[PATCH 04/12] clocksource: Sched clock source for Versatile Express

[Pawel Moll]

On Wed, 2014-04-16 at 15:45 +0100, Rob Herring wrote:
> >> > +       setup_sched_clock(vexpress_sys_24mhz_read, 32, 24000000);
> >>
> >> This frequency should come from a DT clock binding. You will have to
> >> fallback to 24MHz for backwards compatibility though.
> >
> > I don't see why would it go to the binding. You may have noticed the
> > register is called "SYS_24MHZ", not "SYS_RANDOMCLOCK". The driver
> > *knows* what the frequency is.
> 
> A 24MHz clock is fed to this h/w block to be used by the counter in
> the block. The DT should describe that.

I even have a clk24mhz in the motherboard's dtsi, so adding a phandle to
the node is trivial, can do it. But my point is that, whatever this
clock is (24MHz, 12MHz then multiplied by 2 inside sysreg, 48MHz divided
by 2 etc.), the value in this register is incremented every 1/24 uS by
definition. And the driver can rely on this. Otherwise we're really
talking about a generic mmio-counter-based clock source, with a
clock-frequency property or a clock phandle (and a multiplier/divider
then?).

> >> > +}
> >>
> >> Wouldn't this code work for Versatile and Realview ARM reference
> >> boards? Even the register offset is the same.
> >>
> >> > +CLOCKSOURCE_OF_DECLARE(vexpress, "arm,vexpress-sysreg",
> >> > +               vexpress_sched_clock_init);
> >
> > I guess it would, yes. The sysregs are annoyingly similar and different
> > at the same time.
> >
> > One could of course try to come up with a "generic mmio clock source"
> > binding, taking the frequency as a property, but don't count on me doing
> > this... ;-)
> 
> I'm not asking for that. Just take care of all ARM Ltd boards which
> have the exact same 24MHz counter at offset 0x5C.

Yes, this is doable by all means. I'll rename it to
"clocksource/versatile.c" so we can add relevant compatible values.

Paweł