[PATCH 4/4] ARM: shmobile: marzen: Use disabled variant of clock workaround for scif devices

[Simon Horman]

On Wed, Apr 30, 2014 at 04:49:57AM +0000, Magnus Damm wrote:
> On Wed, Apr 30, 2014 at 1:50 AM, Simon Horman <horms at verge.net.au> wrote:
> > On Wed, Apr 30, 2014 at 02:47:30AM +0200, Laurent Pinchart wrote:
> >> Hi Simon,
> >>
> >> On Wednesday 30 April 2014 08:51:11 Simon Horman wrote:
> >> > On Tue, Apr 29, 2014 at 01:40:08PM +0000, Magnus Damm wrote:
> >> > > On Tue, Apr 29, 2014 at 7:43 AM, Simon Horman wrote:
> >> > > > Now that SCIF devices are initialised using DT it should
> >> > > > be sufficient to use the disabled variant of the clock workaround.
> >> > > >
> >> > > > Signed-off-by: Simon Horman <horms+renesas at verge.net.au>
> >> > > > ---
> >> > > >
> >> > > >  arch/arm/mach-shmobile/board-marzen-reference.c | 12 ++++++------
> >> > > >  1 file changed, 6 insertions(+), 6 deletions(-)
> >> > > >
> >> > > > diff --git a/arch/arm/mach-shmobile/board-marzen-reference.c
> >> > > > b/arch/arm/mach-shmobile/board-marzen-reference.c index
> >> > > > 6d34baf..1f14d80 100644
> >> > > > --- a/arch/arm/mach-shmobile/board-marzen-reference.c
> >> > > > +++ b/arch/arm/mach-shmobile/board-marzen-reference.c
> >> > > > @@ -40,12 +40,6 @@ static void __init marzen_init_timer(void)
> >> > > >
> >> > > >   * devices until they get moved to DT.
> >> > > >   */
> >> > > >
> >> > > >  static const struct clk_name clk_names[] __initconst = {
> >> > > >
> >> > > > -       { "scif0", NULL, "sh-sci.0" },
> >> > > > -       { "scif1", NULL, "sh-sci.1" },
> >> > > > -       { "scif2", NULL, "sh-sci.2" },
> >> > > > -       { "scif3", NULL, "sh-sci.3" },
> >> > > > -       { "scif4", NULL, "sh-sci.4" },
> >> > > > -       { "scif5", NULL, "sh-sci.5" },
> >> > > >
> >> > > >         { "tmu0", NULL, "sh_tmu.0" },
> >> > > >         { "tmu1", NULL, "sh_tmu.1" },
> >> > > >         { "tmu2", NULL, "sh_tmu.2" },
> >> > > >
> >> > > > @@ -55,6 +49,12 @@ static const struct clk_name clk_names[] __initconst
> >> > > > = {
> >> > > >
> >> > > >   * This is a really crude hack to work around core platform clock
> >> > > >   issues
> >> > > >   */
> >> > > >
> >> > > >  static const struct clk_name clk_enables[] __initconst = {
> >> > > >
> >> > > > +       { "scif0", NULL, "sh-sci.0" },
> >> > > > +       { "scif1", NULL, "sh-sci.1" },
> >> > > > +       { "scif2", NULL, "sh-sci.2" },
> >> > > > +       { "scif3", NULL, "sh-sci.3" },
> >> > > > +       { "scif4", NULL, "sh-sci.4" },
> >> > > > +       { "scif5", NULL, "sh-sci.5" },
> >> > > >
> >> > > >         { "sdhi0", NULL, "ffe4c000.sd" },
> >> > > >         { "thermal", NULL, "ffc48000.thermal" },
> >> > > >
> >> > > >  };
> >> > >
> >> > > Hi Simon,
> >> > >
> >> > > Thanks for your SCIF DT patches for r8a7779. They look good and clean
> >> > > in general I think.
> >> > >
> >> > > Regarding this specific patch, I'm not sure why you need to perform
> >> > > this kind of change. Is it following the same style as other SoCs?
> >> > >
> >> > > In my mind the SCIF driver at least used to rely on both Runtime PM
> >> > > and the clock framework for clock control so only relying on clock
> >> > > framework should be enough for now. The driver may have been updated
> >> > > though, but if so we should make sure we follow the same pattern on
> >> > > other SoCs as well.
> >> >
> >> > Thanks, I agree this is not needed.
> >>
> >> Shouldn't we remove the scif clock entries from the clk_names array completely
> >> ?
> >
> > To be honest I am somewhat confused.
> >
> > My understanding was that clk_enables were used for devices
> > that are initialised using DT but whose clocks need to be forcibly
> > enabled because the interaction with runtime PM isn't handled properly
> > yet.
> >
> > And my reading of Magnus's email was that SCI devices do not need
> > to be enabled as the driver interacts with runtime PM correctly.
> >
> > So I tend to agree with you that we ought to be able to simply remove
> > the SCI devices from clk_names now they are initialised via DT
> > (unlike TMU which is still initialised as a platform device).
> 
> I suspect that simply removing the SCIF bits from clk_names makes sense.

Thanks, I now suspect so too.

I'll update this patch accordingly.