Re: [PATCH 2/2] ARM: dt: tegra: cardhu: register core regulator tps65911

Re: [PATCH 2/2] ARM: dt: tegra: cardhu: register core regulator tps65911

[Olof Johansson]

[+devicetree-discuss and grant/rob]

On Fri, Jun 1, 2012 at 2:04 PM, Mark Brown
<broonie-yzvPICuk2AATkU/dhu1WVueM+bqZidxxQQ4Iyu8u01E@public.gmane.org> wrote:
> On Fri, Jun 01, 2012 at 02:44:00PM -0600, Stephen Warren wrote:
>
>> Could you expand on "named property" a bit; I'm not quite sure what
>> you're getting at - literally a property with name "named" (which
>> would be the same as regulator-id under just a different property
>> name), or ...?
>
> Just a property where we only care about a name (ie, that the property
> is present).
>
>> > Can't we use the right hand side of this?  It appears to just be
>> > syntactic sugar without any current meaning.
>
>> The stuff to the right of @ is the "unit address" and must match the
>> value in the reg property. Using that was the first proposal I had
>> above (which I also didn't like as much)
>
> The stuff to the left of the @ is just noise right now, though - it has
> no meaning currently.  It's filled in with "regulator" because we need
> to put something there AFAICT.

Right. In general (and historically) in the device tree, names of the
nodes should have meaning for the person reading the device tree, but
it's not meant to be used for software to figure out the hardware
configuration -- that should instead be handled through compatible +
other properties.

Names are generally kept fairly generic (ethernet, cpus, memory, pci, etc).

Where it starts to become gray area is when it comes down to specific
bindings, and essentially the device nodes underneath of those
devices. It's been generally accepted that we can put meaning to the
names there if needed, but it's still better to avoid it.

I was originally OK with the regulator binding where names have
meaning, but after having looked at it a bit recently when looking at
bindings for some new boards we have, I realized that the original
suggestion for regulator bindings doesn't necessarily isolate the
naming dependencies to only be under the regulators in question. In
particular, for things such as fixed regulators, they can be located
at other places in the device tree.

Maybe the solution to that case is to just aggregate them in one place
and make a pseudo-binding for that (or those, in case of multiple
locations).

On the rest of the name-has-meaning discussion, I think it would be
cleaner to move away from it now while there's relatively few users of
it (with a migratin path), rather than revise it later. But I'll leave
it to Grant and Rob to decide which way the prefer things to be. I
think they might both be travelling around LC/LinuxCon events at the
moment though.


-Olof

Re: [PATCH 2/2] ARM: dt: tegra: cardhu: register core regulator tps65911

[Rob Herring]

On 06/02/2012 04:19 PM, Olof Johansson wrote:
> [+devicetree-discuss and grant/rob]
> 
> On Fri, Jun 1, 2012 at 2:04 PM, Mark Brown
> <broonie-yzvPICuk2AATkU/dhu1WVueM+bqZidxxQQ4Iyu8u01E@public.gmane.org> wrote:
>> On Fri, Jun 01, 2012 at 02:44:00PM -0600, Stephen Warren wrote:
>>
>>> Could you expand on "named property" a bit; I'm not quite sure what
>>> you're getting at - literally a property with name "named" (which
>>> would be the same as regulator-id under just a different property
>>> name), or ...?
>>
>> Just a property where we only care about a name (ie, that the property
>> is present).
>>
>>>> Can't we use the right hand side of this?  It appears to just be
>>>> syntactic sugar without any current meaning.
>>
>>> The stuff to the right of @ is the "unit address" and must match the
>>> value in the reg property. Using that was the first proposal I had
>>> above (which I also didn't like as much)
>>
>> The stuff to the left of the @ is just noise right now, though - it has
>> no meaning currently.  It's filled in with "regulator" because we need
>> to put something there AFAICT.
> 
> Right. In general (and historically) in the device tree, names of the
> nodes should have meaning for the person reading the device tree, but
> it's not meant to be used for software to figure out the hardware
> configuration -- that should instead be handled through compatible +
> other properties.
> 
> Names are generally kept fairly generic (ethernet, cpus, memory, pci, etc).
> 
> Where it starts to become gray area is when it comes down to specific
> bindings, and essentially the device nodes underneath of those
> devices. It's been generally accepted that we can put meaning to the
> names there if needed, but it's still better to avoid it.
> 
> I was originally OK with the regulator binding where names have
> meaning, but after having looked at it a bit recently when looking at
> bindings for some new boards we have, I realized that the original
> suggestion for regulator bindings doesn't necessarily isolate the
> naming dependencies to only be under the regulators in question. In
> particular, for things such as fixed regulators, they can be located
> at other places in the device tree.
> 
> Maybe the solution to that case is to just aggregate them in one place
> and make a pseudo-binding for that (or those, in case of multiple
> locations).
> 
> On the rest of the name-has-meaning discussion, I think it would be
> cleaner to move away from it now while there's relatively few users of
> it (with a migratin path), rather than revise it later. But I'll leave
> it to Grant and Rob to decide which way the prefer things to be. I
> think they might both be travelling around LC/LinuxCon events at the
> moment though.

I tend to agree with Steven's and Olof's comments in this thread. As the
node names generally don't have much meaning, I don't think we should
start now. We've already got multiple styles of bindings and I don't
think we need more.

Rob


> 
> -Olof
> _______________________________________________
> devicetree-discuss mailing list
> devicetree-discuss-uLR06cmDAlY/bJ5BZ2RsiQ@public.gmane.org
> https://lists.ozlabs.org/listinfo/devicetree-discuss

Re: [PATCH 2/2] ARM: dt: tegra: cardhu: register core regulator tps65911

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 386 bytes --]

On Sat, Jun 02, 2012 at 02:19:57PM -0700, Olof Johansson wrote:

> naming dependencies to only be under the regulators in question. In
> particular, for things such as fixed regulators, they can be located
> at other places in the device tree.

I'm sorry, I can't parse this at all.  What "naming dependencies" are
you talking about?  This is purely an issue for multi-regulator chips.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 2/2] ARM: dt: tegra: cardhu: register core regulator tps65911

[Mark Brown]

[-- Attachment #1.1: Type: text/plain, Size: 1055 bytes --]

On Sat, Jun 02, 2012 at 09:45:10PM -0500, Rob Herring wrote:

> I tend to agree with Steven's and Olof's comments in this thread. As the
> node names generally don't have much meaning, I don't think we should
> start now. We've already got multiple styles of bindings and I don't
> think we need more.

Well, if we're going to go with an existing idiom the normal thing would
be an ordered array which is absolutely abysmal from a usability
standpoint.  Compatible properties don't work as the whole reason we
have an issue here is that people want to have a single node
representing a group of regulators - for regulators which we can add a
compatible property to we're already doing that and have no issue.

What device tree seems to need rather badly is a way of representing
key/value pairs - aside from the legacy bindings that seems to be the
major source of pain when trying to contort things into DT.

Using the "regulator" string that we have to put in the binding (which
is currently totally meaningless) does seem like a good way forward
here.

[-- Attachment #1.2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

[-- Attachment #2: Type: text/plain, Size: 192 bytes --]

_______________________________________________
devicetree-discuss mailing list
devicetree-discuss-uLR06cmDAlY/bJ5BZ2RsiQ@public.gmane.org
https://lists.ozlabs.org/listinfo/devicetree-discuss

Re: [PATCH 2/2] ARM: dt: tegra: cardhu: register core regulator tps65911

[Mitch Bradley]

[-- Attachment #1.1: Type: text/plain, Size: 1435 bytes --]

On 6/3/2012 2:05 AM, Mark Brown wrote:
> On Sat, Jun 02, 2012 at 09:45:10PM -0500, Rob Herring wrote:
>
>> I tend to agree with Steven's and Olof's comments in this thread. As the
>> node names generally don't have much meaning, I don't think we should
>> start now. We've already got multiple styles of bindings and I don't
>> think we need more.
>
> Well, if we're going to go with an existing idiom the normal thing would
> be an ordered array which is absolutely abysmal from a usability
> standpoint.  Compatible properties don't work as the whole reason we
> have an issue here is that people want to have a single node
> representing a group of regulators - for regulators which we can add a
> compatible property to we're already doing that and have no issue.
>
> What device tree seems to need rather badly is a way of representing
> key/value pairs -

Perhaps ironically, the fundamental device tree construct - the 
"property" - is a key/value pair.

> aside from the legacy bindings that seems to be the
> major source of pain when trying to contort things into DT.
>
> Using the "regulator" string that we have to put in the binding (which
> is currently totally meaningless) does seem like a good way forward
> here.
>
>
> _______________________________________________
> devicetree-discuss mailing list
> devicetree-discuss-uLR06cmDAlY/bJ5BZ2RsiQ@public.gmane.org
> https://lists.ozlabs.org/listinfo/devicetree-discuss

[-- Attachment #1.2: Type: text/html, Size: 2443 bytes --]

[-- Attachment #2: Type: text/plain, Size: 192 bytes --]

_______________________________________________
devicetree-discuss mailing list
devicetree-discuss-uLR06cmDAlY/bJ5BZ2RsiQ@public.gmane.org
https://lists.ozlabs.org/listinfo/devicetree-discuss

Re: [PATCH 2/2] ARM: dt: tegra: cardhu: register core regulator tps65911

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 471 bytes --]

On Sun, Jun 03, 2012 at 06:11:21AM -1000, Mitch Bradley wrote:

> >What device tree seems to need rather badly is a way of representing
> >key/value pairs -

> Perhaps ironically, the fundamental device tree construct - the
> "property" - is a key/value pair.

Yeah, I know - the problem is when the value is itself a node, at least
as far as I can tell.  It's also frustrating that nodes have this user
bit of identifying text that we can't seem to use for some reason.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]