[PATCH v3 01/11] dt-bindings: thermal: Describe Armada AP806 and CP110

[robh@kernel.org (Rob Herring)]

On Fri, Dec 15, 2017 at 11:52:30AM +0100, Miquel RAYNAL wrote:
> Hello Baruch and Gregory,
> 
> On Fri, 15 Dec 2017 09:44:19 +0100
> Gregory CLEMENT <gregory.clement@free-electrons.com> wrote:
> 
> > Hi Miquel,
> >  
> >  On ven., d?c. 15 2017, Miquel RAYNAL
> > <miquel.raynal@free-electrons.com> wrote:
> > 
> > > Hello Baruch,
> > >
> > > On Fri, 15 Dec 2017 10:27:59 +0200
> > > Baruch Siach <baruch@tkos.co.il> wrote:
> > >  
> > >> Hi Miquel
> > >> 
> > >> On Thu, Dec 14, 2017 at 11:30:01AM +0100, Miquel Raynal wrote:  
> > >> > +- marvell,thermal-zone-name: The name to identify the thermal
> > >> > zone
> > >> > +                             within the sysfs, useful when
> > >> > multiple
> > >> > +                             thermal zones are registered (AP,
> > >> > CPx...).    
> > >> 
> > >> I don't think that would be acceptable. DT is about describing the
> > >> hardware. sysfs is a Linux implementation detail which is not tied
> > >> to any specific hardware. If this is accepted, the property should
> > >> be named 'linux,thermal-zone-name'.  
> > >
> > > You are right the sysfs mention should not appear in the
> > > description.
> 
> Actually, you are right for all of it, this property should not
> exist, sorry for my too quick answer.
> 
> > >
> > > Otherwise for the naming I'm not sure "linux," is a valid prefix in
> > > that case.  
> 
> Thank you both for your explanations, I was also wrong about the prefix.
> 
> > 
> > Actually the choice between linux or marvell make me realize that
> > there is something wrong. Having a name associated to a device is
> > something pretty usual with the device tree, however it is as the
> > class device level, such as clock-names, line-name, or
> > regulator-name. So in my opinion if we want to support naming from
> > device tree it would be done for all the thermal device not just for
> > the Marvell one.
> > 
> > However I don't think we need it. For example for the clocks we
> > created the name dynamically using of the base address of the
> > register to keep them unique.
> 
> I was convinced that dev_name's would be the same but after trying it on
> a 8040-DB, using dev_name(&pdev->dev) gives:
> 
>     f06f808c.thermal
>     f2400078.thermal
>     f4400078.thermal
> 
> which I found meaningful enough.
> 
> I will drop the property and use dev_name instead. I still need your
> help to solve one problem though: how to make the distinction between
> using "armada_thermal" (the previous name) and dev_name() ? If I don't
> it kind of breaks userspace, doesn't it ?

No. The /sys/devices/... or /sys/bus/platform/... paths and names are 
not guaranteed to be stable. These changed for every platform converted 
to DT for example. Userspace should be accessing things through 
/sys/class/... (or deal with changes).

Rob