On Mon, Jan 30, 2017 at 04:13:07PM +0300, Alexander Kochetkov wrote:

quoted

30 ???. 2017 ?., ? 15:04, Daniel Lezcano [off-list ref] ???????(?):

There is no case when the rockchip timer is used for the clockevent.

The is already timer entry for rk3228 in the DT. And it act as clockevent. I guess it work as backup,
but I cannot test it also. In order to not break DT compatibility I had to provide one timer to be
initialized as clockevent. And implemented implicit rule the driver (first DT timer - clockevent,
second DT timer - clocksource) already exists for other timers.

quoted

If I'm not wrong, you can check /proc/interrupts and see there is no interrupt
on the rockchip timer.

Yes, you right here. I?ve temporary disable smp_twd to test rockchip timer interrupts.
There is no interrupt on the rockchip timer during normal work.

quoted

- when the CPU enters a deep idle state. But such state does not exist on rk3188

ARM chip/revision specific?

rk3188 specific. Applies also for rk3288 AFAICT. Without entering in the
details, the SoC can't handle that.

quoted

- when the system goes to suspend. But the timers are stopped in any case and
  CPU0 is always on.

There is timer for rk3228 in the DT. I guess there are situations where it can be used.
May be the situations are also acceptable for rk3188? May be something like suspend to RAM or
suspend to HDD? I am not expert in that questions. I can do some tests if you give hint/link.

Nope, rockchip clockevents won't be used, smp_twd will instead.

You can check by commenting the local-timer in the DT, and you should see the
/proc/interrupts line for the rockchip incrementing while the 'twd' are zero.

So, as I understood you suggest to leave only one timer what can be used as clocksource
only. I can implement that, but there should be DT rule what allow to setup timer
as clocksource only. I cannot do more without timer framework support.
Looks, like I have to wait your patch to implement that.

No, actually I meant the rockchip clockevents won't be used for both rk3188 and
rk3288, they are not needed.

It is for information, this patch is ok.

  -- Daniel

-- 

 <http://www.linaro.org/> Linaro.org ? Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog