Re: [PATCH] PCI: add missing DT binding for linux,pci-domain property

[Arnd Bergmann <arnd@arndb.de>]

On Thursday 06 November 2014 16:08:57 Lucas Stach wrote:
> Am Dienstag, den 04.11.2014, 13:07 +0100 schrieb Arnd Bergmann:
> > On Tuesday 04 November 2014 12:00:52 Liviu Dudau wrote:
> > > 
> > > While the description is potentially correct, what it fails to explain is that the
> > > choice of using the property or generating an unstable (across boots) unique
> > > number is actually the choice of the host bridge driver at the moment. I know that
> > > my earlier implementations were defaulting to the automatic numbering, but that has
> > > been dropped from the final series as Rob Herring was objecting to it.
> > > 
> > > There is still scope to adopt a wide policy here, but for now it should say something
> > > to the tune:
> > > 
> > >    If present this property assigns a fixed PCI domain number to a host bridge,
> > >    otherwise an unstable (across boots) unique number will be assigned.
> > >    If you decide to use the property to assign a fixed PCI domain number to a host
> > >    bridge you have to ensure that all the host bridge drivers present in the system
> > >    follow the same policy. Otherwise, potentially conflicting domain numbers
> > >    may be assigned to root busses behind different host bridges.
> > 
> > But with the latest change to the domain handling, all drivers would implement
> > this. I would just mention that Linux kernels older than 3.19 are probably
> > going to ignore this property.
> > 
> Hm, I don't think we should stick those things into the binding docs, as
> those should not be Linux specific. IMHO the time when the parsing of a
> property gets implemented is a implementation detail that has nothing to
> do with the binding. Besides vendors always screw with this timeframes
> by doing backports.

Well, unlike most properties, this one is explicitly Linux-specific,
at least that is what the name implies.

	Arnd