Re: [PATCH 3/9] clk: sunxi: Add prcm mod0 clock driver

[Maxime Ripard <maxime.ripard-wi1+55ScJUtKEb57/3fJTNBPR1lH4CV8@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 2659 bytes --]

Hi,

On Thu, Nov 27, 2014 at 11:10:51AM +0100, Hans de Goede wrote:
> Hi,
> 
> On 11/27/2014 10:28 AM, Chen-Yu Tsai wrote:
> >Hi,
> >
> >On Thu, Nov 27, 2014 at 4:41 PM, Hans de Goede <hdegoede-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote:
> 
> <snip>
> 
> >>I notice that you've not responded to my proposal to simple make the clock
> >>node a child node of the clocks node in the dt, that should work nicely, and
> >>avoid the need for any kernel level changes to support it, I'm beginning to
> >>think that that is probably the best solution.
> >
> >Would that not cause an overlap of the io regions, and cause one of them
> >to fail? AFAIK the OF subsystem doesn't like overlapping resources.
> 
> No the overlap check is done by the platform dev resource code, and of_clk_declare
> does not use that. So the overlap would be there, but not an issue (in theory
> I did not test this).

An overlap is always an issue.

> Thinking more about this, I believe that using the MFD framework for the prcm seems
> like a mistake to me. It gains us nothing, since we have no irq to de-multiplex or
> some such. We're not using MFD for the CMU, why use it for CMU2 (which the prcm
> effectively is) ?

Because the PRCM is much more than that. It also handles the power
domains for example.

And also because the 1 node per clock is no longer the current trend
and that Mike discourages to use that model nowadays.

> So I think it would be best to remove the prcm node from the dt, and simply put the
> different blocks inside it directly under the soc node, this will work fine with
> current kernels, since as said we do not use any MFD features, we use it to
> create platform devs and assign resources, something which will happen automatically
> if we put the nodes directly under the soc node, since then simple-bus will do the
> work for us.
> 
> And then in a release or 2 we can remove the mfd prcm driver from the kernel, and we
> have the same functionality we have now with less code.
> 
> We could then also chose to move the existing prcm clock nodes to of_clk_declare
> (this will work once they are nodes directly under soc with a proper reg property).
> and the ir-clk can use allwinner,sun4i-a10-mod0-clk compatible and can live under
> either clocks or soc, depending on what we want to do with the other prcm clocks.

Have you considered the SMP code in that smooth plan?

The one that is in neither a platform driver, nor a clock, nor does it
have a compatible of its own, but still needs to access some of the
PRCM registers?

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com