RE: [PATCH] tick: add check for the existence of broadcast clock event device

[Thomas Gleixner <tglx@linutronix.de>]

On Mon, 8 Jun 2009, Pan, Jacob jun wrote:
> >I understand that, but HPET does not rely on some magic events
> >happening. That's what I'm worried about. The boot code is fragile and
> >I prefer some explicit setup call over a fragile solution which
> >happens to work.
> >
> >There is not much to complain about a platform specific function call
> >to set up special devices if there is a requirement for a call order.
> >
> >In fact you can avoid setting up the local APIC timer at all. So what
> >you want is something like the patch below. You can set the
> >setup_secondary_clock pointer in the quirks structure when you detect
> >that you are running on such a system.
> >

> [[JPAN]] Hi Thomas, I have been developing the APB timer driver as
> Feng mentioned, thank you for the suggestions.  I agree with you
> that direct setting up secondary clockevent is a better solution.
> But maybe I misunderstand the HPET code, isn't it true that per CPU
> HPET timer also rely on IPI? In hpet_cpuhp_notify().

Well the difference is, that the HPET setup always has a functional
clock event device (local APIC + the broadcast fallback) and it does
not depend on that IPI to make progress.
 
> Also for using x86_quirks, I think the default quirks are more
> generic in the sense of x86 platform, but here we are really
> choosing timer device, so can we switch based on availability of the
> timer devices instead of quirks? Otherwise, if more platforms share
> the same setup_secondary_clock() quirk, we would have to check for
> timer devices anyway.

The quirks mechanism is exactly the right thing. It provides platforms
and I consider your system a platform which has neither PIT nor HPET a
mechanism to override the default implementation. And overriding the
default local APIC timer setup is exaclty what you want to do.

Thanks,

	tglx