[Valgrind-users] i386 memory fence from libev

[Sean M. <se...@ro...>]

Hi all,

I'm trying to decide if a valgrind warning is a false positive or not.

The code comes from libev's memory fence implementation and reduces to:

-------------------------------------
#include <stdio.h>

int main (void)
{
#if __i386 || __i386__

	printf ("hello 32 bit world \n");
	__asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory");
	printf ("goodbye 32 bit world \n");

#endif

	return 0;
}
-------------------------------------

valgrind 3.9.0 complains (on linux, in i386 only of course):

"Invalid read of size 1"
"Address 0x4f760ff is just below the stack ptr."

The libev author believes it is a false positive from valgrind:

<http://lists.schmorp.de/pipermail/libev/2013q2/002173.html>

I've searched the valgrind bug list, but can't seem to find anything related.  I don't really know Intel assembly, but I guess that's doing an OR of constant zero and one byte away from the stack pointer.  Seems dubious to me.

Anyone have thoughts on the snippet's correctness?  Is valgrind indeed wrong to complain?

Thanks,

-- 
____________________________________________________________
Sean McBride, B. Eng                 se...@ro...
Rogue Research                        www.rogue-research.com 
Mac Software Developer              Montréal, Québec, Canada