On Mon, Sep 02, 2013 at 03:38:18PM +0000, Opensource [Adam Thomson] wrote:
On Mon, Sep 02, 2013 at 11:39, Mark Brown wrote:
Please fix your mailer to wrap within 80 columns, it makes your mails very hard to read if you don't do this.
Yeah, sorry for that. Having to use Outlook and of course it doesn't seem to have that feature (at least not that I could find), so trying to do it manually.
It's always been in there, can't remember where exactly and I don't have access to Outlook any more.
They are not separate, they are soldered to the board as part of the same package - quite a few other devices use a similar scheme and are also handled in this fashion (the TI TWL devices are one example).
The difference here is that our combined devices can also be separate chips, not just one HW package containing logically separate devices. I don't believe that's the case with say the TI devices.
That doesn't seem like a unique feature.
This is roughly what ends up happening, you do need to instantiate another I2C client no matter what. The important thing here is that the CODEC does not need to be separately registered by the user, if it really is only the I2C client that needs creating that's probably OK so long as the user doesn't need to worry about that implementation detail.
The only thing that would need populating is some small platform data for the codec (MIC bias voltages, and such). You'd still have to do this, combined or separate, as this is platform specific. Other than this, the I2C client initialisation for the codec is simple, which is a reason why I don't think the PMIC needs to initialise it, and you can just as simply do it from machine code.
It's the bit where the board has to register the CODEC separately at all that's the thing. Think about it from the point of view of people writing and reviewing the machine bindings - they end up with this odd chip that appears twice with two names and registration schemas.
The reasoning is simply that if the chip design solders a single device to the board then the software system integration should register a single device with the system.
Ok, but what about the scenario where the devices start life as separate chips and are then later also packaged together as one chip but still with no internal connection, like DA9055. The drivers were already written and accepted as separate entities in the kernel, without chained initialisation. What would be the approach there? To me, logically it makes sense to leave them separate.
It doesn't seem to make much difference what order the drivers are added in here? You're going to need to add a new device IDs for the SIP anyway since it'd presumably be badged as something new - if it wasn't then that's a bit different.