On 2020-02-06 11:46 am, Mark Brown wrote:
On Thu, Feb 06, 2020 at 01:07:12AM +0000, Robin Murphy wrote:
The RK3328 reference design uses an external line driver IC as a buffer on the analog codec output, enabled by the GPIO_MUTE pin, and such a configuration is currently assumed in the codec driver's direct poking of GRF_SOC_CON10 to control the GPIO_MUTE output value. However, some
This makes sense but it is an ABI break so is going to need quirking for existing boards that unfortunately rely on the existing behaviour.
Yeah, that's where it gets tricky - there doesn't seem to be a nice way to differentiate between "no GPIO because old DT" and "no GPIO because the enable is hard-wired/irrelevant and GPIO_MUTE doesn't do what you think it does", and it seems really improper to introduce a DT property for the sole purpose of telling a Linux driver not to assume something it shouldn't really have in the first place.
My opinion fell on the side of a minor ABI break being the lesser of two evils, given that the worst case once people start enabling this codec on Renegade/ROC-CC boards (which I was only anticipating, but have just discovered is happening already[1]) results in unexpectedly stuffing 3.3V into the SD card and SoC I/O domain while both are in 1.8V mode, and that the change would only really affect one other current board (Rock64), where most mainline users are likely to be upgrading their DTB in lock-step with the kernel anyway.
I guess the existing (mis)behaviour could be predicated on an of_machine_is_compatible() check for Rock64 boards - it's ugly, but should do the job if you feel it's more important to be 100% strict about not regressing supported systems for any possible kernel/DTB combination.
Thanks, Robin.
[1] https://github.com/armbian/build/commit/18b24717be9639b65b86db3dbcf2b42fe73c...