Re: [PATCH 1/2] i2c: designware: register clkdev during acpi device configuration

[Mika Westerberg <mika.westerberg@linux.intel.com>]

On Wed, Oct 21, 2015 at 05:50:02PM +0800, Ken Xue wrote:
> On Wed, 2015-10-21 at 12:49 +0300, Mika Westerberg wrote:
> > On Wed, Oct 21, 2015 at 05:37:53PM +0800, Ken Xue wrote:
> > > On Wed, 2015-10-21 at 12:25 +0300, Mika Westerberg wrote:
> > > > On Wed, Oct 21, 2015 at 04:42:23PM +0800, Ken Xue wrote:
> > > > > On Wed, 2015-10-21 at 10:28 +0300, Mika Westerberg wrote:
> > > > > > On Wed, Oct 21, 2015 at 09:11:33AM +0800, Ken Xue wrote:
> > > > > > > On Tue, 2015-10-20 at 14:17 +0300, Mika Westerberg wrote:
> > > > > > > > On Tue, Oct 20, 2015 at 02:38:01PM +0800, Ken Xue wrote:
> > > > > > > > > DW I2C driver tries to register a clk from id->driver_data as an
> > > > > > > > > alternative way besides intel lpss. But code doesn't register the
> > > > > > > > > clk to clkdev. So, devm_clk_get will fail during probe.
> > > > > > > > > 
> > > > > > > > > The patch can fix this issue.
> > > > > > > > 
> > > > > > > > Since you now have drivers/acpi/acpi_apd.c for AMD ACPI stuff, can you
> > > > > > > > create the clock there just like we do for Intel stuff?
> > > > > > > Sure. APD already creates the clock for AMD0010 as you expected. And the
> > > > > > > next patch([PATCH 2/2] i2c: designware: remove freq definition for
> > > > > > > "AMD0010" in acpi_device_id) is dropping the old way for getting freq.
> > > > > > 
> > > > > > So this patch is not necessary, right?
> > > > > Even though there is no use case that getting freq from id->driver_data,
> > > > > But if we want to keep this design, then we should use current patch for
> > > > > fixing the potential issue. So, the patch is nice to have.
> > > > 
> > > > What potential issue?
> > > devm_clk_get will fail during probe for AMD0010 without current patch.
> > 
> > How can it fail if you provide the very clock from drivers/acpi/acpi_apd.c?
> After apd was accept in kernel V4.1, there is no issue. But between 3.18
> and V4.1, there will be a problem.

Ah, now I got it.

You are saying that the original commit a445900c906092 ("i2c:
designware: Add support for AMD I2C controller") actually never worked
because it failed to register the clock with clkdev? In that case it is
not even a regression ;-) Oh my...

In that case I don't think we need to fix that for 3.18+ because it
never worked in the first place.