Re: [PATCH] powerpc/fsl: Add cache properties for T2080/T2081

[Chris Packham]

On Wed, 2020-03-25 at 15:38 +1300, Chris Packham wrote:
> On Tue, 2020-03-24 at 21:08 -0500, Scott Wood wrote:
> > On Wed, 2020-03-25 at 12:59 +1100, Michael Ellerman wrote:
> > > Chris Packham <chris.packham@xxxxxxxxxxxxxxxxxxx> writes:
> > > > Add the d-cache/i-cache properties for the T208x SoCs. The L1
> > > > cache on
> > > > these SoCs is 32KiB and is split into 64 byte blocks (lines).
> > > > 
> > > > Signed-off-by: Chris Packham <chris.packham@xxxxxxxxxxxxxxxxxxx
> > > > >
> > > > ---
> > > >  arch/powerpc/boot/dts/fsl/t208xsi-pre.dtsi | 16
> > > > ++++++++++++++++
> > > >  1 file changed, 16 insertions(+)
> > > 
> > > LGTM.
> > > 
> > > I'll wait a few days to see if Scott wants to ack it.
> > > 
> > > cheers
> > > 
> > > 
> > > > diff --git a/arch/powerpc/boot/dts/fsl/t208xsi-pre.dtsi
> > > > b/arch/powerpc/boot/dts/fsl/t208xsi-pre.dtsi
> > > > index 3f745de44284..2ad27e16ac16 100644
> > > > --- a/arch/powerpc/boot/dts/fsl/t208xsi-pre.dtsi
> > > > +++ b/arch/powerpc/boot/dts/fsl/t208xsi-pre.dtsi
> > > > @@ -81,6 +81,10 @@ cpus {
> > > >  		cpu0: PowerPC,e6500@0 {
> > > >  			device_type = "cpu";
> > > >  			reg = <0 1>;
> > > > +			d-cache-line-size = <64>;
> > > > +			i-cache-line-size = <64>;
> > > > +			d-cache-size = <32768>;
> > > > +			i-cache-size = <32768>;
> > > >  			clocks = <&clockgen 1 0>;
> > > >  			next-level-cache = <&L2_1>;
> > > >  			fsl,portid-mapping = <0x80000000>;
> > 
> > U-Boot should be setting d/i-cache-size and d/i-cache-block-size --
> > are you
> > using something else?
> 
> Nope it is u-boot specifically
> 
> U-Boot 2017.01-rc3-dirty
> 
> I'm pretty sure the '-dirty' is just a change to use a different
> cross-
> compiler but I can't confirm and I'm a little hesitant to try
> updating
> as I've only got remote access to the board right now.
> 
> > 
> > The line size is the same as the block size so we don't need a
> > separate d/i-
> > cache-line-size.
> > 
> 
> I'm not sure that'll work looking at the code[1]. It has logic to set
> bsizep to lsizep if no block size is set but not the other way round.
> Looking at the spec from devicetree.org this actually seems wrong.
> Perhaps that is the real source of the error.

Sure enough without my change

# ls /sys/firmware/devicetree/base/cpus/PowerPC,e6500@0/
bus-frequency       d-cache-size        name
cache-stash-id      device_type         next-level-cache
clock-frequency     enable-method       phandle
clocks              fsl,portid-mapping  reg
cpu-release-addr    i-cache-block-size  status
d-cache-block-size  i-cache-sets        timebase-frequency
d-cache-sets        i-cache-size

So it's the lack of handling the optional line-size. Different patch
incomming.

> 
> --
> [1] - 
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/powerpc/kernel/setup_64.c#n510
> 
>