Re: [PATCH 1/1] dt-bindings: pci: layerscape-pci: Convert to yaml file

[Conor Dooley]

On Thu, Feb 08, 2024 at 04:34:19PM -0500, Frank Li wrote:
> On Thu, Feb 08, 2024 at 09:20:08PM +0000, Conor Dooley wrote:
> > On Thu, Feb 08, 2024 at 03:20:50PM -0500, Frank Li wrote:
> > 
> > > > > > > > > +  reg:
> > > > > > > > > +    maxItems: 2
> > > > > > > > > +
> > > > > > > > > +  reg-names:
> > > > > > > > > +    items:
> > > > > > > > > +      - const: regs
> > > > > > > > > +      - const: addr_space
> > > > > > > > 
> > > > > > > > The example uses "regs" and "config". Where did addr_space come from?
> > > > > > > 
> > > > > > > Example just show pcie-host part. Not show pcie-ep part.
> > > > > > > pcie-ep part need 'addr_space'.
> > > > > > 
> > > > > > Okay. Again, please mention where this is coming from.
> > > > > 
> > > > > Ideally it comes from snsp,dwc-pcie-ep.yaml. but it is use 'dbi' instead
> > > > > of 'regs'. It needs extra effort to make driver code algin common
> > > > > snps,dwc-pcie-ep.yaml, and update exist all dts files.
> > > > > 
> > > > > I think it will be deleted soon. 
> > > > 
> > > > What I am looking for here is you to explain in the commit message that
> > > > the endpoint driver in linux and the dts have always used "addr_space".
> > > > Checking that there's not a u-boot or *bsd that uses "config" would also
> > > > be very helpful.
> > > 
> > > I confused. Actually this two part PCIE-RC and PCIE-EP.
> > > PCIE-RC using 'config'
> > > PCIE-EP using 'addr_spcae'
> > 
> > Yeah, I get this. The text binding makes it seem like "config" should be
> > used for both RC and EP, so I am just asking you to check that there are
> > no drivers in other kernels or bootloaders that use "config" for EP
> > mode.
> 
> There are not 'config' concept for EP mode. Only RC mode have 'config'
> space concept to get PCIe device's config space. EP mode only have
> "add_space" for outbound windows. If other place using "config" for EP, it
> is totally wrong, they should fix it.

It might be a totally wrong concept, but it is what the binding said, so
they are within their rights to use that name.

> 
> > 
> > > I check old txt file, which have not mention it. I can remove it.
> > 
> > if you drop "addr_space", you'll need to update the endpoint driver so
> > that it supports both "addr_space" and "config". If there are no
> > endpoint drivers using "config" in other operating systems, and all the
> > dts files use "addr_space", documenting "reg" and "addr_space" for
> > endpoint mode seems fair to me.
> 
> It is up to how to create patches. "addr_space" needs. If you want me to
> create one version, which 100% match original txt. I can do that. Then
> create increment patch to fix the problem.
> 
> If want to create a basic work version like this, which included some minus
> fixes.  
> 
> The both method is fine for me. Second method just need more efforts.

I don't have a problem with you making it "addr_space" as long as you
check to make sure that there are no users of "config" outside the
kernel and you mention in the commit message that this is a difference
from the text binding (and why).

Thanks,
Conor.

Attachment: signature.asc
Description: PGP signature