Re: [BUG -rt] -rt16 hang w/ realtime thread test

[john stultz <johnstul@us.ibm.com>]

On Sat, 2006-02-11 at 10:34 -0500, Steven Rostedt wrote:
> On Fri, 10 Feb 2006, john stultz wrote:
> 
> > Hey Ingo,
> > 	I've been hunting a report that lower priority realtime threads are not
> > preempting higher priority realtime threads. However, in generating test
> > cases, I found I was locking the system quite frequently.
> >
> > The attached test runs to completion on 2.6.15, but with 2.6.15-rt16, it
> > hangs the box. It could very well be a test issue, but I'm not sure I
> > see where the problem is.
> >
> 
> Hi John,
> 
> Have you turned on nmi_watchdog and softlockup detect?  Just so we can see
> where it is hung.

Ugh. Ok, I think I've found the issue.

The systems I'm testing w/ all use the ACPI PM timer for the
clocksource. On a whim I forced the TSC to be used and the hang went
away.

It appears the issue is that the ACPI PM wraps every ~5 seconds.  Since
the test takes longer the 5 seconds to run, the
timeofday_periodic_hook() function gets starved and we never accumulate
time. Then as the counter wraps, time starts wrapping thus timers do not
expire and the test never completes, effectively hanging the box.

I believe this issue was hit before back when cycle_t was 32bits long,
thus causing the TSC to wrap every ~4seconds on a 1Ghz box.

So whats the solution here? Do we need to do something to
timeofday_periodic_hook is guaranteed to run with some frequency? Or is
the test just bunk because realtime threads must give up the cpu in
order for the kernel to function?

Thoughts?
-john



-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/