Re: 2.6.33: ftrace triggers soft lockup

[Frederic Weisbecker <fweisbec@gmail.com>]

On Thu, Mar 04, 2010 at 01:10:16PM +0800, Américo Wang wrote:
> On Thu, Mar 4, 2010 at 11:18 AM, Steven Rostedt <rostedt@goodmis.org> wrote:
> > On Thu, 2010-03-04 at 11:01 +0800, Américo Wang wrote:
> >
> >> >
> >> > So it is stuck in stop machine. I wonder where exactly. I see some do_exit
> >> > at the top but I wonder how much they are reliable.
> >>
> >> Well, I think 'kstop' is just random, sometimes I got 'watchdog' or some other
> >> process.
> >>
> >> >
> >> > Anyway, as Steve said, we really need a full config to reproduce it.
> >> >
> >>
> >> Done in another reply.
> >
> > Thanks!
> >
> > Frederic, I notice that lockdep is on, did anything change that might
> > slow down the code in lockdep, or is the function graph tracer doing
> > more locking?
> >
> > I'm betting that we are hitting a live lock. That is, an interrupt goes
> > off, it is being traced, and the function graph is tracing it, but some
> > locking is happening (although it also tracks disabling of interrupts)
> > and this slows the interrupt handler down enough that when it finishes,
> > another interrupt goes off.
> >
> > Américo,
> >
> > Could you disable LOCKDEP and see if you still encounter this lockup?
> >
> 
> Sure, after disabling LOCKDEP, I can't see the warning, _but_ the system
> is still as unacceptablly slow as when LOCKDEP was enabled.



Looks like a progress. It doesn't appear to be a true lockup but more a
starvation or a livelock.

I'm building your config, hopefully I could reproduce.

Thanks.