On Saturday, August 25, 2012, David Faure wrote:
> How do I tell helgrind that some atomic operations on integers are OK?
> [...]
Without looking in more detail at this, it is difficult to tell if that
is really the right question to ask.
Helgrind treats atomic operations (atomic read-modify-write) as if they
were simply reads, both for x86/x86_64, and similarly for LL/SC pairs on
ARM, PPC, MIPS, etc. So it should not report races on code that uses
atomic operations.
Now, that said, a reason it might be reporting races like this is
that Qt is using the atomic ops to create inter-thread synchronisation
that Helgrind cannot see, hence it reports a race where there is none.
Or maybe Qt really is racey -- perhaps harmlessly. In which case H is
correct to complain.
Without looking further, it's not really possible to say what's happening.
I would be happy to do that if you make it easy (viz, send commands
starting with "wget qt5.tarball" etc) to build a simple test case.
One way to hide the complaints is use
VALGRIND_{DISABLE,ENABLE}_ERROR_REPORTING
around the affected areas. Read the docs in valgrind.h about that.
Or you can use VALGRIND_HG_{DISABLE,ENABLE}_CHECKING on the address
range (in helgrind.h) to tell H not to check the particular address
range. That's quite dangerous though because you need be sure to
re-enable checking on the range when the memory is freed.
Both of those approaches are basically bad though, because they
might hide real bugs, both in Qt and in apps linked against it.
Better to investigate what is going on here, fix any bugs found, and
mark up Qt to notify Helgrind/DRD of any inter-thread synchronisation
that the normal pthread intercepts cannot see. Once that is done,
it should be possible to use Helgrind/DRD on Qt without false errors.
That could be complex though, because it may involve describing the
synchronisation induced by thread-safe reference counting. That's
doable -- I marked up Firefox's XPCOM implementation so it can be
Helgrinded, so I know it is possible.
LMK if I can help more.
J
------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and
threat landscape has changed and how IT managers can respond. Discussions
will include endpoint security, mobile security and the latest in malware
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Valgrind-users mailing list
Valgrind-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/valgrind-users
