On Thu, Aug 01, 2013 at 08:15:50AM +0200, John Crispin wrote:
> >Whatever works for you. I still would like to understand why plat_time_init()
> >is not suitable for John's specific use case.
> Hi Florian,
> the reason is that fixing it in plat_time_init() works around the
> real problem. the double request of the irq is a symptom of the
> actual problem, which is, that the cevt-r4k sets up the timer during
> init and not during setup. additionally, plat_time_init is used to
> probe the cevt drivers from OF already. currently the mips code just
> assumes that on a r4k we always have and want to run the cevt-r4k.
> this assumption is wrong and can quickly be fixed by making the
> cevt-r4k use the correct api.
> also fixing it this way allows the user to control the clocksource
> and change it at runtime via sysfs, a feature als not working
> currently on r4k as the cevt driver did not implement the set_mode()
> handler correctly. to be quite honest, i cannot think of a single
> way in which this can be fixed cleanly in the ralink
> plat_time_init() without using some weird heuristic. also if i fix
> this inside ralink plat_time_init() it is fixed only on ralink SoC
> and not on any other platform.
setup_irq() may fail but set_mode doesn't have a way to communicate an
error - other than leaving back a half-wrecked system so set_mode is not
a good place to do that kind of job.
How about using get_c0_compare_int() for a solution? Currently
get_c0_compare_int() can not return an error. If it could return a
negative value to indicate the unavailability of an interrupt for
cevt-r4k's use, that interrupt would be available for alternative use.