* Joel Fernandes <joelf@xxxxxx> [131126 11:53]:
> On 11/26/2013 12:29 PM, Tony Lindgren wrote:
> [..]
> >>> We can implement an irqchip and a clocksource in the dmtimer code for the
> >>> client drivers to use, and after that we only have a couple of dmtimer
> >>> specific functions left to export.
> >>>
> >>> I'm thinkging some thing like this for the public API:
> >>>
> >>> omap_dm_timer_request request_irq
> >>> omap_dm_timer_request_specific request_irq
> >>> omap_dm_timer_get_irq request_irq
> >>> omap_dm_timer_set_source clk_set_rate
> >>
> >> For clk_set_rate, how would one directly access the timer node if we've hidden
> >> it behind an irq chip abstraction?
> >>
> >> per your suggestion, one would have something like:
> >>
> >> dsp {
> >> interrupt-parent = <&timer1>;
> >> }
> >>
> >> so how do you clk_set_rate rate something like this given the dsp node?
> >
> > All you have to do is implement a clocksource driver in dmtimer.c code.
> >
> >> If the suggestion is to get the timer1 node from the interrupt-parent property,
> >> if I may say- that's a bit ugly because now you're breaking the irq chip
> >> abstraction just to access the timer node..
> >
> > Hmm sorry I don't follow you here.
>
> I assumed above that you were suggesting implementing interrupt chaining like
> gpio-omap driver.
>
> Can you give an example workflow to explain your suggestion?
>
> I'll tell you how I understand what you were suggesting and then you can correct
> me, and maybe we can meet somewhere:
>
> Typically *without* irqchip or chaining what you suggested, we would have
> something like (purely an example):
> dsp {
> timer = <&timer1>;
> }
>
> There is an API:
> omap_dm_timer *omap_dm_timer_request_by_node(struct device_node *np); that one
> would use.
>
> Now moving to your suggestion, the dts would look like:
> dsp {
> interrupt-parrent = <&timer1>;
> interrupts = <1>;
> }
Yes something like that.
> Naturally some APIs will not fit into the IRQ framework, so these subset of
> dmtimer API may need to be exposed as you pointed. To use the API, the timer has
> to first be retrieved from interrupt-parent property of "dsp" here to get a
> device_node, and then the timer has to be requested and subset API used on it.
> This is "hack" that's not acceptable according to me...
Well it's really the same story with the GPIO framework if you follow the
request_irq analogy for a GPIO pin. You may still need to configure some things
manually that don't fit into the IRQ framework using let's say pinctrl framework.
Sure some of that can be automated eventually for GPIO and also for hardware
timers when we eventually have some generic hardware timer framework,
but the request_irq and clock_set_rate parts will stay valid.
Regards,
Tony
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
