tsan does not recognize QMutex

[GoogleCodeExporter]

What steps will reproduce the problem?

Compile and run the test program below.

What is the expected output? What do you see instead?

Tsan should recognize the QMutex as a synchronizing mechanism and not print a 
warning. The example works like expected using ab boost:mutex instead.

What version of the product are you using? On what operating system?

llvm-3.4 on centos-6.5 x86_64

Please provide any additional information below.

    #include <QtConcurrentRun>
    #include <QMutex>
    #include <iostream>
    #include <boost/thread/mutex.hpp>

    class TestClass
    {
    public:
     void fun()
     {
      QMutexLocker autoMutex(&mutex);
      //boost::mutex::scoped_lock bautoMutex(bmutex);
      data = 1;
     }
    private:
     int data;
     QMutex mutex;
     boost::mutex bmutex;
    };

    int main()
    {
     TestClass *tc = new TestClass;

     QFuture<void> t1 = QtConcurrent::run(tc, &TestClass::fun);
     QFuture<void> t2 = QtConcurrent::run(tc, &TestClass::fun);

     t1.waitForFinished();
     t2.waitForFinished();

     return 0;
    }

Original issue reported on code.google.com by benjamin...@wincor-nixdorf.com on 10 Mar 2014 at 1:10

[GoogleCodeExporter]

Hi,

Thanks for the report.

Does QT accept patches from community?

Please also file an issue against QT. Let's see what interest are there.

Original comment by dvyu...@google.com on 11 Mar 2014 at 6:08

• Added labels: Type-Enhancement
• Removed labels: Type-Defect

[GoogleCodeExporter]

The Qt bug report was closed because it's "not our issue": 
https://bugreports.qt.io/browse/QTBUG-37402

However they do accept patches if someone is able to fix the issue in the way 
they want you to...

Original comment by cbpar...@gmail.com on 2 Feb 2015 at 8:37

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

Is there a way to annotate QMutex for ThreadSanitizer to understand it? If so, 
then I'd add such code to Qt Base when defined(__has_feature) && 
__has_feature(thread_sanitizer). 

If there is no such thing, and I cannot find any headers with macros I could 
use for that purpose, can someone point me to the sources in TSan (in 
data-race-test, right?) that need to be patched to add support for QMutex?

Original comment by milian.w...@kdab.com on 16 Jul 2015 at 4:29

[GoogleCodeExporter]

Hi milian.wolff,

Yes, tsan supports annotations for mutexes:
https://code.google.com/p/data-race-test/source/browse/trunk/dynamic_annotations
/dynamic_annotations.h
They are implemented in tsan here:
http://llvm.org/viewvc/llvm-project/compiler-rt/trunk/lib/tsan/rtl/tsan_interfac
e_ann.cc?view=markup

In particular you want:
AnnotateRWLockCreate(char *f, int l, uptr m)
AnnotateRWLockDestroy(char *f, int l, uptr m)
AnnotateRWLockAcquired(char *f, int l, uptr m, uptr is_w)
AnnotateRWLockReleased(char *f, int l, uptr m, uptr is_w)

f is file name (__FILE__)
l is line (__LINE__)
m is mutex address
is_w is whether mutexes is locked in write mode or in read mode

AnnotateRWLockAcquired must be called after the actual acquire.
AnnotateRWLockReleased must be called right before the mutex is actually 
unlocked.

Original comment by dvyu...@google.com on 16 Jul 2015 at 8:46

[GoogleCodeExporter]

GCC does not seem to support the __has_feature check for TSan - is there an 
alternative to detect it there?

Original comment by milian.w...@kdab.com on 20 Jul 2015 at 10:31

[GoogleCodeExporter]

The attached file works fine with clang++, but in GCC it does not work at all. 
Note that I rely instead on a define (QT_SANITIZE_THREAD) to be set, instead of 
the __has_feature which is also not supported by GCC.

I thought GCC and clang share the same TSan, but this here shows that 
apparently this is it not the case. Is there any way to fix this also for GCC?

Original comment by milian.w...@kdab.com on 20 Jul 2015 at 12:53

[GoogleCodeExporter]

Internally at Google we define THREAD_SANITIZER macro when tsan is enabled, and 
check that macro.

You can file a bug against gcc, but as far as I remember they intentionally did 
not implement it.

Original comment by dvyu...@google.com on 20 Jul 2015 at 6:09

[GoogleCodeExporter]

Yuri, do you know any way to detect if a sanitizer is enabled or not in gcc?

Original comment by dvyu...@google.com on 20 Jul 2015 at 6:10

[GoogleCodeExporter]

OK, I can add a similar macro. But the annotations don't work there - should 
they? I'm using GCC 5.1.0.

Original comment by milian.w...@kdab.com on 21 Jul 2015 at 2:26

[GoogleCodeExporter]

gcc should support these annotations as well.
Please show how you annotate, and how exactly annotation don't work (don't 
compile? don't link?).

Original comment by dvyu...@google.com on 22 Jul 2015 at 6:25

[GoogleCodeExporter]

Adding Project:ThreadSanitizer as part of GitHub migration.

Original comment by gli...@google.com on 30 Jul 2015 at 9:21

• Added labels: ProjectThreadSanitizer