 ==15370== a global variable declared at varinfo1.c:41
910 ==15370==
911 ==15370== Uninitialised byte(s) found during client check request
912 ==15370== at 0x80484A9: croak (varinfo1.c:28)
913 ==15370== by 0x8048550: main (varinfo1.c:56)
914 ==15370== Location 0xbeb4a0cc is 0 bytes inside local var "local"
915 ==15370== declared at varinfo1.c:46, in frame #1 of thread 1
916 --vgdb-poll=<number> [default: 5000]
918 As part of its main loop, the Valgrind scheduler will poll to check
919 if some activity (such as an external command or some input from a
920 gdb) has to be handled by gdbserver. This activity poll will be
921 done after having run the given number of basic blocks (or slightly
922 more than the given number of basic blocks). This poll is quite
923 cheap so the default value is set relatively low. You might further
924 decrease this value if vgdb cannot use ptrace system call to
925 interrupt Valgrind if all threads are (most of the time) blocked in
926 a system call.
927 --vgdb-shadow-registers=no|yes [default: no]
929 When activated, gdbserver will expose the Valgrind shadow registers
930 to GDB. With this, the value of the Valgrind shadow registers can
931 be examined or changed using GDB. Exposing shadow registers only
932 works with GDB version 7.1 or later.
933 --vgdb-prefix=<prefix> [default: /tmp/vgdb-pipe]
935 To communicate with gdb/vgdb, the Valgrind gdbserver creates 3
936 files (2 named FIFOs and a mmap shared memory file). The prefix
937 option controls the directory and prefix for the creation of these
938 files.
939 --run-libc-freeres=<yes|no> [default: yes]
941 This option is only relevant when running Valgrind on Linux.
942 The GNU C library (libc.so), which is used by all programs, may
944 allocate memory for its own uses. Usually it doesn't bother to free
945 that memory when the program ends—there would be no point, since
946 the Linux kernel reclaims all process resources when a process
947 exits anyway, so it would just slow things down.
948 The glibc authors realised that this behaviour causes leak
950 checkers, such as Valgrind, to falsely report leaks in glibc, when
951 a leak check is done at exit. In order to avoid this, they provided
952 a routine called __libc_freeres specifically to make glibc release
953 all memory it has allocated. Memcheck therefore tries to run
954 __libc_freeres at exit.
955 Unfortunately, in some very old versions of glibc, __libc_freeres
957 is sufficiently buggy to cause segmentation faults. This was
958 particularly noticeable on Red Hat 7.1. So this option is provided
959 in order to inhibit the run of __libc_freeres. If your program
960 seems to run fine on Valgrind, but segfaults at exit, you may find
961 that --run-libc-freeres=no fixes that, although at the cost of
962 possibly falsely reporting space leaks in libc.so.
963 --run-cxx-freeres=<yes|no> [default: yes]
965 This option is only relevant when running Valgrind on Linux or
966 Solaris C++ programs.
967 The GNU Standard C++ library (libstdc++.so), which is used by all
969 C++ programs compiled with g++, may allocate memory for its own
970 uses. Usually it doesn't bother to free that memory when the
971 program ends—there would be no point, since the kernel reclaims all
972 process resources when a process exits anyway, so it would just
973 slow things down.
974 The gcc authors realised that this behaviour causes leak checkers,
976 such as Valgrind, to falsely report leaks in libstdc++, when a leak
977 check is done at exit. In order to avoid this, they provided a
978 routine called __gnu_cxx::__freeres specifically to make libstdc++
979 release all memory it has allocated. Memcheck therefore tries to
980 run __gnu_cxx::__freeres at exit.
981 For the sake of flexibility and unforeseen problems with
983 __gnu_cxx::__freeres, option --run-cxx-freeres=no exists, although
984 at the cost of possibly falsely reporting space leaks in
985 libstdc++.so.
986 --sim-hints=hint1,hint2,...
988 Pass miscellaneous hints to Valgrind which slightly modify the
989 simulated behaviour in nonstandard or dangerous ways, possibly to
990 help the simulation of strange features. By default no hints are
991 enabled. Use with caution! Currently known hints are:
992 • lax-ioctls: Be very lax about ioctl handling; the only
994 assumption is that the size is correct. Doesn't require the
995 full buffer to be initialised when writing. Without this, using
996 some device drivers with a large number of strange ioctl
997 commands becomes very tiresome.
998 • fuse-compatible: Enable special handling for certain system
1000 calls that may block in a FUSE file-system. This may be
1001 necessary when running Valgrind on a multi-threaded program
1002 that uses one thread to manage a FUSE file-system and another
1003 thread to access that file-system.
1004 • enable-outer: Enable some special magic needed when the program
1006 being run is itself Valgrind.
1007 • no-inner-prefix: Disable printing a prefix > in front of each
1009 stdout or stderr output line in an inner Valgrind being run by
1010 an outer Valgrind. This is useful when running Valgrind
1011 regression tests in an outer/inner setup. Note that the prefix
1012 > will always be printed in front of the inner debug logging
1013 lines.
1014 • no-nptl-pthread-stackcache: This hint is only relevant when
1016 running Valgrind on Linux; it is ignored on Solaris and Mac OS
1017 X.
1018 The GNU glibc pthread library (libpthread.so), which is used by
1020 pthread programs, maintains a cache of pthread stacks. When a
1021 pthread terminates, the memory used for the pthread stack and
1022 some thread local storage related data structure are not always
1023 directly released. This memory is kept in a cache (up to a
1024 certain size), and is re-used if a new thread is started.
1025 This cache causes the helgrind tool to report some false
1027 positive race condition errors on this cached memory, as
1028 helgrind does not understand the internal glibc cache
1029 synchronisation primitives. So, when using helgrind, disabling
1030 the cache helps to avoid false positive race conditions, in
1031 particular when using thread local storage variables (e.g.
1032 variables using the __thread qualifier).
1033 When using the memcheck tool, disabling the cache ensures the
1035 memory used by glibc to handle __thread variables is directly
1036 released when a thread terminates.
1037 Note: Valgrind disables the cache using some internal knowledge
1039 of the glibc stack cache implementation and by examining the
1040 debug information of the pthread library. This technique is
1041 thus somewhat fragile and might not work for all glibc
1042 versions. This has been successfully tested with various glibc
1043 versions (e.g. 2.11, 2.16, 2.18) on various platforms.
1044 • lax-doors: (Solaris only) Be very lax about door syscall
1046 handling over unrecognised door file descriptors. Does not
1047 require that full buffer is initialised when writing. Without
1048 this, programs using libdoor(3LIB) functionality with
1049 completely proprietary semantics may report large number of
1050 false positives.
1051 • fallback-llsc: (MIPS and ARM64 only): Enables an alternative
1053 implementation of Load-Linked (LL) and Store-Conditional (SC)
1054 instructions. The standard implementation gives more correct
1055 behaviour, but can cause indefinite looping on certain
1056 processor implementations that are intolerant of extra memory
1057 references between LL and SC. So far this is known only to
1058 happen on Cavium 3 cores. You should not need to use this flag,
1059 since the relevant cores are detected at startup and the
1060 alternative implementation is automatically enabled if
1061 necessary. There is no equivalent anti-flag: you cannot
1062 force-disable the alternative implementation, if it is
1063 automatically enabled. The underlying problem exists because
1064 the "standard" implementation of LL and SC is done by copying
1065 through LL and SC instructions into the instrumented code.
1066 However, tools may insert extra instrumentation memory
1067 references in between the LL and SC instructions. These memory
1068 references are not present in the original uninstrumented code,
1069 and their presence in the instrumented code can cause the SC
1070 instructions to persistently fail, leading to indefinite
1071 looping in LL-SC blocks. The alternative implementation gives
1072 correct behaviour of LL and SC instructions between threads in
1073 a process, up to and including the ABA scenario. It also gives
1074 correct behaviour between a Valgrinded thread and a
1075 non-Valgrinded thread running in a different process, that
1076 communicate via shared memory, but only up to and including
1077 correct CAS behaviour -- in this case the ABA scenario may not
1078 be correctly handled.
1079 --fair-sched=<no|yes|try> [default: no]
1081 The --fair-sched option controls the locking mechanism used by
1082 Valgrind to serialise thread execution. The locking mechanism
1083 controls the way the threads are scheduled, and different settings
1084 give different trade-offs between fairness and performance. For
1085 more details about the Valgrind thread serialisation scheme and its
1086 impact on performance and thread scheduling, see Scheduling and
1087 Multi-Thread Performance.
1088 • The value --fair-sched=yes activates a fair scheduler. In
1090 short, if multiple threads are ready to run, the threads will
1091 be scheduled in a round robin fashion. This mechanism is not
1092 available on all platforms or Linux versions. If not available,
1093 using --fair-sched=yes will cause Valgrind to terminate with an
1094 error.
1095 You may find this setting improves overall responsiveness if
1097 you are running an interactive multithreaded program, for
1098 example a web browser, on Valgrind.
1099 • The value --fair-sched=try activates fair scheduling if
1101 available on the platform. Otherwise, it will automatically
1102 fall back to --fair-sched=no.
1103 • The value --fair-sched=no activates a scheduler which does not
1105 guarantee fairness between threads ready to run, but which in
1106 general gives the highest performance.
1107 --kernel-variant=variant1,variant2,...
1109 Handle system calls and ioctls arising from minor variants of the
1110 default kernel for this platform. This is useful for running on
1111 hacked kernels or with kernel modules which support nonstandard
1112 ioctls, for example. Use with caution. If you don't understand what
1113 this option does then you almost certainly don't need it. Currently
1114 known variants are:
1115 • bproc: support the sys_broc system call on x86. This is for
1117 running on BProc, which is a minor variant of standard Linux
1118 which is sometimes used for building clusters.
1119 • android-no-hw-tls: some versions of the Android emulator for
1121 ARM do not provide a hardware TLS (thread-local state)
1122 register, and Valgrind crashes at startup. Use this variant to
1123 select software support for TLS.
1124 • android-gpu-sgx5xx: use this to support handling of proprietary
1126 ioctls for the PowerVR SGX 5XX series of GPUs on Android
1127 devices. Failure to select this does not cause stability
1128 problems, but may cause Memcheck to report false errors after
1129 the program performs GPU-specific ioctls.
1130 • android-gpu-adreno3xx: similarly, use this to support handling
1132 of proprietary ioctls for the Qualcomm Adreno 3XX series of
1133 GPUs on Android devices.
1134 --merge-recursive-frames=<number> [default: 0]
1136 Some recursive algorithms, for example balanced binary tree
1137 implementations, create many different stack traces, each
1138 containing cycles of calls. A cycle is defined as two identical
1139 program counter values separated by zero or more other program
1140 counter values. Valgrind may then use a lot of memory to store all
1141 these stack traces. This is a poor use of memory considering that
1142 such stack traces contain repeated uninteresting recursive calls
1143 instead of more interesting information such as the function that
1144 has initiated the recursive call.
1145 The option --merge-recursive-frames=<number> instructs Valgrind to
1147 detect and merge recursive call cycles having a size of up to
1148 <number> frames. When such a cycle is detected, Valgrind records
1149 the cycle in the stack trace as a unique program counter.
1150 The value 0 (the default) causes no recursive call merging. A value
1152 of 1 will cause stack traces of simple recursive algorithms (for
1153 example, a factorial implementation) to be collapsed. A value of 2
1154 will usually be needed to collapse stack traces produced by
1155 recursive algorithms such as binary trees, quick sort, etc. Higher
1156 values might be needed for more complex recursive algorithms.
1157 Note: recursive calls are detected by analysis of program counter
1159 values. They are not detected by looking at function names.
1160 --num-transtab-sectors=<number> [default: 6 for Android platforms, 16
1162 for all others]
1163 Valgrind translates and instruments your program's machine code in
1164 small fragments (basic blocks). The translations are stored in a
1165 translation cache that is divided into a number of sections
1166 (sectors). If the cache is full, the sector containing the oldest
1167 translations is emptied and reused. If these old translations are
1168 needed again, Valgrind must re-translate and re-instrument the
1169 corresponding machine code, which is expensive. If the "executed
1170 instructions" working set of a program is big, increasing the
1171 number of sectors may improve performance by reducing the number of
1172 re-translations needed. Sectors are allocated on demand. Once
1173 allocated, a sector can never be freed, and occupies considerable
1174 space, depending on the tool and the value of
1175 --avg-transtab-entry-size (about 40 MB per sector for Memcheck).
1176 Use the option --stats=yes to obtain precise information about the
1177 memory used by a sector and the allocation and recycling of
1178 sectors.
1179 --avg-transtab-entry-size=<number> [default: 0, meaning use tool
1181 provided default]
1182 Average size of translated basic block. This average size is used
1183 to dimension the size of a sector. Each tool provides a default
1184 value to be used. If this default value is too small, the
1185 translation sectors will become full too quickly. If this default
1186 value is too big, a significant part of the translation sector
1187 memory will be unused. Note that the average size of a basic block
1188 translation depends on the tool, and might depend on tool options.
1189 For example, the memcheck option --track-origins=yes increases the
1190 size of the basic block translations. Use --avg-transtab-entry-size
1191 to tune the size of the sectors, either to gain memory or to avoid
1192 too many retranslations.
1193 --aspace-minaddr=<address> [default: depends on the platform]
1195 To avoid potential conflicts with some system libraries, Valgrind
1196 does not use the address space below --aspace-minaddr value,
1197 keeping it reserved in case a library specifically requests memory
1198 in this region. So, some "pessimistic" value is guessed by Valgrind
1199 depending on the platform. On linux, by default, Valgrind avoids
1200 using the first 64MB even if typically there is no conflict in this
1201 complete zone. You can use the option --aspace-minaddr to have your
1202 memory hungry application benefitting from more of this lower
1203 memory. On the other hand, if you encounter a conflict, increasing
1204 aspace-minaddr value might solve it. Conflicts will typically
1205 manifest themselves with mmap failures in the low range of the
1206 address space. The provided address must be page aligned and must
1207 be equal or bigger to 0x1000 (4KB). To find the default value on
1208 your platform, do something such as valgrind -d -d date 2>&1 | grep
1209 -i minaddr. Values lower than 0x10000 (64KB) are known to create
1210 problems on some distributions.
1211 --valgrind-stacksize=<number> [default: 1MB]
1213 For each thread, Valgrind needs its own 'private' stack. The
1214 default size for these stacks is largely dimensioned, and so should
1215 be sufficient in most cases. In case the size is too small,
1216 Valgrind will segfault. Before segfaulting, a warning might be
1217 produced by Valgrind when approaching the limit.
1218 Use the option --valgrind-stacksize if such an (unlikely) warning
1220 is produced, or Valgrind dies due to a segmentation violation. Such
1221 segmentation violations have been seen when demangling huge C++
1222 symbols.
1223 If your application uses many threads and needs a lot of memory,
1225 you can gain some memory by reducing the size of these Valgrind
1226 stacks using the option --valgrind-stacksize.
1227 --show-emwarns=<yes|no> [default: no]
1229 When enabled, Valgrind will emit warnings about its CPU emulation
1230 in certain cases. These are usually not interesting.
1231 --require-text-symbol=:sonamepatt:fnnamepatt
1233 When a shared object whose soname matches sonamepatt is loaded into
1234 the process, examine all the text symbols it exports. If none of
1235 those match fnnamepatt, print an error message and abandon the run.
1236 This makes it possible to ensure that the run does not continue
1237 unless a given shared object contains a particular function name.
1238 Both sonamepatt and fnnamepatt can be written using the usual ?
1240 and * wildcards. For example: ":*libc.so*:foo?bar". You may use
1241 characters other than a colon to separate the two patterns. It is
1242 only important that the first character and the separator character
1243 are the same. For example, the above example could also be written
1244 "Q*libc.so*Qfoo?bar". Multiple
1245 --require-text-symbol flags are allowed, in which case shared
1246 objects that are loaded into the process will be checked against
1247 all of them.
1248 The purpose of this is to support reliable usage of marked-up
1250 libraries. For example, suppose we have a version of GCC's
1251 libgomp.so which has been marked up with annotations to support
1252 Helgrind. It is only too easy and confusing to load the wrong,
1253 un-annotated libgomp.so into the application. So the idea is: add a
1254 text symbol in the marked-up library, for example
1255 annotated_for_helgrind_3_6, and then give the flag
1256 --require-text-symbol=:*libgomp*so*:annotated_for_helgrind_3_6 so
1257 that when libgomp.so is loaded, Valgrind scans its symbol table,
1258 and if the symbol isn't present the run is aborted, rather than
1259 continuing silently with the un-marked-up library. Note that you
1260 should put the entire flag in quotes to stop shells expanding up
1261 the * and ? wildcards.
1262 --soname-synonyms=syn1=pattern1,syn2=pattern2,...
1264 When a shared library is loaded, Valgrind checks for functions in
1265 the library that must be replaced or wrapped. For example, Memcheck
1266 replaces some string and memory functions (strchr, strlen, strcpy,
1267 memchr, memcpy, memmove, etc.) with its own versions. Such
1268 replacements are normally done only in shared libraries whose
1269 soname matches a predefined soname pattern (e.g. libc.so* on
1270 linux). By default, no replacement is done for a statically linked
1271 binary or for alternative libraries, except for the allocation
1272 functions (malloc, free, calloc, memalign, realloc, operator new,
1273 operator delete, etc.) Such allocation functions are intercepted by
1274 default in any shared library or in the executable if they are
1275 exported as global symbols. This means that if a replacement
1276 allocation library such as tcmalloc is found, its functions are
1277 also intercepted by default. In some cases, the replacements allow
1278 --soname-synonyms to specify one additional synonym pattern, giving
1279 flexibility in the replacement. Or to prevent interception of all
1280 public allocation symbols.
1281 Currently, this flexibility is only allowed for the malloc related
1283 functions, using the synonym somalloc. This synonym is usable for
1284 all tools doing standard replacement of malloc related functions
1285 (e.g. memcheck, helgrind, drd, massif, dhat).
1286 • Alternate malloc library: to replace the malloc related
1288 functions in a specific alternate library with soname
1289 mymalloclib.so (and not in any others), give the option
1290 --soname-synonyms=somalloc=mymalloclib.so. A pattern can be
1291 used to match multiple libraries sonames. For example,
1292 --soname-synonyms=somalloc=*tcmalloc* will match the soname of
1293 all variants of the tcmalloc library (native, debug, profiled,
1294 ... tcmalloc variants).
1295 Note: the soname of a elf shared library can be retrieved using
1297 the readelf utility.
1298 • Replacements in a statically linked library are done by using
1300 the NONE pattern. For example, if you link with libtcmalloc.a,
1301 and only want to intercept the malloc related functions in the
1302 executable (and standard libraries) themselves, but not any
1303 other shared libraries, you can give the option
1304 --soname-synonyms=somalloc=NONE. Note that a NONE pattern will
1305 match the main executable and any shared library having no
1306 soname.
1307 • To run a "default" Firefox build for Linux, in which JEMalloc
1309 is linked in to the main executable, use
1310 --soname-synonyms=somalloc=NONE.
1311 • To only intercept allocation symbols in the default system
1313 libraries, but not in any other shared library or the
1314 executable defining public malloc or operator new related
1315 functions use a non-existing library name like
1316 --soname-synonyms=somalloc=nouserintercepts (where
1317 nouserintercepts can be any non-existing library name).
1318 • Shared library of the dynamic (runtime) linker is excluded from
1320 searching for global public symbols, such as those for the
1321 malloc related functions (identified by somalloc synonym).
1322 --progress-interval=<number> [default: 0, meaning 'disabled']
1324 This is an enhancement to Valgrind's debugging output. It is
1325 unlikely to be of interest to end users.
1326 When number is set to a non-zero value, Valgrind will print a
1328 one-line progress summary every number seconds. Valid settings for
1329 number are between 0 and 3600 inclusive. Here's some example output
1330 with number set to 10:
1331 PROGRESS: U 110s, W 113s, 97.3% CPU, EvC 414.79M, TIn 616.7k, TOut 0.5k, #thr 67
1333 PROGRESS: U 120s, W 124s, 96.8% CPU, EvC 505.27M, TIn 636.6k, TOut 3.0k, #thr 64
1334 PROGRESS: U 130s, W 134s, 97.0% CPU, EvC 574.90M, TIn 657.5k, TOut 3.0k, #thr 63
1335 Each line shows:
1337 • U: total user time
1339 • W: total wallclock time
1341 • CPU: overall average cpu use
1343 • EvC: number of event checks. An event check is a backwards
1345 branch in the simulated program, so this is a measure of
1346 forward progress of the program
1347 • TIn: number of code blocks instrumented by the JIT
1349 • TOut: number of instrumented code blocks that have been thrown
1351 away
1352 • #thr: number of threads in the program
1354 From the progress of these, it is possible to observe:
1356 • when the program is compute bound (TIn rises slowly, EvC rises
1358 rapidly)
1359 • when the program is in a spinloop (TIn/TOut fixed, EvC rises
1361 rapidly)
1362 • when the program is JIT-bound (TIn rises rapidly)
1364 • when the program is rapidly discarding code (TOut rises
1366 rapidly)
1367 • when the program is about to achieve some expected state (EvC
1369 arrives at some value you expect)
1370 • when the program is idling (U rises more slowly than W)
1372
1375 There are also some options for debugging Valgrind itself. You
1376 shouldn't need to use them in the normal run of things. If you wish to
1377 see the list, use the --help-debug option.
1378
1380 --leak-check=<no|summary|yes|full> [default: summary]
1381 When enabled, search for memory leaks when the client program
1382 finishes. If set to summary, it says how many leaks occurred. If
1383 set to full or yes, each individual leak will be shown in detail
1384 and/or counted as an error, as specified by the options
1385 --show-leak-kinds and --errors-for-leak-kinds.
1386 If --xml=yes is given, memcheck will automatically use the value
1388 --leak-check=full. You can use --show-leak-kinds=none to reduce the
1389 size of the xml output if you are not interested in the leak
1390 results.
1391 --leak-resolution=<low|med|high> [default: high]
1393 When doing leak checking, determines how willing Memcheck is to
1394 consider different backtraces to be the same for the purposes of
1395 merging multiple leaks into a single leak report. When set to low,
1396 only the first two entries need match. When med, four entries have
1397 to match. When high, all entries need to match.
1398 For hardcore leak debugging, you probably want to use
1400 --leak-resolution=high together with --num-callers=40 or some such
1401 large number.
1402 Note that the --leak-resolution setting does not affect Memcheck's
1404 ability to find leaks. It only changes how the results are
1405 presented.
1406 --show-leak-kinds=<set> [default: definite,possible]
1408 Specifies the leak kinds to show in a full leak search, in one of
1409 the following ways:
1410 • a comma separated list of one or more of definite indirect
1412 possible reachable.
1413 • all to specify the complete set (all leak kinds). It is
1415 equivalent to
1416 --show-leak-kinds=definite,indirect,possible,reachable.
1417 • none for the empty set.
1419 --errors-for-leak-kinds=<set> [default: definite,possible]
1421 Specifies the leak kinds to count as errors in a full leak search.
1422 The <set> is specified similarly to --show-leak-kinds
1423 --leak-check-heuristics=<set> [default: all]
1425 Specifies the set of leak check heuristics to be used during leak
1426 searches. The heuristics control which interior pointers to a block
1427 cause it to be considered as reachable. The heuristic set is
1428 specified in one of the following ways:
1429 • a comma separated list of one or more of stdstring length64
1431 newarray multipleinheritance.
1432 • all to activate the complete set of heuristics. It is
1434 equivalent to
1435 --leak-check-heuristics=stdstring,length64,newarray,multipleinheritance.
1436 • none for the empty set.
1438 Note that these heuristics are dependent on the layout of the
1440 objects produced by the C++ compiler. They have been tested with
1441 some gcc versions (e.g. 4.4 and 4.7). They might not work properly
1442 with other C++ compilers.
1443 --show-reachable=<yes|no> , --show-possibly-lost=<yes|no>
1445 These options provide an alternative way to specify the leak kinds
1446 to show:
1447 • --show-reachable=no --show-possibly-lost=yes is equivalent to
1449 --show-leak-kinds=definite,possible.
1450 • --show-reachable=no --show-possibly-lost=no is equivalent to
1452 --show-leak-kinds=definite.
1453 • --show-reachable=yes is equivalent to --show-leak-kinds=all.
1455 Note that --show-possibly-lost=no has no effect if
1457 --show-reachable=yes is specified.
1458 --xtree-leak=<no|yes> [no]
1460 If set to yes, the results for the leak search done at exit will be
1461 output in a 'Callgrind Format' execution tree file. Note that this
1462 automatically sets the options --leak-check=full and
1463 --show-leak-kinds=all, to allow xtree visualisation tools such as
1464 kcachegrind to select what kind to leak to visualise. The produced
1465 file will contain the following events:
1466 • RB : Reachable Bytes
1468 • PB : Possibly lost Bytes
1470 • IB : Indirectly lost Bytes
1472 • DB : Definitely lost Bytes (direct plus indirect)
1474 • DIB : Definitely Indirectly lost Bytes (subset of DB)
1476 • RBk : reachable Blocks
1478 • PBk : Possibly lost Blocks
1480 • IBk : Indirectly lost Blocks
1482 • DBk : Definitely lost Blocks
1484 The increase or decrease for all events above will also be output
1486 in the file to provide the delta (increase or decrease) between 2
1487 successive leak searches. For example, iRB is the increase of the
1488 RB event, dPBk is the decrease of PBk event. The values for the
1489 increase and decrease events will be zero for the first leak search
1490 done.
1491 See Execution Trees for a detailed explanation about execution
1493 trees.
1494 --xtree-leak-file=<filename> [default: xtleak.kcg.%p]
1496 Specifies that Valgrind should produce the xtree leak report in the
1497 specified file. Any %p, %q or %n sequences appearing in the
1498 filename are expanded in exactly the same way as they are for
1499 --log-file. See the description of --log-file for details.
1500 See Execution Trees for a detailed explanation about execution
1502 trees formats.
1503 --undef-value-errors=<yes|no> [default: yes]
1505 Controls whether Memcheck reports uses of undefined value errors.
1506 Set this to no if you don't want to see undefined value errors. It
1507 also has the side effect of speeding up Memcheck somewhat.
1508 AddrCheck (removed in Valgrind 3.1.0) functioned like Memcheck with
1509 --undef-value-errors=no.
1510 --track-origins=<yes|no> [default: no]
1512 Controls whether Memcheck tracks the origin of uninitialised
1513 values. By default, it does not, which means that although it can
1514 tell you that an uninitialised value is being used in a dangerous
1515 way, it cannot tell you where the uninitialised value came from.
1516 This often makes it difficult to track down the root problem.
1517 When set to yes, Memcheck keeps track of the origins of all
1519 uninitialised values. Then, when an uninitialised value error is
1520 reported, Memcheck will try to show the origin of the value. An
1521 origin can be one of the following four places: a heap block, a
1522 stack allocation, a client request, or miscellaneous other sources
1523 (eg, a call to brk).
1524 For uninitialised values originating from a heap block, Memcheck
1526 shows where the block was allocated. For uninitialised values
1527 originating from a stack allocation, Memcheck can tell you which
1528 function allocated the value, but no more than that -- typically it
1529 shows you the source location of the opening brace of the function.
1530 So you should carefully check that all of the function's local
1531 variables are initialised properly.
1532 Performance overhead: origin tracking is expensive. It halves
1534 Memcheck's speed and increases memory use by a minimum of 100MB,
1535 and possibly more. Nevertheless it can drastically reduce the
1536 effort required to identify the root cause of uninitialised value
1537 errors, and so is often a programmer productivity win, despite
1538 running more slowly.
1539 Accuracy: Memcheck tracks origins quite accurately. To avoid very
1541 large space and time overheads, some approximations are made. It is
1542 possible, although unlikely, that Memcheck will report an incorrect
1543 origin, or not be able to identify any origin.
1544 Note that the combination --track-origins=yes and
1546 --undef-value-errors=no is nonsensical. Memcheck checks for and
1547 rejects this combination at startup.
1548 --partial-loads-ok=<yes|no> [default: yes]
1550 Controls how Memcheck handles 32-, 64-, 128- and 256-bit naturally
1551 aligned loads from addresses for which some bytes are addressable
1552 and others are not. When yes, such loads do not produce an address
1553 error. Instead, loaded bytes originating from illegal addresses are
1554 marked as uninitialised, and those corresponding to legal addresses
1555 are handled in the normal way.
1556 When no, loads from partially invalid addresses are treated the
1558 same as loads from completely invalid addresses: an illegal-address
1559 error is issued, and the resulting bytes are marked as initialised.
1560 Note that code that behaves in this way is in violation of the ISO
1562 C/C++ standards, and should be considered broken. If at all
1563 possible, such code should be fixed.
1564 --expensive-definedness-checks=<no|auto|yes> [default: auto]
1566 Controls whether Memcheck should employ more precise but also more
1567 expensive (time consuming) instrumentation when checking the
1568 definedness of certain values. In particular, this affects the
1569 instrumentation of integer adds, subtracts and equality
1570 comparisons.
1571 Selecting --expensive-definedness-checks=yes causes Memcheck to use
1573 the most accurate analysis possible. This minimises false error
1574 rates but can cause up to 30% performance degradation.
1575 Selecting --expensive-definedness-checks=no causes Memcheck to use
1577 the cheapest instrumentation possible. This maximises performance
1578 but will normally give an unusably high false error rate.
1579 The default setting, --expensive-definedness-checks=auto, is
1581 strongly recommended. This causes Memcheck to use the minimum of
1582 expensive instrumentation needed to achieve the same false error
1583 rate as --expensive-definedness-checks=yes. It also enables an
1584 instrumentation-time analysis pass which aims to further reduce the
1585 costs of accurate instrumentation. Overall, the performance loss is
1586 generally around 5% relative to --expensive-definedness-checks=no,
1587 although this is strongly workload dependent. Note that the exact
1588 instrumentation settings in this mode are architecture dependent.
1589 --keep-stacktraces=alloc|free|alloc-and-free|alloc-then-free|none
1591 [default: alloc-and-free]
1592 Controls which stack trace(s) to keep for malloc'd and/or free'd
1593 blocks.
1594 With alloc-then-free, a stack trace is recorded at allocation time,
1596 and is associated with the block. When the block is freed, a second
1597 stack trace is recorded, and this replaces the allocation stack
1598 trace. As a result, any "use after free" errors relating to this
1599 block can only show a stack trace for where the block was freed.
1600 With alloc-and-free, both allocation and the deallocation stack
1602 traces for the block are stored. Hence a "use after free" error
1603 will show both, which may make the error easier to diagnose.
1604 Compared to alloc-then-free, this setting slightly increases
1605 Valgrind's memory use as the block contains two references instead
1606 of one.
1607 With alloc, only the allocation stack trace is recorded (and
1609 reported). With free, only the deallocation stack trace is recorded
1610 (and reported). These values somewhat decrease Valgrind's memory
1611 and cpu usage. They can be useful depending on the error types you
1612 are searching for and the level of detail you need to analyse them.
1613 For example, if you are only interested in memory leak errors, it
1614 is sufficient to record the allocation stack traces.
1615 With none, no stack traces are recorded for malloc and free
1617 operations. If your program allocates a lot of blocks and/or
1618 allocates/frees from many different stack traces, this can
1619 significantly decrease cpu and/or memory required. Of course, few
1620 details will be reported for errors related to heap blocks.
1621 Note that once a stack trace is recorded, Valgrind keeps the stack
1623 trace in memory even if it is not referenced by any block. Some
1624 programs (for example, recursive algorithms) can generate a huge
1625 number of stack traces. If Valgrind uses too much memory in such
1626 circumstances, you can reduce the memory required with the options
1627 --keep-stacktraces and/or by using a smaller value for the option
1628 --num-callers.
1629 If you want to use --xtree-memory=full memory profiling (see
1631 Execution Trees), then you cannot specify --keep-stacktraces=free
1632 or --keep-stacktraces=none.
1633 --freelist-vol=<number> [default: 20000000]
1635 When the client program releases memory using free (in C) or delete
1636 (C++), that memory is not immediately made available for
1637 re-allocation. Instead, it is marked inaccessible and placed in a
1638 queue of freed blocks. The purpose is to defer as long as possible
1639 the point at which freed-up memory comes back into circulation.
1640 This increases the chance that Memcheck will be able to detect
1641 invalid accesses to blocks for some significant period of time
1642 after they have been freed.
1643 This option specifies the maximum total size, in bytes, of the
1645 blocks in the queue. The default value is twenty million bytes.
1646 Increasing this increases the total amount of memory used by
1647 Memcheck but may detect invalid uses of freed blocks which would
1648 otherwise go undetected.
1649 --freelist-big-blocks=<number> [default: 1000000]
1651 When making blocks from the queue of freed blocks available for
1652 re-allocation, Memcheck will in priority re-circulate the blocks
1653 with a size greater or equal to --freelist-big-blocks. This ensures
1654 that freeing big blocks (in particular freeing blocks bigger than
1655 --freelist-vol) does not immediately lead to a re-circulation of
1656 all (or a lot of) the small blocks in the free list. In other
1657 words, this option increases the likelihood to discover dangling
1658 pointers for the "small" blocks, even when big blocks are freed.
1659 Setting a value of 0 means that all the blocks are re-circulated in
1661 a FIFO order.
1662 --workaround-gcc296-bugs=<yes|no> [default: no]
1664 When enabled, assume that reads and writes some small distance
1665 below the stack pointer are due to bugs in GCC 2.96, and does not
1666 report them. The "small distance" is 256 bytes by default. Note
1667 that GCC 2.96 is the default compiler on some ancient Linux
1668 distributions (RedHat 7.X) and so you may need to use this option.
1669 Do not use it if you do not have to, as it can cause real errors to
1670 be overlooked. A better alternative is to use a more recent GCC in
1671 which this bug is fixed.
1672 You may also need to use this option when working with GCC 3.X or
1674 4.X on 32-bit PowerPC Linux. This is because GCC generates code
1675 which occasionally accesses below the stack pointer, particularly
1676 for floating-point to/from integer conversions. This is in
1677 violation of the 32-bit PowerPC ELF specification, which makes no
1678 provision for locations below the stack pointer to be accessible.
1679 This option is deprecated as of version 3.12 and may be removed
1681 from future versions. You should instead use
1682 --ignore-range-below-sp to specify the exact range of offsets below
1683 the stack pointer that should be ignored. A suitable equivalent is
1684 --ignore-range-below-sp=1024-1.
1685 --ignore-range-below-sp=<number>-<number>
1687 This is a more general replacement for the deprecated
1688 --workaround-gcc296-bugs option. When specified, it causes Memcheck
1689 not to report errors for accesses at the specified offsets below
1690 the stack pointer. The two offsets must be positive decimal numbers
1691 and -- somewhat counterintuitively -- the first one must be larger,
1692 in order to imply a non-wraparound address range to ignore. For
1693 example, to ignore 4 byte accesses at 8192 bytes below the stack
1694 pointer, use --ignore-range-below-sp=8192-8189. Only one range may
1695 be specified.
1696 --show-mismatched-frees=<yes|no> [default: yes]
1698 When enabled, Memcheck checks that heap blocks are deallocated
1699 using a function that matches the allocating function. That is, it
1700 expects free to be used to deallocate blocks allocated by malloc,
1701 delete for blocks allocated by new, and delete[] for blocks
1702 allocated by new[]. If a mismatch is detected, an error is
1703 reported. This is in general important because in some
1704 environments, freeing with a non-matching function can cause
1705 crashes.
1706 There is however a scenario where such mismatches cannot be
1708 avoided. That is when the user provides implementations of
1709 new/new[] that call malloc and of delete/delete[] that call free,
1710 and these functions are asymmetrically inlined. For example,
1711 imagine that delete[] is inlined but new[] is not. The result is
1712 that Memcheck "sees" all delete[] calls as direct calls to free,
1713 even when the program source contains no mismatched calls.
1714 This causes a lot of confusing and irrelevant error reports.
1716 --show-mismatched-frees=no disables these checks. It is not
1717 generally advisable to disable them, though, because you may miss
1718 real errors as a result.
1719 --ignore-ranges=0xPP-0xQQ[,0xRR-0xSS]
1721 Any ranges listed in this option (and multiple ranges can be
1722 specified, separated by commas) will be ignored by Memcheck's
1723 addressability checking.
1724 --malloc-fill=<hexnumber>
1726 Fills blocks allocated by malloc, new, etc, but not by calloc, with
1727 the specified byte. This can be useful when trying to shake out
1728 obscure memory corruption problems. The allocated area is still
1729 regarded by Memcheck as undefined -- this option only affects its
1730 contents. Note that --malloc-fill does not affect a block of memory
1731 when it is used as argument to client requests
1732 VALGRIND_MEMPOOL_ALLOC or VALGRIND_MALLOCLIKE_BLOCK.
1733 --free-fill=<hexnumber>
1735 Fills blocks freed by free, delete, etc, with the specified byte
1736 value. This can be useful when trying to shake out obscure memory
1737 corruption problems. The freed area is still regarded by Memcheck
1738 as not valid for access -- this option only affects its contents.
1739 Note that --free-fill does not affect a block of memory when it is
1740 used as argument to client requests VALGRIND_MEMPOOL_FREE or
1741 VALGRIND_FREELIKE_BLOCK.
1742
1744 --I1=<size>,<associativity>,<line size>
1745 Specify the size, associativity and line size of the level 1
1746 instruction cache.
1747 --D1=<size>,<associativity>,<line size>
1749 Specify the size, associativity and line size of the level 1 data
1750 cache.
1751 --LL=<size>,<associativity>,<line size>
1753 Specify the size, associativity and line size of the last-level
1754 cache.
1755 --cache-sim=no|yes [yes]
1757 Enables or disables collection of cache access and miss counts.
1758 --branch-sim=no|yes [no]
1760 Enables or disables collection of branch instruction and
1761 misprediction counts. By default this is disabled as it slows
1762 Cachegrind down by approximately 25%. Note that you cannot specify
1763 --cache-sim=no and --branch-sim=no together, as that would leave
1764 Cachegrind with no information to collect.
1765 --cachegrind-out-file=<file>
1767 Write the profile data to file rather than to the default output
1768 file, cachegrind.out.<pid>. The %p and %q format specifiers can be
1769 used to embed the process ID and/or the contents of an environment
1770 variable in the name, as is the case for the core option
1771 --log-file.
1772
1774 --callgrind-out-file=<file>
1775 Write the profile data to file rather than to the default output
1776 file, callgrind.out.<pid>. The %p and %q format specifiers can be
1777 used to embed the process ID and/or the contents of an environment
1778 variable in the name, as is the case for the core option
1779 --log-file. When multiple dumps are made, the file name is modified
1780 further; see below.
1781 --dump-line=<no|yes> [default: yes]
1783 This specifies that event counting should be performed at source
1784 line granularity. This allows source annotation for sources which
1785 are compiled with debug information (-g).
1786 --dump-instr=<no|yes> [default: no]
1788 This specifies that event counting should be performed at
1789 per-instruction granularity. This allows for assembly code
1790 annotation. Currently the results can only be displayed by
1791 KCachegrind.
1792 --compress-strings=<no|yes> [default: yes]
1794 This option influences the output format of the profile data. It
1795 specifies whether strings (file and function names) should be
1796 identified by numbers. This shrinks the file, but makes it more
1797 difficult for humans to read (which is not recommended in any
1798 case).
1799 --compress-pos=<no|yes> [default: yes]
1801 This option influences the output format of the profile data. It
1802 specifies whether numerical positions are always specified as
1803 absolute values or are allowed to be relative to previous numbers.
1804 This shrinks the file size.
1805 --combine-dumps=<no|yes> [default: no]
1807 When enabled, when multiple profile data parts are to be generated
1808 these parts are appended to the same output file. Not recommended.
1809 --dump-every-bb=<count> [default: 0, never]
1811 Dump profile data every count basic blocks. Whether a dump is
1812 needed is only checked when Valgrind's internal scheduler is run.
1813 Therefore, the minimum setting useful is about 100000. The count is
1814 a 64-bit value to make long dump periods possible.
1815 --dump-before=<function>
1817 Dump when entering function.
1818 --zero-before=<function>
1820 Zero all costs when entering function.
1821 --dump-after=<function>
1823 Dump when leaving function.
1824 --instr-atstart=<yes|no> [default: yes]
1826 Specify if you want Callgrind to start simulation and profiling
1827 from the beginning of the program. When set to no, Callgrind will
1828 not be able to collect any information, including calls, but it
1829 will have at most a slowdown of around 4, which is the minimum
1830 Valgrind overhead. Instrumentation can be interactively enabled via
1831 callgrind_control -i on.
1832 Note that the resulting call graph will most probably not contain
1834 main, but will contain all the functions executed after
1835 instrumentation was enabled. Instrumentation can also be
1836 programmatically enabled/disabled. See the Callgrind include file
1837 callgrind.h for the macro you have to use in your source code.
1838 For cache simulation, results will be less accurate when switching
1840 on instrumentation later in the program run, as the simulator
1841 starts with an empty cache at that moment. Switch on event
1842 collection later to cope with this error.
1843 --collect-atstart=<yes|no> [default: yes]
1845 Specify whether event collection is enabled at beginning of the
1846 profile run.
1847 To only look at parts of your program, you have two possibilities:
1849 1. Zero event counters before entering the program part you want
1851 to profile, and dump the event counters to a file after leaving
1852 that program part.
1853 2. Switch on/off collection state as needed to only see event
1855 counters happening while inside of the program part you want to
1856 profile.
1857 The second option can be used if the program part you want to
1859 profile is called many times. Option 1, i.e. creating a lot of
1860 dumps is not practical here.
1861 Collection state can be toggled at entry and exit of a given
1863 function with the option --toggle-collect. If you use this option,
1864 collection state should be disabled at the beginning. Note that the
1865 specification of --toggle-collect implicitly sets
1866 --collect-state=no.
1867 Collection state can be toggled also by inserting the client
1869 request CALLGRIND_TOGGLE_COLLECT ; at the needed code positions.
1870 --toggle-collect=<function>
1872 Toggle collection on entry/exit of function.
1873 --collect-jumps=<no|yes> [default: no]
1875 This specifies whether information for (conditional) jumps should
1876 be collected. As above, callgrind_annotate currently is not able to
1877 show you the data. You have to use KCachegrind to get jump arrows
1878 in the annotated code.
1879 --collect-systime=<no|yes|msec|usec|nsec> [default: no]
1881 This specifies whether information for system call times should be
1882 collected.
1883 The value no indicates to record no system call information.
1885 The other values indicate to record the number of system calls done
1887 (sysCount event) and the elapsed time (sysTime event) spent in
1888 system calls. The --collect-systime value gives the unit used for
1889 sysTime : milli seconds, micro seconds or nano seconds. With the
1890 value nsec, callgrind also records the cpu time spent during system
1891 calls (sysCpuTime).
1892 The value yes is a synonym of msec. The value nsec is not supported
1894 on Darwin.
1895 --collect-bus=<no|yes> [default: no]
1897 This specifies whether the number of global bus events executed
1898 should be collected. The event type "Ge" is used for these events.
1899 --cache-sim=<yes|no> [default: no]
1901 Specify if you want to do full cache simulation. By default, only
1902 instruction read accesses will be counted ("Ir"). With cache
1903 simulation, further event counters are enabled: Cache misses on
1904 instruction reads ("I1mr"/"ILmr"), data read accesses ("Dr") and
1905 related cache misses ("D1mr"/"DLmr"), data write accesses ("Dw")
1906 and related cache misses ("D1mw"/"DLmw"). For more information, see
1907 Cachegrind: a cache and branch-prediction profiler.
1908 --branch-sim=<yes|no> [default: no]
1910 Specify if you want to do branch prediction simulation. Further
1911 event counters are enabled: Number of executed conditional branches
1912 and related predictor misses ("Bc"/"Bcm"), executed indirect jumps
1913 and related misses of the jump address predictor ("Bi"/"Bim").
1914
1916 --free-is-write=no|yes [default: no]
1917 When enabled (not the default), Helgrind treats freeing of heap
1918 memory as if the memory was written immediately before the free.
1919 This exposes races where memory is referenced by one thread, and
1920 freed by another, but there is no observable synchronisation event
1921 to ensure that the reference happens before the free.
1922 This functionality is new in Valgrind 3.7.0, and is regarded as
1924 experimental. It is not enabled by default because its interaction
1925 with custom memory allocators is not well understood at present.
1926 User feedback is welcomed.
1927 --track-lockorders=no|yes [default: yes]
1929 When enabled (the default), Helgrind performs lock order
1930 consistency checking. For some buggy programs, the large number of
1931 lock order errors reported can become annoying, particularly if
1932 you're only interested in race errors. You may therefore find it
1933 helpful to disable lock order checking.
1934 --history-level=none|approx|full [default: full]
1936 --history-level=full (the default) causes Helgrind collects enough
1937 information about "old" accesses that it can produce two stack
1938 traces in a race report -- both the stack trace for the current
1939 access, and the trace for the older, conflicting access. To limit
1940 memory usage, "old" accesses stack traces are limited to a maximum
1941 of 8 entries, even if --num-callers value is bigger.
1942 Collecting such information is expensive in both speed and memory,
1944 particularly for programs that do many inter-thread synchronisation
1945 events (locks, unlocks, etc). Without such information, it is more
1946 difficult to track down the root causes of races. Nonetheless, you
1947 may not need it in situations where you just want to check for the
1948 presence or absence of races, for example, when doing regression
1949 testing of a previously race-free program.
1950 --history-level=none is the opposite extreme. It causes Helgrind
1952 not to collect any information about previous accesses. This can be
1953 dramatically faster than --history-level=full.
1954 --history-level=approx provides a compromise between these two
1956 extremes. It causes Helgrind to show a full trace for the later
1957 access, and approximate information regarding the earlier access.
1958 This approximate information consists of two stacks, and the
1959 earlier access is guaranteed to have occurred somewhere between
1960 program points denoted by the two stacks. This is not as useful as
1961 showing the exact stack for the previous access (as
1962 --history-level=full does), but it is better than nothing, and it
1963 is almost as fast as --history-level=none.
1964 --delta-stacktrace=no|yes [default: yes on linux amd64/x86]
1966 This flag only has any effect at --history-level=full.
1967 --delta-stacktrace configures the way Helgrind captures the
1969 stacktraces for the option --history-level=full. Such a stacktrace
1970 is typically needed each time a new piece of memory is read or
1971 written in a basic block of instructions.
1972 --delta-stacktrace=no causes Helgrind to compute a full history
1974 stacktrace from the unwind info each time a stacktrace is needed.
1975 --delta-stacktrace=yes indicates to Helgrind to derive a new
1977 stacktrace from the previous stacktrace, as long as there was no
1978 call instruction, no return instruction, or any other instruction
1979 changing the call stack since the previous stacktrace was captured.
1980 If no such instruction was executed, the new stacktrace can be
1981 derived from the previous stacktrace by just changing the top frame
1982 to the current program counter. This option can speed up Helgrind
1983 by 25% when using --history-level=full.
1984 The following aspects have to be considered when using
1986 --delta-stacktrace=yes :
1987 • In some cases (for example in a function prologue), the
1989 valgrind unwinder might not properly unwind the stack, due to
1990 some limitations and/or due to wrong unwind info. When using
1991 --delta-stacktrace=yes, the wrong stack trace captured in the
1992 function prologue will be kept till the next call or return.
1993 • On the other hand, --delta-stacktrace=yes sometimes helps to
1995 obtain a correct stacktrace, for example when the unwind info
1996 allows a correct stacktrace to be done in the beginning of the
1997 sequence, but not later on in the instruction sequence.
1998 • Determining which instructions are changing the callstack is
2000 partially based on platform dependent heuristics, which have to
2001 be tuned/validated specifically for the platform. Also,
2002 unwinding in a function prologue must be good enough to allow
2003 using --delta-stacktrace=yes. Currently, the option
2004 --delta-stacktrace=yes has been reasonably validated only on
2005 linux x86 32 bits and linux amd64 64 bits. For more details
2006 about how to validate --delta-stacktrace=yes, see debug option
2007 --hg-sanity-flags and the function check_cached_rcec_ok in
2008 libhb_core.c.
2009 --conflict-cache-size=N [default: 1000000]
2012 This flag only has any effect at --history-level=full.
2013 Information about "old" conflicting accesses is stored in a cache
2015 of limited size, with LRU-style management. This is necessary
2016 because it isn't practical to store a stack trace for every single
2017 memory access made by the program. Historical information on not
2018 recently accessed locations is periodically discarded, to free up
2019 space in the cache.
2020 This option controls the size of the cache, in terms of the number
2022 of different memory addresses for which conflicting access
2023 information is stored. If you find that Helgrind is showing race
2024 errors with only one stack instead of the expected two stacks, try
2025 increasing this value.
2026 The minimum value is 10,000 and the maximum is 30,000,000 (thirty
2028 times the default value). Increasing the value by 1 increases
2029 Helgrind's memory requirement by very roughly 100 bytes, so the
2030 maximum value will easily eat up three extra gigabytes or so of
2031 memory.
2032 --check-stack-refs=no|yes [default: yes]
2034 By default Helgrind checks all data memory accesses made by your
2035 program. This flag enables you to skip checking for accesses to
2036 thread stacks (local variables). This can improve performance, but
2037 comes at the cost of missing races on stack-allocated data.
2038 --ignore-thread-creation=<yes|no> [default: no]
2040 Controls whether all activities during thread creation should be
2041 ignored. By default enabled only on Solaris. Solaris provides
2042 higher throughput, parallelism and scalability than other operating
2043 systems, at the cost of more fine-grained locking activity. This
2044 means for example that when a thread is created under glibc, just
2045 one big lock is used for all thread setup. Solaris libc uses
2046 several fine-grained locks and the creator thread resumes its
2047 activities as soon as possible, leaving for example stack and TLS
2048 setup sequence to the created thread. This situation confuses
2049 Helgrind as it assumes there is some false ordering in place
2050 between creator and created thread; and therefore many types of
2051 race conditions in the application would not be reported. To
2052 prevent such false ordering, this command line option is set to yes
2053 by default on Solaris. All activity (loads, stores, client
2054 requests) is therefore ignored during:
2055 • pthread_create() call in the creator thread
2057 • thread creation phase (stack and TLS setup) in the created
2059 thread
2060 Also new memory allocated during thread creation is untracked, that
2062 is race reporting is suppressed there. DRD does the same thing
2063 implicitly. This is necessary because Solaris libc caches many
2064 objects and reuses them for different threads and that confuses
2065 Helgrind.
2066
2068 --check-stack-var=<yes|no> [default: no]
2069 Controls whether DRD detects data races on stack variables.
2070 Verifying stack variables is disabled by default because most
2071 programs do not share stack variables over threads.
2072 --exclusive-threshold=<n> [default: off]
2074 Print an error message if any mutex or writer lock has been held
2075 longer than the time specified in milliseconds. This option enables
2076 the detection of lock contention.
2077 --join-list-vol=<n> [default: 10]
2079 Data races that occur between a statement at the end of one thread
2080 and another thread can be missed if memory access information is
2081 discarded immediately after a thread has been joined. This option
2082 allows one to specify for how many joined threads memory access
2083 information should be retained.
2084 --first-race-only=<yes|no> [default: no]
2086 Whether to report only the first data race that has been detected
2087 on a memory location or all data races that have been detected on a
2088 memory location.
2089 --free-is-write=<yes|no> [default: no]
2091 Whether to report races between accessing memory and freeing
2092 memory. Enabling this option may cause DRD to run slightly slower.
2093 Notes:
2094 • Don't enable this option when using custom memory allocators
2096 that use the VG_USERREQ__MALLOCLIKE_BLOCK and
2097 VG_USERREQ__FREELIKE_BLOCK because that would result in false
2098 positives.
2099 • Don't enable this option when using reference-counted objects
2101 because that will result in false positives, even when that
2102 code has been annotated properly with ANNOTATE_HAPPENS_BEFORE
2103 and ANNOTATE_HAPPENS_AFTER. See e.g. the output of the
2104 following command for an example: valgrind --tool=drd
2105 --free-is-write=yes drd/tests/annotate_smart_pointer.
2106 --report-signal-unlocked=<yes|no> [default: yes]
2108 Whether to report calls to pthread_cond_signal and
2109 pthread_cond_broadcast where the mutex associated with the signal
2110 through pthread_cond_wait or pthread_cond_timed_waitis not locked
2111 at the time the signal is sent. Sending a signal without holding a
2112 lock on the associated mutex is a common programming error which
2113 can cause subtle race conditions and unpredictable behavior. There
2114 exist some uncommon synchronization patterns however where it is
2115 safe to send a signal without holding a lock on the associated
2116 mutex.
2117 --segment-merging=<yes|no> [default: yes]
2119 Controls segment merging. Segment merging is an algorithm to limit
2120 memory usage of the data race detection algorithm. Disabling
2121 segment merging may improve the accuracy of the so-called 'other
2122 segments' displayed in race reports but can also trigger an out of
2123 memory error.
2124 --segment-merging-interval=<n> [default: 10]
2126 Perform segment merging only after the specified number of new
2127 segments have been created. This is an advanced configuration
2128 option that allows one to choose whether to minimize DRD's memory
2129 usage by choosing a low value or to let DRD run faster by choosing
2130 a slightly higher value. The optimal value for this parameter
2131 depends on the program being analyzed. The default value works well
2132 for most programs.
2133 --shared-threshold=<n> [default: off]
2135 Print an error message if a reader lock has been held longer than
2136 the specified time (in milliseconds). This option enables the
2137 detection of lock contention.
2138 --show-confl-seg=<yes|no> [default: yes]
2140 Show conflicting segments in race reports. Since this information
2141 can help to find the cause of a data race, this option is enabled
2142 by default. Disabling this option makes the output of DRD more
2143 compact.
2144 --show-stack-usage=<yes|no> [default: no]
2146 Print stack usage at thread exit time. When a program creates a
2147 large number of threads it becomes important to limit the amount of
2148 virtual memory allocated for thread stacks. This option makes it
2149 possible to observe how much stack memory has been used by each
2150 thread of the client program. Note: the DRD tool itself allocates
2151 some temporary data on the client thread stack. The space necessary
2152 for this temporary data must be allocated by the client program
2153 when it allocates stack memory, but is not included in stack usage
2154 reported by DRD.
2155 --ignore-thread-creation=<yes|no> [default: no]
2157 Controls whether all activities during thread creation should be
2158 ignored. By default enabled only on Solaris. Solaris provides
2159 higher throughput, parallelism and scalability than other operating
2160 systems, at the cost of more fine-grained locking activity. This
2161 means for example that when a thread is created under glibc, just
2162 one big lock is used for all thread setup. Solaris libc uses
2163 several fine-grained locks and the creator thread resumes its
2164 activities as soon as possible, leaving for example stack and TLS
2165 setup sequence to the created thread. This situation confuses DRD
2166 as it assumes there is some false ordering in place between creator
2167 and created thread; and therefore many types of race conditions in
2168 the application would not be reported. To prevent such false
2169 ordering, this command line option is set to yes by default on
2170 Solaris. All activity (loads, stores, client requests) is therefore
2171 ignored during:
2172 • pthread_create() call in the creator thread
2174 • thread creation phase (stack and TLS setup) in the created
2176 thread
2177 --trace-addr=<address> [default: none]
2179 Trace all load and store activity for the specified address. This
2180 option may be specified more than once.
2181 --ptrace-addr=<address> [default: none]
2183 Trace all load and store activity for the specified address and
2184 keep doing that even after the memory at that address has been
2185 freed and reallocated.
2186 --trace-alloc=<yes|no> [default: no]
2188 Trace all memory allocations and deallocations. May produce a huge
2189 amount of output.
2190 --trace-barrier=<yes|no> [default: no]
2192 Trace all barrier activity.
2193 --trace-cond=<yes|no> [default: no]
2195 Trace all condition variable activity.
2196 --trace-fork-join=<yes|no> [default: no]
2198 Trace all thread creation and all thread termination events.
2199 --trace-hb=<yes|no> [default: no]
2201 Trace execution of the ANNOTATE_HAPPENS_BEFORE(),
2202 ANNOTATE_HAPPENS_AFTER() and ANNOTATE_HAPPENS_DONE() client
2203 requests.
2204 --trace-mutex=<yes|no> [default: no]
2206 Trace all mutex activity.
2207 --trace-rwlock=<yes|no> [default: no]
2209 Trace all reader-writer lock activity.
2210 --trace-semaphore=<yes|no> [default: no]
2212 Trace all semaphore activity.
2213
2215 --heap=<yes|no> [default: yes]
2216 Specifies whether heap profiling should be done.
2217 --heap-admin=<size> [default: 8]
2219 If heap profiling is enabled, gives the number of administrative
2220 bytes per block to use. This should be an estimate of the average,
2221 since it may vary. For example, the allocator used by glibc on
2222 Linux requires somewhere between 4 to 15 bytes per block, depending
2223 on various factors. That allocator also requires admin space for
2224 freed blocks, but Massif cannot account for this.
2225 --stacks=<yes|no> [default: no]
2227 Specifies whether stack profiling should be done. This option slows
2228 Massif down greatly, and so is off by default. Note that Massif
2229 assumes that the main stack has size zero at start-up. This is not
2230 true, but doing otherwise accurately is difficult. Furthermore,
2231 starting at zero better indicates the size of the part of the main
2232 stack that a user program actually has control over.
2233 --pages-as-heap=<yes|no> [default: no]
2235 Tells Massif to profile memory at the page level rather than at the
2236 malloc'd block level. See above for details.
2237 --depth=<number> [default: 30]
2239 Maximum depth of the allocation trees recorded for detailed
2240 snapshots. Increasing it will make Massif run somewhat more slowly,
2241 use more memory, and produce bigger output files.
2242 --alloc-fn=<name>
2244 Functions specified with this option will be treated as though they
2245 were a heap allocation function such as malloc. This is useful for
2246 functions that are wrappers to malloc or new, which can fill up the
2247 allocation trees with uninteresting information. This option can be
2248 specified multiple times on the command line, to name multiple
2249 functions.
2250 Note that the named function will only be treated this way if it is
2252 the top entry in a stack trace, or just below another function
2253 treated this way. For example, if you have a function malloc1 that
2254 wraps malloc, and malloc2 that wraps malloc1, just specifying
2255 --alloc-fn=malloc2 will have no effect. You need to specify
2256 --alloc-fn=malloc1 as well. This is a little inconvenient, but the
2257 reason is that checking for allocation functions is slow, and it
2258 saves a lot of time if Massif can stop looking through the stack
2259 trace entries as soon as it finds one that doesn't match rather
2260 than having to continue through all the entries.
2261 Note that C++ names are demangled. Note also that overloaded C++
2263 names must be written in full. Single quotes may be necessary to
2264 prevent the shell from breaking them up. For example:
2265 --alloc-fn='operator new(unsigned, std::nothrow_t const&)'
2267 --ignore-fn=<name>
2270 Any direct heap allocation (i.e. a call to malloc, new, etc, or a
2271 call to a function named by an --alloc-fn option) that occurs in a
2272 function specified by this option will be ignored. This is mostly
2273 useful for testing purposes. This option can be specified multiple
2274 times on the command line, to name multiple functions.
2275 Any realloc of an ignored block will also be ignored, even if the
2277 realloc call does not occur in an ignored function. This avoids the
2278 possibility of negative heap sizes if ignored blocks are shrunk
2279 with realloc.
2280 The rules for writing C++ function names are the same as for
2282 --alloc-fn above.
2283 --threshold=<m.n> [default: 1.0]
2285 The significance threshold for heap allocations, as a percentage of
2286 total memory size. Allocation tree entries that account for less
2287 than this will be aggregated. Note that this should be specified in
2288 tandem with ms_print's option of the same name.
2289 --peak-inaccuracy=<m.n> [default: 1.0]
2291 Massif does not necessarily record the actual global memory
2292 allocation peak; by default it records a peak only when the global
2293 memory allocation size exceeds the previous peak by at least 1.0%.
2294 This is because there can be many local allocation peaks along the
2295 way, and doing a detailed snapshot for every one would be expensive
2296 and wasteful, as all but one of them will be later discarded. This
2297 inaccuracy can be changed (even to 0.0%) via this option, but
2298 Massif will run drastically slower as the number approaches zero.
2299 --time-unit=<i|ms|B> [default: i]
2301 The time unit used for the profiling. There are three
2302 possibilities: instructions executed (i), which is good for most
2303 cases; real (wallclock) time (ms, i.e. milliseconds), which is
2304 sometimes useful; and bytes allocated/deallocated on the heap
2305 and/or stack (B), which is useful for very short-run programs, and
2306 for testing purposes, because it is the most reproducible across
2307 different machines.
2308 --detailed-freq=<n> [default: 10]
2310 Frequency of detailed snapshots. With --detailed-freq=1, every
2311 snapshot is detailed.
2312 --max-snapshots=<n> [default: 100]
2314 The maximum number of snapshots recorded. If set to N, for all
2315 programs except very short-running ones, the final number of
2316 snapshots will be between N/2 and N.
2317 --massif-out-file=<file> [default: massif.out.%p]
2319 Write the profile data to file rather than to the default output
2320 file, massif.out.<pid>. The %p and %q format specifiers can be used
2321 to embed the process ID and/or the contents of an environment
2322 variable in the name, as is the case for the core option
2323 --log-file.
2324
2326 --bb-out-file=<name> [default: bb.out.%p]
2327 This option selects the name of the basic block vector file. The %p
2328 and %q format specifiers can be used to embed the process ID and/or
2329 the contents of an environment variable in the name, as is the case
2330 for the core option --log-file.
2331 --pc-out-file=<name> [default: pc.out.%p]
2333 This option selects the name of the PC file. This file holds
2334 program counter addresses and function name info for the various
2335 basic blocks. This can be used in conjunction with the basic block
2336 vector file to fast-forward via function names instead of just
2337 instruction counts. The %p and %q format specifiers can be used to
2338 embed the process ID and/or the contents of an environment variable
2339 in the name, as is the case for the core option --log-file.
2340 --interval-size=<number> [default: 100000000]
2342 This option selects the size of the interval to use. The default is
2343 100 million instructions, which is a commonly used value. Other
2344 sizes can be used; smaller intervals can help programs with
2345 finer-grained phases. However smaller interval size can lead to
2346 accuracy issues due to warm-up effects (When fast-forwarding the
2347 various architectural features will be un-initialized, and it will
2348 take some number of instructions before they "warm up" to the state
2349 a full simulation would be at without the fast-forwarding. Large
2350 interval sizes tend to mitigate this.)
2351 --instr-count-only [default: no]
2353 This option tells the tool to only display instruction count
2354 totals, and to not generate the actual basic block vector file.
2355 This is useful for debugging, and for gathering instruction count
2356 info without generating the large basic block vector files.
2357
2359 --basic-counts=<no|yes> [default: yes]
2360 When enabled, Lackey prints the following statistics and
2361 information about the execution of the client program:
2362 1. The number of calls to the function specified by the --fnname
2364 option (the default is main). If the program has had its
2365 symbols stripped, the count will always be zero.
2366 2. The number of conditional branches encountered and the number
2368 and proportion of those taken.
2369 3. The number of superblocks entered and completed by the program.
2371 Note that due to optimisations done by the JIT, this is not at
2372 all an accurate value.
2373 4. The number of guest (x86, amd64, ppc, etc.) instructions and IR
2375 statements executed. IR is Valgrind's RISC-like intermediate
2376 representation via which all instrumentation is done.
2377 5. Ratios between some of these counts.
2379 6. The exit code of the client program.
2381 --detailed-counts=<no|yes> [default: no]
2383 When enabled, Lackey prints a table containing counts of loads,
2384 stores and ALU operations, differentiated by their IR types. The IR
2385 types are identified by their IR name ("I1", "I8", ... "I128",
2386 "F32", "F64", and "V128").
2387 --trace-mem=<no|yes> [default: no]
2389 When enabled, Lackey prints the size and address of almost every
2390 memory access made by the program. See the comments at the top of
2391 the file lackey/lk_main.c for details about the output format, how
2392 it works, and inaccuracies in the address trace. Note that this
2393 option produces immense amounts of output.
2394 --trace-superblocks=<no|yes> [default: no]
2396 When enabled, Lackey prints out the address of every superblock (a
2397 single entry, multiple exit, linear chunk of code) executed by the
2398 program. This is primarily of interest to Valgrind developers. See
2399 the comments at the top of the file lackey/lk_main.c for details
2400 about the output format. Note that this option produces large
2401 amounts of output.
2402 --fnname=<name> [default: main]
2404 Changes the function for which calls are counted when
2405 --basic-counts=yes is specified.
2406
2408 Valgrind supports the downloading of debuginfo files via debuginfod, an
2409 HTTP server for distributing ELF/DWARF debugging information. When a
2410 debuginfo file cannot be found locally, Valgrind is able to query
2411 debuginfod servers for the file using the file's build-id.
2412 In order to use this feature debuginfod-find must be installed and the
2414 $DEBUGINFOD_URLS environment variable must contain space-separated URLs
2415 of debuginfod servers. Valgrind does not support debuginfod-find
2416 verbose output that is normally enabled with $DEBUGINFOD_PROGRESS and
2417 $DEBUGINFOD_VERBOSE. These environment variables will be ignored. This
2418 feature is supported on Linux only.
2419 For more information regarding debuginfod, see Elfutils Debuginfod[1] .
2421
2423 cg_annotate(1), callgrind_annotate(1), callgrind_control(1),
2424 ms_print(1), $INSTALL/share/doc/valgrind/html/index.html or
2425 http://www.valgrind.org/docs/manual/index.html, Debugging your program
2426 using Valgrind's gdbserver and GDB[2] vgdb[3], Valgrind monitor
2427 commands[4], The Commentary[5], Scheduling and Multi-Thread
2428 Performance[6], Cachegrind: a cache and branch-prediction profiler[7].
2429 Execution Trees[8]
2430
2432 See the AUTHORS file in the valgrind distribution for a comprehensive
2433 list of authors.
2434 This manpage was written by Andres Roldan <aroldan@debian.org> and the
2436 Valgrind developers.
2437
2439 1. Elfutils Debuginfod
2440 https://sourceware.org/elfutils/Debuginfod.html
2441 2. Debugging your program using Valgrind's gdbserver and GDB
2443 http://www.valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.gdbserver
2444 3. vgdb
2446 http://www.valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.vgdb
2447 4. Valgrind monitor commands
2449 http://www.valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.valgrind-monitor-commands
2450 5. The Commentary
2452 http://www.valgrind.org/docs/manual/manual-core.html#manual-core.comment
2453 6. Scheduling and Multi-Thread Performance
2455 http://www.valgrind.org/docs/manual/manual-core.html#manual-core.pthreads_perf_sched
2456 7. Cachegrind: a cache and branch-prediction profiler
2458 http://www.valgrind.org/docs/manual/cg-manual.html
2459 8. Execution Trees
2461 http://www.valgrind.org/docs/manual/manual-core.html#manual-core.xtree
2462Release 3.20.0 10/24/2022 VALGRIND(1)