use of aliases in device trees

use of aliases in device trees

[Kumar Gala]

In some discussion on the u-boot dev list it became clear that having  
aliases in the device tree might be useful as a common way to deal  
with finding specific nodes that need fixing up by the firmware.   
This problem also exists in the kernel bootwrappers.

The common example is how to associate a given MAC address with the  
proper ethernet node.  In u-boot an explicit path is hard coded into  
the u-boot build for each ethernet device.  In the bootwrapper we use  
"linux,network-index = <N>" in the given ethernet node.

One common solution would be having a top level aliases like the pmac  
tree's have:

aliases {
	enet0 = "...";
	enet1 = "...";
	pci0 = "...";
	pci1 = "...";
};

I wanted to see what people think of this idea and about trying to  
use common names for the aliases?  If nothing else I believe we will  
look at doing this on the FSL boards/parts.

- k

Re: [U-Boot-Users] use of aliases in device trees

[Grant Likely]

On 11/3/07, Kumar Gala <galak@kernel.crashing.org> wrote:
> In some discussion on the u-boot dev list it became clear that having
> aliases in the device tree might be useful as a common way to deal
> with finding specific nodes that need fixing up by the firmware.
> This problem also exists in the kernel bootwrappers.
>
> The common example is how to associate a given MAC address with the
> proper ethernet node.  In u-boot an explicit path is hard coded into
> the u-boot build for each ethernet device.  In the bootwrapper we use
> "linux,network-index = <N>" in the given ethernet node.
>
> One common solution would be having a top level aliases like the pmac
> tree's have:
>
> aliases {
>         enet0 = "...";
>         enet1 = "...";
>         pci0 = "...";
>         pci1 = "...";
> };

One question to ask is do we use full paths or phandles to point at nodes?

OF, of course, uses full paths, but that does require more memory and
processing power, but it might not be significant enough to worry
about.

If we use phandles, then we should use names that don't conflict with
full path alias names.  It would also be desirable to be able to
generate a phandle alias from the full path alias in order to maintain
some level of compatibility with OF.  It also think it will make
maintaining aliases in .dts files simpler because trivial changes to
node paths won't break the phandle alias.

Maybe something like:
aliases {
      enet0,phandle = <&enet0>;
      enet1,phandle = <&enet1>;
      ...
};

Cheers,
g.

>
> I wanted to see what people think of this idea and about trying to
> use common names for the aliases?  If nothing else I believe we will
> look at doing this on the FSL boards/parts.
>
> - k
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc.
> Still grepping through log files to find problems?  Stop.
> Now Search log events and configuration files using AJAX and a browser.
> Download your FREE copy of Splunk now >> http://get.splunk.com/
> _______________________________________________
> U-Boot-Users mailing list
> U-Boot-Users@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/u-boot-users
>


-- 
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.
grant.likely@secretlab.ca
(403) 399-0195

Re: [U-Boot-Users] use of aliases in device trees

[David Gibson]

On Sat, Nov 03, 2007 at 09:29:48PM -0600, Grant Likely wrote:
> On 11/3/07, Kumar Gala <galak@kernel.crashing.org> wrote:
> > In some discussion on the u-boot dev list it became clear that having
> > aliases in the device tree might be useful as a common way to deal
> > with finding specific nodes that need fixing up by the firmware.
> > This problem also exists in the kernel bootwrappers.
> >
> > The common example is how to associate a given MAC address with the
> > proper ethernet node.  In u-boot an explicit path is hard coded into
> > the u-boot build for each ethernet device.  In the bootwrapper we use
> > "linux,network-index = <N>" in the given ethernet node.
> >
> > One common solution would be having a top level aliases like the pmac
> > tree's have:
> >
> > aliases {
> >         enet0 = "...";
> >         enet1 = "...";
> >         pci0 = "...";
> >         pci1 = "...";
> > };
> 
> One question to ask is do we use full paths or phandles to point at nodes?
> 
> OF, of course, uses full paths, but that does require more memory and
> processing power, but it might not be significant enough to worry
> about.
> 
> If we use phandles, then we should use names that don't conflict with
> full path alias names.  It would also be desirable to be able to
> generate a phandle alias from the full path alias in order to maintain
> some level of compatibility with OF.  It also think it will make
> maintaining aliases in .dts files simpler because trivial changes to
> node paths won't break the phandle alias.
> 
> Maybe something like:
> aliases {
>       enet0,phandle = <&enet0>;
>       enet1,phandle = <&enet1>;
>       ...
> };

The fairly small advantages of using phandles here are not worth the
hassle of defining a new alias format, when the OF one with full paths
already exists.

I would, however, like to extend dtc so that we can do something like:
	aliases {
		enet0 = &enet0;
	};
and the reference in non-cell context will expand to the full path.
That could be used for things like linux,stdout-path in /chosen, too.

I'd also like to extend dtc so that the label can be tagged somehow, say:
	*enet0: { ... };
and as well as creating a normal dtc label, an alias will also be
automatically added to /aliases.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

Re: [U-Boot-Users] use of aliases in device trees

[Grant Likely]

On 11/4/07, David Gibson <david@gibson.dropbear.id.au> wrote:
> On Sat, Nov 03, 2007 at 09:29:48PM -0600, Grant Likely wrote:
> > Maybe something like:
> > aliases {
> >       enet0,phandle = <&enet0>;
> >       enet1,phandle = <&enet1>;
> >       ...
> > };
>
> The fairly small advantages of using phandles here are not worth the
> hassle of defining a new alias format, when the OF one with full paths
> already exists.
>
> I would, however, like to extend dtc so that we can do something like:
>         aliases {
>                 enet0 = &enet0;
>         };
> and the reference in non-cell context will expand to the full path.
> That could be used for things like linux,stdout-path in /chosen, too.
>
> I'd also like to extend dtc so that the label can be tagged somehow, say:
>         *enet0: { ... };
> and as well as creating a normal dtc label, an alias will also be
> automatically added to /aliases.

Yeah, okay!  I like both of those ideas.  That certainly addresses my
ease of maintenance concerns.

Cheers,
g.

-- 
Grant Likely, B.Sc., P.Eng.
Secret Lab Technologies Ltd.
grant.likely@secretlab.ca
(403) 399-0195

Re: use of aliases in device trees

[Segher Boessenkool]

> In some discussion on the u-boot dev list it became clear that having
> aliases in the device tree might be useful as a common way to deal
> with finding specific nodes that need fixing up by the firmware.

> One common solution would be having a top level aliases like the pmac
> tree's have:
>
> aliases {
> 	enet0 = "...";
> 	enet1 = "...";
> 	pci0 = "...";
> 	pci1 = "...";
> };

It's not just Apple that uses this; this is standard OF stuff.

> I wanted to see what people think of this idea and about trying to
> use common names for the aliases?

I'm obviously all for it, having suggested it myself a few times.

For common names, I think a good starting point would be to use
the generic names (as defined in the "generic names" recommended
practice; this should be the name of the nodes pointed to as well),
followed by a number; and/or a generic name without any suffix,
which points to the "main" device of that type.

This is a lot like your example, except that "enet" isn't a generic
name (that would be "network" I believe).  It is perfectly fine to
have to or more aliases pointing to the same node though.

This won't solve all identification problems of course, it only gives
an ordering per class of devices, but in most cases this will be enough,
esp. when platform code uses this.  We'll likely need a few extra names
for special cases, but let's use some restraint there, and esp. not
define something as a generic solution when it isn't.


Segher