1VALGRIND(1) Release 3.13.0 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, lackey, none, exp-sgcheck,
33 exp-bbv, exp-dhat, etc.
34
36 These options work with all tools.
37 -h --help
39 Show help for all options, both for the core and for the selected
40 tool. If the option is repeated it is equivalent to giving
41 --help-debug.
42 --help-debug
44 Same as --help, but also lists debugging options which usually are
45 only of use to Valgrind's developers.
46 --version
48 Show the version number of the Valgrind core. Tools can have their
49 own version numbers. There is a scheme in place to ensure that
50 tools only execute when the core version is one they are known to
51 work with. This was done to minimise the chances of strange
52 problems arising from tool-vs-core version incompatibilities.
53 -q, --quiet
55 Run silently, and only print error messages. Useful if you are
56 running regression tests or have some other automated test
57 machinery.
58 -v, --verbose
60 Be more verbose. Gives extra information on various aspects of your
61 program, such as: the shared objects loaded, the suppressions used,
62 the progress of the instrumentation and execution engines, and
63 warnings about unusual behaviour. Repeating the option increases
64 the verbosity level.
65 --trace-children=<yes|no> [default: no]
67 When enabled, Valgrind will trace into sub-processes initiated via
68 the exec system call. This is necessary for multi-process programs.
69 Note that Valgrind does trace into the child of a fork (it would be
71 difficult not to, since fork makes an identical copy of a process),
72 so this option is arguably badly named. However, most children of
73 fork calls immediately call exec anyway.
74 --trace-children-skip=patt1,patt2,...
76 This option only has an effect when --trace-children=yes is
77 specified. It allows for some children to be skipped. The option
78 takes a comma separated list of patterns for the names of child
79 executables that Valgrind should not trace into. Patterns may
80 include the metacharacters ? and *, which have the usual meaning.
81 This can be useful for pruning uninteresting branches from a tree
83 of processes being run on Valgrind. But you should be careful when
84 using it. When Valgrind skips tracing into an executable, it
85 doesn't just skip tracing that executable, it also skips tracing
86 any of that executable's child processes. In other words, the flag
87 doesn't merely cause tracing to stop at the specified executables
88 -- it skips tracing of entire process subtrees rooted at any of the
89 specified executables.
90 --trace-children-skip-by-arg=patt1,patt2,...
92 This is the same as --trace-children-skip, with one difference: the
93 decision as to whether to trace into a child process is made by
94 examining the arguments to the child process, rather than the name
95 of its executable.
96 --child-silent-after-fork=<yes|no> [default: no]
98 When enabled, Valgrind will not show any debugging or logging
99 output for the child process resulting from a fork call. This can
100 make the output less confusing (although more misleading) when
101 dealing with processes that create children. It is particularly
102 useful in conjunction with --trace-children=. Use of this option is
103 also strongly recommended if you are requesting XML output
104 (--xml=yes), since otherwise the XML from child and parent may
105 become mixed up, which usually makes it useless.
106 --vgdb=<no|yes|full> [default: yes]
108 Valgrind will provide "gdbserver" functionality when --vgdb=yes or
109 --vgdb=full is specified. This allows an external GNU GDB debugger
110 to control and debug your program when it runs on Valgrind.
111 --vgdb=full incurs significant performance overheads, but provides
112 more precise breakpoints and watchpoints. See Debugging your
113 program using Valgrind's gdbserver and GDB for a detailed
114 description.
115 If the embedded gdbserver is enabled but no gdb is currently being
117 used, the vgdb command line utility can send "monitor commands" to
118 Valgrind from a shell. The Valgrind core provides a set of Valgrind
119 monitor commands. A tool can optionally provide tool specific
120 monitor commands, which are documented in the tool specific
121 chapter.
122 --vgdb-error=<number> [default: 999999999]
124 Use this option when the Valgrind gdbserver is enabled with
125 --vgdb=yes or --vgdb=full. Tools that report errors will wait for
126 "number" errors to be reported before freezing the program and
127 waiting for you to connect with GDB. It follows that a value of
128 zero will cause the gdbserver to be started before your program is
129 executed. This is typically used to insert GDB breakpoints before
130 execution, and also works with tools that do not report errors,
131 such as Massif.
132 --vgdb-stop-at=<set> [default: none]
134 Use this option when the Valgrind gdbserver is enabled with
135 --vgdb=yes or --vgdb=full. The Valgrind gdbserver will be invoked
136 for each error after --vgdb-error have been reported. You can
137 additionally ask the Valgrind gdbserver to be invoked for other
138 events, specified in one of the following ways:
139 · a comma separated list of one or more of startup exit
141 valgrindabexit.
142 The values startup exit valgrindabexit respectively indicate to
144 invoke gdbserver before your program is executed, after the
145 last instruction of your program, on Valgrind abnormal exit
146 (e.g. internal error, out of memory, ...).
147 Note: startup and --vgdb-error=0 will both cause Valgrind
149 gdbserver to be invoked before your program is executed. The
150 --vgdb-error=0 will in addition cause your program to stop on
151 all subsequent errors.
152 · all to specify the complete set. It is equivalent to
154 --vgdb-stop-at=startup,exit,valgrindabexit.
155 · none for the empty set.
157 --track-fds=<yes|no> [default: no]
159 When enabled, Valgrind will print out a list of open file
160 descriptors on exit or on request, via the gdbserver monitor
161 command v.info open_fds. Along with each file descriptor is printed
162 a stack backtrace of where the file was opened and any details
163 relating to the file descriptor such as the file name or socket
164 details.
165 --time-stamp=<yes|no> [default: no]
167 When enabled, each message is preceded with an indication of the
168 elapsed wallclock time since startup, expressed as days, hours,
169 minutes, seconds and milliseconds.
170 --log-fd=<number> [default: 2, stderr]
172 Specifies that Valgrind should send all of its messages to the
173 specified file descriptor. The default, 2, is the standard error
174 channel (stderr). Note that this may interfere with the client's
175 own use of stderr, as Valgrind's output will be interleaved with
176 any output that the client sends to stderr.
177 --log-file=<filename>
179 Specifies that Valgrind should send all of its messages to the
180 specified file. If the file name is empty, it causes an abort.
181 There are three special format specifiers that can be used in the
182 file name.
183 %p is replaced with the current process ID. This is very useful for
185 program that invoke multiple processes. WARNING: If you use
186 --trace-children=yes and your program invokes multiple processes OR
187 your program forks without calling exec afterwards, and you don't
188 use this specifier (or the %q specifier below), the Valgrind output
189 from all those processes will go into one file, possibly jumbled
190 up, and possibly incomplete. Note: If the program forks and calls
191 exec afterwards, Valgrind output of the child from the period
192 between fork and exec will be lost. Fortunately this gap is really
193 tiny for most programs; and modern programs use posix_spawn anyway.
194 %n is replaced with a file sequence number unique for this process.
196 This is useful for processes that produces several files from the
197 same filename template.
198 %q{FOO} is replaced with the contents of the environment variable
200 FOO. If the {FOO} part is malformed, it causes an abort. This
201 specifier is rarely needed, but very useful in certain
202 circumstances (eg. when running MPI programs). The idea is that you
203 specify a variable which will be set differently for each process
204 in the job, for example BPROC_RANK or whatever is applicable in
205 your MPI setup. If the named environment variable is not set, it
206 causes an abort. Note that in some shells, the { and } characters
207 may need to be escaped with a backslash.
208 %% is replaced with %.
210 If an % is followed by any other character, it causes an abort.
212 If the file name specifies a relative file name, it is put in the
214 program's initial working directory: this is the current directory
215 when the program started its execution after the fork or after the
216 exec. If it specifies an absolute file name (ie. starts with '/')
217 then it is put there.
218 --log-socket=<ip-address:port-number>
220 Specifies that Valgrind should send all of its messages to the
221 specified port at the specified IP address. The port may be
222 omitted, in which case port 1500 is used. If a connection cannot be
223 made to the specified socket, Valgrind falls back to writing output
224 to the standard error (stderr). This option is intended to be used
225 in conjunction with the valgrind-listener program. For further
226 details, see the commentary in the manual.
227
229 These options are used by all tools that can report errors, e.g.
230 Memcheck, but not Cachegrind.
231 --xml=<yes|no> [default: no]
233 When enabled, the important parts of the output (e.g. tool error
234 messages) will be in XML format rather than plain text.
235 Furthermore, the XML output will be sent to a different output
236 channel than the plain text output. Therefore, you also must use
237 one of --xml-fd, --xml-file or --xml-socket to specify where the
238 XML is to be sent.
239 Less important messages will still be printed in plain text, but
241 because the XML output and plain text output are sent to different
242 output channels (the destination of the plain text output is still
243 controlled by --log-fd, --log-file and --log-socket) this should
244 not cause problems.
245 This option is aimed at making life easier for tools that consume
247 Valgrind's output as input, such as GUI front ends. Currently this
248 option works with Memcheck, Helgrind, DRD and SGcheck. The output
249 format is specified in the file
250 docs/internals/xml-output-protocol4.txt in the source tree for
251 Valgrind 3.5.0 or later.
252 The recommended options for a GUI to pass, when requesting XML
254 output, are: --xml=yes to enable XML output, --xml-file to send the
255 XML output to a (presumably GUI-selected) file, --log-file to send
256 the plain text output to a second GUI-selected file,
257 --child-silent-after-fork=yes, and -q to restrict the plain text
258 output to critical error messages created by Valgrind itself. For
259 example, failure to read a specified suppressions file counts as a
260 critical error message. In this way, for a successful run the text
261 output file will be empty. But if it isn't empty, then it will
262 contain important information which the GUI user should be made
263 aware of.
264 --xml-fd=<number> [default: -1, disabled]
266 Specifies that Valgrind should send its XML output to the specified
267 file descriptor. It must be used in conjunction with --xml=yes.
268 --xml-file=<filename>
270 Specifies that Valgrind should send its XML output to the specified
271 file. It must be used in conjunction with --xml=yes. Any %p or %q
272 sequences appearing in the filename are expanded in exactly the
273 same way as they are for --log-file. See the description of --log-
274 file for details.
275 --xml-socket=<ip-address:port-number>
277 Specifies that Valgrind should send its XML output the specified
278 port at the specified IP address. It must be used in conjunction
279 with --xml=yes. The form of the argument is the same as that used
280 by --log-socket. See the description of --log-socket for further
281 details.
282 --xml-user-comment=<string>
284 Embeds an extra user comment string at the start of the XML output.
285 Only works when --xml=yes is specified; ignored otherwise.
286 --demangle=<yes|no> [default: yes]
288 Enable/disable automatic demangling (decoding) of C++ names.
289 Enabled by default. When enabled, Valgrind will attempt to
290 translate encoded C++ names back to something approaching the
291 original. The demangler handles symbols mangled by g++ versions
292 2.X, 3.X and 4.X.
293 An important fact about demangling is that function names mentioned
295 in suppressions files should be in their mangled form. Valgrind
296 does not demangle function names when searching for applicable
297 suppressions, because to do otherwise would make suppression file
298 contents dependent on the state of Valgrind's demangling machinery,
299 and also slow down suppression matching.
300 --num-callers=<number> [default: 12]
302 Specifies the maximum number of entries shown in stack traces that
303 identify program locations. Note that errors are commoned up using
304 only the top four function locations (the place in the current
305 function, and that of its three immediate callers). So this doesn't
306 affect the total number of errors reported.
307 The maximum value for this is 500. Note that higher settings will
309 make Valgrind run a bit more slowly and take a bit more memory, but
310 can be useful when working with programs with deeply-nested call
311 chains.
312 --unw-stack-scan-thresh=<number> [default: 0] ,
314 --unw-stack-scan-frames=<number> [default: 5]
315 Stack-scanning support is available only on ARM targets.
316 These flags enable and control stack unwinding by stack scanning.
318 When the normal stack unwinding mechanisms -- usage of Dwarf CFI
319 records, and frame-pointer following -- fail, stack scanning may be
320 able to recover a stack trace.
321 Note that stack scanning is an imprecise, heuristic mechanism that
323 may give very misleading results, or none at all. It should be used
324 only in emergencies, when normal unwinding fails, and it is
325 important to nevertheless have stack traces.
326 Stack scanning is a simple technique: the unwinder reads words from
328 the stack, and tries to guess which of them might be return
329 addresses, by checking to see if they point just after ARM or Thumb
330 call instructions. If so, the word is added to the backtrace.
331 The main danger occurs when a function call returns, leaving its
333 return address exposed, and a new function is called, but the new
334 function does not overwrite the old address. The result of this is
335 that the backtrace may contain entries for functions which have
336 already returned, and so be very confusing.
337 A second limitation of this implementation is that it will scan
339 only the page (4KB, normally) containing the starting stack
340 pointer. If the stack frames are large, this may result in only a
341 few (or not even any) being present in the trace. Also, if you are
342 unlucky and have an initial stack pointer near the end of its
343 containing page, the scan may miss all interesting frames.
344 By default stack scanning is disabled. The normal use case is to
346 ask for it when a stack trace would otherwise be very short. So, to
347 enable it, use --unw-stack-scan-thresh=number. This requests
348 Valgrind to try using stack scanning to "extend" stack traces which
349 contain fewer than number frames.
350 If stack scanning does take place, it will only generate at most
352 the number of frames specified by --unw-stack-scan-frames.
353 Typically, stack scanning generates so many garbage entries that
354 this value is set to a low value (5) by default. In no case will a
355 stack trace larger than the value specified by --num-callers be
356 created.
357 --error-limit=<yes|no> [default: yes]
359 When enabled, Valgrind stops reporting errors after 10,000,000 in
360 total, or 1,000 different ones, have been seen. This is to stop the
361 error tracking machinery from becoming a huge performance overhead
362 in programs with many errors.
363 --error-exitcode=<number> [default: 0]
365 Specifies an alternative exit code to return if Valgrind reported
366 any errors in the run. When set to the default value (zero), the
367 return value from Valgrind will always be the return value of the
368 process being simulated. When set to a nonzero value, that value is
369 returned instead, if Valgrind detects any errors. This is useful
370 for using Valgrind as part of an automated test suite, since it
371 makes it easy to detect test cases for which Valgrind has reported
372 errors, just by inspecting return codes.
373 --error-markers=<begin>,<end> [default: none]
375 When errors are output as plain text (i.e. XML not used),
376 --error-markers instructs to output a line containing the begin
377 (end) string before (after) each error.
378 Such marker lines facilitate searching for errors and/or extracting
380 errors in an output file that contain valgrind errors mixed with
381 the program output.
382 Note that empty markers are accepted. So, only using a begin (or an
384 end) marker is possible.
385 --sigill-diagnostics=<yes|no> [default: yes]
387 Enable/disable printing of illegal instruction diagnostics. Enabled
388 by default, but defaults to disabled when --quiet is given. The
389 default can always be explicitly overridden by giving this option.
390 When enabled, a warning message will be printed, along with some
392 diagnostics, whenever an instruction is encountered that Valgrind
393 cannot decode or translate, before the program is given a SIGILL
394 signal. Often an illegal instruction indicates a bug in the program
395 or missing support for the particular instruction in Valgrind. But
396 some programs do deliberately try to execute an instruction that
397 might be missing and trap the SIGILL signal to detect processor
398 features. Using this flag makes it possible to avoid the diagnostic
399 output that you would otherwise get in such cases.
400 --show-below-main=<yes|no> [default: no]
402 By default, stack traces for errors do not show any functions that
403 appear beneath main because most of the time it's uninteresting C
404 library stuff and/or gobbledygook. Alternatively, if main is not
405 present in the stack trace, stack traces will not show any
406 functions below main-like functions such as glibc's
407 __libc_start_main. Furthermore, if main-like functions are present
408 in the trace, they are normalised as (below main), in order to make
409 the output more deterministic.
410 If this option is enabled, all stack trace entries will be shown
412 and main-like functions will not be normalised.
413 --fullpath-after=<string> [default: don't show source paths]
415 By default Valgrind only shows the filenames in stack traces, but
416 not full paths to source files. When using Valgrind in large
417 projects where the sources reside in multiple different
418 directories, this can be inconvenient. --fullpath-after provides a
419 flexible solution to this problem. When this option is present, the
420 path to each source file is shown, with the following all-important
421 caveat: if string is found in the path, then the path up to and
422 including string is omitted, else the path is shown unmodified.
423 Note that string is not required to be a prefix of the path.
424 For example, consider a file named
426 /home/janedoe/blah/src/foo/bar/xyzzy.c. Specifying
427 --fullpath-after=/home/janedoe/blah/src/ will cause Valgrind to
428 show the name as foo/bar/xyzzy.c.
429 Because the string is not required to be a prefix,
431 --fullpath-after=src/ will produce the same output. This is useful
432 when the path contains arbitrary machine-generated characters. For
433 example, the path /my/build/dir/C32A1B47/blah/src/foo/xyzzy can be
434 pruned to foo/xyzzy using --fullpath-after=/blah/src/.
435 If you simply want to see the full path, just specify an empty
437 string: --fullpath-after=. This isn't a special case, merely a
438 logical consequence of the above rules.
439 Finally, you can use --fullpath-after multiple times. Any
441 appearance of it causes Valgrind to switch to producing full paths
442 and applying the above filtering rule. Each produced path is
443 compared against all the --fullpath-after-specified strings, in the
444 order specified. The first string to match causes the path to be
445 truncated as described above. If none match, the full path is
446 shown. This facilitates chopping off prefixes when the sources are
447 drawn from a number of unrelated directories.
448 --extra-debuginfo-path=<path> [default: undefined and unused]
450 By default Valgrind searches in several well-known paths for debug
451 objects, such as /usr/lib/debug/.
452 However, there may be scenarios where you may wish to put debug
454 objects at an arbitrary location, such as external storage when
455 running Valgrind on a mobile device with limited local storage.
456 Another example might be a situation where you do not have
457 permission to install debug object packages on the system where you
458 are running Valgrind.
459 In these scenarios, you may provide an absolute path as an extra,
461 final place for Valgrind to search for debug objects by specifying
462 --extra-debuginfo-path=/path/to/debug/objects. The given path will
463 be prepended to the absolute path name of the searched-for object.
464 For example, if Valgrind is looking for the debuginfo for
465 /w/x/y/zz.so and --extra-debuginfo-path=/a/b/c is specified, it
466 will look for a debug object at /a/b/c/w/x/y/zz.so.
467 This flag should only be specified once. If it is specified
469 multiple times, only the last instance is honoured.
470 --debuginfo-server=ipaddr:port [default: undefined and unused]
472 This is a new, experimental, feature introduced in version 3.9.0.
473 In some scenarios it may be convenient to read debuginfo from
475 objects stored on a different machine. With this flag, Valgrind
476 will query a debuginfo server running on ipaddr and listening on
477 port port, if it cannot find the debuginfo object in the local
478 filesystem.
479 The debuginfo server must accept TCP connections on port port. The
481 debuginfo server is contained in the source file
482 auxprogs/valgrind-di-server.c. It will only serve from the
483 directory it is started in. port defaults to 1500 in both client
484 and server if not specified.
485 If Valgrind looks for the debuginfo for /w/x/y/zz.so by using the
487 debuginfo server, it will strip the pathname components and merely
488 request zz.so on the server. That in turn will look only in its
489 current working directory for a matching debuginfo object.
490 The debuginfo data is transmitted in small fragments (8 KB) as
492 requested by Valgrind. Each block is compressed using LZO to reduce
493 transmission time. The implementation has been tuned for best
494 performance over a single-stage 802.11g (WiFi) network link.
495 Note that checks for matching primary vs debug objects, using GNU
497 debuglink CRC scheme, are performed even when using the debuginfo
498 server. To disable such checking, you need to also specify
499 --allow-mismatched-debuginfo=yes.
500 By default the Valgrind build system will build valgrind-di-server
502 for the target platform, which is almost certainly not what you
503 want. So far we have been unable to find out how to get
504 automake/autoconf to build it for the build platform. If you want
505 to use it, you will have to recompile it by hand using the command
506 shown at the top of auxprogs/valgrind-di-server.c.
507 --allow-mismatched-debuginfo=no|yes [no]
509 When reading debuginfo from separate debuginfo objects, Valgrind
510 will by default check that the main and debuginfo objects match,
511 using the GNU debuglink mechanism. This guarantees that it does not
512 read debuginfo from out of date debuginfo objects, and also ensures
513 that Valgrind can't crash as a result of mismatches.
514 This check can be overridden using
516 --allow-mismatched-debuginfo=yes. This may be useful when the
517 debuginfo and main objects have not been split in the proper way.
518 Be careful when using this, though: it disables all consistency
519 checking, and Valgrind has been observed to crash when the main and
520 debuginfo objects don't match.
521 --suppressions=<filename> [default: $PREFIX/lib/valgrind/default.supp]
523 Specifies an extra file from which to read descriptions of errors
524 to suppress. You may use up to 100 extra suppression files.
525 --gen-suppressions=<yes|no|all> [default: no]
527 When set to yes, Valgrind will pause after every error shown and
528 print the line:
529 ---- Print suppression ? --- [Return/N/n/Y/y/C/c] ----
531 Pressing Ret, or N Ret or n Ret, causes Valgrind continue execution
533 without printing a suppression for this error.
534 Pressing Y Ret or y Ret causes Valgrind to write a suppression for
536 this error. You can then cut and paste it into a suppression file
537 if you don't want to hear about the error in the future.
538 When set to all, Valgrind will print a suppression for every
540 reported error, without querying the user.
541 This option is particularly useful with C++ programs, as it prints
543 out the suppressions with mangled names, as required.
544 Note that the suppressions printed are as specific as possible. You
546 may want to common up similar ones, by adding wildcards to function
547 names, and by using frame-level wildcards. The wildcarding
548 facilities are powerful yet flexible, and with a bit of careful
549 editing, you may be able to suppress a whole family of related
550 errors with only a few suppressions.
551 Sometimes two different errors are suppressed by the same
553 suppression, in which case Valgrind will output the suppression
554 more than once, but you only need to have one copy in your
555 suppression file (but having more than one won't cause problems).
556 Also, the suppression name is given as <insert a suppression name
557 here>; the name doesn't really matter, it's only used with the -v
558 option which prints out all used suppression records.
559 --input-fd=<number> [default: 0, stdin]
561 When using --gen-suppressions=yes, Valgrind will stop so as to read
562 keyboard input from you when each error occurs. By default it reads
563 from the standard input (stdin), which is problematic for programs
564 which close stdin. This option allows you to specify an alternative
565 file descriptor from which to read input.
566 --dsymutil=no|yes [yes]
568 This option is only relevant when running Valgrind on Mac OS X.
569 Mac OS X uses a deferred debug information (debuginfo) linking
571 scheme. When object files containing debuginfo are linked into a
572 .dylib or an executable, the debuginfo is not copied into the final
573 file. Instead, the debuginfo must be linked manually by running
574 dsymutil, a system-provided utility, on the executable or .dylib.
575 The resulting combined debuginfo is placed in a directory alongside
576 the executable or .dylib, but with the extension .dSYM.
577 With --dsymutil=no, Valgrind will detect cases where the .dSYM
579 directory is either missing, or is present but does not appear to
580 match the associated executable or .dylib, most likely because it
581 is out of date. In these cases, Valgrind will print a warning
582 message but take no further action.
583 With --dsymutil=yes, Valgrind will, in such cases, automatically
585 run dsymutil as necessary to bring the debuginfo up to date. For
586 all practical purposes, if you always use --dsymutil=yes, then
587 there is never any need to run dsymutil manually or as part of your
588 applications's build system, since Valgrind will run it as
589 necessary.
590 Valgrind will not attempt to run dsymutil on any executable or
592 library in /usr/, /bin/, /sbin/, /opt/, /sw/, /System/, /Library/
593 or /Applications/ since dsymutil will always fail in such
594 situations. It fails both because the debuginfo for such
595 pre-installed system components is not available anywhere, and also
596 because it would require write privileges in those directories.
597 Be careful when using --dsymutil=yes, since it will cause
599 pre-existing .dSYM directories to be silently deleted and
600 re-created. Also note that dsymutil is quite slow, sometimes
601 excessively so.
602 --max-stackframe=<number> [default: 2000000]
604 The maximum size of a stack frame. If the stack pointer moves by
605 more than this amount then Valgrind will assume that the program is
606 switching to a different stack.
607 You may need to use this option if your program has large
609 stack-allocated arrays. Valgrind keeps track of your program's
610 stack pointer. If it changes by more than the threshold amount,
611 Valgrind assumes your program is switching to a different stack,
612 and Memcheck behaves differently than it would for a stack pointer
613 change smaller than the threshold. Usually this heuristic works
614 well. However, if your program allocates large structures on the
615 stack, this heuristic will be fooled, and Memcheck will
616 subsequently report large numbers of invalid stack accesses. This
617 option allows you to change the threshold to a different value.
618 You should only consider use of this option if Valgrind's debug
620 output directs you to do so. In that case it will tell you the new
621 threshold you should specify.
622 In general, allocating large structures on the stack is a bad idea,
624 because you can easily run out of stack space, especially on
625 systems with limited memory or which expect to support large
626 numbers of threads each with a small stack, and also because the
627 error checking performed by Memcheck is more effective for
628 heap-allocated data than for stack-allocated data. If you have to
629 use this option, you may wish to consider rewriting your code to
630 allocate on the heap rather than on the stack.
631 --main-stacksize=<number> [default: use current 'ulimit' value]
633 Specifies the size of the main thread's stack.
634 To simplify its memory management, Valgrind reserves all required
636 space for the main thread's stack at startup. That means it needs
637 to know the required stack size at startup.
638 By default, Valgrind uses the current "ulimit" value for the stack
640 size, or 16 MB, whichever is lower. In many cases this gives a
641 stack size in the range 8 to 16 MB, which almost never overflows
642 for most applications.
643 If you need a larger total stack size, use --main-stacksize to
645 specify it. Only set it as high as you need, since reserving far
646 more space than you need (that is, hundreds of megabytes more than
647 you need) constrains Valgrind's memory allocators and may reduce
648 the total amount of memory that Valgrind can use. This is only
649 really of significance on 32-bit machines.
650 On Linux, you may request a stack of size up to 2GB. Valgrind will
652 stop with a diagnostic message if the stack cannot be allocated.
653 --main-stacksize only affects the stack size for the program's
655 initial thread. It has no bearing on the size of thread stacks, as
656 Valgrind does not allocate those.
657 You may need to use both --main-stacksize and --max-stackframe
659 together. It is important to understand that --main-stacksize sets
660 the maximum total stack size, whilst --max-stackframe specifies the
661 largest size of any one stack frame. You will have to work out the
662 --main-stacksize value for yourself (usually, if your applications
663 segfaults). But Valgrind will tell you the needed --max-stackframe
664 size, if necessary.
665 As discussed further in the description of --max-stackframe, a
667 requirement for a large stack is a sign of potential portability
668 problems. You are best advised to place all large data in
669 heap-allocated memory.
670 --max-threads=<number> [default: 500]
672 By default, Valgrind can handle to up to 500 threads. Occasionally,
673 that number is too small. Use this option to provide a different
674 limit. E.g. --max-threads=3000.
675
677 For tools that use their own version of malloc (e.g. Memcheck, Massif,
678 Helgrind, DRD), the following options apply.
679 --alignment=<number> [default: 8 or 16, depending on the platform]
681 By default Valgrind's malloc, realloc, etc, return a block whose
682 starting address is 8-byte aligned or 16-byte aligned (the value
683 depends on the platform and matches the platform default). This
684 option allows you to specify a different alignment. The supplied
685 value must be greater than or equal to the default, less than or
686 equal to 4096, and must be a power of two.
687 --redzone-size=<number> [default: depends on the tool]
689 Valgrind's malloc, realloc, etc, add padding blocks before and
690 after each heap block allocated by the program being run. Such
691 padding blocks are called redzones. The default value for the
692 redzone size depends on the tool. For example, Memcheck adds and
693 protects a minimum of 16 bytes before and after each block
694 allocated by the client. This allows it to detect block underruns
695 or overruns of up to 16 bytes.
696 Increasing the redzone size makes it possible to detect overruns of
698 larger distances, but increases the amount of memory used by
699 Valgrind. Decreasing the redzone size will reduce the memory needed
700 by Valgrind but also reduces the chances of detecting
701 over/underruns, so is not recommended.
702 --xtree-memory=none|allocs|full [none]
704 Tools replacing Valgrind's malloc, realloc, etc, can optionally
705 produce an execution tree detailing which piece of code is
706 responsible for heap memory usage. See ??? for a detailed
707 explanation about execution trees.
708 When set to none, no memory execution tree is produced.
710 When set to allocs, the memory execution tree gives the current
712 number of allocated bytes and the current number of allocated
713 blocks.
714 When set to full, the memory execution tree gives 6 different
716 measurements : the current number of allocated bytes and blocks
717 (same values as for allocs), the total number of allocated bytes
718 and blocks, the total number of freed bytes and blocks.
719 Note that the overhead in cpu and memory to produce an xtree
721 depends on the tool. The overhead in cpu is small for the value
722 allocs, as the information needed to produce this report is
723 maintained in any case by the tool. For massif and helgrind,
724 specifying full implies to capture a stack trace for each free
725 operation, while normally these tools only capture an allocation
726 stack trace. For memcheck, the cpu overhead for the value full is
727 small, as this can only be used in combination with
728 --keep-stacktraces=alloc-and-free or
729 --keep-stacktraces=alloc-then-free, which already records a stack
730 trace for each free operation. The memory overhead varies between 5
731 and 10 words per unique stacktrace in the xtree, plus the memory
732 needed to record the stack trace for the free operations, if needed
733 specifically for the xtree.
734 --xtree-memory-file=<filename> [default: xtmemory.kcg.%p]
736 Specifies that Valgrind should produce the xtree memory report in
737 the specified file. Any %p or %q sequences appearing in the
738 filename are expanded in exactly the same way as they are for
739 --log-file. See the description of --log-file for details.
740 If the filename contains the extension .ms, then the produced file
742 format will be a massif output file format. If the filename
743 contains the extension .kcg or no extension is provided or
744 recognised, then the produced file format will be a callgrind
745 output format.
746 See ??? for a detailed explanation about execution trees formats.
748
750 These options apply to all tools, as they affect certain obscure
751 workings of the Valgrind core. Most people won't need to use them.
752 --smc-check=<none|stack|all|all-non-file> [default: all-non-file for
754 x86/amd64/s390x, stack for other archs]
755 This option controls Valgrind's detection of self-modifying code.
756 If no checking is done, when a program executes some code, then
757 overwrites it with new code, and executes the new code, Valgrind
758 will continue to execute the translations it made for the old code.
759 This will likely lead to incorrect behaviour and/or crashes.
760 For "modern" architectures -- anything that's not x86, amd64 or
762 s390x -- the default is stack. This is because a correct program
763 must take explicit action to reestablish D-I cache coherence
764 following code modification. Valgrind observes and honours such
765 actions, with the result that self-modifying code is transparently
766 handled with zero extra cost.
767 For x86, amd64 and s390x, the program is not required to notify the
769 hardware of required D-I coherence syncing. Hence the default is
770 all-non-file, which covers the normal case of generating code into
771 an anonymous (non-file-backed) mmap'd area.
772 The meanings of the four available settings are as follows. No
774 detection (none), detect self-modifying code on the stack (which is
775 used by GCC to implement nested functions) (stack), detect
776 self-modifying code everywhere (all), and detect self-modifying
777 code everywhere except in file-backed mappings (all-non-file).
778 Running with all will slow Valgrind down noticeably. Running with
780 none will rarely speed things up, since very little code gets
781 dynamically generated in most programs. The
782 VALGRIND_DISCARD_TRANSLATIONS client request is an alternative to
783 --smc-check=all and --smc-check=all-non-file that requires more
784 programmer effort but allows Valgrind to run your program faster,
785 by telling it precisely when translations need to be re-made.
786 --smc-check=all-non-file provides a cheaper but more limited
788 version of --smc-check=all. It adds checks to any translations that
789 do not originate from file-backed memory mappings. Typical
790 applications that generate code, for example JITs in web browsers,
791 generate code into anonymous mmaped areas, whereas the "fixed" code
792 of the browser always lives in file-backed mappings.
793 --smc-check=all-non-file takes advantage of this observation,
794 limiting the overhead of checking to code which is likely to be JIT
795 generated.
796 --read-inline-info=<yes|no> [default: see below]
798 When enabled, Valgrind will read information about inlined function
799 calls from DWARF3 debug info. This slows Valgrind startup and makes
800 it use more memory (typically for each inlined piece of code, 6
801 words and space for the function name), but it results in more
802 descriptive stacktraces. For the 3.10.0 release, this functionality
803 is enabled by default only for Linux, Android and Solaris targets
804 and only for the tools Memcheck, Helgrind and DRD. Here is an
805 example of some stacktraces with --read-inline-info=no:
806 ==15380== Conditional jump or move depends on uninitialised value(s)
808 ==15380== at 0x80484EA: main (inlinfo.c:6)
809 ==15380==
810 ==15380== Conditional jump or move depends on uninitialised value(s)
811 ==15380== at 0x8048550: fun_noninline (inlinfo.c:6)
812 ==15380== by 0x804850E: main (inlinfo.c:34)
813 ==15380==
814 ==15380== Conditional jump or move depends on uninitialised value(s)
815 ==15380== at 0x8048520: main (inlinfo.c:6)
816 And here are the same errors with --read-inline-info=yes:
818 ==15377== Conditional jump or move depends on uninitialised value(s)
820 ==15377== at 0x80484EA: fun_d (inlinfo.c:6)
821 ==15377== by 0x80484EA: fun_c (inlinfo.c:14)
822 ==15377== by 0x80484EA: fun_b (inlinfo.c:20)
823 ==15377== by 0x80484EA: fun_a (inlinfo.c:26)
824 ==15377== by 0x80484EA: main (inlinfo.c:33)
825 ==15377==
826 ==15377== Conditional jump or move depends on uninitialised value(s)
827 ==15377== at 0x8048550: fun_d (inlinfo.c:6)
828 ==15377== by 0x8048550: fun_noninline (inlinfo.c:41)
829 ==15377== by 0x804850E: main (inlinfo.c:34)
830 ==15377==
831 ==15377== Conditional jump or move depends on uninitialised value(s)
832 ==15377== at 0x8048520: fun_d (inlinfo.c:6)
833 ==15377== by 0x8048520: main (inlinfo.c:35)
834 --read-var-info=<yes|no> [default: no]
836 When enabled, Valgrind will read information about variable types
837 and locations from DWARF3 debug info. This slows Valgrind startup
838 significantly and makes it use significantly more memory, but for
839 the tools that can take advantage of it (Memcheck, Helgrind, DRD)
840 it can result in more precise error messages. For example, here are
841 some standard errors issued by Memcheck:
842 ==15363== Uninitialised byte(s) found during client check request
844 ==15363== at 0x80484A9: croak (varinfo1.c:28)
845 ==15363== by 0x8048544: main (varinfo1.c:55)
846 ==15363== Address 0x80497f7 is 7 bytes inside data symbol "global_i2"
847 ==15363==
848 ==15363== Uninitialised byte(s) found during client check request
849 ==15363== at 0x80484A9: croak (varinfo1.c:28)
850 ==15363== by 0x8048550: main (varinfo1.c:56)
851 ==15363== Address 0xbea0d0cc is on thread 1's stack
852 ==15363== in frame #1, created by main (varinfo1.c:45)
853 And here are the same errors with --read-var-info=yes:
855 ==15370== Uninitialised byte(s) found during client check request
857 ==15370== at 0x80484A9: croak (varinfo1.c:28)
858 ==15370== by 0x8048544: main (varinfo1.c:55)
859 ==15370== Location 0x80497f7 is 0 bytes inside global_i2[7],
860 ==15370== a global variable declared at varinfo1.c:41
861 ==15370==
862 ==15370== Uninitialised byte(s) found during client check request
863 ==15370== at 0x80484A9: croak (varinfo1.c:28)
864 ==15370== by 0x8048550: main (varinfo1.c:56)
865 ==15370== Location 0xbeb4a0cc is 0 bytes inside local var "local"
866 ==15370== declared at varinfo1.c:46, in frame #1 of thread 1
867 --vgdb-poll=<number> [default: 5000]
869 As part of its main loop, the Valgrind scheduler will poll to check
870 if some activity (such as an external command or some input from a
871 gdb) has to be handled by gdbserver. This activity poll will be
872 done after having run the given number of basic blocks (or slightly
873 more than the given number of basic blocks). This poll is quite
874 cheap so the default value is set relatively low. You might further
875 decrease this value if vgdb cannot use ptrace system call to
876 interrupt Valgrind if all threads are (most of the time) blocked in
877 a system call.
878 --vgdb-shadow-registers=no|yes [default: no]
880 When activated, gdbserver will expose the Valgrind shadow registers
881 to GDB. With this, the value of the Valgrind shadow registers can
882 be examined or changed using GDB. Exposing shadow registers only
883 works with GDB version 7.1 or later.
884 --vgdb-prefix=<prefix> [default: /tmp/vgdb-pipe]
886 To communicate with gdb/vgdb, the Valgrind gdbserver creates 3
887 files (2 named FIFOs and a mmap shared memory file). The prefix
888 option controls the directory and prefix for the creation of these
889 files.
890 --run-libc-freeres=<yes|no> [default: yes]
892 This option is only relevant when running Valgrind on Linux.
893 The GNU C library (libc.so), which is used by all programs, may
895 allocate memory for its own uses. Usually it doesn't bother to free
896 that memory when the program ends—there would be no point, since
897 the Linux kernel reclaims all process resources when a process
898 exits anyway, so it would just slow things down.
899 The glibc authors realised that this behaviour causes leak
901 checkers, such as Valgrind, to falsely report leaks in glibc, when
902 a leak check is done at exit. In order to avoid this, they provided
903 a routine called __libc_freeres specifically to make glibc release
904 all memory it has allocated. Memcheck therefore tries to run
905 __libc_freeres at exit.
906 Unfortunately, in some very old versions of glibc, __libc_freeres
908 is sufficiently buggy to cause segmentation faults. This was
909 particularly noticeable on Red Hat 7.1. So this option is provided
910 in order to inhibit the run of __libc_freeres. If your program
911 seems to run fine on Valgrind, but segfaults at exit, you may find
912 that --run-libc-freeres=no fixes that, although at the cost of
913 possibly falsely reporting space leaks in libc.so.
914 --run-cxx-freeres=<yes|no> [default: yes]
916 This option is only relevant when running Valgrind on Linux or
917 Solaris C++ programs.
918 The GNU Standard C++ library (libstdc++.so), which is used by all
920 C++ programs compiled with g++, may allocate memory for its own
921 uses. Usually it doesn't bother to free that memory when the
922 program ends—there would be no point, since the kernel reclaims all
923 process resources when a process exits anyway, so it would just
924 slow things down.
925 The gcc authors realised that this behaviour causes leak checkers,
927 such as Valgrind, to falsely report leaks in libstdc++, when a leak
928 check is done at exit. In order to avoid this, they provided a
929 routine called __gnu_cxx::__freeres specifically to make libstdc++
930 release all memory it has allocated. Memcheck therefore tries to
931 run __gnu_cxx::__freeres at exit.
932 For the sake of flexibility and unforeseen problems with
934 __gnu_cxx::__freeres, option --run-cxx-freeres=no exists, although
935 at the cost of possibly falsely reporting space leaks in
936 libstdc++.so.
937 --sim-hints=hint1,hint2,...
939 Pass miscellaneous hints to Valgrind which slightly modify the
940 simulated behaviour in nonstandard or dangerous ways, possibly to
941 help the simulation of strange features. By default no hints are
942 enabled. Use with caution! Currently known hints are:
943 · lax-ioctls: Be very lax about ioctl handling; the only
945 assumption is that the size is correct. Doesn't require the
946 full buffer to be initialised when writing. Without this, using
947 some device drivers with a large number of strange ioctl
948 commands becomes very tiresome.
949 · fuse-compatible: Enable special handling for certain system
951 calls that may block in a FUSE file-system. This may be
952 necessary when running Valgrind on a multi-threaded program
953 that uses one thread to manage a FUSE file-system and another
954 thread to access that file-system.
955 · enable-outer: Enable some special magic needed when the program
957 being run is itself Valgrind.
958 · no-inner-prefix: Disable printing a prefix > in front of each
960 stdout or stderr output line in an inner Valgrind being run by
961 an outer Valgrind. This is useful when running Valgrind
962 regression tests in an outer/inner setup. Note that the prefix
963 > will always be printed in front of the inner debug logging
964 lines.
965 · no-nptl-pthread-stackcache: This hint is only relevant when
967 running Valgrind on Linux.
968 The GNU glibc pthread library (libpthread.so), which is used by
970 pthread programs, maintains a cache of pthread stacks. When a
971 pthread terminates, the memory used for the pthread stack and
972 some thread local storage related data structure are not always
973 directly released. This memory is kept in a cache (up to a
974 certain size), and is re-used if a new thread is started.
975 This cache causes the helgrind tool to report some false
977 positive race condition errors on this cached memory, as
978 helgrind does not understand the internal glibc cache
979 synchronisation primitives. So, when using helgrind, disabling
980 the cache helps to avoid false positive race conditions, in
981 particular when using thread local storage variables (e.g.
982 variables using the __thread qualifier).
983 When using the memcheck tool, disabling the cache ensures the
985 memory used by glibc to handle __thread variables is directly
986 released when a thread terminates.
987 Note: Valgrind disables the cache using some internal knowledge
989 of the glibc stack cache implementation and by examining the
990 debug information of the pthread library. This technique is
991 thus somewhat fragile and might not work for all glibc
992 versions. This has been successfully tested with various glibc
993 versions (e.g. 2.11, 2.16, 2.18) on various platforms.
994 · lax-doors: (Solaris only) Be very lax about door syscall
996 handling over unrecognised door file descriptors. Does not
997 require that full buffer is initialised when writing. Without
998 this, programs using libdoor(3LIB) functionality with
999 completely proprietary semantics may report large number of
1000 false positives.
1001 · fallback-llsc: (MIPS and ARM64 only): Enables an alternative
1003 implementation of Load-Linked (LL) and Store-Conditional (SC)
1004 instructions. The standard implementation gives more correct
1005 behaviour, but can cause indefinite looping on certain
1006 processor implementations that are intolerant of extra memory
1007 references between LL and SC. So far this is known only to
1008 happen on Cavium 3 cores. You should not need to use this flag,
1009 since the relevant cores are detected at startup and the
1010 alternative implementation is automatically enabled if
1011 necessary. There is no equivalent anti-flag: you cannot
1012 force-disable the alternative implementation, if it is
1013 automatically enabled. The underlying problem exists because
1014 the "standard" implementation of LL and SC is done by copying
1015 through LL and SC instructions into the instrumented code.
1016 However, tools may insert extra instrumentation memory
1017 references in between the LL and SC instructions. These memory
1018 references are not present in the original uninstrumented code,
1019 and their presence in the instrumented code can cause the SC
1020 instructions to persistently fail, leading to indefinite
1021 looping in LL-SC blocks. The alternative implementation gives
1022 correct behaviour of LL and SC instructions between threads in
1023 a process, up to and including the ABA scenario. It also gives
1024 correct behaviour between a Valgrinded thread and a
1025 non-Valgrinded thread running in a different process, that
1026 communicate via shared memory, but only up to and including
1027 correct CAS behaviour -- in this case the ABA scenario may not
1028 be correctly handled.
1029 --fair-sched=<no|yes|try> [default: no]
1031 The --fair-sched option controls the locking mechanism used by
1032 Valgrind to serialise thread execution. The locking mechanism
1033 controls the way the threads are scheduled, and different settings
1034 give different trade-offs between fairness and performance. For
1035 more details about the Valgrind thread serialisation scheme and its
1036 impact on performance and thread scheduling, see Scheduling and
1037 Multi-Thread Performance.
1038 · The value --fair-sched=yes activates a fair scheduler. In
1040 short, if multiple threads are ready to run, the threads will
1041 be scheduled in a round robin fashion. This mechanism is not
1042 available on all platforms or Linux versions. If not available,
1043 using --fair-sched=yes will cause Valgrind to terminate with an
1044 error.
1045 You may find this setting improves overall responsiveness if
1047 you are running an interactive multithreaded program, for
1048 example a web browser, on Valgrind.
1049 · The value --fair-sched=try activates fair scheduling if
1051 available on the platform. Otherwise, it will automatically
1052 fall back to --fair-sched=no.
1053 · The value --fair-sched=no activates a scheduler which does not
1055 guarantee fairness between threads ready to run, but which in
1056 general gives the highest performance.
1057 --kernel-variant=variant1,variant2,...
1059 Handle system calls and ioctls arising from minor variants of the
1060 default kernel for this platform. This is useful for running on
1061 hacked kernels or with kernel modules which support nonstandard
1062 ioctls, for example. Use with caution. If you don't understand what
1063 this option does then you almost certainly don't need it. Currently
1064 known variants are:
1065 · bproc: support the sys_broc system call on x86. This is for
1067 running on BProc, which is a minor variant of standard Linux
1068 which is sometimes used for building clusters.
1069 · android-no-hw-tls: some versions of the Android emulator for
1071 ARM do not provide a hardware TLS (thread-local state)
1072 register, and Valgrind crashes at startup. Use this variant to
1073 select software support for TLS.
1074 · android-gpu-sgx5xx: use this to support handling of proprietary
1076 ioctls for the PowerVR SGX 5XX series of GPUs on Android
1077 devices. Failure to select this does not cause stability
1078 problems, but may cause Memcheck to report false errors after
1079 the program performs GPU-specific ioctls.
1080 · android-gpu-adreno3xx: similarly, use this to support handling
1082 of proprietary ioctls for the Qualcomm Adreno 3XX series of
1083 GPUs on Android devices.
1084 --merge-recursive-frames=<number> [default: 0]
1086 Some recursive algorithms, for example balanced binary tree
1087 implementations, create many different stack traces, each
1088 containing cycles of calls. A cycle is defined as two identical
1089 program counter values separated by zero or more other program
1090 counter values. Valgrind may then use a lot of memory to store all
1091 these stack traces. This is a poor use of memory considering that
1092 such stack traces contain repeated uninteresting recursive calls
1093 instead of more interesting information such as the function that
1094 has initiated the recursive call.
1095 The option --merge-recursive-frames=<number> instructs Valgrind to
1097 detect and merge recursive call cycles having a size of up to
1098 <number> frames. When such a cycle is detected, Valgrind records
1099 the cycle in the stack trace as a unique program counter.
1100 The value 0 (the default) causes no recursive call merging. A value
1102 of 1 will cause stack traces of simple recursive algorithms (for
1103 example, a factorial implementation) to be collapsed. A value of 2
1104 will usually be needed to collapse stack traces produced by
1105 recursive algorithms such as binary trees, quick sort, etc. Higher
1106 values might be needed for more complex recursive algorithms.
1107 Note: recursive calls are detected by analysis of program counter
1109 values. They are not detected by looking at function names.
1110 --num-transtab-sectors=<number> [default: 6 for Android platforms, 16
1112 for all others]
1113 Valgrind translates and instruments your program's machine code in
1114 small fragments (basic blocks). The translations are stored in a
1115 translation cache that is divided into a number of sections
1116 (sectors). If the cache is full, the sector containing the oldest
1117 translations is emptied and reused. If these old translations are
1118 needed again, Valgrind must re-translate and re-instrument the
1119 corresponding machine code, which is expensive. If the "executed
1120 instructions" working set of a program is big, increasing the
1121 number of sectors may improve performance by reducing the number of
1122 re-translations needed. Sectors are allocated on demand. Once
1123 allocated, a sector can never be freed, and occupies considerable
1124 space, depending on the tool and the value of
1125 --avg-transtab-entry-size (about 40 MB per sector for Memcheck).
1126 Use the option --stats=yes to obtain precise information about the
1127 memory used by a sector and the allocation and recycling of
1128 sectors.
1129 --avg-transtab-entry-size=<number> [default: 0, meaning use tool
1131 provided default]
1132 Average size of translated basic block. This average size is used
1133 to dimension the size of a sector. Each tool provides a default
1134 value to be used. If this default value is too small, the
1135 translation sectors will become full too quickly. If this default
1136 value is too big, a significant part of the translation sector
1137 memory will be unused. Note that the average size of a basic block
1138 translation depends on the tool, and might depend on tool options.
1139 For example, the memcheck option --track-origins=yes increases the
1140 size of the basic block translations. Use --avg-transtab-entry-size
1141 to tune the size of the sectors, either to gain memory or to avoid
1142 too many retranslations.
1143 --aspace-minaddr=<address> [default: depends on the platform]
1145 To avoid potential conflicts with some system libraries, Valgrind
1146 does not use the address space below --aspace-minaddr value,
1147 keeping it reserved in case a library specifically requests memory
1148 in this region. So, some "pessimistic" value is guessed by Valgrind
1149 depending on the platform. On linux, by default, Valgrind avoids
1150 using the first 64MB even if typically there is no conflict in this
1151 complete zone. You can use the option --aspace-minaddr to have your
1152 memory hungry application benefitting from more of this lower
1153 memory. On the other hand, if you encounter a conflict, increasing
1154 aspace-minaddr value might solve it. Conflicts will typically
1155 manifest themselves with mmap failures in the low range of the
1156 address space. The provided address must be page aligned and must
1157 be equal or bigger to 0x1000 (4KB). To find the default value on
1158 your platform, do something such as valgrind -d -d date 2>&1 | grep
1159 -i minaddr. Values lower than 0x10000 (64KB) are known to create
1160 problems on some distributions.
1161 --valgrind-stacksize=<number> [default: 1MB]
1163 For each thread, Valgrind needs its own 'private' stack. The
1164 default size for these stacks is largely dimensioned, and so should
1165 be sufficient in most cases. In case the size is too small,
1166 Valgrind will segfault. Before segfaulting, a warning might be
1167 produced by Valgrind when approaching the limit.
1168 Use the option --valgrind-stacksize if such an (unlikely) warning
1170 is produced, or Valgrind dies due to a segmentation violation. Such
1171 segmentation violations have been seen when demangling huge C++
1172 symbols.
1173 If your application uses many threads and needs a lot of memory,
1175 you can gain some memory by reducing the size of these Valgrind
1176 stacks using the option --valgrind-stacksize.
1177 --show-emwarns=<yes|no> [default: no]
1179 When enabled, Valgrind will emit warnings about its CPU emulation
1180 in certain cases. These are usually not interesting.
1181 --require-text-symbol=:sonamepatt:fnnamepatt
1183 When a shared object whose soname matches sonamepatt is loaded into
1184 the process, examine all the text symbols it exports. If none of
1185 those match fnnamepatt, print an error message and abandon the run.
1186 This makes it possible to ensure that the run does not continue
1187 unless a given shared object contains a particular function name.
1188 Both sonamepatt and fnnamepatt can be written using the usual ?
1190 and * wildcards. For example: ":*libc.so*:foo?bar". You may use
1191 characters other than a colon to separate the two patterns. It is
1192 only important that the first character and the separator character
1193 are the same. For example, the above example could also be written
1194 "Q*libc.so*Qfoo?bar". Multiple
1195 --require-text-symbol flags are allowed, in which case shared
1196 objects that are loaded into the process will be checked against
1197 all of them.
1198 The purpose of this is to support reliable usage of marked-up
1200 libraries. For example, suppose we have a version of GCC's
1201 libgomp.so which has been marked up with annotations to support
1202 Helgrind. It is only too easy and confusing to load the wrong,
1203 un-annotated libgomp.so into the application. So the idea is: add a
1204 text symbol in the marked-up library, for example
1205 annotated_for_helgrind_3_6, and then give the flag
1206 --require-text-symbol=:*libgomp*so*:annotated_for_helgrind_3_6 so
1207 that when libgomp.so is loaded, Valgrind scans its symbol table,
1208 and if the symbol isn't present the run is aborted, rather than
1209 continuing silently with the un-marked-up library. Note that you
1210 should put the entire flag in quotes to stop shells expanding up
1211 the * and ? wildcards.
1212 --soname-synonyms=syn1=pattern1,syn2=pattern2,...
1214 When a shared library is loaded, Valgrind checks for functions in
1215 the library that must be replaced or wrapped. For example, Memcheck
1216 replaces some string and memory functions (strchr, strlen, strcpy,
1217 memchr, memcpy, memmove, etc.) with its own versions. Such
1218 replacements are normally done only in shared libraries whose
1219 soname matches a predefined soname pattern (e.g. libc.so* on
1220 linux). By default, no replacement is done for a statically linked
1221 binary or for alternative libraries, except for the allocation
1222 functions (malloc, free, calloc, memalign, realloc, operator new,
1223 operator delete, etc.) Such allocation functions are intercepted by
1224 default in any shared library or in the executable if they are
1225 exported as global symbols. This means that if a replacement
1226 allocation library such as tcmalloc is found, its functions are
1227 also intercepted by default. In some cases, the replacements allow
1228 --soname-synonyms to specify one additional synonym pattern, giving
1229 flexibility in the replacement. Or to prevent interception of all
1230 public allocation symbols.
1231 Currently, this flexibility is only allowed for the malloc related
1233 functions, using the synonym somalloc. This synonym is usable for
1234 all tools doing standard replacement of malloc related functions
1235 (e.g. memcheck, massif, drd, helgrind, exp-dhat, exp-sgcheck).
1236 · Alternate malloc library: to replace the malloc related
1238 functions in a specific alternate library with soname
1239 mymalloclib.so (and not in any others), give the option
1240 --soname-synonyms=somalloc=mymalloclib.so. A pattern can be
1241 used to match multiple libraries sonames. For example,
1242 --soname-synonyms=somalloc=*tcmalloc* will match the soname of
1243 all variants of the tcmalloc library (native, debug, profiled,
1244 ... tcmalloc variants).
1245 Note: the soname of a elf shared library can be retrieved using
1247 the readelf utility.
1248 · Replacements in a statically linked library are done by using
1250 the NONE pattern. For example, if you link with libtcmalloc.a,
1251 and only want to intercept the malloc related functions in the
1252 executable (and standard libraries) themselves, but not any
1253 other shared libraries, you can give the option
1254 --soname-synonyms=somalloc=NONE. Note that a NONE pattern will
1255 match the main executable and any shared library having no
1256 soname.
1257 · To run a "default" Firefox build for Linux, in which JEMalloc
1259 is linked in to the main executable, use
1260 --soname-synonyms=somalloc=NONE.
1261 · To only intercept allocation symbols in the default system
1263 libraries, but not in any other shared library or the
1264 executable defining public malloc or operator new related
1265 functions use a non-existing library name like
1266 --soname-synonyms=somalloc=nouserintercepts (where
1267 nouserintercepts can be any non-existing library name).
1268 · Shared library of the dynamic (runtime) linker is excluded from
1270 searching for global public symbols, such as those for the
1271 malloc related functions (identified by somalloc synonym).
1272
1274 There are also some options for debugging Valgrind itself. You
1275 shouldn't need to use them in the normal run of things. If you wish to
1276 see the list, use the --help-debug option.
1277
1279 --leak-check=<no|summary|yes|full> [default: summary]
1280 When enabled, search for memory leaks when the client program
1281 finishes. If set to summary, it says how many leaks occurred. If
1282 set to full or yes, each individual leak will be shown in detail
1283 and/or counted as an error, as specified by the options
1284 --show-leak-kinds and --errors-for-leak-kinds.
1285 --leak-resolution=<low|med|high> [default: high]
1287 When doing leak checking, determines how willing Memcheck is to
1288 consider different backtraces to be the same for the purposes of
1289 merging multiple leaks into a single leak report. When set to low,
1290 only the first two entries need match. When med, four entries have
1291 to match. When high, all entries need to match.
1292 For hardcore leak debugging, you probably want to use
1294 --leak-resolution=high together with --num-callers=40 or some such
1295 large number.
1296 Note that the --leak-resolution setting does not affect Memcheck's
1298 ability to find leaks. It only changes how the results are
1299 presented.
1300 --show-leak-kinds=<set> [default: definite,possible]
1302 Specifies the leak kinds to show in a full leak search, in one of
1303 the following ways:
1304 · a comma separated list of one or more of definite indirect
1306 possible reachable.
1307 · all to specify the complete set (all leak kinds). It is
1309 equivalent to
1310 --show-leak-kinds=definite,indirect,possible,reachable.
1311 · none for the empty set.
1313 --errors-for-leak-kinds=<set> [default: definite,possible]
1315 Specifies the leak kinds to count as errors in a full leak search.
1316 The <set> is specified similarly to --show-leak-kinds
1317 --leak-check-heuristics=<set> [default: all]
1319 Specifies the set of leak check heuristics to be used during leak
1320 searches. The heuristics control which interior pointers to a block
1321 cause it to be considered as reachable. The heuristic set is
1322 specified in one of the following ways:
1323 · a comma separated list of one or more of stdstring length64
1325 newarray multipleinheritance.
1326 · all to activate the complete set of heuristics. It is
1328 equivalent to
1329 --leak-check-heuristics=stdstring,length64,newarray,multipleinheritance.
1330 · none for the empty set.
1332 Note that these heuristics are dependent on the layout of the
1334 objects produced by the C++ compiler. They have been tested with
1335 some gcc versions (e.g. 4.4 and 4.7). They might not work properly
1336 with other C++ compilers.
1337 --show-reachable=<yes|no> , --show-possibly-lost=<yes|no>
1339 These options provide an alternative way to specify the leak kinds
1340 to show:
1341 · --show-reachable=no --show-possibly-lost=yes is equivalent to
1343 --show-leak-kinds=definite,possible.
1344 · --show-reachable=no --show-possibly-lost=no is equivalent to
1346 --show-leak-kinds=definite.
1347 · --show-reachable=yes is equivalent to --show-leak-kinds=all.
1349 Note that --show-possibly-lost=no has no effect if
1351 --show-reachable=yes is specified.
1352 --xtree-leak=<no|yes> [no]
1354 If set to yes, the results for the leak search done at exit will be
1355 output in a 'Callgrind Format' execution tree file. Note that this
1356 automatically sets the option --leak-check=full. The produced file
1357 will contain the following events:
1358 · RB : Reachable Bytes
1360 · PB : Possibly lost Bytes
1362 · IB : Indirectly lost Bytes
1364 · DB : Definitely lost Bytes (direct plus indirect)
1366 · DIB : Definitely Indirectly lost Bytes (subset of DB)
1368 · RBk : reachable Blocks
1370 · PBk : Possibly lost Blocks
1372 · IBk : Indirectly lost Blocks
1374 · DBk : Definitely lost Blocks
1376 The increase or decrease for all events above will also be output
1378 in the file to provide the delta (increase or decreaseà between 2
1379 successive leak searches. For example, iRB is the increase of the
1380 RB event, dPBk is the decrease of PBk event. The values for the
1381 increase and decrease events will be zero for the first leak search
1382 done.
1383 See ??? for a detailed explanation about execution trees.
1385 --xtree-leak-file=<filename> [default: xtleak.kcg.%p]
1387 Specifies that Valgrind should produce the xtree leak report in the
1388 specified file. Any %p, %q or %n sequences appearing in the
1389 filename are expanded in exactly the same way as they are for
1390 --log-file. See the description of --log-file for details.
1391 See ??? for a detailed explanation about execution trees formats.
1393 --undef-value-errors=<yes|no> [default: yes]
1395 Controls whether Memcheck reports uses of undefined value errors.
1396 Set this to no if you don't want to see undefined value errors. It
1397 also has the side effect of speeding up Memcheck somewhat.
1398 --track-origins=<yes|no> [default: no]
1400 Controls whether Memcheck tracks the origin of uninitialised
1401 values. By default, it does not, which means that although it can
1402 tell you that an uninitialised value is being used in a dangerous
1403 way, it cannot tell you where the uninitialised value came from.
1404 This often makes it difficult to track down the root problem.
1405 When set to yes, Memcheck keeps track of the origins of all
1407 uninitialised values. Then, when an uninitialised value error is
1408 reported, Memcheck will try to show the origin of the value. An
1409 origin can be one of the following four places: a heap block, a
1410 stack allocation, a client request, or miscellaneous other sources
1411 (eg, a call to brk).
1412 For uninitialised values originating from a heap block, Memcheck
1414 shows where the block was allocated. For uninitialised values
1415 originating from a stack allocation, Memcheck can tell you which
1416 function allocated the value, but no more than that -- typically it
1417 shows you the source location of the opening brace of the function.
1418 So you should carefully check that all of the function's local
1419 variables are initialised properly.
1420 Performance overhead: origin tracking is expensive. It halves
1422 Memcheck's speed and increases memory use by a minimum of 100MB,
1423 and possibly more. Nevertheless it can drastically reduce the
1424 effort required to identify the root cause of uninitialised value
1425 errors, and so is often a programmer productivity win, despite
1426 running more slowly.
1427 Accuracy: Memcheck tracks origins quite accurately. To avoid very
1429 large space and time overheads, some approximations are made. It is
1430 possible, although unlikely, that Memcheck will report an incorrect
1431 origin, or not be able to identify any origin.
1432 Note that the combination --track-origins=yes and
1434 --undef-value-errors=no is nonsensical. Memcheck checks for and
1435 rejects this combination at startup.
1436 --partial-loads-ok=<yes|no> [default: yes]
1438 Controls how Memcheck handles 32-, 64-, 128- and 256-bit naturally
1439 aligned loads from addresses for which some bytes are addressable
1440 and others are not. When yes, such loads do not produce an address
1441 error. Instead, loaded bytes originating from illegal addresses are
1442 marked as uninitialised, and those corresponding to legal addresses
1443 are handled in the normal way.
1444 When no, loads from partially invalid addresses are treated the
1446 same as loads from completely invalid addresses: an illegal-address
1447 error is issued, and the resulting bytes are marked as initialised.
1448 Note that code that behaves in this way is in violation of the ISO
1450 C/C++ standards, and should be considered broken. If at all
1451 possible, such code should be fixed.
1452 --expensive-definedness-checks=<yes|no> [default: no]
1454 Controls whether Memcheck should employ more precise but also more
1455 expensive (time consuming) algorithms when checking the definedness
1456 of a value. The default setting is not to do that and it is usually
1457 sufficient. However, for highly optimised code valgrind may
1458 sometimes incorrectly complain. Invoking valgrind with
1459 --expensive-definedness-checks=yes helps but comes at a performance
1460 cost. Runtime degradation of 25% have been observed but the extra
1461 cost depends a lot on the application at hand.
1462 --keep-stacktraces=alloc|free|alloc-and-free|alloc-then-free|none
1464 [default: alloc-and-free]
1465 Controls which stack trace(s) to keep for malloc'd and/or free'd
1466 blocks.
1467 With alloc-then-free, a stack trace is recorded at allocation time,
1469 and is associated with the block. When the block is freed, a second
1470 stack trace is recorded, and this replaces the allocation stack
1471 trace. As a result, any "use after free" errors relating to this
1472 block can only show a stack trace for where the block was freed.
1473 With alloc-and-free, both allocation and the deallocation stack
1475 traces for the block are stored. Hence a "use after free" error
1476 will show both, which may make the error easier to diagnose.
1477 Compared to alloc-then-free, this setting slightly increases
1478 Valgrind's memory use as the block contains two references instead
1479 of one.
1480 With alloc, only the allocation stack trace is recorded (and
1482 reported). With free, only the deallocation stack trace is recorded
1483 (and reported). These values somewhat decrease Valgrind's memory
1484 and cpu usage. They can be useful depending on the error types you
1485 are searching for and the level of detail you need to analyse them.
1486 For example, if you are only interested in memory leak errors, it
1487 is sufficient to record the allocation stack traces.
1488 With none, no stack traces are recorded for malloc and free
1490 operations. If your program allocates a lot of blocks and/or
1491 allocates/frees from many different stack traces, this can
1492 significantly decrease cpu and/or memory required. Of course, few
1493 details will be reported for errors related to heap blocks.
1494 Note that once a stack trace is recorded, Valgrind keeps the stack
1496 trace in memory even if it is not referenced by any block. Some
1497 programs (for example, recursive algorithms) can generate a huge
1498 number of stack traces. If Valgrind uses too much memory in such
1499 circumstances, you can reduce the memory required with the options
1500 --keep-stacktraces and/or by using a smaller value for the option
1501 --num-callers.
1502 If you want to use --xtree-memory=full memory profiling (see ???
1504 ), then you cannot specify --keep-stacktraces=free or
1505 --keep-stacktraces=none.
1506 --freelist-vol=<number> [default: 20000000]
1508 When the client program releases memory using free (in C) or delete
1509 (C++), that memory is not immediately made available for
1510 re-allocation. Instead, it is marked inaccessible and placed in a
1511 queue of freed blocks. The purpose is to defer as long as possible
1512 the point at which freed-up memory comes back into circulation.
1513 This increases the chance that Memcheck will be able to detect
1514 invalid accesses to blocks for some significant period of time
1515 after they have been freed.
1516 This option specifies the maximum total size, in bytes, of the
1518 blocks in the queue. The default value is twenty million bytes.
1519 Increasing this increases the total amount of memory used by
1520 Memcheck but may detect invalid uses of freed blocks which would
1521 otherwise go undetected.
1522 --freelist-big-blocks=<number> [default: 1000000]
1524 When making blocks from the queue of freed blocks available for
1525 re-allocation, Memcheck will in priority re-circulate the blocks
1526 with a size greater or equal to --freelist-big-blocks. This ensures
1527 that freeing big blocks (in particular freeing blocks bigger than
1528 --freelist-vol) does not immediately lead to a re-circulation of
1529 all (or a lot of) the small blocks in the free list. In other
1530 words, this option increases the likelihood to discover dangling
1531 pointers for the "small" blocks, even when big blocks are freed.
1532 Setting a value of 0 means that all the blocks are re-circulated in
1534 a FIFO order.
1535 --workaround-gcc296-bugs=<yes|no> [default: no]
1537 When enabled, assume that reads and writes some small distance
1538 below the stack pointer are due to bugs in GCC 2.96, and does not
1539 report them. The "small distance" is 256 bytes by default. Note
1540 that GCC 2.96 is the default compiler on some ancient Linux
1541 distributions (RedHat 7.X) and so you may need to use this option.
1542 Do not use it if you do not have to, as it can cause real errors to
1543 be overlooked. A better alternative is to use a more recent GCC in
1544 which this bug is fixed.
1545 You may also need to use this option when working with GCC 3.X or
1547 4.X on 32-bit PowerPC Linux. This is because GCC generates code
1548 which occasionally accesses below the stack pointer, particularly
1549 for floating-point to/from integer conversions. This is in
1550 violation of the 32-bit PowerPC ELF specification, which makes no
1551 provision for locations below the stack pointer to be accessible.
1552 This option is deprecated as of version 3.12 and may be removed
1554 from future versions. You should instead use
1555 --ignore-range-below-sp to specify the exact range of offsets below
1556 the stack pointer that should be ignored. A suitable equivalent is
1557 --ignore-range-below-sp=1024-1.
1558 --ignore-range-below-sp=<number>-<number>
1560 This is a more general replacement for the deprecated
1561 --workaround-gcc296-bugs option. When specified, it causes Memcheck
1562 not to report errors for accesses at the specified offsets below
1563 the stack pointer. The two offsets must be positive decimal numbers
1564 and -- somewhat counterintuitively -- the first one must be larger,
1565 in order to imply a non-wraparound address range to ignore. For
1566 example, to ignore 4 byte accesses at 8192 bytes below the stack
1567 pointer, use --ignore-range-below-sp=8192-8189. Only one range may
1568 be specified.
1569 --show-mismatched-frees=<yes|no> [default: yes]
1571 When enabled, Memcheck checks that heap blocks are deallocated
1572 using a function that matches the allocating function. That is, it
1573 expects free to be used to deallocate blocks allocated by malloc,
1574 delete for blocks allocated by new, and delete[] for blocks
1575 allocated by new[]. If a mismatch is detected, an error is
1576 reported. This is in general important because in some
1577 environments, freeing with a non-matching function can cause
1578 crashes.
1579 There is however a scenario where such mismatches cannot be
1581 avoided. That is when the user provides implementations of
1582 new/new[] that call malloc and of delete/delete[] that call free,
1583 and these functions are asymmetrically inlined. For example,
1584 imagine that delete[] is inlined but new[] is not. The result is
1585 that Memcheck "sees" all delete[] calls as direct calls to free,
1586 even when the program source contains no mismatched calls.
1587 This causes a lot of confusing and irrelevant error reports.
1589 --show-mismatched-frees=no disables these checks. It is not
1590 generally advisable to disable them, though, because you may miss
1591 real errors as a result.
1592 --ignore-ranges=0xPP-0xQQ[,0xRR-0xSS]
1594 Any ranges listed in this option (and multiple ranges can be
1595 specified, separated by commas) will be ignored by Memcheck's
1596 addressability checking.
1597 --malloc-fill=<hexnumber>
1599 Fills blocks allocated by malloc, new, etc, but not by calloc, with
1600 the specified byte. This can be useful when trying to shake out
1601 obscure memory corruption problems. The allocated area is still
1602 regarded by Memcheck as undefined -- this option only affects its
1603 contents. Note that --malloc-fill does not affect a block of memory
1604 when it is used as argument to client requests
1605 VALGRIND_MEMPOOL_ALLOC or VALGRIND_MALLOCLIKE_BLOCK.
1606 --free-fill=<hexnumber>
1608 Fills blocks freed by free, delete, etc, with the specified byte
1609 value. This can be useful when trying to shake out obscure memory
1610 corruption problems. The freed area is still regarded by Memcheck
1611 as not valid for access -- this option only affects its contents.
1612 Note that --free-fill does not affect a block of memory when it is
1613 used as argument to client requests VALGRIND_MEMPOOL_FREE or
1614 VALGRIND_FREELIKE_BLOCK.
1615
1617 --I1=<size>,<associativity>,<line size>
1618 Specify the size, associativity and line size of the level 1
1619 instruction cache.
1620 --D1=<size>,<associativity>,<line size>
1622 Specify the size, associativity and line size of the level 1 data
1623 cache.
1624 --LL=<size>,<associativity>,<line size>
1626 Specify the size, associativity and line size of the last-level
1627 cache.
1628 --cache-sim=no|yes [yes]
1630 Enables or disables collection of cache access and miss counts.
1631 --branch-sim=no|yes [no]
1633 Enables or disables collection of branch instruction and
1634 misprediction counts. By default this is disabled as it slows
1635 Cachegrind down by approximately 25%. Note that you cannot specify
1636 --cache-sim=no and --branch-sim=no together, as that would leave
1637 Cachegrind with no information to collect.
1638 --cachegrind-out-file=<file>
1640 Write the profile data to file rather than to the default output
1641 file, cachegrind.out.<pid>. The %p and %q format specifiers can be
1642 used to embed the process ID and/or the contents of an environment
1643 variable in the name, as is the case for the core option --log-
1644 file.
1645
1647 --callgrind-out-file=<file>
1648 Write the profile data to file rather than to the default output
1649 file, callgrind.out.<pid>. The %p and %q format specifiers can be
1650 used to embed the process ID and/or the contents of an environment
1651 variable in the name, as is the case for the core option --log-
1652 file. When multiple dumps are made, the file name is modified
1653 further; see below.
1654 --dump-line=<no|yes> [default: yes]
1656 This specifies that event counting should be performed at source
1657 line granularity. This allows source annotation for sources which
1658 are compiled with debug information (-g).
1659 --dump-instr=<no|yes> [default: no]
1661 This specifies that event counting should be performed at
1662 per-instruction granularity. This allows for assembly code
1663 annotation. Currently the results can only be displayed by
1664 KCachegrind.
1665 --compress-strings=<no|yes> [default: yes]
1667 This option influences the output format of the profile data. It
1668 specifies whether strings (file and function names) should be
1669 identified by numbers. This shrinks the file, but makes it more
1670 difficult for humans to read (which is not recommended in any
1671 case).
1672 --compress-pos=<no|yes> [default: yes]
1674 This option influences the output format of the profile data. It
1675 specifies whether numerical positions are always specified as
1676 absolute values or are allowed to be relative to previous numbers.
1677 This shrinks the file size.
1678 --combine-dumps=<no|yes> [default: no]
1680 When enabled, when multiple profile data parts are to be generated
1681 these parts are appended to the same output file. Not recommended.
1682 --dump-every-bb=<count> [default: 0, never]
1684 Dump profile data every count basic blocks. Whether a dump is
1685 needed is only checked when Valgrind's internal scheduler is run.
1686 Therefore, the minimum setting useful is about 100000. The count is
1687 a 64-bit value to make long dump periods possible.
1688 --dump-before=<function>
1690 Dump when entering function.
1691 --zero-before=<function>
1693 Zero all costs when entering function.
1694 --dump-after=<function>
1696 Dump when leaving function.
1697 --instr-atstart=<yes|no> [default: yes]
1699 Specify if you want Callgrind to start simulation and profiling
1700 from the beginning of the program. When set to no, Callgrind will
1701 not be able to collect any information, including calls, but it
1702 will have at most a slowdown of around 4, which is the minimum
1703 Valgrind overhead. Instrumentation can be interactively enabled via
1704 callgrind_control -i on.
1705 Note that the resulting call graph will most probably not contain
1707 main, but will contain all the functions executed after
1708 instrumentation was enabled. Instrumentation can also
1709 programatically enabled/disabled. See the Callgrind include file
1710 callgrind.h for the macro you have to use in your source code.
1711 For cache simulation, results will be less accurate when switching
1713 on instrumentation later in the program run, as the simulator
1714 starts with an empty cache at that moment. Switch on event
1715 collection later to cope with this error.
1716 --collect-atstart=<yes|no> [default: yes]
1718 Specify whether event collection is enabled at beginning of the
1719 profile run.
1720 To only look at parts of your program, you have two possibilities:
1722 1. Zero event counters before entering the program part you want
1724 to profile, and dump the event counters to a file after leaving
1725 that program part.
1726 2. Switch on/off collection state as needed to only see event
1728 counters happening while inside of the program part you want to
1729 profile.
1730 The second option can be used if the program part you want to
1732 profile is called many times. Option 1, i.e. creating a lot of
1733 dumps is not practical here.
1734 Collection state can be toggled at entry and exit of a given
1736 function with the option --toggle-collect. If you use this option,
1737 collection state should be disabled at the beginning. Note that the
1738 specification of --toggle-collect implicitly sets
1739 --collect-state=no.
1740 Collection state can be toggled also by inserting the client
1742 request CALLGRIND_TOGGLE_COLLECT ; at the needed code positions.
1743 --toggle-collect=<function>
1745 Toggle collection on entry/exit of function.
1746 --collect-jumps=<no|yes> [default: no]
1748 This specifies whether information for (conditional) jumps should
1749 be collected. As above, callgrind_annotate currently is not able to
1750 show you the data. You have to use KCachegrind to get jump arrows
1751 in the annotated code.
1752 --collect-systime=<no|yes> [default: no]
1754 This specifies whether information for system call times should be
1755 collected.
1756 --collect-bus=<no|yes> [default: no]
1758 This specifies whether the number of global bus events executed
1759 should be collected. The event type "Ge" is used for these events.
1760 --cache-sim=<yes|no> [default: no]
1762 Specify if you want to do full cache simulation. By default, only
1763 instruction read accesses will be counted ("Ir"). With cache
1764 simulation, further event counters are enabled: Cache misses on
1765 instruction reads ("I1mr"/"ILmr"), data read accesses ("Dr") and
1766 related cache misses ("D1mr"/"DLmr"), data write accesses ("Dw")
1767 and related cache misses ("D1mw"/"DLmw"). For more information, see
1768 Cachegrind: a cache and branch-prediction profiler.
1769 --branch-sim=<yes|no> [default: no]
1771 Specify if you want to do branch prediction simulation. Further
1772 event counters are enabled: Number of executed conditional branches
1773 and related predictor misses ("Bc"/"Bcm"), executed indirect jumps
1774 and related misses of the jump address predictor ("Bi"/"Bim").
1775
1777 --free-is-write=no|yes [default: no]
1778 When enabled (not the default), Helgrind treats freeing of heap
1779 memory as if the memory was written immediately before the free.
1780 This exposes races where memory is referenced by one thread, and
1781 freed by another, but there is no observable synchronisation event
1782 to ensure that the reference happens before the free.
1783 This functionality is new in Valgrind 3.7.0, and is regarded as
1785 experimental. It is not enabled by default because its interaction
1786 with custom memory allocators is not well understood at present.
1787 User feedback is welcomed.
1788 --track-lockorders=no|yes [default: yes]
1790 When enabled (the default), Helgrind performs lock order
1791 consistency checking. For some buggy programs, the large number of
1792 lock order errors reported can become annoying, particularly if
1793 you're only interested in race errors. You may therefore find it
1794 helpful to disable lock order checking.
1795 --history-level=none|approx|full [default: full]
1797 --history-level=full (the default) causes Helgrind collects enough
1798 information about "old" accesses that it can produce two stack
1799 traces in a race report -- both the stack trace for the current
1800 access, and the trace for the older, conflicting access. To limit
1801 memory usage, "old" accesses stack traces are limited to a maximum
1802 of 8 entries, even if --num-callers value is bigger.
1803 Collecting such information is expensive in both speed and memory,
1805 particularly for programs that do many inter-thread synchronisation
1806 events (locks, unlocks, etc). Without such information, it is more
1807 difficult to track down the root causes of races. Nonetheless, you
1808 may not need it in situations where you just want to check for the
1809 presence or absence of races, for example, when doing regression
1810 testing of a previously race-free program.
1811 --history-level=none is the opposite extreme. It causes Helgrind
1813 not to collect any information about previous accesses. This can be
1814 dramatically faster than --history-level=full.
1815 --history-level=approx provides a compromise between these two
1817 extremes. It causes Helgrind to show a full trace for the later
1818 access, and approximate information regarding the earlier access.
1819 This approximate information consists of two stacks, and the
1820 earlier access is guaranteed to have occurred somewhere between
1821 program points denoted by the two stacks. This is not as useful as
1822 showing the exact stack for the previous access (as
1823 --history-level=full does), but it is better than nothing, and it
1824 is almost as fast as --history-level=none.
1825 --conflict-cache-size=N [default: 1000000]
1827 This flag only has any effect at --history-level=full.
1828 Information about "old" conflicting accesses is stored in a cache
1830 of limited size, with LRU-style management. This is necessary
1831 because it isn't practical to store a stack trace for every single
1832 memory access made by the program. Historical information on not
1833 recently accessed locations is periodically discarded, to free up
1834 space in the cache.
1835 This option controls the size of the cache, in terms of the number
1837 of different memory addresses for which conflicting access
1838 information is stored. If you find that Helgrind is showing race
1839 errors with only one stack instead of the expected two stacks, try
1840 increasing this value.
1841 The minimum value is 10,000 and the maximum is 30,000,000 (thirty
1843 times the default value). Increasing the value by 1 increases
1844 Helgrind's memory requirement by very roughly 100 bytes, so the
1845 maximum value will easily eat up three extra gigabytes or so of
1846 memory.
1847 --check-stack-refs=no|yes [default: yes]
1849 By default Helgrind checks all data memory accesses made by your
1850 program. This flag enables you to skip checking for accesses to
1851 thread stacks (local variables). This can improve performance, but
1852 comes at the cost of missing races on stack-allocated data.
1853 --ignore-thread-creation=<yes|no> [default: no]
1855 Controls whether all activities during thread creation should be
1856 ignored. By default enabled only on Solaris. Solaris provides
1857 higher throughput, parallelism and scalability than other operating
1858 systems, at the cost of more fine-grained locking activity. This
1859 means for example that when a thread is created under glibc, just
1860 one big lock is used for all thread setup. Solaris libc uses
1861 several fine-grained locks and the creator thread resumes its
1862 activities as soon as possible, leaving for example stack and TLS
1863 setup sequence to the created thread. This situation confuses
1864 Helgrind as it assumes there is some false ordering in place
1865 between creator and created thread; and therefore many types of
1866 race conditions in the application would not be reported. To
1867 prevent such false ordering, this command line option is set to yes
1868 by default on Solaris. All activity (loads, stores, client
1869 requests) is therefore ignored during:
1870 · pthread_create() call in the creator thread
1872 · thread creation phase (stack and TLS setup) in the created
1874 thread
1875 Also new memory allocated during thread creation is untracked, that
1877 is race reporting is suppressed there. DRD does the same thing
1878 implicitly. This is necessary because Solaris libc caches many
1879 objects and reuses them for different threads and that confuses
1880 Helgrind.
1881
1883 --check-stack-var=<yes|no> [default: no]
1884 Controls whether DRD detects data races on stack variables.
1885 Verifying stack variables is disabled by default because most
1886 programs do not share stack variables over threads.
1887 --exclusive-threshold=<n> [default: off]
1889 Print an error message if any mutex or writer lock has been held
1890 longer than the time specified in milliseconds. This option enables
1891 the detection of lock contention.
1892 --join-list-vol=<n> [default: 10]
1894 Data races that occur between a statement at the end of one thread
1895 and another thread can be missed if memory access information is
1896 discarded immediately after a thread has been joined. This option
1897 allows to specify for how many joined threads memory access
1898 information should be retained.
1899 --first-race-only=<yes|no> [default: no]
1901 Whether to report only the first data race that has been detected
1902 on a memory location or all data races that have been detected on a
1903 memory location.
1904 --free-is-write=<yes|no> [default: no]
1906 Whether to report races between accessing memory and freeing
1907 memory. Enabling this option may cause DRD to run slightly slower.
1908 Notes:
1909 · Don't enable this option when using custom memory allocators
1911 that use the VG_USERREQ__MALLOCLIKE_BLOCK and
1912 VG_USERREQ__FREELIKE_BLOCK because that would result in false
1913 positives.
1914 · Don't enable this option when using reference-counted objects
1916 because that will result in false positives, even when that
1917 code has been annotated properly with ANNOTATE_HAPPENS_BEFORE
1918 and ANNOTATE_HAPPENS_AFTER. See e.g. the output of the
1919 following command for an example: valgrind --tool=drd
1920 --free-is-write=yes drd/tests/annotate_smart_pointer.
1921 --report-signal-unlocked=<yes|no> [default: yes]
1923 Whether to report calls to pthread_cond_signal and
1924 pthread_cond_broadcast where the mutex associated with the signal
1925 through pthread_cond_wait or pthread_cond_timed_waitis not locked
1926 at the time the signal is sent. Sending a signal without holding a
1927 lock on the associated mutex is a common programming error which
1928 can cause subtle race conditions and unpredictable behavior. There
1929 exist some uncommon synchronization patterns however where it is
1930 safe to send a signal without holding a lock on the associated
1931 mutex.
1932 --segment-merging=<yes|no> [default: yes]
1934 Controls segment merging. Segment merging is an algorithm to limit
1935 memory usage of the data race detection algorithm. Disabling
1936 segment merging may improve the accuracy of the so-called 'other
1937 segments' displayed in race reports but can also trigger an out of
1938 memory error.
1939 --segment-merging-interval=<n> [default: 10]
1941 Perform segment merging only after the specified number of new
1942 segments have been created. This is an advanced configuration
1943 option that allows to choose whether to minimize DRD's memory usage
1944 by choosing a low value or to let DRD run faster by choosing a
1945 slightly higher value. The optimal value for this parameter depends
1946 on the program being analyzed. The default value works well for
1947 most programs.
1948 --shared-threshold=<n> [default: off]
1950 Print an error message if a reader lock has been held longer than
1951 the specified time (in milliseconds). This option enables the
1952 detection of lock contention.
1953 --show-confl-seg=<yes|no> [default: yes]
1955 Show conflicting segments in race reports. Since this information
1956 can help to find the cause of a data race, this option is enabled
1957 by default. Disabling this option makes the output of DRD more
1958 compact.
1959 --show-stack-usage=<yes|no> [default: no]
1961 Print stack usage at thread exit time. When a program creates a
1962 large number of threads it becomes important to limit the amount of
1963 virtual memory allocated for thread stacks. This option makes it
1964 possible to observe how much stack memory has been used by each
1965 thread of the client program. Note: the DRD tool itself allocates
1966 some temporary data on the client thread stack. The space necessary
1967 for this temporary data must be allocated by the client program
1968 when it allocates stack memory, but is not included in stack usage
1969 reported by DRD.
1970 --ignore-thread-creation=<yes|no> [default: no]
1972 Controls whether all activities during thread creation should be
1973 ignored. By default enabled only on Solaris. Solaris provides
1974 higher throughput, parallelism and scalability than other operating
1975 systems, at the cost of more fine-grained locking activity. This
1976 means for example that when a thread is created under glibc, just
1977 one big lock is used for all thread setup. Solaris libc uses
1978 several fine-grained locks and the creator thread resumes its
1979 activities as soon as possible, leaving for example stack and TLS
1980 setup sequence to the created thread. This situation confuses DRD
1981 as it assumes there is some false ordering in place between creator
1982 and created thread; and therefore many types of race conditions in
1983 the application would not be reported. To prevent such false
1984 ordering, this command line option is set to yes by default on
1985 Solaris. All activity (loads, stores, client requests) is therefore
1986 ignored during:
1987 · pthread_create() call in the creator thread
1989 · thread creation phase (stack and TLS setup) in the created
1991 thread
1992 --trace-addr=<address> [default: none]
1994 Trace all load and store activity for the specified address. This
1995 option may be specified more than once.
1996 --ptrace-addr=<address> [default: none]
1998 Trace all load and store activity for the specified address and
1999 keep doing that even after the memory at that address has been
2000 freed and reallocated.
2001 --trace-alloc=<yes|no> [default: no]
2003 Trace all memory allocations and deallocations. May produce a huge
2004 amount of output.
2005 --trace-barrier=<yes|no> [default: no]
2007 Trace all barrier activity.
2008 --trace-cond=<yes|no> [default: no]
2010 Trace all condition variable activity.
2011 --trace-fork-join=<yes|no> [default: no]
2013 Trace all thread creation and all thread termination events.
2014 --trace-hb=<yes|no> [default: no]
2016 Trace execution of the ANNOTATE_HAPPENS_BEFORE(),
2017 ANNOTATE_HAPPENS_AFTER() and ANNOTATE_HAPPENS_DONE() client
2018 requests.
2019 --trace-mutex=<yes|no> [default: no]
2021 Trace all mutex activity.
2022 --trace-rwlock=<yes|no> [default: no]
2024 Trace all reader-writer lock activity.
2025 --trace-semaphore=<yes|no> [default: no]
2027 Trace all semaphore activity.
2028
2030 --heap=<yes|no> [default: yes]
2031 Specifies whether heap profiling should be done.
2032 --heap-admin=<size> [default: 8]
2034 If heap profiling is enabled, gives the number of administrative
2035 bytes per block to use. This should be an estimate of the average,
2036 since it may vary. For example, the allocator used by glibc on
2037 Linux requires somewhere between 4 to 15 bytes per block, depending
2038 on various factors. That allocator also requires admin space for
2039 freed blocks, but Massif cannot account for this.
2040 --stacks=<yes|no> [default: no]
2042 Specifies whether stack profiling should be done. This option slows
2043 Massif down greatly, and so is off by default. Note that Massif
2044 assumes that the main stack has size zero at start-up. This is not
2045 true, but doing otherwise accurately is difficult. Furthermore,
2046 starting at zero better indicates the size of the part of the main
2047 stack that a user program actually has control over.
2048 --pages-as-heap=<yes|no> [default: no]
2050 Tells Massif to profile memory at the page level rather than at the
2051 malloc'd block level. See above for details.
2052 --depth=<number> [default: 30]
2054 Maximum depth of the allocation trees recorded for detailed
2055 snapshots. Increasing it will make Massif run somewhat more slowly,
2056 use more memory, and produce bigger output files.
2057 --alloc-fn=<name>
2059 Functions specified with this option will be treated as though they
2060 were a heap allocation function such as malloc. This is useful for
2061 functions that are wrappers to malloc or new, which can fill up the
2062 allocation trees with uninteresting information. This option can be
2063 specified multiple times on the command line, to name multiple
2064 functions.
2065 Note that the named function will only be treated this way if it is
2067 the top entry in a stack trace, or just below another function
2068 treated this way. For example, if you have a function malloc1 that
2069 wraps malloc, and malloc2 that wraps malloc1, just specifying
2070 --alloc-fn=malloc2 will have no effect. You need to specify
2071 --alloc-fn=malloc1 as well. This is a little inconvenient, but the
2072 reason is that checking for allocation functions is slow, and it
2073 saves a lot of time if Massif can stop looking through the stack
2074 trace entries as soon as it finds one that doesn't match rather
2075 than having to continue through all the entries.
2076 Note that C++ names are demangled. Note also that overloaded C++
2078 names must be written in full. Single quotes may be necessary to
2079 prevent the shell from breaking them up. For example:
2080 --alloc-fn='operator new(unsigned, std::nothrow_t const&)'
2082 --ignore-fn=<name>
2085 Any direct heap allocation (i.e. a call to malloc, new, etc, or a
2086 call to a function named by an --alloc-fn option) that occurs in a
2087 function specified by this option will be ignored. This is mostly
2088 useful for testing purposes. This option can be specified multiple
2089 times on the command line, to name multiple functions.
2090 Any realloc of an ignored block will also be ignored, even if the
2092 realloc call does not occur in an ignored function. This avoids the
2093 possibility of negative heap sizes if ignored blocks are shrunk
2094 with realloc.
2095 The rules for writing C++ function names are the same as for
2097 --alloc-fn above.
2098 --threshold=<m.n> [default: 1.0]
2100 The significance threshold for heap allocations, as a percentage of
2101 total memory size. Allocation tree entries that account for less
2102 than this will be aggregated. Note that this should be specified in
2103 tandem with ms_print's option of the same name.
2104 --peak-inaccuracy=<m.n> [default: 1.0]
2106 Massif does not necessarily record the actual global memory
2107 allocation peak; by default it records a peak only when the global
2108 memory allocation size exceeds the previous peak by at least 1.0%.
2109 This is because there can be many local allocation peaks along the
2110 way, and doing a detailed snapshot for every one would be expensive
2111 and wasteful, as all but one of them will be later discarded. This
2112 inaccuracy can be changed (even to 0.0%) via this option, but
2113 Massif will run drastically slower as the number approaches zero.
2114 --time-unit=<i|ms|B> [default: i]
2116 The time unit used for the profiling. There are three
2117 possibilities: instructions executed (i), which is good for most
2118 cases; real (wallclock) time (ms, i.e. milliseconds), which is
2119 sometimes useful; and bytes allocated/deallocated on the heap
2120 and/or stack (B), which is useful for very short-run programs, and
2121 for testing purposes, because it is the most reproducible across
2122 different machines.
2123 --detailed-freq=<n> [default: 10]
2125 Frequency of detailed snapshots. With --detailed-freq=1, every
2126 snapshot is detailed.
2127 --max-snapshots=<n> [default: 100]
2129 The maximum number of snapshots recorded. If set to N, for all
2130 programs except very short-running ones, the final number of
2131 snapshots will be between N/2 and N.
2132 --massif-out-file=<file> [default: massif.out.%p]
2134 Write the profile data to file rather than to the default output
2135 file, massif.out.<pid>. The %p and %q format specifiers can be used
2136 to embed the process ID and/or the contents of an environment
2137 variable in the name, as is the case for the core option --log-
2138 file.
2139
2141 There are no SGCheck-specific command-line options at present.
2142
2144 --bb-out-file=<name> [default: bb.out.%p]
2145 This option selects the name of the basic block vector file. The %p
2146 and %q format specifiers can be used to embed the process ID and/or
2147 the contents of an environment variable in the name, as is the case
2148 for the core option --log-file.
2149 --pc-out-file=<name> [default: pc.out.%p]
2151 This option selects the name of the PC file. This file holds
2152 program counter addresses and function name info for the various
2153 basic blocks. This can be used in conjunction with the basic block
2154 vector file to fast-forward via function names instead of just
2155 instruction counts. The %p and %q format specifiers can be used to
2156 embed the process ID and/or the contents of an environment variable
2157 in the name, as is the case for the core option --log-file.
2158 --interval-size=<number> [default: 100000000]
2160 This option selects the size of the interval to use. The default is
2161 100 million instructions, which is a commonly used value. Other
2162 sizes can be used; smaller intervals can help programs with
2163 finer-grained phases. However smaller interval size can lead to
2164 accuracy issues due to warm-up effects (When fast-forwarding the
2165 various architectural features will be un-initialized, and it will
2166 take some number of instructions before they "warm up" to the state
2167 a full simulation would be at without the fast-forwarding. Large
2168 interval sizes tend to mitigate this.)
2169 --instr-count-only [default: no]
2171 This option tells the tool to only display instruction count
2172 totals, and to not generate the actual basic block vector file.
2173 This is useful for debugging, and for gathering instruction count
2174 info without generating the large basic block vector files.
2175
2177 --basic-counts=<no|yes> [default: yes]
2178 When enabled, Lackey prints the following statistics and
2179 information about the execution of the client program:
2180 1. The number of calls to the function specified by the --fnname
2182 option (the default is main). If the program has had its
2183 symbols stripped, the count will always be zero.
2184 2. The number of conditional branches encountered and the number
2186 and proportion of those taken.
2187 3. The number of superblocks entered and completed by the program.
2189 Note that due to optimisations done by the JIT, this is not at
2190 all an accurate value.
2191 4. The number of guest (x86, amd64, ppc, etc.) instructions and IR
2193 statements executed. IR is Valgrind's RISC-like intermediate
2194 representation via which all instrumentation is done.
2195 5. Ratios between some of these counts.
2197 6. The exit code of the client program.
2199 --detailed-counts=<no|yes> [default: no]
2201 When enabled, Lackey prints a table containing counts of loads,
2202 stores and ALU operations, differentiated by their IR types. The IR
2203 types are identified by their IR name ("I1", "I8", ... "I128",
2204 "F32", "F64", and "V128").
2205 --trace-mem=<no|yes> [default: no]
2207 When enabled, Lackey prints the size and address of almost every
2208 memory access made by the program. See the comments at the top of
2209 the file lackey/lk_main.c for details about the output format, how
2210 it works, and inaccuracies in the address trace. Note that this
2211 option produces immense amounts of output.
2212 --trace-superblocks=<no|yes> [default: no]
2214 When enabled, Lackey prints out the address of every superblock (a
2215 single entry, multiple exit, linear chunk of code) executed by the
2216 program. This is primarily of interest to Valgrind developers. See
2217 the comments at the top of the file lackey/lk_main.c for details
2218 about the output format. Note that this option produces large
2219 amounts of output.
2220 --fnname=<name> [default: main]
2222 Changes the function for which calls are counted when
2223 --basic-counts=yes is specified.
2224
2226 cg_annotate(1), callgrind_annotate(1), callgrind_control(1),
2227 ms_print(1), $INSTALL/share/doc/valgrind/html/index.html or
2228 http://www.valgrind.org/docs/manual/index.html, Debugging your program
2229 using Valgrind's gdbserver and GDB[1] vgdb[2], Valgrind monitor
2230 commands[3], The Commentary[4], Scheduling and Multi-Thread
2231 Performance[5], Cachegrind: a cache and branch-prediction profiler[6].
2232
2234 See the AUTHORS file in the valgrind distribution for a comprehensive
2235 list of authors.
2236 This manpage was written by Andres Roldan <aroldan@debian.org> and the
2238 Valgrind developers.
2239
2241 1. Debugging your program using Valgrind's gdbserver and GDB
2242 http://www.valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.gdbserver
2243 2. vgdb
2245 http://www.valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.vgdb
2246 3. Valgrind monitor commands
2248 http://www.valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.valgrind-monitor-commands
2249 4. The Commentary
2251 http://www.valgrind.org/docs/manual/manual-core.html#manual-core.comment
2252 5. Scheduling and Multi-Thread Performance
2254 http://www.valgrind.org/docs/manual/manual-core.html#manual-core.pthreads_perf_sched
2255 6. Cachegrind: a cache and branch-prediction profiler
2257 http://www.valgrind.org/docs/manual/cg-manual.html
2258Release 3.13.0 06/15/2017 VALGRIND(1)