Re: [PATCH v2 1/4] clk: core: introduce clk_hw_set_parent()

[Stephen Boyd]

Quoting Jerome Brunet (2019-08-06 01:28:19)
> On Wed 31 Jul 2019 at 10:40, Neil Armstrong <narmstrong@xxxxxxxxxxxx> wrote:
> 
> > Introduce the clk_hw_set_parent() provider call to change parent of
> > a clock by using the clk_hw pointers.
> >
> > This eases the clock reparenting from clock rate notifiers and
> > implementing DVFS with simpler code avoiding the boilerplates
> > functions as __clk_lookup(clk_hw_get_name()) then clk_set_parent().
> >
> > Signed-off-by: Neil Armstrong <narmstrong@xxxxxxxxxxxx>
> > Acked-by: Martin Blumenstingl <martin.blumenstingl@xxxxxxxxxxxxxx>
> 
> Looks ok to me but we will obviously need Stephen's ack to apply it

Acked-by: Stephen Boyd <sboyd@xxxxxxxxxx>

> 
> > ---
> >  drivers/clk/clk.c            | 6 ++++++
> >  include/linux/clk-provider.h | 1 +
> >  2 files changed, 7 insertions(+)
> >
> > diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
> > index c0990703ce54..c11b1781d24a 100644
> > --- a/drivers/clk/clk.c
> > +++ b/drivers/clk/clk.c
> > @@ -2487,6 +2487,12 @@ static int clk_core_set_parent_nolock(struct clk_core *core,
> >       return ret;
> >  }
> >  
> > +int clk_hw_set_parent(struct clk_hw *hw, struct clk_hw *parent)
> > +{
> > +     return clk_core_set_parent_nolock(hw->core, parent->core);

I wonder if you really want to call all those things in
clk_core_set_parent_nolock(). For example, do you want notifiers to run
again and for rates to be speculated? Maybe all you want to do is
overwrite some value for the clk's parent by calling into the ops for
the clk and generically parse the value that's passed as the "parent"
here.

I ask because it may be good to avoid doing all that work and updating
bookkeeping when we're deep in a notifier. If we can clearly understand
what the driver wants to do from the notifier then it's simpler to
change in the future to use things such as the coordinated clk rate
vaporware.