Re: [dtc PATCH V2] Warn on node name unit-address presence/absence mismatch

[Kumar Gala <galak@codeaurora.org>]

On Sep 27, 2013, at 12:17 AM, David Gibson wrote:

> On Fri, Sep 27, 2013 at 11:30:38AM +1000, Benjamin Herrenschmidt wrote:
>> On Thu, 2013-09-26 at 17:12 -0600, Stephen Warren wrote:
>>> Well, ePAPR seems pretty specific that unit address and reg are
>>> related,
>>> but says nothing about ranges in the section on node naming, nor about
>>> node naming in the section about ranges.
>>> 
>>> I'd claim that the existing PPC trees are nonconforming, and should be
>>> fixed too:-)
>> 
>> This is tricky, we should probably fix ePAPR here.
>> 
>> If you have a "soc" bus covering a given range of addresses which it
>> forwards to its children devices but doesn't have per-se its own
>> registers in that area, then it wouldn't have a "reg" property. I would
>> thus argue that in the absence of a "reg" property, if a "ranges" one is
>> present, the "parent address" entry in there is an acceptable substitute
>> for the "reg" property as far as unit addresses are concerned.
> 
> So, that's been accepted practice in fdt world for a while; I think
> ePAPR already permits that, in fact.

Are you saying that the bus binding would cover this case or something else?

>> Also don't forget that in real OFW land, the unit address is something
>> that's somewhat bus specific ... for example, PCI uses "dev,fn" rather
>> than the full 96-bit number of the "reg" entry :-)
>> 
>> Another option which would more strictly conform to ePAPR and in fact to
>> of1275 would be to require such bus nodes to have a "reg" property with
>> the address value set to the beginning of the range and the size value
>> set to 0 :-)

Uugh, that's a bit ugly.  I wonder what breaks if we had reg w/size 0.

- k

-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation