Re: [PATCH] clk: clk-gpio: Add dt option to propagate rate change to parent

[Ardelean, Alexandru]

On Thu, 2019-11-07 at 14:53 -0800, Stephen Boyd wrote:
> Quoting Ardelean, Alexandru (2019-11-07 05:25:38)
> > On Wed, 2019-11-06 at 14:43 -0800, Stephen Boyd wrote:
> > > Quoting Alexandru Ardelean (2019-11-06 03:35:51)
> > > > From: Michael Hennerich <michael.hennerich@xxxxxxxxxx>
> > > > 
> > > > For certain setups/boards it's useful to propagate the rate change
> > > > of
> > > > the
> > > > clock up one level to the parent clock.
> > > > 
> > > > This change implements this by defining a `clk-set-rate-parent`
> > > > device-
> > > > tree
> > > > property which sets the `CLK_SET_RATE_PARENT` flag to the clock
> > > > (when
> > > > set).
> > > > 
> > > > Signed-off-by: Michael Hennerich <michael.hennerich@xxxxxxxxxx>
> > > > Signed-off-by: Alexandru Ardelean <alexandru.ardelean@xxxxxxxxxx>
> > > > ---
> > > >  drivers/clk/clk-gpio.c | 8 ++++++--
> > > >  1 file changed, 6 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git a/drivers/clk/clk-gpio.c b/drivers/clk/clk-gpio.c
> > > > index 9d930edd6516..6dfbc4b952fe 100644
> > > > --- a/drivers/clk/clk-gpio.c
> > > > +++ b/drivers/clk/clk-gpio.c
> > > > @@ -241,6 +241,7 @@ static int gpio_clk_driver_probe(struct
> > > > platform_device *pdev)
> > > >         struct device_node *node = pdev->dev.of_node;
> > > >         const char **parent_names, *gpio_name;
> > > >         unsigned int num_parents;
> > > > +       unsigned long clk_flags;
> > > >         struct gpio_desc *gpiod;
> > > >         struct clk *clk;
> > > >         bool is_mux;
> > > > @@ -274,13 +275,16 @@ static int gpio_clk_driver_probe(struct
> > > > platform_device *pdev)
> > > >                 return ret;
> > > >         }
> > > >  
> > > > +       clk_flags = of_property_read_bool(node, "clk-set-rate-
> > > > parent")
> > > > ?
> > > > +                       CLK_SET_RATE_PARENT : 0;
> > > 
> > > Is there a DT binding update somewhere? It looks like a linux-ism
> > > from
> > 
> > Good point. I did not think about the DT, and I guess I didn't search
> > it
> > thoroughly enough. Found DT files now.
> > 
> > > the DT perspective. I wonder if we can somehow figure out that it's
> > > OK
> > > to call clk_set_rate() on the parent here? Or is it safe to assume
> > > that
> > > we can just always call set rate on the parent? I think for a gate
> > > it's
> > > good and we can just do so, but for a mux maybe not. Care to describe
> > > your scenario a little more so we can understand why you want to set
> > > this flag? Is it for a mux or a gate type gpio?
> > > 
> > 
> > For our case we are using it here [with a slight name variation in the
> > prop
> > name]:
> > https://github.com/analogdevicesinc/linux/blob/master/arch/arm/boot/dts/zynq-adrv9361-z7035.dtsi#L43
> > 
> > And on this board:
> > https://wiki.analog.com/resources/eval/user-guides/adrv936x_rfsom/user-guide/introduction
> > 
> > It's for a gate-type GPIO.
> > The clock defined in that DT (ad9361_clkin) is attached to a clock that
> > has
> > a fixed rate (xo_40mhz_fixed_clk), that can become a variable rate
> > [from an
> > external source, when it is provided].
> > 
> > Our understanding, is that a GPIO gate clock should propagate the rate
> > change to the parent clock. The same goes for the GPIO MUX clock.
> > So, the default mode would be to always set CLK_SET_RATE_PARENT.
> > 
> > But, given that there are users of this driver, such a behavior change
> > could break other users, so we are using this DT prop.
> > 
> > There seems to be only one user of gpio-mux-clock in
> > arch/arm64/boot/dts/renesas/ulcb-kf.dtsi
> > 
> > Whereas for gpio-gate-clock, there are multiple users. I can't say
> > whether
> > this change would break anything for them, or it would be fine.
> 
> I think we can just add CLK_SET_RATE_PARENT for all gpio gate clks then
> and you'll be happy? Care to send that patch and Cc Geert and other
> Renesas maintainers? That avoids any DT property and should be fine for
> anyone as far as I can tell. Maybe the inverse will be desired, but not
> wanting rate to propagate is an edge case that we may want to call out
> explicitly in DT at some point in the future, but shouldn't worry about
> until then.
> 

Will resend, thanks.
Geert is CC-ed here as well [unless there is another Geert].

I'll send a V2 in-reply-to this thread.

Thanks
Alex