Re: [PATCH v1] watchdog: Use a reference cycle counter to avoid scaling issues

[Thomas Gleixner]

On Thu, 23 Apr 2015, Andi Kleen wrote:
> > We can just detect the deviation in the callback itself:
> > 
> >        u64 now = ktime_get_mono_fast_ns();
> > 
> >        if (now - __this_cpu_read(nmi_timestamp) < period)
> >        	       return;
> > 
> >        __this_cpu_write(nmi_timestamp, now);
> > 
> > It's that simple.
> 
> It's a simple short term hac^wsolution.

Yes, and way simpler and less complex for pushing into stable.

> But if we had a (hypothetical) system with let's say 10*TSC max you
> may end up with quite a few false ticks, as in unnecessary
> interrupts. With 100*TSC it would be really bad.

And hypothetical systems with 100*TSC justify all that?
 
> There were systems in the past that ran TSC at a much slower frequency,
> such as the early AMD Barcelona systems.
> 
> So the problem may eventually come back if not solved properly.

There are better ways to do that than using heuristics. We have to
deal with 3 variants of the reference counter:

1) Core and Atom: counts bus cycles and we know that frequency already
   	    	  from the local apic calibration

2) Nehalem, Westmere: Same as TSC

3) Sandybridge and later:  XCLK which is 100MHz

No magic calibration, just use the information which we have on our
hands already.

Thanks,

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