On Tue, Mar 28, 2017 at 6:24 PM, Mark Brown [off-list ref] wrote:

On Tue, Mar 28, 2017 at 12:53:06PM +0100, Charles Keepax wrote:

quoted

On Tue, Mar 28, 2017 at 01:47:04PM +0300, Daniel Baluta wrote:

quoted

quoted

quoted

quoted

-     codec_clk = devm_clk_get(&codec_dev->dev, NULL);
+     codec_clk = clk_get(&codec_dev->dev, NULL);

quoted

quoted

is different from &codec_dev->dev.

quoted

I get that they are different, I just don't get why changing
from a devm_clk_get to a clk_get is a better fix than changing
&codec->dev to &pdev->dev.

This should be clear from the semantics of clk_get(): you're looking up
the clock in the context of the supplied device and the clock is
attached to the CODEC so you need to look up in the CODEC context.  What
would be even better would be to move the allocation of the clock into
the CODEC driver...

If I read the code correctly, both machine and codec driver are doing:

* codec_clk = clk_get(...)

I guess that the codec driver is the first to be loaded and it will call:

* __clk_create_clk

which will do the allocation. Then when machine driver is loaded, it
will only take a reference to the allocated clock.

I might be very wrong on this. Can you clarify what does moving allocation
of the clock into codec driver implies? Machine driver only needs a reference
to codec_clk to just get the clock's rate.

Daniel.