Re: [RFC] Reentrant clock sources

[Ingo Molnar <mingo@elte.hu>]

* john stultz <johnstul@us.ibm.com> wrote:

> On Wed, 2008-11-26 at 04:07 +0100, Ingo Molnar wrote:
> > * Thomas Gleixner <tglx@linutronix.de> wrote:
> > 
> > > > +	cycle_t (*vread)(struct clocksource *cs);
> > > 
> > > This is crap. vread can not access the clocksource.
> > 
> > i think 'reentrant' in the sense of creating self-sufficient driver 
> > entities. vread wont (and shouldnt) call ->vread() recursively - but 
> > it might want to access fields on the clocksource.
> 
> I think Thomas' issue is that vread() *cannot* access fields on the 
> clocksource (since vread has to be careful not to access any 
> non-vsyscall mapped memory).

ah, yeah - i was thinking about ->read().

in a more dynamic driver model the clocksource driver could store 
dynamic data like target port/memory address next to the clocksource 
driver, and access that - if the driver pointer is passed in.

> However not all clocksources use vread(), but really I'm not quite 
> clear on what one would want to access in the clocksource structure 
> when making a ->read() call.
> 
> So maybe a further description of what specific need motivates this 
> change would be helpful? The brief description of power management 
> doesn't quite click in my head yet.

yeah, that would be nice.

	Ingo