Re: [RFC PATCH v2 1/5] clk: meson: axg: move reset controller's code to separate module

[Stephen Boyd]

Quoting Philipp Zabel (2024-04-08 01:21:47)
> On So, 2024-04-07 at 19:39 -0700, Stephen Boyd wrote:
> > Quoting Jerome Brunet (2024-04-02 07:52:38)
> > > 
> > > On Thu 28 Mar 2024 at 04:08, Jan Dakinevich <jan.dakinevich@xxxxxxxxxxxxxxxxx> wrote:
> > > 
> > > > This code will by reused by A1 SoC.
> > > 
> > > Could expand a bit please ?
> > > 
> > > > 
> > > > Signed-off-by: Jan Dakinevich <jan.dakinevich@xxxxxxxxxxxxxxxxx>
> > > 
> > > In general, I like the idea.
> > > 
> > > We do have a couple a reset registers lost in middle of clocks and this
> > > change makes it possible to re-use the code instead duplicating it.
> > > 
> > > The exported function would be used by audio clock controllers, but the
> > > module created would be purely about reset.
> > > 
> > > One may wonder how it ended up in the clock tree, especially since the
> > > kernel as a reset tree too.
> > > 
> > > I'm not sure if this should move to the reset framework or if it would
> > > be an unnecessary churn. Stephen, Philipp, do you have an opinion on
> > > this ?
> > > 
> > 
> > I'd prefer it be made into an auxiliary device and the driver put in
> > drivers/reset/ so we can keep reset code in the reset directory.
> 
> Seconded, the clk-mpfs/reset-mpfs and clk-starfive-jh7110-sys/reset-
> starfive-jh7110 drivers are examples of this.
> 
> > The auxiliary device creation function can also be in the
> > drivers/reset/ directory so that the clk driver calls some function
> > to create and register the device.
> 
> I'm undecided about this, do you think mpfs_reset_controller_register()
> and jh7110_reset_controller_register() should rather live with the
> reset aux drivers in drivers/reset/ ?

Yes, and also mpfs_reset_read() and friends. We should pass the base
iomem pointer and parent device to mpfs_reset_adev_alloc() instead and
then move all that code into drivers/reset with some header file
exported function to call. That way the clk driver hands over the data
without having to implement half the implementation.