Re: [BISECTED] 2.6.35.*: horrible (exponential? >linear) slowdown to unusability (HPET)

[Thomas Gleixner <tglx@linutronix.de>]

On Tue, 14 Sep 2010, Nix wrote:

> On 14 Sep 2010, Thomas Gleixner said:
> 
> > On Fri, 10 Sep 2010, Thomas Gleixner wrote:
> >> On Fri, 10 Sep 2010, Nix wrote:
> >> > 
> >> > I suspect current_clocksource doesn't do what we think it does, or the
> >> > clocksource and 'clock event device' are not the same.)
> >> 
> >> Right, they are not the same. clocksource provides us a read out
> >> device for timekeeping (usually a simple increasing counter). clock
> >> event device is used to generate timer interrupts.
> >> 
> >> HPET provides both functionalities.
> >> 
> >> The patch you bisected is affecting the clock events part of the
> >> HPET. So yes, it's not a clock source problem.
> >
> > Another question. If you add hpet=verbose (or revert that commit) does
> > the printk_once in hpet_set_next_event() show up in dmesg ?
> >
> >     "hpet: compare register read back failed."
> 
> Well, given that I reverted that commit in the past (during the
> bisection) and also added hpet=verbose[1], I'd have expected it to turn
> up in my past syslogs. There is no sign of it, in syslogs back to 2009.
> 
> 
> [1] and indeed am running 2.6.35.4 with hpet=verbose and no symptoms
>     on the affected system *now* -- a mystery because I thought the
>     hpet=verbose-turns-it-off code was removed, but turning on verbose
>     still seems to quash the bug?! or perhaps it's slightly
>     intermittent and every N boots I'm lucky enough to miss it. I'm
>     slightly chary of rebooting to find out ;}

hpet=verbose forces the read back.

FYI, we got a reasonable explanation now what's going on in hardware
land. Patch is in the works.

Thanks,

	tglx