Re: [PATCH] RFC: nomadik: expand timesource to 63 bits

[Thomas Gleixner]

On Fri, 12 Nov 2010, Linus Walleij wrote:

> Thomas Gleixner wrote:
> > On Fri, 12 Nov 2010, Linus Walleij wrote:
> >> Uwe Kleine-König wrote:
> >>
> >>> Instead of implementing sched_clock for each architecture seperatly,
> >>> wouldn't it be nice to have a generic sched_clock that uses the
> >>> architecture's clocksource?  I tried to implement that some time ago,
> >>> but tglx shoot it down because of locking problems.
> >> I was and still am a big fan of this approach, I am willing to help
> >> testing it if you want to dust off this patch set...
> > 
> > sched_clock is not necessarily the same as the current clocksource.
> 
> IIRC Uwe:s approach was that if and only if you would want to use
> the clocksource for sched_clock() you provide a special flag on
> the clocksource, and it will be attempted to be used for sched_clock().

I still don't like the special flag approach. It's mixing
concepts. clock->read() != sched_clock().

I can understand that you want to reuse the conversion factors etc in
struct clocksource, but we have to be more clever than special flags
and magic functions to install sched_clock.

I could accept a solution where we have generic infrastructure which
uses clock->read() and does the magic 63 bit expansion and you'd
simply do sched_clock_install(struct clocksource *) to explicitely
assign that clocksource as sched_clock. No magic, straight forward and
simple.

> Incidentally the dual use of a single free-running timer as both
> single clocksource and sched_clock() baseline seem to creep up in
> a lot of embedded platforms...

No argument about that.

Thanks,

	tglx