Re: [PATCH v4 13/21] ARM: omap: convert wakeupgen to stacked domains

[Tony Lindgren <tony@atomide.com>]

* santosh shilimkar <santosh.shilimkar@oracle.com> [150121 13:31]:
> On 1/21/2015 12:43 PM, Tony Lindgren wrote:
> >* santosh shilimkar <santosh.shilimkar@oracle.com> [150121 12:16]:
> >>
> >>TWD is useless on this machine since single core and TWD
> >>as know die in low power states. All the broadcast stuff
> >>is for SMP machines.
> >
> >Hmm it seems we should still use TWD during runtime and
> >swich over to the gptimer for idle states for wake-up
> >events.
> >
> Well timer wheel code don't support it so if you are serious,
> some one needs to do that. For me, it is not worth at all.
> You will have more to loose than gain with these time switching
> schemes since you have to keep 2 times alive, do switching, loose
> the idle time.
> 
> All of that is to save few CPU cycles since TWD is closer
> compared to other SOC timer.
> 
> Anyways I will let you fight it out but IIRC, I had a
> discussion a while back with tglx in one of the conference
> and the conclusion was it not worth doing.
> Rather TWD hardware on SOC should be made wakeup capable
> and then everything is good.
> 
> Till you have support, using TWD on AM43XX will break CPUIDLE.
> Not sure if it is supported or some one cares about it. Just
> keep that aspect in mind.

Yes sure I'm aware of this. It should be easy to profile the
speed gain to see if it would make much of a difference
before starting to tinker with that.

The way I think it's possible to do would be to copy the TWD
timer value to a wake-up capable gptimer before hitting any
deeper idle state. Of course some aux timer support might be
still needed :)

Regards,

Tony