On Wed, Apr 23, 2014 at 02:33:50PM +0200, Arnd Bergmann wrote:
> On Thursday 17 April 2014, Maxime Ripard wrote:
> > This will allow to add per-SoC hooks more easily.
> > 
> > Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
> > ---
> >  arch/arm/mach-sunxi/Makefile  |   6 +-
> >  arch/arm/mach-sunxi/restart.c | 104 +++++++++++++++++++++++++++
> >  arch/arm/mach-sunxi/restart.h |  20 ++++++
> >  arch/arm/mach-sunxi/sun4i.c   |  36 ++++++++++
> >  arch/arm/mach-sunxi/sun5i.c   |  37 ++++++++++
> >  arch/arm/mach-sunxi/sun6i.c   |  49 +++++++++++++
> >  arch/arm/mach-sunxi/sun7i.c   |  36 ++++++++++
> >  arch/arm/mach-sunxi/sunxi.c   | 164 ------------------------------------------
> >  8 files changed, 287 insertions(+), 165 deletions(-)
> 
> This doesn't look like a move in the right direction, and I don't see
> the connection with the DMA driver.

This is actually not adding any code at all, just reorganizing it.

This was also something that would allow to have per-family
configuration options, and hence not select drivers like the pinctrl
one that are pretty huge if they're not needed.

> 
> > diff --git a/arch/arm/mach-sunxi/restart.c b/arch/arm/mach-sunxi/restart.c
> > new file mode 100644
> > index 000000000000..3b89727ee943
> > --- /dev/null
> > +++ b/arch/arm/mach-sunxi/restart.c
> 
> Splitting out the restart code is good, but please complete this by
> moving it into a driver directory. My preferred location would be
> within the watchdog driver, since it's really the same registers.
> You can turn on the watchdog driver as built-in in the defconfig.
> While it's not ideal that you can end up without a restart function
> if the driver does not get loaded, I would still prefer that over
> having the driver in the mach-sunxi directory.

Ok. The thing is, we don't have any watchdog driver for the A31 for
the moment, so the restart code will have to stay there. But I can
totally do that for the A10/A20.

> An alternative would be to move the restart code into
> drivers/power/reset.

That would be a good place for the A31 code then.

> > diff --git a/arch/arm/mach-sunxi/sunxi.c b/arch/arm/mach-sunxi/sunxi.c
> > deleted file mode 100644
> > index 460b5a4962ef..000000000000
> > --- a/arch/arm/mach-sunxi/sunxi.c
> 
> Instead of splitting up this file, I think it would be better to
> reduce the number of special hacks required per machine. Note that
> for arm64 we require platforms not to have any code outside of
> the drivers directory, and I'd like to get to the same situation
> for much of arch/arm/mach-* as well.

I think it's fair.

How would you enforce default clock parenting and/or rate then?

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com