[PATCH] SPI: bcm2835: move to the transfer_one driver model

[Stephen Warren]

On 03/26/2015 04:08 AM, Martin Sperl wrote:
...
> ---
> 
> Note that there is quite a bit of complexity involved to make the native 
> CS work correctly.
> Also a few future optimizations in the pipeline will only work reliably 
> with gpio CS.

Can you expand on that a bit more?

Are you planning on implementing code in the driver so it always uses
GPIO CS even when GPIOs aren't specified in the DT, or disabling those
optimizations when native CS is in use?

> So the question is if we should depreciate native chip-selects for this
> driver with one of those future improvements listed below.

Only if you can make the driver transparently use GPIO CS mode even when
no GPIOs are specified in the DT. DT is an ABI, and old DTs need to
continue to work on newer kernels.

I haven't had a chance to look at the code in this patch yet.

> As for testing: I have also tried to test with mmc_spi, but I have not
> been able to make that driver work reliably in any recent kernel
> versions.
> Most of the time I see timeouts - and with lots of different SD-cards...
> 
> IIRC the last time I tested it successfully was with 3.12.

It'd be great if you could use "git bisect" to track down the change
that broke this.