Re: trinity finds ftrace/perf bug. Film at 11.

[Ingo Molnar <mingo@kernel.org>]

* Peter Zijlstra <peterz@infradead.org> wrote:

> On Fri, Sep 13, 2013 at 10:16:38AM -0400, Dave Jones wrote:
> > without the diff, it'll use all kinds of fds. this diff hardwires it to just use
> > fd's from perf_event_open.  I'll hack up a command line switch to do this in a better
> > way at some point.
> > 
> > 	Dave
> > 
> > 
> > diff --git a/fds.c b/fds.c
> > index cc1ca6d..f48c95e 100644
> > --- a/fds.c
> > +++ b/fds.c
> > @@ -119,6 +119,8 @@ static int get_new_random_fd(void)
> >  retry:
> >  	i = rand() % 6;
> >  
> > +	i = 3;
> > +
> >  	if (do_specific_proto == TRUE)
> >  		i = 1;
> 
> I'm just not having much luck it seems, even after actually enabling
> list debugging (D0H).

Maybe Dave could send you his kernel config, to make sure you two run 
comparable kernel images?

> What kind of hardware are you running this on?

That's a possible variation too indeed.

Also, Dave, do you run a single copy of Trinity, or multiple ones? Any 
special options for Peter to reproduce your crashes?

Thanks,

	Ingo