Re[2]: [PATCH] ARM: dts: imx: fix clocks for cspi

[Alexander Shiyan]

> > > On Thu, May 23, 2013 at 04:10:59PM +0400, Alexander Shiyan wrote:
> > > > > > The CSPI controller has only one clock, but the driver spi-imx.c needs clock "per" to calculate bitrate divisor.
> > > > > > 
> > > > > > Signed-off-by: Jonas Andersson <jonas at microbit.se>
> > > > > > ---
> > > > > >  arch/arm/boot/dts/imx25.dtsi | 12 ++++++------
> > > > > >  arch/arm/boot/dts/imx27.dtsi |  6 +++---
> > > > > >  arch/arm/boot/dts/imx51.dtsi |  2 +-
> > > > > >  arch/arm/boot/dts/imx53.dtsi |  2 +-
> > > > > >  4 files changed, 11 insertions(+), 11 deletions(-)
> > > > > > 
> > > > > > diff --git a/arch/arm/boot/dts/imx25.dtsi b/arch/arm/boot/dts/imx25.dtsi
> > > > > > index d2550e0..7011539 100644
> > > > > > --- a/arch/arm/boot/dts/imx25.dtsi
> > > > > > +++ b/arch/arm/boot/dts/imx25.dtsi
> > > > > > @@ -141,8 +141,8 @@
> > > > > >  				#size-cells = <0>;
> > > > > >  				compatible = "fsl,imx25-cspi", "fsl,imx35-cspi";
> > > > > >  				reg = <0x43fa4000 0x4000>;
> > > > > > -				clocks = <&clks 62>;
> > > > > > -				clock-names = "ipg";
> > > > > > +				clocks = <&clks 62>, <&clks 62>;
> > > > > > +				clock-names = "ipg", "per";
> > > > > >  				interrupts = <14>;
> > > > > >  				status = "disabled";
> > > > > >  			};
> > > > 
> > > > Sorry, typo.
> > > > 
> > > > > Why you do not use "dummy" clock for "per" here?
> > > > 
> > > > *for "ipg"
> > > 
> > > Because the same clock is used for the register clock and the baudrate
> > > clock. You have to enable the ipg clock to access registers.
> > 
> > I am not see any "ipg" clock usage in the spi driver.
> 
> It's the register clock. The driver enables it in order to access the
> registers.
> 
> > If "ipg" clock is
> > a dependency for "per" clock, it should be registered in ccm as parent.
> 
> Look, the eCSPI unit has two clock inputs, one for accessing the
> registers (ipg) and one for generating the SPI bit clock (per). Now the
> CSPI unit only has a single clock input, so the 100% correct way would
> be to specify only a single clock for this unit. However, since we
> handle both units with the same driver the simplest way to cope with
> it is to provide the very same clock twice, once for the register access
> and once for generating the SPI bit clock.
> And yes, we have to specify the real clock twice and can't replace one
> with a dummy clock, because if we specify a dummy clock for the ipg
> clock, then the driver couldn't access the registers even if it enabled
> the clock. If we would replace the SPI bit clock with a dummy clock then
> the device couldn't send data even if the driver enabled the bit clock.
> (That of course only becomes relevant when the driver actually
> en/disables the clocks during runtime and not only once during probe).

Well, I completely forgot about the clock for registers.
Everything looks right then.
However, it would be nice to add a few words about the clocks in the
Documentation/devicetree/bindings/spi/fsl-imx~spi.txt
Thanks for the clarification.

---