[PATCH v2 2/3] soc: dove: add legacy support to PMU driver

[Arnd Bergmann]

On Monday 07 December 2015 22:24:58 Gregory CLEMENT wrote:
> Hi Russell King,
>  
>  On lun., déc. 07 2015, Russell King - ARM Linux <linux at arm.linux.org.uk> wrote:
> 
> > On Mon, Dec 07, 2015 at 10:09:27PM +0100, Gregory CLEMENT wrote:
> >> Hi Russell,
> >>  
> >>  On lun., déc. 07 2015, Gregory CLEMENT <gregory.clement at free-electrons.com> wrote:
> >> 
> >> >  On lun., déc. 07 2015, Gregory CLEMENT <gregory.clement at free-electrons.com> wrote:
> >> >
> >> >> Hi Russell,
> >> >>  
> >> >>  On lun., déc. 07 2015, Russell King <rmk+kernel at arm.linux.org.uk> wrote:
> >> >>
> >> >>> Add support for legacy non-DT Dove to the PMU driver, so that we can
> >> >>> transition the legacy support over.
> >> >>>
> >> >>> Signed-off-by: Russell King <rmk+kernel at arm.linux.org.uk>
> >> >>
> >> >> Applied on mvebu/soc (and fixed a conflict in drivers/soc/Makefile)
> >> >
> >> > Eventually I applied it on mvebu/drivers rather than mvebu/soc
> >> 
> >> Actually I was too optimitic. It doesn't build on mvebu/soc (based on
> >> 4.4-rc1).
> >> 
> >> First pm_genpd_poweroff_unused() had been removed since bb4b72fc63d4 "PM
> >> / Domains: Remove pm_genpd_poweroff_unused() API ".
> >> 
> >> Then, even by removing the call this function to go furthetr I got other
> >> errors such as: drivers/soc/dove/pmu.c:344:46: error:
> >> ‘IRQ_DOVE_PMU_START’ undeclared (first use in this function)
> >
> > I guess that's with Arnd's patches?  Sigh, okay, it'll have to wait yet
> > _another_ kernel cycle then.
> 
> Yes I think the breakage comes from "ARM: orion: use SPARSE_IRQ
> everywhere". With this one mach/irqs.h is no more visible by
> drivers/soc/dove/pmu.c. But I am sure we can find a solution. It could
> not be the first time there is such issue when moving to multiplatform.
> 
> Arnd, do you have some suggestion?

It should be enough to add

#ifdef ARCH_DOVE
#include <mach/irqs.h>
#endif

in the file to avoid the error, and that can be done with or without
my "ARM: orion: use SPARSE_IRQ everywhere" patch.

Alternatively, we can avoid the header dependency in some way, e.g.
by passing the interrupt range (or just the first IRQ) as a resource
or through dove_pmu_domain_initdata.

	Arnd