On Fri, May 29, 2020 at 11:14 PM Tony Lindgren <tony@xxxxxxxxxxx> wrote:
> * Arnd Bergmann <arnd@xxxxxxxx> [200529 21:09]:
> >
> > #ifdef CONFIG_SOC_HAS_REALTIME_COUNTER
> > extern void omap5_realtime_timer_init(void);
> > #else
> > static inline void omap5_realtime_timer_init(void)
> > {
> > }
> > #endif
> >
> > In fact, the inline stub is what that caused the regression,
> > so I think it's ok with my patch.
>
> To me it seems not having SOC_HAS_REALTIME_COUNTER will
> cause omap5_realtime_timer_init() not get called?
Correct, this looked to me like it was the intention of that
symbol. Unfortunately there is no help text but it is user
selectable:
config SOC_HAS_REALTIME_COUNTER
bool "Real time free running counter"
depends on SOC_OMAP5 || SOC_DRA7XX
default y
> That initializes clocks and calls timer_probe(). So this
> will result in non-booting system AFAIK, the header
> file stub should no rely CONFIG_SOC_HAS_REALTIME_COUNTER
> also, but rather ! CONFIG_SOC_OMAP5 || CONFIG_SOC_DRA7XX.
>
> Also the Makefile change at least seems wrong, that
> can't rely on CONFIG_SOC_HAS_REALTIME_COUNTER.
How about just removing the prompt on
CONFIG_SOC_HAS_REALTIME_COUNTER but keeping the
rest of my patch? That way it's just always enabled when
there is a chip that needs it enabled in the kernel config.
The only other usage of the symbol is
#ifdef CONFIG_SOC_HAS_REALTIME_COUNTER
void set_cntfreq(void);
#else
static inline void set_cntfreq(void)
{
}
#endif
Alternatively, we could just remove the Kconfig symbol
altogether and rely on (SOC_OMAP5 || SOC_DRA7XX)
everywhere, but that seems a little more fragile in case
there is going to be another chip that needs it.
Arnd
