Re: [PATCH 1/4] clk: sunxi-ng: a10/a20: rewrite init code to a platform driver

[Maxime Ripard]

On Thu, Apr 30, 2020 at 06:21:37AM +0000, Priit Laes wrote:
> On Wed, Apr 29, 2020 at 04:35:10PM +0200, Maxime Ripard wrote:
> > On Mon, Apr 20, 2020 at 08:32:28PM +0000, Priit Laes wrote:
> > > On Mon, Apr 20, 2020 at 02:49:35PM +0200, Maxime Ripard wrote:
> > > > On Sat, Apr 18, 2020 at 01:17:27AM +0300, Priit Laes wrote:
> > > > > In order to register regmap for sun7i CCU, there needs to be
> > > > > a device structure already bound to the CCU device node.
> > > > > 
> > > > > Convert the sun4i/sun7i CCU setup to platform driver to use
> > > > > it later as platform device.
> > > > > 
> > > > > Signed-off-by: Priit Laes <plaes@xxxxxxxxx>
> > > > 
> > > > You can't relly do that though. We have timers that need those clocks before the
> > > > device model is initialized.
> > > 
> > > Ok, I'm somewhat lost now... are these the affected timers on sun7i following:
> > > - allwinner,sun4i-a10-timer (timer@1c20c00)
> > > - allwinner,sun7i-a20-hstimer (hstimer@1c60000)
> > 
> > Yep
> > 
> > > Any ideas on what approach I could actually use?
> > 
> > I guess you could keep the CLK_OF_DECLARE registration, and then have a
> > platform_driver probe and register the regmap?
> > 
> 
> Thanks this did the trick.
> 
> > > Also, similar timer dependency would affect then sun6i-a31 and sun9i-a80
> > > platforms too...
> 
> I didn't check this before, but sun9i-a80 CCU is initialized currently via
> platform device. Should it be converted first to clock driver (CLK_OF_DECLARE)?

I guess we could just remove the timer node on the A80. It has never been tested
and never worked if the clock driver is probed through a platform device.

Maxime

Attachment: signature.asc
Description: PGP signature