On Wed, Dec 06, 2017 at 03:03:18PM +0900, Katsuhiro Suzuki wrote:

...

I'd expect this code to be structured more like a library - have a driver that handles the specific IPs then have it call into a shared block of code that does the generic bits. Though in this case the device specific bit looks like a couple of tiny data tables so I'm not sure it's worth making it conditional or separate at all.

Sorry... I agree your opinion, but I can't imagine the detail.

I think my driver has structure as follows (ex. startup): DAI: uniphier_aio_startup()@aio-core.c Lib: uniphier_aio_init()@aio-regctrl.c SoC specific: uniphier_aio_ld11_spec@aio-ld11.c

Am I wrong? Would you mean split the functions in aio-regctl.[ch] to other kernel module? I wonder if you could tell me the example from existing drivers. I'll try to fix my driver like as it.

One example is how all the drivers that use the generic dmaengine code instantiate their DMA drivers, or how all the drivers for CODECs that have both I2C and SPIi control interfaces instantiate - given that the device specific code here seems to be mostly data tables that's probably the closest thing.

...

At least. I do think we need to get to the bottom of how flexible the hardware is first though.

Yes, indeed. This hardware is more flexible and complex, but now I (and our company) don't use it. Of course, I don't want to hide some features of this hardware from ALSA people. I should try to upstream all features in the future, I think.

My main concern here is to make sure that when you decide you need to use the more complex hardware that this can be done without too much pain to existing machines (and that they can benefit from as much of the enhanced functionality as is possible).