Re: [Valgrind-users] Memcheck with Jikes RVM

[Karl C. <ka...@cs...>]

On 07/08/2014 02:04 PM, Philippe Waroquiers wrote:
> On Tue, 2014-07-08 at 03:49 -0400, Karl Cronburg wrote:
>> Any idea as to what's going on, or how I might debug the problem further?
>> Any Valgrind flags or options I might be missing? The RVM runs fine
>> on its' own, as well as in Nulgrind (and cachegrind and callgrind):
> 
> From what I can see, the above happens when Valgrind is not able to
> track properly the stack used by a thread.
> The 'switching stacks' messages very probably indicate that the jikes
> rvm does something special with stack.
> If that is the case, then usually the problem can be solved by adding
> client requests in the application code (the various VALGRIND_STACK_*
> client requests in valgrind.h).
> 
> 
> This might also happen due to the use of sigaltstack.
> It is unclear to me how to properly inform Valgrind of such alternate
> stack.
> 
> In both case, basically, the problem is that valgrind believes that the
> stack goes from the current stack pointer value till the "last known
> stack base". When changing stack; this can include non addressable
> segments. The stack unwind code assumes however that everything is
> accessible between the current sp and the stack base.

You're right, this was the problem. I verified that the stack base pointer
was way to large (right next to the SIGSEGV address). I've also found that
everything runs fine when I add the "--keep-stacktraces=none" flag, which
is acceptable for now since I'm more interested in the Java stack traces
inside the RVM anyways.

Thanks!
-Karl-