Dear Russell King - ARM Linux,

On Wed, 28 Aug 2013 13:42:55 +0100, Russell King - ARM Linux wrote:

quoted

Using the name of the oldest SoC in the family that had the IP block is
the norm, because it's really what "compatible" means: the IP block in
Dove is *compatible* with the one that was originally introduced in
Kirkwood.

See what Rob Herring (one of the DT maintainer) says in
http://lists.infradead.org/pipermail/linux-mtd/2012-March/040417.html:

"""
There is no reason all machines can't use "st,spear600-smi" in their
dts. It doesn't have to be a spear600, just compatible with it. Really
you want the string to be the oldest SOC the block is in and then newer
SOCs can claim compatibility with the old version.
"""

The thread was precisely about replacing a SoC-specific compatible
string "st,spear600-smi" by a more generic "st,spear-smi" and Rob
Herring (above) was opposing to that.

We're not talking about replacing a pre-existing string, we're talking
about adding one, which is a different situation.

I don't see how this makes this a different situation. See for example
http://lists.infradead.org/pipermail/linux-arm-kernel/2013-April/161065.html
and
http://lists.infradead.org/pipermail/linux-arm-kernel/2012-March/087702.html
where Arnd also said using the oldest SoC that has the same IP block as
the compatible string was the right thing to do.

quoted

quoted

... which means there's no problem with using marvell,mvebu-audio as the
compatible string if you're going to use properties to describe what
facilities are available.

I disagree, because how do you know if a future "mvebu" SOC such as
Armada 370, or one that doesn't exist yet, will not have a different
audio IP block?

The Dove already contains _three_ audio blocks, two of which are this
one, and another which is block for driving an AC'97 codec (which doesn't
have a driver.)  That's no problem because you won't call that one an
"audio" block but an AC'97 block.  So...

And? If that's a different IP block, it'll have a different compatible
string, that's it.

That doesn't change my point: using "marvell,mvebu-audio" as the
compatible string is stupid, because you have absolutely no idea what
the future of audio in mvebu SOCs will be. However, you do know,
*today* that Kirkwood and Dove have compatible IP blocks for audio, and
that they were first introduced with Kirkwood.

quoted

It will still be audio, it will still be mvebu, but it
will not be able to use a "marvell,mvebu-audio" driver. Or maybe it can
use the same driver, but with a few variations, so a different
compatible string will be needed to identify the original IP
("marvell,kirkwood-audio", used on Kirkwood/Dove) and slightly newer
versions of the IP ("marvell,some-funky-soc-audio").

I don't think this really applies.

It does. We're exactly in this situation, as I will soon be working on
Armada 370 audio support, and while the IP looks similar, I have
checked all the details to see if it's exactly identical.

And Armada 370 is really a mvebu architecture: it's even supported in
mach-mvebu/, while Kirkwood and Dove are not (yet).

quoted

quoted

In any case "marvell,has-spdif" is too generic - as I've indicated above,
there's versions with spdif out, and other versions with spdif in and
out.

Right, the above was just an example to illustrate that we can have
additional properties to encode the differences between each instance
of the audio devices.

I think this is a mistake too: these properties will just tell us what
may be possible, and the driver will take no real action on them.  I
suppose that a property specifying whether there is a SPDIF output could
be used to control whether the IEC958 channel status controls are
registered.  However...

What's more important is which outputs are actually wired up, and
therefore which bits of this hardware are enabled.  Even then, we
wouldn't want to expose (eg) the IEC958 channel status controls if
the SPDIF output isn't wired.  So all in all, I don't see any point
to a set of properties saying "we have SPDIF" etc.  That information
should come solely from whether the SPDIF output has been "wired up".

Let me put that another way: we _can_ provide those properties to
indicate what facilities the hardware has, we just wouldn't use them
at all - and to provide them seems like over-design to me.

I am not arguing about the properties, as I haven't looked at the
specific problem that needs to be solved. By suggesting properties, I
was merely suggesting one possible solution to the problem that
Sebastian was raising, where the different instances of the IP block
don't have the same capabilities.

What I am however strongly arguing on is the choice of the compatible
string. marvell,mvebu-audio is a wrong choice.

Best regards,

Thomas
-- 
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com