Re: [PATCH v1 3/3] ARM64 LPC: update binding doc

[Arnd Bergmann]

On Wednesday 13 January 2016 14:34:47 Rongrong Zou wrote:
> On 2016/1/13 13:53, Benjamin Herrenschmidt wrote:
> > On Tue, 2016-01-12 at 23:52 +0100, Arnd Bergmann wrote:
> >> On Tuesday 12 January 2016 15:13:35 liviu.dudau@xxxxxxx wrote:
> >>>> int of_address_to_resource(struct device_node *dev, int index,
> >>>>                            struct resource *r)
> >>>> {
> >>>>         ...
> >>>>         /* flags can be get here, without ranges property reqired.
> >>>>          * if the reg = <0x0 0xe4 4>, I can get flag of
> >> IORESOURCE_MEM,
> >>>>          * if the reg = <0x1 0xe4 4>, I can get flag of
> >> IORESOURCE_IO,
> >>>
> >>> That is strange, the parent node has #address-cells = <2> so the
> >> first two numbers should be part
> >>> of the address and not influence the flags. Can you add some
> >> debugging in of_get_address() and
> >>> try to figure out what bus is used in  *flags = bus-
> >>> get_flags(prop) ?
> >>>
> >>>
> >>
> >> This is the standard ISA binding. The first cell is the address space
> >> (IO or MEM), the second cell is the address within that space. This
> >> is similar to how PCI works.
> >
> > Picking up that mid-way, I have LPC busses on power and am using a
> > similar binding. I'll try to grab some examples and review the
> > document tomorrow (only just noticed it while unpiling emails post-
> > vacation).

I really should have thought of that, as you mentioned already that
there is an ast2400 on those machines, and no I/O space on the PCI
bus.

Too bad we have to keep the I/O workarounds alive on PowerPC now,
I was already hoping they could go away after spider-pci gets phased
out.

> Thanks for reviewing this, I found a similar implementation at arch/powerpc/
> platform/powernv/opal-lpc.c and I had get some ideas from your work. It is
> nice to me. I'm expecting your suggestion.Thanks in advance.

Unfortunately, the way that PCI host bridges on PowerPC are handled
is a bit different from what we do on ARM64, otherwise the obvious
solution would be to move the I/O workarounds to an architecture
independent location. Maybe it's still possible, but that also requires
some refactoring then.

	Arnd