On Tue, Jan 08, 2013 at 11:52:33PM +0100, Lucas Stach wrote:

Am Dienstag, den 08.01.2013, 15:17 -0700 schrieb Stephen Warren:

...

Hmmm. I thought that could be auto-instantiated based on probing the AC'97 bus for whatever was there?

AC97 is in principle probeable by reading the first regs of the codec, but all the infrastructure to do proper hotplugging on the AC97 bus is missing in the kernel. As I didn't want to be the one to add all this infrastructure to a dying audio platform, I opted to do the same thing as all the other platform using AC97 and instantiate the codec from the machine driver.

That's about the size of it, yes. We should write the device tree as though we'd got better infrastructure here since this isn't something that ought to be required but right now this is how you do AC'97 in ASoC. It'd be really nice if someone had the time to work on the AC'97 infrastructure here but I'd not hold my breath given that it's rarely used in new designs.