1VALGRIND(1) valgrind VALGRIND(1)
2
6 valgrind - a suite of tools for debugging and profiling programs
7
9 valgrind [valgrind-options] [your-program] [your-program-options]
10
12 Valgrind is a flexible program for debugging and profiling Linux
13 executables. It consists of a core, which provides a synthetic CPU in
14 software, and a series of debugging and profiling tools. The
15 architecture is modular, so that new tools can be created easily and
16 without disturbing the existing structure.
17 Some of the options described below work with all Valgrind tools, and
19 some only work with a few or one. The section MEMCHECK OPTIONS and
20 those below it describe tool-specific options.
21 This manual page covers only basic usage and options. For more
23 comprehensive information, please see the HTML documentation on your
24 system: $INSTALL/share/doc/valgrind/html/index.html, or online:
25 http://www.valgrind.org/docs/manual/index.html.
26
28 The single most important option.
29 --tool=<toolname> [default: memcheck]
31 Run the Valgrind tool called toolname, e.g. memcheck, cachegrind,
32 callgrind, helgrind, drd, massif, dhat, lackey, none, exp-bbv, etc.
33
35 These options work with all tools.
36 -h --help
38 Show help for all options, both for the core and for the selected
39 tool. If the option is repeated it is equivalent to giving
40 --help-debug.
41 --help-debug
43 Same as --help, but also lists debugging options which usually are
44 only of use to Valgrind's developers.
45 --version
47 Show the version number of the Valgrind core. Tools can have their
48 own version numbers. There is a scheme in place to ensure that
49 tools only execute when the core version is one they are known to
50 work with. This was done to minimise the chances of strange
51 problems arising from tool-vs-core version incompatibilities.
52 -q, --quiet
54 Run silently, and only print error messages. Useful if you are
55 running regression tests or have some other automated test
56 machinery.
57 -v, --verbose
59 Be more verbose. Gives extra information on various aspects of your
60 program, such as: the shared objects loaded, the suppressions used,
61 the progress of the instrumentation and execution engines, and
62 warnings about unusual behaviour. Repeating the option increases
63 the verbosity level.
64 --trace-children=<yes|no> [default: no]
66 When enabled, Valgrind will trace into sub-processes initiated via
67 the exec system call. This is necessary for multi-process programs.
68 Note that Valgrind does trace into the child of a fork (it would be
70 difficult not to, since fork makes an identical copy of a process),
71 so this option is arguably badly named. However, most children of
72 fork calls immediately call exec anyway.
73 --trace-children-skip=patt1,patt2,...
75 This option only has an effect when --trace-children=yes is
76 specified. It allows for some children to be skipped. The option
77 takes a comma separated list of patterns for the names of child
78 executables that Valgrind should not trace into. Patterns may
79 include the metacharacters ? and *, which have the usual meaning.
80 This can be useful for pruning uninteresting branches from a tree
82 of processes being run on Valgrind. But you should be careful when
83 using it. When Valgrind skips tracing into an executable, it
84 doesn't just skip tracing that executable, it also skips tracing
85 any of that executable's child processes. In other words, the flag
86 doesn't merely cause tracing to stop at the specified executables
87 -- it skips tracing of entire process subtrees rooted at any of the
88 specified executables.
89 --trace-children-skip-by-arg=patt1,patt2,...
91 This is the same as --trace-children-skip, with one difference: the
92 decision as to whether to trace into a child process is made by
93 examining the arguments to the child process, rather than the name
94 of its executable.
95 --child-silent-after-fork=<yes|no> [default: no]
97 When enabled, Valgrind will not show any debugging or logging
98 output for the child process resulting from a fork call. This can
99 make the output less confusing (although more misleading) when
100 dealing with processes that create children. It is particularly
101 useful in conjunction with --trace-children=. Use of this option is
102 also strongly recommended if you are requesting XML output
103 (--xml=yes), since otherwise the XML from child and parent may
104 become mixed up, which usually makes it useless.
105 --vgdb=<no|yes|full> [default: yes]
107 Valgrind will provide "gdbserver" functionality when --vgdb=yes or
108 --vgdb=full is specified. This allows an external GNU GDB debugger
109 to control and debug your program when it runs on Valgrind.
110 --vgdb=full incurs significant performance overheads, but provides
111 more precise breakpoints and watchpoints. See Debugging your
112 program using Valgrind's gdbserver and GDB for a detailed
113 description.
114 If the embedded gdbserver is enabled but no gdb is currently being
116 used, the vgdb command line utility can send "monitor commands" to
117 Valgrind from a shell. The Valgrind core provides a set of Valgrind
118 monitor commands. A tool can optionally provide tool specific
119 monitor commands, which are documented in the tool specific
120 chapter.
121 --vgdb-error=<number> [default: 999999999]
123 Use this option when the Valgrind gdbserver is enabled with
124 --vgdb=yes or --vgdb=full. Tools that report errors will wait for
125 "number" errors to be reported before freezing the program and
126 waiting for you to connect with GDB. It follows that a value of
127 zero will cause the gdbserver to be started before your program is
128 executed. This is typically used to insert GDB breakpoints before
129 execution, and also works with tools that do not report errors,
130 such as Massif.
131 --vgdb-stop-at=<set> [default: none]
133 Use this option when the Valgrind gdbserver is enabled with
134 --vgdb=yes or --vgdb=full. The Valgrind gdbserver will be invoked
135 for each error after --vgdb-error have been reported. You can
136 additionally ask the Valgrind gdbserver to be invoked for other
137 events, specified in one of the following ways:
138 · a comma separated list of one or more of startup exit
140 valgrindabexit.
141 The values startup exit valgrindabexit respectively indicate to
143 invoke gdbserver before your program is executed, after the
144 last instruction of your program, on Valgrind abnormal exit
145 (e.g. internal error, out of memory, ...).
146 Note: startup and --vgdb-error=0 will both cause Valgrind
148 gdbserver to be invoked before your program is executed. The
149 --vgdb-error=0 will in addition cause your program to stop on
150 all subsequent errors.
151 · all to specify the complete set. It is equivalent to
153 --vgdb-stop-at=startup,exit,valgrindabexit.
154 · none for the empty set.
156 --track-fds=<yes|no> [default: no]
158 When enabled, Valgrind will print out a list of open file
159 descriptors on exit or on request, via the gdbserver monitor
160 command v.info open_fds. Along with each file descriptor is printed
161 a stack backtrace of where the file was opened and any details
162 relating to the file descriptor such as the file name or socket
163 details.
164 --time-stamp=<yes|no> [default: no]
166 When enabled, each message is preceded with an indication of the
167 elapsed wallclock time since startup, expressed as days, hours,
168 minutes, seconds and milliseconds.
169 --log-fd=<number> [default: 2, stderr]
171 Specifies that Valgrind should send all of its messages to the
172 specified file descriptor. The default, 2, is the standard error
173 channel (stderr). Note that this may interfere with the client's
174 own use of stderr, as Valgrind's output will be interleaved with
175 any output that the client sends to stderr.
176 --log-file=<filename>
178 Specifies that Valgrind should send all of its messages to the
179 specified file. If the file name is empty, it causes an abort.
180 There are three special format specifiers that can be used in the
181 file name.
182 %p is replaced with the current process ID. This is very useful for
184 program that invoke multiple processes. WARNING: If you use
185 --trace-children=yes and your program invokes multiple processes OR
186 your program forks without calling exec afterwards, and you don't
187 use this specifier (or the %q specifier below), the Valgrind output
188 from all those processes will go into one file, possibly jumbled
189 up, and possibly incomplete. Note: If the program forks and calls
190 exec afterwards, Valgrind output of the child from the period
191 between fork and exec will be lost. Fortunately this gap is really
192 tiny for most programs; and modern programs use posix_spawn anyway.
193 %n is replaced with a file sequence number unique for this process.
195 This is useful for processes that produces several files from the
196 same filename template.
197 %q{FOO} is replaced with the contents of the environment variable
199 FOO. If the {FOO} part is malformed, it causes an abort. This
200 specifier is rarely needed, but very useful in certain
201 circumstances (eg. when running MPI programs). The idea is that you
202 specify a variable which will be set differently for each process
203 in the job, for example BPROC_RANK or whatever is applicable in
204 your MPI setup. If the named environment variable is not set, it
205 causes an abort. Note that in some shells, the { and } characters
206 may need to be escaped with a backslash.
207 %% is replaced with %.
209 If an % is followed by any other character, it causes an abort.
211 If the file name specifies a relative file name, it is put in the
213 program's initial working directory: this is the current directory
214 when the program started its execution after the fork or after the
215 exec. If it specifies an absolute file name (ie. starts with '/')
216 then it is put there.
217 --log-socket=<ip-address:port-number>
219 Specifies that Valgrind should send all of its messages to the
220 specified port at the specified IP address. The port may be
221 omitted, in which case port 1500 is used. If a connection cannot be
222 made to the specified socket, Valgrind falls back to writing output
223 to the standard error (stderr). This option is intended to be used
224 in conjunction with the valgrind-listener program. For further
225 details, see the commentary in the manual.
226
228 These options are used by all tools that can report errors, e.g.
229 Memcheck, but not Cachegrind.
230 --xml=<yes|no> [default: no]
232 When enabled, the important parts of the output (e.g. tool error
233 messages) will be in XML format rather than plain text.
234 Furthermore, the XML output will be sent to a different output
235 channel than the plain text output. Therefore, you also must use
236 one of --xml-fd, --xml-file or --xml-socket to specify where the
237 XML is to be sent.
238 Less important messages will still be printed in plain text, but
240 because the XML output and plain text output are sent to different
241 output channels (the destination of the plain text output is still
242 controlled by --log-fd, --log-file and --log-socket) this should
243 not cause problems.
244 This option is aimed at making life easier for tools that consume
246 Valgrind's output as input, such as GUI front ends. Currently this
247 option works with Memcheck, Helgrind and DRD. The output format is
248 specified in the file docs/internals/xml-output-protocol4.txt in
249 the source tree for Valgrind 3.5.0 or later.
250 The recommended options for a GUI to pass, when requesting XML
252 output, are: --xml=yes to enable XML output, --xml-file to send the
253 XML output to a (presumably GUI-selected) file, --log-file to send
254 the plain text output to a second GUI-selected file,
255 --child-silent-after-fork=yes, and -q to restrict the plain text
256 output to critical error messages created by Valgrind itself. For
257 example, failure to read a specified suppressions file counts as a
258 critical error message. In this way, for a successful run the text
259 output file will be empty. But if it isn't empty, then it will
260 contain important information which the GUI user should be made
261 aware of.
262 --xml-fd=<number> [default: -1, disabled]
264 Specifies that Valgrind should send its XML output to the specified
265 file descriptor. It must be used in conjunction with --xml=yes.
266 --xml-file=<filename>
268 Specifies that Valgrind should send its XML output to the specified
269 file. It must be used in conjunction with --xml=yes. Any %p or %q
270 sequences appearing in the filename are expanded in exactly the
271 same way as they are for --log-file. See the description of --log-
272 file for details.
273 --xml-socket=<ip-address:port-number>
275 Specifies that Valgrind should send its XML output the specified
276 port at the specified IP address. It must be used in conjunction
277 with --xml=yes. The form of the argument is the same as that used
278 by --log-socket. See the description of --log-socket for further
279 details.
280 --xml-user-comment=<string>
282 Embeds an extra user comment string at the start of the XML output.
283 Only works when --xml=yes is specified; ignored otherwise.
284 --demangle=<yes|no> [default: yes]
286 Enable/disable automatic demangling (decoding) of C++ names.
287 Enabled by default. When enabled, Valgrind will attempt to
288 translate encoded C++ names back to something approaching the
289 original. The demangler handles symbols mangled by g++ versions
290 2.X, 3.X and 4.X.
291 An important fact about demangling is that function names mentioned
293 in suppressions files should be in their mangled form. Valgrind
294 does not demangle function names when searching for applicable
295 suppressions, because to do otherwise would make suppression file
296 contents dependent on the state of Valgrind's demangling machinery,
297 and also slow down suppression matching.
298 --num-callers=<number> [default: 12]
300 Specifies the maximum number of entries shown in stack traces that
301 identify program locations. Note that errors are commoned up using
302 only the top four function locations (the place in the current
303 function, and that of its three immediate callers). So this doesn't
304 affect the total number of errors reported.
305 The maximum value for this is 500. Note that higher settings will
307 make Valgrind run a bit more slowly and take a bit more memory, but
308 can be useful when working with programs with deeply-nested call
309 chains.
310 --unw-stack-scan-thresh=<number> [default: 0] ,
312 --unw-stack-scan-frames=<number> [default: 5]
313 Stack-scanning support is available only on ARM targets.
314 These flags enable and control stack unwinding by stack scanning.
316 When the normal stack unwinding mechanisms -- usage of Dwarf CFI
317 records, and frame-pointer following -- fail, stack scanning may be
318 able to recover a stack trace.
319 Note that stack scanning is an imprecise, heuristic mechanism that
321 may give very misleading results, or none at all. It should be used
322 only in emergencies, when normal unwinding fails, and it is
323 important to nevertheless have stack traces.
324 Stack scanning is a simple technique: the unwinder reads words from
326 the stack, and tries to guess which of them might be return
327 addresses, by checking to see if they point just after ARM or Thumb
328 call instructions. If so, the word is added to the backtrace.
329 The main danger occurs when a function call returns, leaving its
331 return address exposed, and a new function is called, but the new
332 function does not overwrite the old address. The result of this is
333 that the backtrace may contain entries for functions which have
334 already returned, and so be very confusing.
335 A second limitation of this implementation is that it will scan
337 only the page (4KB, normally) containing the starting stack
338 pointer. If the stack frames are large, this may result in only a
339 few (or not even any) being present in the trace. Also, if you are
340 unlucky and have an initial stack pointer near the end of its
341 containing page, the scan may miss all interesting frames.
342 By default stack scanning is disabled. The normal use case is to
344 ask for it when a stack trace would otherwise be very short. So, to
345 enable it, use --unw-stack-scan-thresh=number. This requests
346 Valgrind to try using stack scanning to "extend" stack traces which
347 contain fewer than number frames.
348 If stack scanning does take place, it will only generate at most
350 the number of frames specified by --unw-stack-scan-frames.
351 Typically, stack scanning generates so many garbage entries that
352 this value is set to a low value (5) by default. In no case will a
353 stack trace larger than the value specified by --num-callers be
354 created.
355 --error-limit=<yes|no> [default: yes]
357 When enabled, Valgrind stops reporting errors after 10,000,000 in
358 total, or 1,000 different ones, have been seen. This is to stop the
359 error tracking machinery from becoming a huge performance overhead
360 in programs with many errors.
361 --error-exitcode=<number> [default: 0]
363 Specifies an alternative exit code to return if Valgrind reported
364 any errors in the run. When set to the default value (zero), the
365 return value from Valgrind will always be the return value of the
366 process being simulated. When set to a nonzero value, that value is
367 returned instead, if Valgrind detects any errors. This is useful
368 for using Valgrind as part of an automated test suite, since it
369 makes it easy to detect test cases for which Valgrind has reported
370 errors, just by inspecting return codes.
371 --exit-on-first-error=<yes|no> [default: no]
373 If this option is enabled, Valgrind exits on the first error. A
374 nonzero exit value must be defined using --error-exitcode option.
375 Useful if you are running regression tests or have some other
376 automated test machinery.
377 --error-markers=<begin>,<end> [default: none]
379 When errors are output as plain text (i.e. XML not used),
380 --error-markers instructs to output a line containing the begin
381 (end) string before (after) each error.
382 Such marker lines facilitate searching for errors and/or extracting
384 errors in an output file that contain valgrind errors mixed with
385 the program output.
386 Note that empty markers are accepted. So, only using a begin (or an
388 end) marker is possible.
389 --show-error-list=no|yes [default: no]
391 If this option is enabled, for tools that report errors, valgrind
392 will show the list of detected errors and the list of used
393 suppressions at exit.
394 Note that at verbosity 2 and above, valgrind automatically shows
396 the list of detected errors and the list of used suppressions at
397 exit, unless --show-error-list=no is selected.
398 -s
400 Specifying -s is equivalent to --show-error-list=yes.
401 --sigill-diagnostics=<yes|no> [default: yes]
403 Enable/disable printing of illegal instruction diagnostics. Enabled
404 by default, but defaults to disabled when --quiet is given. The
405 default can always be explicitly overridden by giving this option.
406 When enabled, a warning message will be printed, along with some
408 diagnostics, whenever an instruction is encountered that Valgrind
409 cannot decode or translate, before the program is given a SIGILL
410 signal. Often an illegal instruction indicates a bug in the program
411 or missing support for the particular instruction in Valgrind. But
412 some programs do deliberately try to execute an instruction that
413 might be missing and trap the SIGILL signal to detect processor
414 features. Using this flag makes it possible to avoid the diagnostic
415 output that you would otherwise get in such cases.
416 --keep-debuginfo=<yes|no> [default: no]
418 When enabled, keep ("archive") symbols and all other debuginfo for
419 unloaded code. This allows saved stack traces to include file/line
420 info for code that has been dlclose'd (or similar). Be careful with
421 this, since it can lead to unbounded memory use for programs which
422 repeatedly load and unload shared objects.
423 Some tools and some functionalities have only limited support for
425 archived debug info. Memcheck fully supports it. Generally, tools
426 that report errors can use archived debug info to show the error
427 stack traces. The known limitations are: Helgrind's past access
428 stack trace of a race condition is does not use archived debug
429 info. Massif (and more generally the xtree Massif output format)
430 does not make use of archived debug info. Only Memcheck has been
431 (somewhat) tested with --keep-debuginfo=yes, so other tools may
432 have unknown limitations.
433 --show-below-main=<yes|no> [default: no]
435 By default, stack traces for errors do not show any functions that
436 appear beneath main because most of the time it's uninteresting C
437 library stuff and/or gobbledygook. Alternatively, if main is not
438 present in the stack trace, stack traces will not show any
439 functions below main-like functions such as glibc's
440 __libc_start_main. Furthermore, if main-like functions are present
441 in the trace, they are normalised as (below main), in order to make
442 the output more deterministic.
443 If this option is enabled, all stack trace entries will be shown
445 and main-like functions will not be normalised.
446 --fullpath-after=<string> [default: don't show source paths]
448 By default Valgrind only shows the filenames in stack traces, but
449 not full paths to source files. When using Valgrind in large
450 projects where the sources reside in multiple different
451 directories, this can be inconvenient. --fullpath-after provides a
452 flexible solution to this problem. When this option is present, the
453 path to each source file is shown, with the following all-important
454 caveat: if string is found in the path, then the path up to and
455 including string is omitted, else the path is shown unmodified.
456 Note that string is not required to be a prefix of the path.
457 For example, consider a file named
459 /home/janedoe/blah/src/foo/bar/xyzzy.c. Specifying
460 --fullpath-after=/home/janedoe/blah/src/ will cause Valgrind to
461 show the name as foo/bar/xyzzy.c.
462 Because the string is not required to be a prefix,
464 --fullpath-after=src/ will produce the same output. This is useful
465 when the path contains arbitrary machine-generated characters. For
466 example, the path /my/build/dir/C32A1B47/blah/src/foo/xyzzy can be
467 pruned to foo/xyzzy using --fullpath-after=/blah/src/.
468 If you simply want to see the full path, just specify an empty
470 string: --fullpath-after=. This isn't a special case, merely a
471 logical consequence of the above rules.
472 Finally, you can use --fullpath-after multiple times. Any
474 appearance of it causes Valgrind to switch to producing full paths
475 and applying the above filtering rule. Each produced path is
476 compared against all the --fullpath-after-specified strings, in the
477 order specified. The first string to match causes the path to be
478 truncated as described above. If none match, the full path is
479 shown. This facilitates chopping off prefixes when the sources are
480 drawn from a number of unrelated directories.
481 --extra-debuginfo-path=<path> [default: undefined and unused]
483 By default Valgrind searches in several well-known paths for debug
484 objects, such as /usr/lib/debug/.
485 However, there may be scenarios where you may wish to put debug
487 objects at an arbitrary location, such as external storage when
488 running Valgrind on a mobile device with limited local storage.
489 Another example might be a situation where you do not have
490 permission to install debug object packages on the system where you
491 are running Valgrind.
492 In these scenarios, you may provide an absolute path as an extra,
494 final place for Valgrind to search for debug objects by specifying
495 --extra-debuginfo-path=/path/to/debug/objects. The given path will
496 be prepended to the absolute path name of the searched-for object.
497 For example, if Valgrind is looking for the debuginfo for
498 /w/x/y/zz.so and --extra-debuginfo-path=/a/b/c is specified, it
499 will look for a debug object at /a/b/c/w/x/y/zz.so.
500 This flag should only be specified once. If it is specified
502 multiple times, only the last instance is honoured.
503 --debuginfo-server=ipaddr:port [default: undefined and unused]
505 This is a new, experimental, feature introduced in version 3.9.0.
506 In some scenarios it may be convenient to read debuginfo from
508 objects stored on a different machine. With this flag, Valgrind
509 will query a debuginfo server running on ipaddr and listening on
510 port port, if it cannot find the debuginfo object in the local
511 filesystem.
512 The debuginfo server must accept TCP connections on port port. The
514 debuginfo server is contained in the source file
515 auxprogs/valgrind-di-server.c. It will only serve from the
516 directory it is started in. port defaults to 1500 in both client
517 and server if not specified.
518 If Valgrind looks for the debuginfo for /w/x/y/zz.so by using the
520 debuginfo server, it will strip the pathname components and merely
521 request zz.so on the server. That in turn will look only in its
522 current working directory for a matching debuginfo object.
523 The debuginfo data is transmitted in small fragments (8 KB) as
525 requested by Valgrind. Each block is compressed using LZO to reduce
526 transmission time. The implementation has been tuned for best
527 performance over a single-stage 802.11g (WiFi) network link.
528 Note that checks for matching primary vs debug objects, using GNU
530 debuglink CRC scheme, are performed even when using the debuginfo
531 server. To disable such checking, you need to also specify
532 --allow-mismatched-debuginfo=yes.
533 By default the Valgrind build system will build valgrind-di-server
535 for the target platform, which is almost certainly not what you
536 want. So far we have been unable to find out how to get
537 automake/autoconf to build it for the build platform. If you want
538 to use it, you will have to recompile it by hand using the command
539 shown at the top of auxprogs/valgrind-di-server.c.
540 --allow-mismatched-debuginfo=no|yes [no]
542 When reading debuginfo from separate debuginfo objects, Valgrind
543 will by default check that the main and debuginfo objects match,
544 using the GNU debuglink mechanism. This guarantees that it does not
545 read debuginfo from out of date debuginfo objects, and also ensures
546 that Valgrind can't crash as a result of mismatches.
547 This check can be overridden using
549 --allow-mismatched-debuginfo=yes. This may be useful when the
550 debuginfo and main objects have not been split in the proper way.
551 Be careful when using this, though: it disables all consistency
552 checking, and Valgrind has been observed to crash when the main and
553 debuginfo objects don't match.
554 --suppressions=<filename> [default: $PREFIX/lib/valgrind/default.supp]
556 Specifies an extra file from which to read descriptions of errors
557 to suppress. You may use up to 100 extra suppression files.
558 --gen-suppressions=<yes|no|all> [default: no]
560 When set to yes, Valgrind will pause after every error shown and
561 print the line:
562 ---- Print suppression ? --- [Return/N/n/Y/y/C/c] ----
564 Pressing Ret, or N Ret or n Ret, causes Valgrind continue execution
566 without printing a suppression for this error.
567 Pressing Y Ret or y Ret causes Valgrind to write a suppression for
569 this error. You can then cut and paste it into a suppression file
570 if you don't want to hear about the error in the future.
571 When set to all, Valgrind will print a suppression for every
573 reported error, without querying the user.
574 This option is particularly useful with C++ programs, as it prints
576 out the suppressions with mangled names, as required.
577 Note that the suppressions printed are as specific as possible. You
579 may want to common up similar ones, by adding wildcards to function
580 names, and by using frame-level wildcards. The wildcarding
581 facilities are powerful yet flexible, and with a bit of careful
582 editing, you may be able to suppress a whole family of related
583 errors with only a few suppressions.
584 Sometimes two different errors are suppressed by the same
586 suppression, in which case Valgrind will output the suppression
587 more than once, but you only need to have one copy in your
588 suppression file (but having more than one won't cause problems).
589 Also, the suppression name is given as <insert a suppression name
590 here>; the name doesn't really matter, it's only used with the -v
591 option which prints out all used suppression records.
592 --input-fd=<number> [default: 0, stdin]
594 When using --gen-suppressions=yes, Valgrind will stop so as to read
595 keyboard input from you when each error occurs. By default it reads
596 from the standard input (stdin), which is problematic for programs
597 which close stdin. This option allows you to specify an alternative
598 file descriptor from which to read input.
599 --dsymutil=no|yes [yes]
601 This option is only relevant when running Valgrind on Mac OS X.
602 Mac OS X uses a deferred debug information (debuginfo) linking
604 scheme. When object files containing debuginfo are linked into a
605 .dylib or an executable, the debuginfo is not copied into the final
606 file. Instead, the debuginfo must be linked manually by running
607 dsymutil, a system-provided utility, on the executable or .dylib.
608 The resulting combined debuginfo is placed in a directory alongside
609 the executable or .dylib, but with the extension .dSYM.
610 With --dsymutil=no, Valgrind will detect cases where the .dSYM
612 directory is either missing, or is present but does not appear to
613 match the associated executable or .dylib, most likely because it
614 is out of date. In these cases, Valgrind will print a warning
615 message but take no further action.
616 With --dsymutil=yes, Valgrind will, in such cases, automatically
618 run dsymutil as necessary to bring the debuginfo up to date. For
619 all practical purposes, if you always use --dsymutil=yes, then
620 there is never any need to run dsymutil manually or as part of your
621 applications's build system, since Valgrind will run it as
622 necessary.
623 Valgrind will not attempt to run dsymutil on any executable or
625 library in /usr/, /bin/, /sbin/, /opt/, /sw/, /System/, /Library/
626 or /Applications/ since dsymutil will always fail in such
627 situations. It fails both because the debuginfo for such
628 pre-installed system components is not available anywhere, and also
629 because it would require write privileges in those directories.
630 Be careful when using --dsymutil=yes, since it will cause
632 pre-existing .dSYM directories to be silently deleted and
633 re-created. Also note that dsymutil is quite slow, sometimes
634 excessively so.
635 --max-stackframe=<number> [default: 2000000]
637 The maximum size of a stack frame. If the stack pointer moves by
638 more than this amount then Valgrind will assume that the program is
639 switching to a different stack.
640 You may need to use this option if your program has large
642 stack-allocated arrays. Valgrind keeps track of your program's
643 stack pointer. If it changes by more than the threshold amount,
644 Valgrind assumes your program is switching to a different stack,
645 and Memcheck behaves differently than it would for a stack pointer
646 change smaller than the threshold. Usually this heuristic works
647 well. However, if your program allocates large structures on the
648 stack, this heuristic will be fooled, and Memcheck will
649 subsequently report large numbers of invalid stack accesses. This
650 option allows you to change the threshold to a different value.
651 You should only consider use of this option if Valgrind's debug
653 output directs you to do so. In that case it will tell you the new
654 threshold you should specify.
655 In general, allocating large structures on the stack is a bad idea,
657 because you can easily run out of stack space, especially on
658 systems with limited memory or which expect to support large
659 numbers of threads each with a small stack, and also because the
660 error checking performed by Memcheck is more effective for
661 heap-allocated data than for stack-allocated data. If you have to
662 use this option, you may wish to consider rewriting your code to
663 allocate on the heap rather than on the stack.
664 --main-stacksize=<number> [default: use current 'ulimit' value]
666 Specifies the size of the main thread's stack.
667 To simplify its memory management, Valgrind reserves all required
669 space for the main thread's stack at startup. That means it needs
670 to know the required stack size at startup.
671 By default, Valgrind uses the current "ulimit" value for the stack
673 size, or 16 MB, whichever is lower. In many cases this gives a
674 stack size in the range 8 to 16 MB, which almost never overflows
675 for most applications.
676 If you need a larger total stack size, use --main-stacksize to
678 specify it. Only set it as high as you need, since reserving far
679 more space than you need (that is, hundreds of megabytes more than
680 you need) constrains Valgrind's memory allocators and may reduce
681 the total amount of memory that Valgrind can use. This is only
682 really of significance on 32-bit machines.
683 On Linux, you may request a stack of size up to 2GB. Valgrind will
685 stop with a diagnostic message if the stack cannot be allocated.
686 --main-stacksize only affects the stack size for the program's
688 initial thread. It has no bearing on the size of thread stacks, as
689 Valgrind does not allocate those.
690 You may need to use both --main-stacksize and --max-stackframe
692 together. It is important to understand that --main-stacksize sets
693 the maximum total stack size, whilst --max-stackframe specifies the
694 largest size of any one stack frame. You will have to work out the
695 --main-stacksize value for yourself (usually, if your applications
696 segfaults). But Valgrind will tell you the needed --max-stackframe
697 size, if necessary.
698 As discussed further in the description of --max-stackframe, a
700 requirement for a large stack is a sign of potential portability
701 problems. You are best advised to place all large data in
702 heap-allocated memory.
703 --max-threads=<number> [default: 500]
705 By default, Valgrind can handle to up to 500 threads. Occasionally,
706 that number is too small. Use this option to provide a different
707 limit. E.g. --max-threads=3000.
708
710 For tools that use their own version of malloc (e.g. Memcheck, Massif,
711 Helgrind, DRD), the following options apply.
712 --alignment=<number> [default: 8 or 16, depending on the platform]
714 By default Valgrind's malloc, realloc, etc, return a block whose
715 starting address is 8-byte aligned or 16-byte aligned (the value
716 depends on the platform and matches the platform default). This
717 option allows you to specify a different alignment. The supplied
718 value must be greater than or equal to the default, less than or
719 equal to 4096, and must be a power of two.
720 --redzone-size=<number> [default: depends on the tool]
722 Valgrind's malloc, realloc, etc, add padding blocks before and
723 after each heap block allocated by the program being run. Such
724 padding blocks are called redzones. The default value for the
725 redzone size depends on the tool. For example, Memcheck adds and
726 protects a minimum of 16 bytes before and after each block
727 allocated by the client. This allows it to detect block underruns
728 or overruns of up to 16 bytes.
729 Increasing the redzone size makes it possible to detect overruns of
731 larger distances, but increases the amount of memory used by
732 Valgrind. Decreasing the redzone size will reduce the memory needed
733 by Valgrind but also reduces the chances of detecting
734 over/underruns, so is not recommended.
735 --xtree-memory=none|allocs|full [none]
737 Tools replacing Valgrind's malloc, realloc, etc, can optionally
738 produce an execution tree detailing which piece of code is
739 responsible for heap memory usage. See Execution Trees for a
740 detailed explanation about execution trees.
741 When set to none, no memory execution tree is produced.
743 When set to allocs, the memory execution tree gives the current
745 number of allocated bytes and the current number of allocated
746 blocks.
747 When set to full, the memory execution tree gives 6 different
749 measurements : the current number of allocated bytes and blocks
750 (same values as for allocs), the total number of allocated bytes
751 and blocks, the total number of freed bytes and blocks.
752 Note that the overhead in cpu and memory to produce an xtree
754 depends on the tool. The overhead in cpu is small for the value
755 allocs, as the information needed to produce this report is
756 maintained in any case by the tool. For massif and helgrind,
757 specifying full implies to capture a stack trace for each free
758 operation, while normally these tools only capture an allocation
759 stack trace. For Memcheck, the cpu overhead for the value full is
760 small, as this can only be used in combination with
761 --keep-stacktraces=alloc-and-free or
762 --keep-stacktraces=alloc-then-free, which already records a stack
763 trace for each free operation. The memory overhead varies between 5
764 and 10 words per unique stacktrace in the xtree, plus the memory
765 needed to record the stack trace for the free operations, if needed
766 specifically for the xtree.
767 --xtree-memory-file=<filename> [default: xtmemory.kcg.%p]
769 Specifies that Valgrind should produce the xtree memory report in
770 the specified file. Any %p or %q sequences appearing in the
771 filename are expanded in exactly the same way as they are for
772 --log-file. See the description of --log-file for details.
773 If the filename contains the extension .ms, then the produced file
775 format will be a massif output file format. If the filename
776 contains the extension .kcg or no extension is provided or
777 recognised, then the produced file format will be a callgrind
778 output format.
779 See Execution Trees for a detailed explanation about execution
781 trees formats.
782
784 These options apply to all tools, as they affect certain obscure
785 workings of the Valgrind core. Most people won't need to use them.
786 --smc-check=<none|stack|all|all-non-file> [default: all-non-file for
788 x86/amd64/s390x, stack for other archs]
789 This option controls Valgrind's detection of self-modifying code.
790 If no checking is done, when a program executes some code, then
791 overwrites it with new code, and executes the new code, Valgrind
792 will continue to execute the translations it made for the old code.
793 This will likely lead to incorrect behaviour and/or crashes.
794 For "modern" architectures -- anything that's not x86, amd64 or
796 s390x -- the default is stack. This is because a correct program
797 must take explicit action to reestablish D-I cache coherence
798 following code modification. Valgrind observes and honours such
799 actions, with the result that self-modifying code is transparently
800 handled with zero extra cost.
801 For x86, amd64 and s390x, the program is not required to notify the
803 hardware of required D-I coherence syncing. Hence the default is
804 all-non-file, which covers the normal case of generating code into
805 an anonymous (non-file-backed) mmap'd area.
806 The meanings of the four available settings are as follows. No
808 detection (none), detect self-modifying code on the stack (which is
809 used by GCC to implement nested functions) (stack), detect
810 self-modifying code everywhere (all), and detect self-modifying
811 code everywhere except in file-backed mappings (all-non-file).
812 Running with all will slow Valgrind down noticeably. Running with
814 none will rarely speed things up, since very little code gets
815 dynamically generated in most programs. The
816 VALGRIND_DISCARD_TRANSLATIONS client request is an alternative to
817 --smc-check=all and --smc-check=all-non-file that requires more
818 programmer effort but allows Valgrind to run your program faster,
819 by telling it precisely when translations need to be re-made.
820 --smc-check=all-non-file provides a cheaper but more limited
822 version of --smc-check=all. It adds checks to any translations that
823 do not originate from file-backed memory mappings. Typical
824 applications that generate code, for example JITs in web browsers,
825 generate code into anonymous mmaped areas, whereas the "fixed" code
826 of the browser always lives in file-backed mappings.
827 --smc-check=all-non-file takes advantage of this observation,
828 limiting the overhead of checking to code which is likely to be JIT
829 generated.
830 --read-inline-info=<yes|no> [default: see below]
832 When enabled, Valgrind will read information about inlined function
833 calls from DWARF3 debug info. This slows Valgrind startup and makes
834 it use more memory (typically for each inlined piece of code, 6
835 words and space for the function name), but it results in more
836 descriptive stacktraces. Currently, this functionality is enabled
837 by default only for Linux, Android and Solaris targets and only for
838 the tools Memcheck, Massif, Helgrind and DRD. Here is an example of
839 some stacktraces with --read-inline-info=no:
840 ==15380== Conditional jump or move depends on uninitialised value(s)
842 ==15380== at 0x80484EA: main (inlinfo.c:6)
843 ==15380==
844 ==15380== Conditional jump or move depends on uninitialised value(s)
845 ==15380== at 0x8048550: fun_noninline (inlinfo.c:6)
846 ==15380== by 0x804850E: main (inlinfo.c:34)
847 ==15380==
848 ==15380== Conditional jump or move depends on uninitialised value(s)
849 ==15380== at 0x8048520: main (inlinfo.c:6)
850 And here are the same errors with --read-inline-info=yes:
852 ==15377== Conditional jump or move depends on uninitialised value(s)
854 ==15377== at 0x80484EA: fun_d (inlinfo.c:6)
855 ==15377== by 0x80484EA: fun_c (inlinfo.c:14)
856 ==15377== by 0x80484EA: fun_b (inlinfo.c:20)
857 ==15377== by 0x80484EA: fun_a (inlinfo.c:26)
858 ==15377== by 0x80484EA: main (inlinfo.c:33)
859 ==15377==
860 ==15377== Conditional jump or move depends on uninitialised value(s)
861 ==15377== at 0x8048550: fun_d (inlinfo.c:6)
862 ==15377== by 0x8048550: fun_noninline (inlinfo.c:41)
863 ==15377== by 0x804850E: main (inlinfo.c:34)
864 ==15377==
865 ==15377== Conditional jump or move depends on uninitialised value(s)
866 ==15377== at 0x8048520: fun_d (inlinfo.c:6)
867 ==15377== by 0x8048520: main (inlinfo.c:35)
868 --read-var-info=<yes|no> [default: no]
870 When enabled, Valgrind will read information about variable types
871 and locations from DWARF3 debug info. This slows Valgrind startup
872 significantly and makes it use significantly more memory, but for
873 the tools that can take advantage of it (Memcheck, Helgrind, DRD)
874 it can result in more precise error messages. For example, here are
875 some standard errors issued by Memcheck:
876 ==15363== Uninitialised byte(s) found during client check request
878 ==15363== at 0x80484A9: croak (varinfo1.c:28)
879 ==15363== by 0x8048544: main (varinfo1.c:55)
880 ==15363== Address 0x80497f7 is 7 bytes inside data symbol "global_i2"
881 ==15363==
882 ==15363== Uninitialised byte(s) found during client check request
883 ==15363== at 0x80484A9: croak (varinfo1.c:28)
884 ==15363== by 0x8048550: main (varinfo1.c:56)
885 ==15363== Address 0xbea0d0cc is on thread 1's stack
886 ==15363== in frame #1, created by main (varinfo1.c:45)
887 And here are the same errors with --read-var-info=yes:
889 ==15370== Uninitialised byte(s) found during client check request
891 ==15370== at 0x80484A9: croak (varinfo1.c:28)
892 ==15370== by 0x8048544: main (varinfo1.c:55)
893 ==15370== Location 0x80497f7 is 0 bytes inside global_i2[7],
894 ==15370== a global variable declared at varinfo1.c:41
895 ==15370==
896 ==15370== Uninitialised byte(s) found during client check request
897 ==15370== at 0x80484A9: croak (varinfo1.c:28)
898 ==15370== by 0x8048550: main (varinfo1.c:56)
899 ==15370== Location 0xbeb4a0cc is 0 bytes inside local var "local"
900 ==15370== declared at varinfo1.c:46, in frame #1 of thread 1
901 --vgdb-poll=<number> [default: 5000]
903 As part of its main loop, the Valgrind scheduler will poll to check
904 if some activity (such as an external command or some input from a
905 gdb) has to be handled by gdbserver. This activity poll will be
906 done after having run the given number of basic blocks (or slightly
907 more than the given number of basic blocks). This poll is quite
908 cheap so the default value is set relatively low. You might further
909 decrease this value if vgdb cannot use ptrace system call to
910 interrupt Valgrind if all threads are (most of the time) blocked in
911 a system call.
912 --vgdb-shadow-registers=no|yes [default: no]
914 When activated, gdbserver will expose the Valgrind shadow registers
915 to GDB. With this, the value of the Valgrind shadow registers can
916 be examined or changed using GDB. Exposing shadow registers only
917 works with GDB version 7.1 or later.
918 --vgdb-prefix=<prefix> [default: /tmp/vgdb-pipe]
920 To communicate with gdb/vgdb, the Valgrind gdbserver creates 3
921 files (2 named FIFOs and a mmap shared memory file). The prefix
922 option controls the directory and prefix for the creation of these
923 files.
924 --run-libc-freeres=<yes|no> [default: yes]
926 This option is only relevant when running Valgrind on Linux.
927 The GNU C library (libc.so), which is used by all programs, may
929 allocate memory for its own uses. Usually it doesn't bother to free
930 that memory when the program ends—there would be no point, since
931 the Linux kernel reclaims all process resources when a process
932 exits anyway, so it would just slow things down.
933 The glibc authors realised that this behaviour causes leak
935 checkers, such as Valgrind, to falsely report leaks in glibc, when
936 a leak check is done at exit. In order to avoid this, they provided
937 a routine called __libc_freeres specifically to make glibc release
938 all memory it has allocated. Memcheck therefore tries to run
939 __libc_freeres at exit.
940 Unfortunately, in some very old versions of glibc, __libc_freeres
942 is sufficiently buggy to cause segmentation faults. This was
943 particularly noticeable on Red Hat 7.1. So this option is provided
944 in order to inhibit the run of __libc_freeres. If your program
945 seems to run fine on Valgrind, but segfaults at exit, you may find
946 that --run-libc-freeres=no fixes that, although at the cost of
947 possibly falsely reporting space leaks in libc.so.
948 --run-cxx-freeres=<yes|no> [default: yes]
950 This option is only relevant when running Valgrind on Linux or
951 Solaris C++ programs.
952 The GNU Standard C++ library (libstdc++.so), which is used by all
954 C++ programs compiled with g++, may allocate memory for its own
955 uses. Usually it doesn't bother to free that memory when the
956 program ends—there would be no point, since the kernel reclaims all
957 process resources when a process exits anyway, so it would just
958 slow things down.
959 The gcc authors realised that this behaviour causes leak checkers,
961 such as Valgrind, to falsely report leaks in libstdc++, when a leak
962 check is done at exit. In order to avoid this, they provided a
963 routine called __gnu_cxx::__freeres specifically to make libstdc++
964 release all memory it has allocated. Memcheck therefore tries to
965 run __gnu_cxx::__freeres at exit.
966 For the sake of flexibility and unforeseen problems with
968 __gnu_cxx::__freeres, option --run-cxx-freeres=no exists, although
969 at the cost of possibly falsely reporting space leaks in
970 libstdc++.so.
971 --sim-hints=hint1,hint2,...
973 Pass miscellaneous hints to Valgrind which slightly modify the
974 simulated behaviour in nonstandard or dangerous ways, possibly to
975 help the simulation of strange features. By default no hints are
976 enabled. Use with caution! Currently known hints are:
977 · lax-ioctls: Be very lax about ioctl handling; the only
979 assumption is that the size is correct. Doesn't require the
980 full buffer to be initialised when writing. Without this, using
981 some device drivers with a large number of strange ioctl
982 commands becomes very tiresome.
983 · fuse-compatible: Enable special handling for certain system
985 calls that may block in a FUSE file-system. This may be
986 necessary when running Valgrind on a multi-threaded program
987 that uses one thread to manage a FUSE file-system and another
988 thread to access that file-system.
989 · enable-outer: Enable some special magic needed when the program
991 being run is itself Valgrind.
992 · no-inner-prefix: Disable printing a prefix > in front of each
994 stdout or stderr output line in an inner Valgrind being run by
995 an outer Valgrind. This is useful when running Valgrind
996 regression tests in an outer/inner setup. Note that the prefix
997 > will always be printed in front of the inner debug logging
998 lines.
999 · no-nptl-pthread-stackcache: This hint is only relevant when
1001 running Valgrind on Linux; it is ignored on Solaris and Mac OS
1002 X.
1003 The GNU glibc pthread library (libpthread.so), which is used by
1005 pthread programs, maintains a cache of pthread stacks. When a
1006 pthread terminates, the memory used for the pthread stack and
1007 some thread local storage related data structure are not always
1008 directly released. This memory is kept in a cache (up to a
1009 certain size), and is re-used if a new thread is started.
1010 This cache causes the helgrind tool to report some false
1012 positive race condition errors on this cached memory, as
1013 helgrind does not understand the internal glibc cache
1014 synchronisation primitives. So, when using helgrind, disabling
1015 the cache helps to avoid false positive race conditions, in
1016 particular when using thread local storage variables (e.g.
1017 variables using the __thread qualifier).
1018 When using the memcheck tool, disabling the cache ensures the
1020 memory used by glibc to handle __thread variables is directly
1021 released when a thread terminates.
1022 Note: Valgrind disables the cache using some internal knowledge
1024 of the glibc stack cache implementation and by examining the
1025 debug information of the pthread library. This technique is
1026 thus somewhat fragile and might not work for all glibc
1027 versions. This has been successfully tested with various glibc
1028 versions (e.g. 2.11, 2.16, 2.18) on various platforms.
1029 · lax-doors: (Solaris only) Be very lax about door syscall
1031 handling over unrecognised door file descriptors. Does not
1032 require that full buffer is initialised when writing. Without
1033 this, programs using libdoor(3LIB) functionality with
1034 completely proprietary semantics may report large number of
1035 false positives.
1036 · fallback-llsc: (MIPS and ARM64 only): Enables an alternative
1038 implementation of Load-Linked (LL) and Store-Conditional (SC)
1039 instructions. The standard implementation gives more correct
1040 behaviour, but can cause indefinite looping on certain
1041 processor implementations that are intolerant of extra memory
1042 references between LL and SC. So far this is known only to
1043 happen on Cavium 3 cores. You should not need to use this flag,
1044 since the relevant cores are detected at startup and the
1045 alternative implementation is automatically enabled if
1046 necessary. There is no equivalent anti-flag: you cannot
1047 force-disable the alternative implementation, if it is
1048 automatically enabled. The underlying problem exists because
1049 the "standard" implementation of LL and SC is done by copying
1050 through LL and SC instructions into the instrumented code.
1051 However, tools may insert extra instrumentation memory
1052 references in between the LL and SC instructions. These memory
1053 references are not present in the original uninstrumented code,
1054 and their presence in the instrumented code can cause the SC
1055 instructions to persistently fail, leading to indefinite
1056 looping in LL-SC blocks. The alternative implementation gives
1057 correct behaviour of LL and SC instructions between threads in
1058 a process, up to and including the ABA scenario. It also gives
1059 correct behaviour between a Valgrinded thread and a
1060 non-Valgrinded thread running in a different process, that
1061 communicate via shared memory, but only up to and including
1062 correct CAS behaviour -- in this case the ABA scenario may not
1063 be correctly handled.
1064 --fair-sched=<no|yes|try> [default: no]
1066 The --fair-sched option controls the locking mechanism used by
1067 Valgrind to serialise thread execution. The locking mechanism
1068 controls the way the threads are scheduled, and different settings
1069 give different trade-offs between fairness and performance. For
1070 more details about the Valgrind thread serialisation scheme and its
1071 impact on performance and thread scheduling, see Scheduling and
1072 Multi-Thread Performance.
1073 · The value --fair-sched=yes activates a fair scheduler. In
1075 short, if multiple threads are ready to run, the threads will
1076 be scheduled in a round robin fashion. This mechanism is not
1077 available on all platforms or Linux versions. If not available,
1078 using --fair-sched=yes will cause Valgrind to terminate with an
1079 error.
1080 You may find this setting improves overall responsiveness if
1082 you are running an interactive multithreaded program, for
1083 example a web browser, on Valgrind.
1084 · The value --fair-sched=try activates fair scheduling if
1086 available on the platform. Otherwise, it will automatically
1087 fall back to --fair-sched=no.
1088 · The value --fair-sched=no activates a scheduler which does not
1090 guarantee fairness between threads ready to run, but which in
1091 general gives the highest performance.
1092 --kernel-variant=variant1,variant2,...
1094 Handle system calls and ioctls arising from minor variants of the
1095 default kernel for this platform. This is useful for running on
1096 hacked kernels or with kernel modules which support nonstandard
1097 ioctls, for example. Use with caution. If you don't understand what
1098 this option does then you almost certainly don't need it. Currently
1099 known variants are:
1100 · bproc: support the sys_broc system call on x86. This is for
1102 running on BProc, which is a minor variant of standard Linux
1103 which is sometimes used for building clusters.
1104 · android-no-hw-tls: some versions of the Android emulator for
1106 ARM do not provide a hardware TLS (thread-local state)
1107 register, and Valgrind crashes at startup. Use this variant to
1108 select software support for TLS.
1109 · android-gpu-sgx5xx: use this to support handling of proprietary
1111 ioctls for the PowerVR SGX 5XX series of GPUs on Android
1112 devices. Failure to select this does not cause stability
1113 problems, but may cause Memcheck to report false errors after
1114 the program performs GPU-specific ioctls.
1115 · android-gpu-adreno3xx: similarly, use this to support handling
1117 of proprietary ioctls for the Qualcomm Adreno 3XX series of
1118 GPUs on Android devices.
1119 --merge-recursive-frames=<number> [default: 0]
1121 Some recursive algorithms, for example balanced binary tree
1122 implementations, create many different stack traces, each
1123 containing cycles of calls. A cycle is defined as two identical
1124 program counter values separated by zero or more other program
1125 counter values. Valgrind may then use a lot of memory to store all
1126 these stack traces. This is a poor use of memory considering that
1127 such stack traces contain repeated uninteresting recursive calls
1128 instead of more interesting information such as the function that
1129 has initiated the recursive call.
1130 The option --merge-recursive-frames=<number> instructs Valgrind to
1132 detect and merge recursive call cycles having a size of up to
1133 <number> frames. When such a cycle is detected, Valgrind records
1134 the cycle in the stack trace as a unique program counter.
1135 The value 0 (the default) causes no recursive call merging. A value
1137 of 1 will cause stack traces of simple recursive algorithms (for
1138 example, a factorial implementation) to be collapsed. A value of 2
1139 will usually be needed to collapse stack traces produced by
1140 recursive algorithms such as binary trees, quick sort, etc. Higher
1141 values might be needed for more complex recursive algorithms.
1142 Note: recursive calls are detected by analysis of program counter
1144 values. They are not detected by looking at function names.
1145 --num-transtab-sectors=<number> [default: 6 for Android platforms, 16
1147 for all others]
1148 Valgrind translates and instruments your program's machine code in
1149 small fragments (basic blocks). The translations are stored in a
1150 translation cache that is divided into a number of sections
1151 (sectors). If the cache is full, the sector containing the oldest
1152 translations is emptied and reused. If these old translations are
1153 needed again, Valgrind must re-translate and re-instrument the
1154 corresponding machine code, which is expensive. If the "executed
1155 instructions" working set of a program is big, increasing the
1156 number of sectors may improve performance by reducing the number of
1157 re-translations needed. Sectors are allocated on demand. Once
1158 allocated, a sector can never be freed, and occupies considerable
1159 space, depending on the tool and the value of
1160 --avg-transtab-entry-size (about 40 MB per sector for Memcheck).
1161 Use the option --stats=yes to obtain precise information about the
1162 memory used by a sector and the allocation and recycling of
1163 sectors.
1164 --avg-transtab-entry-size=<number> [default: 0, meaning use tool
1166 provided default]
1167 Average size of translated basic block. This average size is used
1168 to dimension the size of a sector. Each tool provides a default
1169 value to be used. If this default value is too small, the
1170 translation sectors will become full too quickly. If this default
1171 value is too big, a significant part of the translation sector
1172 memory will be unused. Note that the average size of a basic block
1173 translation depends on the tool, and might depend on tool options.
1174 For example, the memcheck option --track-origins=yes increases the
1175 size of the basic block translations. Use --avg-transtab-entry-size
1176 to tune the size of the sectors, either to gain memory or to avoid
1177 too many retranslations.
1178 --aspace-minaddr=<address> [default: depends on the platform]
1180 To avoid potential conflicts with some system libraries, Valgrind
1181 does not use the address space below --aspace-minaddr value,
1182 keeping it reserved in case a library specifically requests memory
1183 in this region. So, some "pessimistic" value is guessed by Valgrind
1184 depending on the platform. On linux, by default, Valgrind avoids
1185 using the first 64MB even if typically there is no conflict in this
1186 complete zone. You can use the option --aspace-minaddr to have your
1187 memory hungry application benefitting from more of this lower
1188 memory. On the other hand, if you encounter a conflict, increasing
1189 aspace-minaddr value might solve it. Conflicts will typically
1190 manifest themselves with mmap failures in the low range of the
1191 address space. The provided address must be page aligned and must
1192 be equal or bigger to 0x1000 (4KB). To find the default value on
1193 your platform, do something such as valgrind -d -d date 2>&1 | grep
1194 -i minaddr. Values lower than 0x10000 (64KB) are known to create
1195 problems on some distributions.
1196 --valgrind-stacksize=<number> [default: 1MB]
1198 For each thread, Valgrind needs its own 'private' stack. The
1199 default size for these stacks is largely dimensioned, and so should
1200 be sufficient in most cases. In case the size is too small,
1201 Valgrind will segfault. Before segfaulting, a warning might be
1202 produced by Valgrind when approaching the limit.
1203 Use the option --valgrind-stacksize if such an (unlikely) warning
1205 is produced, or Valgrind dies due to a segmentation violation. Such
1206 segmentation violations have been seen when demangling huge C++
1207 symbols.
1208 If your application uses many threads and needs a lot of memory,
1210 you can gain some memory by reducing the size of these Valgrind
1211 stacks using the option --valgrind-stacksize.
1212 --show-emwarns=<yes|no> [default: no]
1214 When enabled, Valgrind will emit warnings about its CPU emulation
1215 in certain cases. These are usually not interesting.
1216 --require-text-symbol=:sonamepatt:fnnamepatt
1218 When a shared object whose soname matches sonamepatt is loaded into
1219 the process, examine all the text symbols it exports. If none of
1220 those match fnnamepatt, print an error message and abandon the run.
1221 This makes it possible to ensure that the run does not continue
1222 unless a given shared object contains a particular function name.
1223 Both sonamepatt and fnnamepatt can be written using the usual ?
1225 and * wildcards. For example: ":*libc.so*:foo?bar". You may use
1226 characters other than a colon to separate the two patterns. It is
1227 only important that the first character and the separator character
1228 are the same. For example, the above example could also be written
1229 "Q*libc.so*Qfoo?bar". Multiple
1230 --require-text-symbol flags are allowed, in which case shared
1231 objects that are loaded into the process will be checked against
1232 all of them.
1233 The purpose of this is to support reliable usage of marked-up
1235 libraries. For example, suppose we have a version of GCC's
1236 libgomp.so which has been marked up with annotations to support
1237 Helgrind. It is only too easy and confusing to load the wrong,
1238 un-annotated libgomp.so into the application. So the idea is: add a
1239 text symbol in the marked-up library, for example
1240 annotated_for_helgrind_3_6, and then give the flag
1241 --require-text-symbol=:*libgomp*so*:annotated_for_helgrind_3_6 so
1242 that when libgomp.so is loaded, Valgrind scans its symbol table,
1243 and if the symbol isn't present the run is aborted, rather than
1244 continuing silently with the un-marked-up library. Note that you
1245 should put the entire flag in quotes to stop shells expanding up
1246 the * and ? wildcards.
1247 --soname-synonyms=syn1=pattern1,syn2=pattern2,...
1249 When a shared library is loaded, Valgrind checks for functions in
1250 the library that must be replaced or wrapped. For example, Memcheck
1251 replaces some string and memory functions (strchr, strlen, strcpy,
1252 memchr, memcpy, memmove, etc.) with its own versions. Such
1253 replacements are normally done only in shared libraries whose
1254 soname matches a predefined soname pattern (e.g. libc.so* on
1255 linux). By default, no replacement is done for a statically linked
1256 binary or for alternative libraries, except for the allocation
1257 functions (malloc, free, calloc, memalign, realloc, operator new,
1258 operator delete, etc.) Such allocation functions are intercepted by
1259 default in any shared library or in the executable if they are
1260 exported as global symbols. This means that if a replacement
1261 allocation library such as tcmalloc is found, its functions are
1262 also intercepted by default. In some cases, the replacements allow
1263 --soname-synonyms to specify one additional synonym pattern, giving
1264 flexibility in the replacement. Or to prevent interception of all
1265 public allocation symbols.
1266 Currently, this flexibility is only allowed for the malloc related
1268 functions, using the synonym somalloc. This synonym is usable for
1269 all tools doing standard replacement of malloc related functions
1270 (e.g. memcheck, helgrind, drd, massif, dhat).
1271 · Alternate malloc library: to replace the malloc related
1273 functions in a specific alternate library with soname
1274 mymalloclib.so (and not in any others), give the option
1275 --soname-synonyms=somalloc=mymalloclib.so. A pattern can be
1276 used to match multiple libraries sonames. For example,
1277 --soname-synonyms=somalloc=*tcmalloc* will match the soname of
1278 all variants of the tcmalloc library (native, debug, profiled,
1279 ... tcmalloc variants).
1280 Note: the soname of a elf shared library can be retrieved using
1282 the readelf utility.
1283 · Replacements in a statically linked library are done by using
1285 the NONE pattern. For example, if you link with libtcmalloc.a,
1286 and only want to intercept the malloc related functions in the
1287 executable (and standard libraries) themselves, but not any
1288 other shared libraries, you can give the option
1289 --soname-synonyms=somalloc=NONE. Note that a NONE pattern will
1290 match the main executable and any shared library having no
1291 soname.
1292 · To run a "default" Firefox build for Linux, in which JEMalloc
1294 is linked in to the main executable, use
1295 --soname-synonyms=somalloc=NONE.
1296 · To only intercept allocation symbols in the default system
1298 libraries, but not in any other shared library or the
1299 executable defining public malloc or operator new related
1300 functions use a non-existing library name like
1301 --soname-synonyms=somalloc=nouserintercepts (where
1302 nouserintercepts can be any non-existing library name).
1303 · Shared library of the dynamic (runtime) linker is excluded from
1305 searching for global public symbols, such as those for the
1306 malloc related functions (identified by somalloc synonym).
1307 --progress-interval=<number> [default: 0, meaning 'disabled']
1309 This is an enhancement to Valgrind's debugging output. It is
1310 unlikely to be of interest to end users.
1311 When number is set to a non-zero value, Valgrind will print a
1313 one-line progress summary every number seconds. Valid settings for
1314 number are between 0 and 3600 inclusive. Here's some example output
1315 with number set to 10:
1316 PROGRESS: U 110s, W 113s, 97.3% CPU, EvC 414.79M, TIn 616.7k, TOut 0.5k, #thr 67
1318 PROGRESS: U 120s, W 124s, 96.8% CPU, EvC 505.27M, TIn 636.6k, TOut 3.0k, #thr 64
1319 PROGRESS: U 130s, W 134s, 97.0% CPU, EvC 574.90M, TIn 657.5k, TOut 3.0k, #thr 63
1320 Each line shows:
1322 · U: total user time
1324 · W: total wallclock time
1326 · CPU: overall average cpu use
1328 · EvC: number of event checks. An event check is a backwards
1330 branch in the simulated program, so this is a measure of
1331 forward progress of the program
1332 · TIn: number of code blocks instrumented by the JIT
1334 · TOut: number of instrumented code blocks that have been thrown
1336 away
1337 · #thr: number of threads in the program
1339 From the progress of these, it is possible to observe:
1341 · when the program is compute bound (TIn rises slowly, EvC rises
1343 rapidly)
1344 · when the program is in a spinloop (TIn/TOut fixed, EvC rises
1346 rapidly)
1347 · when the program is JIT-bound (TIn rises rapidly)
1349 · when the program is rapidly discarding code (TOut rises
1351 rapidly)
1352 · when the program is about to achieve some expected state (EvC
1354 arrives at some value you expect)
1355 · when the program is idling (U rises more slowly than W)
1357
1360 There are also some options for debugging Valgrind itself. You
1361 shouldn't need to use them in the normal run of things. If you wish to
1362 see the list, use the --help-debug option.
1363
1365 --leak-check=<no|summary|yes|full> [default: summary]
1366 When enabled, search for memory leaks when the client program
1367 finishes. If set to summary, it says how many leaks occurred. If
1368 set to full or yes, each individual leak will be shown in detail
1369 and/or counted as an error, as specified by the options
1370 --show-leak-kinds and --errors-for-leak-kinds.
1371 If --xml=yes is given, memcheck will automatically use the value
1373 --leak-check=full. You can use --show-leak-kinds=none to reduce the
1374 size of the xml output if you are not interested in the leak
1375 results.
1376 --leak-resolution=<low|med|high> [default: high]
1378 When doing leak checking, determines how willing Memcheck is to
1379 consider different backtraces to be the same for the purposes of
1380 merging multiple leaks into a single leak report. When set to low,
1381 only the first two entries need match. When med, four entries have
1382 to match. When high, all entries need to match.
1383 For hardcore leak debugging, you probably want to use
1385 --leak-resolution=high together with --num-callers=40 or some such
1386 large number.
1387 Note that the --leak-resolution setting does not affect Memcheck's
1389 ability to find leaks. It only changes how the results are
1390 presented.
1391 --show-leak-kinds=<set> [default: definite,possible]
1393 Specifies the leak kinds to show in a full leak search, in one of
1394 the following ways:
1395 · a comma separated list of one or more of definite indirect
1397 possible reachable.
1398 · all to specify the complete set (all leak kinds). It is
1400 equivalent to
1401 --show-leak-kinds=definite,indirect,possible,reachable.
1402 · none for the empty set.
1404 --errors-for-leak-kinds=<set> [default: definite,possible]
1406 Specifies the leak kinds to count as errors in a full leak search.
1407 The <set> is specified similarly to --show-leak-kinds
1408 --leak-check-heuristics=<set> [default: all]
1410 Specifies the set of leak check heuristics to be used during leak
1411 searches. The heuristics control which interior pointers to a block
1412 cause it to be considered as reachable. The heuristic set is
1413 specified in one of the following ways:
1414 · a comma separated list of one or more of stdstring length64
1416 newarray multipleinheritance.
1417 · all to activate the complete set of heuristics. It is
1419 equivalent to
1420 --leak-check-heuristics=stdstring,length64,newarray,multipleinheritance.
1421 · none for the empty set.
1423 Note that these heuristics are dependent on the layout of the
1425 objects produced by the C++ compiler. They have been tested with
1426 some gcc versions (e.g. 4.4 and 4.7). They might not work properly
1427 with other C++ compilers.
1428 --show-reachable=<yes|no> , --show-possibly-lost=<yes|no>
1430 These options provide an alternative way to specify the leak kinds
1431 to show:
1432 · --show-reachable=no --show-possibly-lost=yes is equivalent to
1434 --show-leak-kinds=definite,possible.
1435 · --show-reachable=no --show-possibly-lost=no is equivalent to
1437 --show-leak-kinds=definite.
1438 · --show-reachable=yes is equivalent to --show-leak-kinds=all.
1440 Note that --show-possibly-lost=no has no effect if
1442 --show-reachable=yes is specified.
1443 --xtree-leak=<no|yes> [no]
1445 If set to yes, the results for the leak search done at exit will be
1446 output in a 'Callgrind Format' execution tree file. Note that this
1447 automatically sets the options --leak-check=full and
1448 --show-leak-kinds=all, to allow xtree visualisation tools such as
1449 kcachegrind to select what kind to leak to visualise. The produced
1450 file will contain the following events:
1451 · RB : Reachable Bytes
1453 · PB : Possibly lost Bytes
1455 · IB : Indirectly lost Bytes
1457 · DB : Definitely lost Bytes (direct plus indirect)
1459 · DIB : Definitely Indirectly lost Bytes (subset of DB)
1461 · RBk : reachable Blocks
1463 · PBk : Possibly lost Blocks
1465 · IBk : Indirectly lost Blocks
1467 · DBk : Definitely lost Blocks
1469 The increase or decrease for all events above will also be output
1471 in the file to provide the delta (increase or decrease) between 2
1472 successive leak searches. For example, iRB is the increase of the
1473 RB event, dPBk is the decrease of PBk event. The values for the
1474 increase and decrease events will be zero for the first leak search
1475 done.
1476 See Execution Trees for a detailed explanation about execution
1478 trees.
1479 --xtree-leak-file=<filename> [default: xtleak.kcg.%p]
1481 Specifies that Valgrind should produce the xtree leak report in the
1482 specified file. Any %p, %q or %n sequences appearing in the
1483 filename are expanded in exactly the same way as they are for
1484 --log-file. See the description of --log-file for details.
1485 See Execution Trees for a detailed explanation about execution
1487 trees formats.
1488 --undef-value-errors=<yes|no> [default: yes]
1490 Controls whether Memcheck reports uses of undefined value errors.
1491 Set this to no if you don't want to see undefined value errors. It
1492 also has the side effect of speeding up Memcheck somewhat.
1493 AddrCheck (removed in Valgrind 3.1.0) functioned like Memcheck with
1494 --undef-value-errors=no.
1495 --track-origins=<yes|no> [default: no]
1497 Controls whether Memcheck tracks the origin of uninitialised
1498 values. By default, it does not, which means that although it can
1499 tell you that an uninitialised value is being used in a dangerous
1500 way, it cannot tell you where the uninitialised value came from.
1501 This often makes it difficult to track down the root problem.
1502 When set to yes, Memcheck keeps track of the origins of all
1504 uninitialised values. Then, when an uninitialised value error is
1505 reported, Memcheck will try to show the origin of the value. An
1506 origin can be one of the following four places: a heap block, a
1507 stack allocation, a client request, or miscellaneous other sources
1508 (eg, a call to brk).
1509 For uninitialised values originating from a heap block, Memcheck
1511 shows where the block was allocated. For uninitialised values
1512 originating from a stack allocation, Memcheck can tell you which
1513 function allocated the value, but no more than that -- typically it
1514 shows you the source location of the opening brace of the function.
1515 So you should carefully check that all of the function's local
1516 variables are initialised properly.
1517 Performance overhead: origin tracking is expensive. It halves
1519 Memcheck's speed and increases memory use by a minimum of 100MB,
1520 and possibly more. Nevertheless it can drastically reduce the
1521 effort required to identify the root cause of uninitialised value
1522 errors, and so is often a programmer productivity win, despite
1523 running more slowly.
1524 Accuracy: Memcheck tracks origins quite accurately. To avoid very
1526 large space and time overheads, some approximations are made. It is
1527 possible, although unlikely, that Memcheck will report an incorrect
1528 origin, or not be able to identify any origin.
1529 Note that the combination --track-origins=yes and
1531 --undef-value-errors=no is nonsensical. Memcheck checks for and
1532 rejects this combination at startup.
1533 --partial-loads-ok=<yes|no> [default: yes]
1535 Controls how Memcheck handles 32-, 64-, 128- and 256-bit naturally
1536 aligned loads from addresses for which some bytes are addressable
1537 and others are not. When yes, such loads do not produce an address
1538 error. Instead, loaded bytes originating from illegal addresses are
1539 marked as uninitialised, and those corresponding to legal addresses
1540 are handled in the normal way.
1541 When no, loads from partially invalid addresses are treated the
1543 same as loads from completely invalid addresses: an illegal-address
1544 error is issued, and the resulting bytes are marked as initialised.
1545 Note that code that behaves in this way is in violation of the ISO
1547 C/C++ standards, and should be considered broken. If at all
1548 possible, such code should be fixed.
1549 --expensive-definedness-checks=<no|auto|yes> [default: auto]
1551 Controls whether Memcheck should employ more precise but also more
1552 expensive (time consuming) instrumentation when checking the
1553 definedness of certain values. In particular, this affects the
1554 instrumentation of integer adds, subtracts and equality
1555 comparisons.
1556 Selecting --expensive-definedness-checks=yes causes Memcheck to use
1558 the most accurate analysis possible. This minimises false error
1559 rates but can cause up to 30% performance degradation.
1560 Selecting --expensive-definedness-checks=no causes Memcheck to use
1562 the cheapest instrumentation possible. This maximises performance
1563 but will normally give an unusably high false error rate.
1564 The default setting, --expensive-definedness-checks=auto, is
1566 strongly recommended. This causes Memcheck to use the minimum of
1567 expensive instrumentation needed to achieve the same false error
1568 rate as --expensive-definedness-checks=yes. It also enables an
1569 instrumentation-time analysis pass which aims to further reduce the
1570 costs of accurate instrumentation. Overall, the performance loss is
1571 generally around 5% relative to --expensive-definedness-checks=no,
1572 although this is strongly workload dependent. Note that the exact
1573 instrumentation settings in this mode are architecture dependent.
1574 --keep-stacktraces=alloc|free|alloc-and-free|alloc-then-free|none
1576 [default: alloc-and-free]
1577 Controls which stack trace(s) to keep for malloc'd and/or free'd
1578 blocks.
1579 With alloc-then-free, a stack trace is recorded at allocation time,
1581 and is associated with the block. When the block is freed, a second
1582 stack trace is recorded, and this replaces the allocation stack
1583 trace. As a result, any "use after free" errors relating to this
1584 block can only show a stack trace for where the block was freed.
1585 With alloc-and-free, both allocation and the deallocation stack
1587 traces for the block are stored. Hence a "use after free" error
1588 will show both, which may make the error easier to diagnose.
1589 Compared to alloc-then-free, this setting slightly increases
1590 Valgrind's memory use as the block contains two references instead
1591 of one.
1592 With alloc, only the allocation stack trace is recorded (and
1594 reported). With free, only the deallocation stack trace is recorded
1595 (and reported). These values somewhat decrease Valgrind's memory
1596 and cpu usage. They can be useful depending on the error types you
1597 are searching for and the level of detail you need to analyse them.
1598 For example, if you are only interested in memory leak errors, it
1599 is sufficient to record the allocation stack traces.
1600 With none, no stack traces are recorded for malloc and free
1602 operations. If your program allocates a lot of blocks and/or
1603 allocates/frees from many different stack traces, this can
1604 significantly decrease cpu and/or memory required. Of course, few
1605 details will be reported for errors related to heap blocks.
1606 Note that once a stack trace is recorded, Valgrind keeps the stack
1608 trace in memory even if it is not referenced by any block. Some
1609 programs (for example, recursive algorithms) can generate a huge
1610 number of stack traces. If Valgrind uses too much memory in such
1611 circumstances, you can reduce the memory required with the options
1612 --keep-stacktraces and/or by using a smaller value for the option
1613 --num-callers.
1614 If you want to use --xtree-memory=full memory profiling (see
1616 Execution Trees), then you cannot specify --keep-stacktraces=free
1617 or --keep-stacktraces=none.
1618 --freelist-vol=<number> [default: 20000000]
1620 When the client program releases memory using free (in C) or delete
1621 (C++), that memory is not immediately made available for
1622 re-allocation. Instead, it is marked inaccessible and placed in a
1623 queue of freed blocks. The purpose is to defer as long as possible
1624 the point at which freed-up memory comes back into circulation.
1625 This increases the chance that Memcheck will be able to detect
1626 invalid accesses to blocks for some significant period of time
1627 after they have been freed.
1628 This option specifies the maximum total size, in bytes, of the
1630 blocks in the queue. The default value is twenty million bytes.
1631 Increasing this increases the total amount of memory used by
1632 Memcheck but may detect invalid uses of freed blocks which would
1633 otherwise go undetected.
1634 --freelist-big-blocks=<number> [default: 1000000]
1636 When making blocks from the queue of freed blocks available for
1637 re-allocation, Memcheck will in priority re-circulate the blocks
1638 with a size greater or equal to --freelist-big-blocks. This ensures
1639 that freeing big blocks (in particular freeing blocks bigger than
1640 --freelist-vol) does not immediately lead to a re-circulation of
1641 all (or a lot of) the small blocks in the free list. In other
1642 words, this option increases the likelihood to discover dangling
1643 pointers for the "small" blocks, even when big blocks are freed.
1644 Setting a value of 0 means that all the blocks are re-circulated in
1646 a FIFO order.
1647 --workaround-gcc296-bugs=<yes|no> [default: no]
1649 When enabled, assume that reads and writes some small distance
1650 below the stack pointer are due to bugs in GCC 2.96, and does not
1651 report them. The "small distance" is 256 bytes by default. Note
1652 that GCC 2.96 is the default compiler on some ancient Linux
1653 distributions (RedHat 7.X) and so you may need to use this option.
1654 Do not use it if you do not have to, as it can cause real errors to
1655 be overlooked. A better alternative is to use a more recent GCC in
1656 which this bug is fixed.
1657 You may also need to use this option when working with GCC 3.X or
1659 4.X on 32-bit PowerPC Linux. This is because GCC generates code
1660 which occasionally accesses below the stack pointer, particularly
1661 for floating-point to/from integer conversions. This is in
1662 violation of the 32-bit PowerPC ELF specification, which makes no
1663 provision for locations below the stack pointer to be accessible.
1664 This option is deprecated as of version 3.12 and may be removed
1666 from future versions. You should instead use
1667 --ignore-range-below-sp to specify the exact range of offsets below
1668 the stack pointer that should be ignored. A suitable equivalent is
1669 --ignore-range-below-sp=1024-1.
1670 --ignore-range-below-sp=<number>-<number>
1672 This is a more general replacement for the deprecated
1673 --workaround-gcc296-bugs option. When specified, it causes Memcheck
1674 not to report errors for accesses at the specified offsets below
1675 the stack pointer. The two offsets must be positive decimal numbers
1676 and -- somewhat counterintuitively -- the first one must be larger,
1677 in order to imply a non-wraparound address range to ignore. For
1678 example, to ignore 4 byte accesses at 8192 bytes below the stack
1679 pointer, use --ignore-range-below-sp=8192-8189. Only one range may
1680 be specified.
1681 --show-mismatched-frees=<yes|no> [default: yes]
1683 When enabled, Memcheck checks that heap blocks are deallocated
1684 using a function that matches the allocating function. That is, it
1685 expects free to be used to deallocate blocks allocated by malloc,
1686 delete for blocks allocated by new, and delete[] for blocks
1687 allocated by new[]. If a mismatch is detected, an error is
1688 reported. This is in general important because in some
1689 environments, freeing with a non-matching function can cause
1690 crashes.
1691 There is however a scenario where such mismatches cannot be
1693 avoided. That is when the user provides implementations of
1694 new/new[] that call malloc and of delete/delete[] that call free,
1695 and these functions are asymmetrically inlined. For example,
1696 imagine that delete[] is inlined but new[] is not. The result is
1697 that Memcheck "sees" all delete[] calls as direct calls to free,
1698 even when the program source contains no mismatched calls.
1699 This causes a lot of confusing and irrelevant error reports.
1701 --show-mismatched-frees=no disables these checks. It is not
1702 generally advisable to disable them, though, because you may miss
1703 real errors as a result.
1704 --ignore-ranges=0xPP-0xQQ[,0xRR-0xSS]
1706 Any ranges listed in this option (and multiple ranges can be
1707 specified, separated by commas) will be ignored by Memcheck's
1708 addressability checking.
1709 --malloc-fill=<hexnumber>
1711 Fills blocks allocated by malloc, new, etc, but not by calloc, with
1712 the specified byte. This can be useful when trying to shake out
1713 obscure memory corruption problems. The allocated area is still
1714 regarded by Memcheck as undefined -- this option only affects its
1715 contents. Note that --malloc-fill does not affect a block of memory
1716 when it is used as argument to client requests
1717 VALGRIND_MEMPOOL_ALLOC or VALGRIND_MALLOCLIKE_BLOCK.
1718 --free-fill=<hexnumber>
1720 Fills blocks freed by free, delete, etc, with the specified byte
1721 value. This can be useful when trying to shake out obscure memory
1722 corruption problems. The freed area is still regarded by Memcheck
1723 as not valid for access -- this option only affects its contents.
1724 Note that --free-fill does not affect a block of memory when it is
1725 used as argument to client requests VALGRIND_MEMPOOL_FREE or
1726 VALGRIND_FREELIKE_BLOCK.
1727
1729 --I1=<size>,<associativity>,<line size>
1730 Specify the size, associativity and line size of the level 1
1731 instruction cache.
1732 --D1=<size>,<associativity>,<line size>
1734 Specify the size, associativity and line size of the level 1 data
1735 cache.
1736 --LL=<size>,<associativity>,<line size>
1738 Specify the size, associativity and line size of the last-level
1739 cache.
1740 --cache-sim=no|yes [yes]
1742 Enables or disables collection of cache access and miss counts.
1743 --branch-sim=no|yes [no]
1745 Enables or disables collection of branch instruction and
1746 misprediction counts. By default this is disabled as it slows
1747 Cachegrind down by approximately 25%. Note that you cannot specify
1748 --cache-sim=no and --branch-sim=no together, as that would leave
1749 Cachegrind with no information to collect.
1750 --cachegrind-out-file=<file>
1752 Write the profile data to file rather than to the default output
1753 file, cachegrind.out.<pid>. The %p and %q format specifiers can be
1754 used to embed the process ID and/or the contents of an environment
1755 variable in the name, as is the case for the core option
1756 --log-file.
1757
1759 --callgrind-out-file=<file>
1760 Write the profile data to file rather than to the default output
1761 file, callgrind.out.<pid>. The %p and %q format specifiers can be
1762 used to embed the process ID and/or the contents of an environment
1763 variable in the name, as is the case for the core option
1764 --log-file. When multiple dumps are made, the file name is modified
1765 further; see below.
1766 --dump-line=<no|yes> [default: yes]
1768 This specifies that event counting should be performed at source
1769 line granularity. This allows source annotation for sources which
1770 are compiled with debug information (-g).
1771 --dump-instr=<no|yes> [default: no]
1773 This specifies that event counting should be performed at
1774 per-instruction granularity. This allows for assembly code
1775 annotation. Currently the results can only be displayed by
1776 KCachegrind.
1777 --compress-strings=<no|yes> [default: yes]
1779 This option influences the output format of the profile data. It
1780 specifies whether strings (file and function names) should be
1781 identified by numbers. This shrinks the file, but makes it more
1782 difficult for humans to read (which is not recommended in any
1783 case).
1784 --compress-pos=<no|yes> [default: yes]
1786 This option influences the output format of the profile data. It
1787 specifies whether numerical positions are always specified as
1788 absolute values or are allowed to be relative to previous numbers.
1789 This shrinks the file size.
1790 --combine-dumps=<no|yes> [default: no]
1792 When enabled, when multiple profile data parts are to be generated
1793 these parts are appended to the same output file. Not recommended.
1794 --dump-every-bb=<count> [default: 0, never]
1796 Dump profile data every count basic blocks. Whether a dump is
1797 needed is only checked when Valgrind's internal scheduler is run.
1798 Therefore, the minimum setting useful is about 100000. The count is
1799 a 64-bit value to make long dump periods possible.
1800 --dump-before=<function>
1802 Dump when entering function.
1803 --zero-before=<function>
1805 Zero all costs when entering function.
1806 --dump-after=<function>
1808 Dump when leaving function.
1809 --instr-atstart=<yes|no> [default: yes]
1811 Specify if you want Callgrind to start simulation and profiling
1812 from the beginning of the program. When set to no, Callgrind will
1813 not be able to collect any information, including calls, but it
1814 will have at most a slowdown of around 4, which is the minimum
1815 Valgrind overhead. Instrumentation can be interactively enabled via
1816 callgrind_control -i on.
1817 Note that the resulting call graph will most probably not contain
1819 main, but will contain all the functions executed after
1820 instrumentation was enabled. Instrumentation can also be
1821 programmatically enabled/disabled. See the Callgrind include file
1822 callgrind.h for the macro you have to use in your source code.
1823 For cache simulation, results will be less accurate when switching
1825 on instrumentation later in the program run, as the simulator
1826 starts with an empty cache at that moment. Switch on event
1827 collection later to cope with this error.
1828 --collect-atstart=<yes|no> [default: yes]
1830 Specify whether event collection is enabled at beginning of the
1831 profile run.
1832 To only look at parts of your program, you have two possibilities:
1834 1. Zero event counters before entering the program part you want
1836 to profile, and dump the event counters to a file after leaving
1837 that program part.
1838 2. Switch on/off collection state as needed to only see event
1840 counters happening while inside of the program part you want to
1841 profile.
1842 The second option can be used if the program part you want to
1844 profile is called many times. Option 1, i.e. creating a lot of
1845 dumps is not practical here.
1846 Collection state can be toggled at entry and exit of a given
1848 function with the option --toggle-collect. If you use this option,
1849 collection state should be disabled at the beginning. Note that the
1850 specification of --toggle-collect implicitly sets
1851 --collect-state=no.
1852 Collection state can be toggled also by inserting the client
1854 request CALLGRIND_TOGGLE_COLLECT ; at the needed code positions.
1855 --toggle-collect=<function>
1857 Toggle collection on entry/exit of function.
1858 --collect-jumps=<no|yes> [default: no]
1860 This specifies whether information for (conditional) jumps should
1861 be collected. As above, callgrind_annotate currently is not able to
1862 show you the data. You have to use KCachegrind to get jump arrows
1863 in the annotated code.
1864 --collect-systime=<no|yes|msec|usec|nsec> [default: no]
1866 This specifies whether information for system call times should be
1867 collected.
1868 The value no indicates to record no system call information.
1870 The other values indicate to record the number of system calls done
1872 (sysCount event) and the elapsed time (sysTime event) spent in
1873 system calls. The --collect-systime value gives the unit used for
1874 sysTime : milli seconds, micro seconds or nano seconds. With the
1875 value nsec, callgrind also records the cpu time spent during system
1876 calls (sysCpuTime).
1877 The value yes is a synonym of msec. The value nsec is not supported
1879 on Darwin.
1880 --collect-bus=<no|yes> [default: no]
1882 This specifies whether the number of global bus events executed
1883 should be collected. The event type "Ge" is used for these events.
1884 --cache-sim=<yes|no> [default: no]
1886 Specify if you want to do full cache simulation. By default, only
1887 instruction read accesses will be counted ("Ir"). With cache
1888 simulation, further event counters are enabled: Cache misses on
1889 instruction reads ("I1mr"/"ILmr"), data read accesses ("Dr") and
1890 related cache misses ("D1mr"/"DLmr"), data write accesses ("Dw")
1891 and related cache misses ("D1mw"/"DLmw"). For more information, see
1892 Cachegrind: a cache and branch-prediction profiler.
1893 --branch-sim=<yes|no> [default: no]
1895 Specify if you want to do branch prediction simulation. Further
1896 event counters are enabled: Number of executed conditional branches
1897 and related predictor misses ("Bc"/"Bcm"), executed indirect jumps
1898 and related misses of the jump address predictor ("Bi"/"Bim").
1899
1901 --free-is-write=no|yes [default: no]
1902 When enabled (not the default), Helgrind treats freeing of heap
1903 memory as if the memory was written immediately before the free.
1904 This exposes races where memory is referenced by one thread, and
1905 freed by another, but there is no observable synchronisation event
1906 to ensure that the reference happens before the free.
1907 This functionality is new in Valgrind 3.7.0, and is regarded as
1909 experimental. It is not enabled by default because its interaction
1910 with custom memory allocators is not well understood at present.
1911 User feedback is welcomed.
1912 --track-lockorders=no|yes [default: yes]
1914 When enabled (the default), Helgrind performs lock order
1915 consistency checking. For some buggy programs, the large number of
1916 lock order errors reported can become annoying, particularly if
1917 you're only interested in race errors. You may therefore find it
1918 helpful to disable lock order checking.
1919 --history-level=none|approx|full [default: full]
1921 --history-level=full (the default) causes Helgrind collects enough
1922 information about "old" accesses that it can produce two stack
1923 traces in a race report -- both the stack trace for the current
1924 access, and the trace for the older, conflicting access. To limit
1925 memory usage, "old" accesses stack traces are limited to a maximum
1926 of 8 entries, even if --num-callers value is bigger.
1927 Collecting such information is expensive in both speed and memory,
1929 particularly for programs that do many inter-thread synchronisation
1930 events (locks, unlocks, etc). Without such information, it is more
1931 difficult to track down the root causes of races. Nonetheless, you
1932 may not need it in situations where you just want to check for the
1933 presence or absence of races, for example, when doing regression
1934 testing of a previously race-free program.
1935 --history-level=none is the opposite extreme. It causes Helgrind
1937 not to collect any information about previous accesses. This can be
1938 dramatically faster than --history-level=full.
1939 --history-level=approx provides a compromise between these two
1941 extremes. It causes Helgrind to show a full trace for the later
1942 access, and approximate information regarding the earlier access.
1943 This approximate information consists of two stacks, and the
1944 earlier access is guaranteed to have occurred somewhere between
1945 program points denoted by the two stacks. This is not as useful as
1946 showing the exact stack for the previous access (as
1947 --history-level=full does), but it is better than nothing, and it
1948 is almost as fast as --history-level=none.
1949 --delta-stacktrace=no|yes [default: yes on linux amd64/x86]
1951 This flag only has any effect at --history-level=full.
1952 --delta-stacktrace configures the way Helgrind captures the
1954 stacktraces for the option --history-level=full. Such a stacktrace
1955 is typically needed each time a new piece of memory is read or
1956 written in a basic block of instructions.
1957 --delta-stacktrace=no causes Helgrind to compute a full history
1959 stacktrace from the unwind info each time a stacktrace is needed.
1960 --delta-stacktrace=yes indicates to Helgrind to derive a new
1962 stacktrace from the previous stacktrace, as long as there was no
1963 call instruction, no return instruction, or any other instruction
1964 changing the call stack since the previous stacktrace was captured.
1965 If no such instruction was executed, the new stacktrace can be
1966 derived from the previous stacktrace by just changing the top frame
1967 to the current program counter. This option can speed up Helgrind
1968 by 25% when using --history-level=full.
1969 The following aspects have to be considered when using
1971 --delta-stacktrace=yes :
1972 · In some cases (for example in a function prologue), the
1974 valgrind unwinder might not properly unwind the stack, due to
1975 some limitations and/or due to wrong unwind info. When using
1976 --delta-stacktrace=yes, the wrong stack trace captured in the
1977 function prologue will be kept till the next call or return.
1978 · On the other hand, --delta-stacktrace=yes sometimes helps to
1980 obtain a correct stacktrace, for example when the unwind info
1981 allows a correct stacktrace to be done in the beginning of the
1982 sequence, but not later on in the instruction sequence.
1983 · Determining which instructions are changing the callstack is
1985 partially based on platform dependent heuristics, which have to
1986 be tuned/validated specifically for the platform. Also,
1987 unwinding in a function prologue must be good enough to allow
1988 using --delta-stacktrace=yes. Currently, the option
1989 --delta-stacktrace=yes has been reasonably validated only on
1990 linux x86 32 bits and linux amd64 64 bits. For more details
1991 about how to validate --delta-stacktrace=yes, see debug option
1992 --hg-sanity-flags and the function check_cached_rcec_ok in
1993 libhb_core.c.
1994 --conflict-cache-size=N [default: 1000000]
1997 This flag only has any effect at --history-level=full.
1998 Information about "old" conflicting accesses is stored in a cache
2000 of limited size, with LRU-style management. This is necessary
2001 because it isn't practical to store a stack trace for every single
2002 memory access made by the program. Historical information on not
2003 recently accessed locations is periodically discarded, to free up
2004 space in the cache.
2005 This option controls the size of the cache, in terms of the number
2007 of different memory addresses for which conflicting access
2008 information is stored. If you find that Helgrind is showing race
2009 errors with only one stack instead of the expected two stacks, try
2010 increasing this value.
2011 The minimum value is 10,000 and the maximum is 30,000,000 (thirty
2013 times the default value). Increasing the value by 1 increases
2014 Helgrind's memory requirement by very roughly 100 bytes, so the
2015 maximum value will easily eat up three extra gigabytes or so of
2016 memory.
2017 --check-stack-refs=no|yes [default: yes]
2019 By default Helgrind checks all data memory accesses made by your
2020 program. This flag enables you to skip checking for accesses to
2021 thread stacks (local variables). This can improve performance, but
2022 comes at the cost of missing races on stack-allocated data.
2023 --ignore-thread-creation=<yes|no> [default: no]
2025 Controls whether all activities during thread creation should be
2026 ignored. By default enabled only on Solaris. Solaris provides
2027 higher throughput, parallelism and scalability than other operating
2028 systems, at the cost of more fine-grained locking activity. This
2029 means for example that when a thread is created under glibc, just
2030 one big lock is used for all thread setup. Solaris libc uses
2031 several fine-grained locks and the creator thread resumes its
2032 activities as soon as possible, leaving for example stack and TLS
2033 setup sequence to the created thread. This situation confuses
2034 Helgrind as it assumes there is some false ordering in place
2035 between creator and created thread; and therefore many types of
2036 race conditions in the application would not be reported. To
2037 prevent such false ordering, this command line option is set to yes
2038 by default on Solaris. All activity (loads, stores, client
2039 requests) is therefore ignored during:
2040 · pthread_create() call in the creator thread
2042 · thread creation phase (stack and TLS setup) in the created
2044 thread
2045 Also new memory allocated during thread creation is untracked, that
2047 is race reporting is suppressed there. DRD does the same thing
2048 implicitly. This is necessary because Solaris libc caches many
2049 objects and reuses them for different threads and that confuses
2050 Helgrind.
2051
2053 --check-stack-var=<yes|no> [default: no]
2054 Controls whether DRD detects data races on stack variables.
2055 Verifying stack variables is disabled by default because most
2056 programs do not share stack variables over threads.
2057 --exclusive-threshold=<n> [default: off]
2059 Print an error message if any mutex or writer lock has been held
2060 longer than the time specified in milliseconds. This option enables
2061 the detection of lock contention.
2062 --join-list-vol=<n> [default: 10]
2064 Data races that occur between a statement at the end of one thread
2065 and another thread can be missed if memory access information is
2066 discarded immediately after a thread has been joined. This option
2067 allows one to specify for how many joined threads memory access
2068 information should be retained.
2069 --first-race-only=<yes|no> [default: no]
2071 Whether to report only the first data race that has been detected
2072 on a memory location or all data races that have been detected on a
2073 memory location.
2074 --free-is-write=<yes|no> [default: no]
2076 Whether to report races between accessing memory and freeing
2077 memory. Enabling this option may cause DRD to run slightly slower.
2078 Notes:
2079 · Don't enable this option when using custom memory allocators
2081 that use the VG_USERREQ__MALLOCLIKE_BLOCK and
2082 VG_USERREQ__FREELIKE_BLOCK because that would result in false
2083 positives.
2084 · Don't enable this option when using reference-counted objects
2086 because that will result in false positives, even when that
2087 code has been annotated properly with ANNOTATE_HAPPENS_BEFORE
2088 and ANNOTATE_HAPPENS_AFTER. See e.g. the output of the
2089 following command for an example: valgrind --tool=drd
2090 --free-is-write=yes drd/tests/annotate_smart_pointer.
2091 --report-signal-unlocked=<yes|no> [default: yes]
2093 Whether to report calls to pthread_cond_signal and
2094 pthread_cond_broadcast where the mutex associated with the signal
2095 through pthread_cond_wait or pthread_cond_timed_waitis not locked
2096 at the time the signal is sent. Sending a signal without holding a
2097 lock on the associated mutex is a common programming error which
2098 can cause subtle race conditions and unpredictable behavior. There
2099 exist some uncommon synchronization patterns however where it is
2100 safe to send a signal without holding a lock on the associated
2101 mutex.
2102 --segment-merging=<yes|no> [default: yes]
2104 Controls segment merging. Segment merging is an algorithm to limit
2105 memory usage of the data race detection algorithm. Disabling
2106 segment merging may improve the accuracy of the so-called 'other
2107 segments' displayed in race reports but can also trigger an out of
2108 memory error.
2109 --segment-merging-interval=<n> [default: 10]
2111 Perform segment merging only after the specified number of new
2112 segments have been created. This is an advanced configuration
2113 option that allows one to choose whether to minimize DRD's memory
2114 usage by choosing a low value or to let DRD run faster by choosing
2115 a slightly higher value. The optimal value for this parameter
2116 depends on the program being analyzed. The default value works well
2117 for most programs.
2118 --shared-threshold=<n> [default: off]
2120 Print an error message if a reader lock has been held longer than
2121 the specified time (in milliseconds). This option enables the
2122 detection of lock contention.
2123 --show-confl-seg=<yes|no> [default: yes]
2125 Show conflicting segments in race reports. Since this information
2126 can help to find the cause of a data race, this option is enabled
2127 by default. Disabling this option makes the output of DRD more
2128 compact.
2129 --show-stack-usage=<yes|no> [default: no]
2131 Print stack usage at thread exit time. When a program creates a
2132 large number of threads it becomes important to limit the amount of
2133 virtual memory allocated for thread stacks. This option makes it
2134 possible to observe how much stack memory has been used by each
2135 thread of the client program. Note: the DRD tool itself allocates
2136 some temporary data on the client thread stack. The space necessary
2137 for this temporary data must be allocated by the client program
2138 when it allocates stack memory, but is not included in stack usage
2139 reported by DRD.
2140 --ignore-thread-creation=<yes|no> [default: no]
2142 Controls whether all activities during thread creation should be
2143 ignored. By default enabled only on Solaris. Solaris provides
2144 higher throughput, parallelism and scalability than other operating
2145 systems, at the cost of more fine-grained locking activity. This
2146 means for example that when a thread is created under glibc, just
2147 one big lock is used for all thread setup. Solaris libc uses
2148 several fine-grained locks and the creator thread resumes its
2149 activities as soon as possible, leaving for example stack and TLS
2150 setup sequence to the created thread. This situation confuses DRD
2151 as it assumes there is some false ordering in place between creator
2152 and created thread; and therefore many types of race conditions in
2153 the application would not be reported. To prevent such false
2154 ordering, this command line option is set to yes by default on
2155 Solaris. All activity (loads, stores, client requests) is therefore
2156 ignored during:
2157 · pthread_create() call in the creator thread
2159 · thread creation phase (stack and TLS setup) in the created
2161 thread
2162 --trace-addr=<address> [default: none]
2164 Trace all load and store activity for the specified address. This
2165 option may be specified more than once.
2166 --ptrace-addr=<address> [default: none]
2168 Trace all load and store activity for the specified address and
2169 keep doing that even after the memory at that address has been
2170 freed and reallocated.
2171 --trace-alloc=<yes|no> [default: no]
2173 Trace all memory allocations and deallocations. May produce a huge
2174 amount of output.
2175 --trace-barrier=<yes|no> [default: no]
2177 Trace all barrier activity.
2178 --trace-cond=<yes|no> [default: no]
2180 Trace all condition variable activity.
2181 --trace-fork-join=<yes|no> [default: no]
2183 Trace all thread creation and all thread termination events.
2184 --trace-hb=<yes|no> [default: no]
2186 Trace execution of the ANNOTATE_HAPPENS_BEFORE(),
2187 ANNOTATE_HAPPENS_AFTER() and ANNOTATE_HAPPENS_DONE() client
2188 requests.
2189 --trace-mutex=<yes|no> [default: no]
2191 Trace all mutex activity.
2192 --trace-rwlock=<yes|no> [default: no]
2194 Trace all reader-writer lock activity.
2195 --trace-semaphore=<yes|no> [default: no]
2197 Trace all semaphore activity.
2198
2200 --heap=<yes|no> [default: yes]
2201 Specifies whether heap profiling should be done.
2202 --heap-admin=<size> [default: 8]
2204 If heap profiling is enabled, gives the number of administrative
2205 bytes per block to use. This should be an estimate of the average,
2206 since it may vary. For example, the allocator used by glibc on
2207 Linux requires somewhere between 4 to 15 bytes per block, depending
2208 on various factors. That allocator also requires admin space for
2209 freed blocks, but Massif cannot account for this.
2210 --stacks=<yes|no> [default: no]
2212 Specifies whether stack profiling should be done. This option slows
2213 Massif down greatly, and so is off by default. Note that Massif
2214 assumes that the main stack has size zero at start-up. This is not
2215 true, but doing otherwise accurately is difficult. Furthermore,
2216 starting at zero better indicates the size of the part of the main
2217 stack that a user program actually has control over.
2218 --pages-as-heap=<yes|no> [default: no]
2220 Tells Massif to profile memory at the page level rather than at the
2221 malloc'd block level. See above for details.
2222 --depth=<number> [default: 30]
2224 Maximum depth of the allocation trees recorded for detailed
2225 snapshots. Increasing it will make Massif run somewhat more slowly,
2226 use more memory, and produce bigger output files.
2227 --alloc-fn=<name>
2229 Functions specified with this option will be treated as though they
2230 were a heap allocation function such as malloc. This is useful for
2231 functions that are wrappers to malloc or new, which can fill up the
2232 allocation trees with uninteresting information. This option can be
2233 specified multiple times on the command line, to name multiple
2234 functions.
2235 Note that the named function will only be treated this way if it is
2237 the top entry in a stack trace, or just below another function
2238 treated this way. For example, if you have a function malloc1 that
2239 wraps malloc, and malloc2 that wraps malloc1, just specifying
2240 --alloc-fn=malloc2 will have no effect. You need to specify
2241 --alloc-fn=malloc1 as well. This is a little inconvenient, but the
2242 reason is that checking for allocation functions is slow, and it
2243 saves a lot of time if Massif can stop looking through the stack
2244 trace entries as soon as it finds one that doesn't match rather
2245 than having to continue through all the entries.
2246 Note that C++ names are demangled. Note also that overloaded C++
2248 names must be written in full. Single quotes may be necessary to
2249 prevent the shell from breaking them up. For example:
2250 --alloc-fn='operator new(unsigned, std::nothrow_t const&)'
2252 --ignore-fn=<name>
2255 Any direct heap allocation (i.e. a call to malloc, new, etc, or a
2256 call to a function named by an --alloc-fn option) that occurs in a
2257 function specified by this option will be ignored. This is mostly
2258 useful for testing purposes. This option can be specified multiple
2259 times on the command line, to name multiple functions.
2260 Any realloc of an ignored block will also be ignored, even if the
2262 realloc call does not occur in an ignored function. This avoids the
2263 possibility of negative heap sizes if ignored blocks are shrunk
2264 with realloc.
2265 The rules for writing C++ function names are the same as for
2267 --alloc-fn above.
2268 --threshold=<m.n> [default: 1.0]
2270 The significance threshold for heap allocations, as a percentage of
2271 total memory size. Allocation tree entries that account for less
2272 than this will be aggregated. Note that this should be specified in
2273 tandem with ms_print's option of the same name.
2274 --peak-inaccuracy=<m.n> [default: 1.0]
2276 Massif does not necessarily record the actual global memory
2277 allocation peak; by default it records a peak only when the global
2278 memory allocation size exceeds the previous peak by at least 1.0%.
2279 This is because there can be many local allocation peaks along the
2280 way, and doing a detailed snapshot for every one would be expensive
2281 and wasteful, as all but one of them will be later discarded. This
2282 inaccuracy can be changed (even to 0.0%) via this option, but
2283 Massif will run drastically slower as the number approaches zero.
2284 --time-unit=<i|ms|B> [default: i]
2286 The time unit used for the profiling. There are three
2287 possibilities: instructions executed (i), which is good for most
2288 cases; real (wallclock) time (ms, i.e. milliseconds), which is
2289 sometimes useful; and bytes allocated/deallocated on the heap
2290 and/or stack (B), which is useful for very short-run programs, and
2291 for testing purposes, because it is the most reproducible across
2292 different machines.
2293 --detailed-freq=<n> [default: 10]
2295 Frequency of detailed snapshots. With --detailed-freq=1, every
2296 snapshot is detailed.
2297 --max-snapshots=<n> [default: 100]
2299 The maximum number of snapshots recorded. If set to N, for all
2300 programs except very short-running ones, the final number of
2301 snapshots will be between N/2 and N.
2302 --massif-out-file=<file> [default: massif.out.%p]
2304 Write the profile data to file rather than to the default output
2305 file, massif.out.<pid>. The %p and %q format specifiers can be used
2306 to embed the process ID and/or the contents of an environment
2307 variable in the name, as is the case for the core option
2308 --log-file.
2309
2311 --bb-out-file=<name> [default: bb.out.%p]
2312 This option selects the name of the basic block vector file. The %p
2313 and %q format specifiers can be used to embed the process ID and/or
2314 the contents of an environment variable in the name, as is the case
2315 for the core option --log-file.
2316 --pc-out-file=<name> [default: pc.out.%p]
2318 This option selects the name of the PC file. This file holds
2319 program counter addresses and function name info for the various
2320 basic blocks. This can be used in conjunction with the basic block
2321 vector file to fast-forward via function names instead of just
2322 instruction counts. The %p and %q format specifiers can be used to
2323 embed the process ID and/or the contents of an environment variable
2324 in the name, as is the case for the core option --log-file.
2325 --interval-size=<number> [default: 100000000]
2327 This option selects the size of the interval to use. The default is
2328 100 million instructions, which is a commonly used value. Other
2329 sizes can be used; smaller intervals can help programs with
2330 finer-grained phases. However smaller interval size can lead to
2331 accuracy issues due to warm-up effects (When fast-forwarding the
2332 various architectural features will be un-initialized, and it will
2333 take some number of instructions before they "warm up" to the state
2334 a full simulation would be at without the fast-forwarding. Large
2335 interval sizes tend to mitigate this.)
2336 --instr-count-only [default: no]
2338 This option tells the tool to only display instruction count
2339 totals, and to not generate the actual basic block vector file.
2340 This is useful for debugging, and for gathering instruction count
2341 info without generating the large basic block vector files.
2342
2344 --basic-counts=<no|yes> [default: yes]
2345 When enabled, Lackey prints the following statistics and
2346 information about the execution of the client program:
2347 1. The number of calls to the function specified by the --fnname
2349 option (the default is main). If the program has had its
2350 symbols stripped, the count will always be zero.
2351 2. The number of conditional branches encountered and the number
2353 and proportion of those taken.
2354 3. The number of superblocks entered and completed by the program.
2356 Note that due to optimisations done by the JIT, this is not at
2357 all an accurate value.
2358 4. The number of guest (x86, amd64, ppc, etc.) instructions and IR
2360 statements executed. IR is Valgrind's RISC-like intermediate
2361 representation via which all instrumentation is done.
2362 5. Ratios between some of these counts.
2364 6. The exit code of the client program.
2366 --detailed-counts=<no|yes> [default: no]
2368 When enabled, Lackey prints a table containing counts of loads,
2369 stores and ALU operations, differentiated by their IR types. The IR
2370 types are identified by their IR name ("I1", "I8", ... "I128",
2371 "F32", "F64", and "V128").
2372 --trace-mem=<no|yes> [default: no]
2374 When enabled, Lackey prints the size and address of almost every
2375 memory access made by the program. See the comments at the top of
2376 the file lackey/lk_main.c for details about the output format, how
2377 it works, and inaccuracies in the address trace. Note that this
2378 option produces immense amounts of output.
2379 --trace-superblocks=<no|yes> [default: no]
2381 When enabled, Lackey prints out the address of every superblock (a
2382 single entry, multiple exit, linear chunk of code) executed by the
2383 program. This is primarily of interest to Valgrind developers. See
2384 the comments at the top of the file lackey/lk_main.c for details
2385 about the output format. Note that this option produces large
2386 amounts of output.
2387 --fnname=<name> [default: main]
2389 Changes the function for which calls are counted when
2390 --basic-counts=yes is specified.
2391
2393 cg_annotate(1), callgrind_annotate(1), callgrind_control(1),
2394 ms_print(1), $INSTALL/share/doc/valgrind/html/index.html or
2395 http://www.valgrind.org/docs/manual/index.html, Debugging your program
2396 using Valgrind's gdbserver and GDB[1] vgdb[2], Valgrind monitor
2397 commands[3], The Commentary[4], Scheduling and Multi-Thread
2398 Performance[5], Cachegrind: a cache and branch-prediction profiler[6].
2399 Execution Trees[7]
2400
2402 See the AUTHORS file in the valgrind distribution for a comprehensive
2403 list of authors.
2404 This manpage was written by Andres Roldan <aroldan@debian.org> and the
2406 Valgrind developers.
2407
2409 1. Debugging your program using Valgrind's gdbserver and GDB
2410 http://www.valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.gdbserver
2411 2. vgdb
2413 http://www.valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.vgdb
2414 3. Valgrind monitor commands
2416 http://www.valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.valgrind-monitor-commands
2417 4. The Commentary
2419 http://www.valgrind.org/docs/manual/manual-core.html#manual-core.comment
2420 5. Scheduling and Multi-Thread Performance
2422 http://www.valgrind.org/docs/manual/manual-core.html#manual-core.pthreads_perf_sched
2423 6. Cachegrind: a cache and branch-prediction profiler
2425 http://www.valgrind.org/docs/manual/cg-manual.html
2426 7. Execution Trees
2428 http://www.valgrind.org/docs/manual/manual-core.html#manual-core.xtree
2429Release 3.16.1 06/22/2020 VALGRIND(1)