Quoting Tony Lindgren (2019-09-08 12:42:41) > * Stephen Boyd <sboyd@xxxxxxxxxx> [190907 03:55]: > > Quoting Tony Lindgren (2019-09-05 14:55:32) > > > We currently have a hidden dependency to the device tree node name for > > > the clkctrl clocks. Instead of using standard node name like "clock", we > > > must use "l4-per-clkctrl" naming so the clock driver can find the > > > > The node name is "clk" though. > > Yes for some it's "clk" and for some it's "l4-per-clkctrl". > > In general, the clock node name should be "clock@foo", rather than > "clk@foo", right? I don't think it really matters but sure, clock is nicer because that's a more standard node name than the vowel-less one. > > > > associated clock domain. Further, if "clk" is specified for a clock node > > > name, the driver sets TI_CLK_CLKCTRL_COMPAT flag that uses different > > > logic with earlier naming for the clock node name. > > > > > > If the clock node naming dependency is not understood, the related > > > clockdomain is not found, or a wrong one can get used if a clock manager > > > instance has multiple domains. > > > > > > As each clkctrl instance represents a single clock domain with it's > > > reg property describing the clocks available in that clock domain, > > > we can simply use "reg-names" property for the clock domain. > > > > > > This simplifies things and removes the hidden dependency to the node > > > name. And then later on, we should be able to drop the related code > > > for parsing the node names. > > > > > > Let's also update the binding to use standard "clock" node naming > > > instead of "clk". > > > > > > Cc: devicetree@xxxxxxxxxxxxxxx > > > Cc: Rob Herring <robh+dt@xxxxxxxxxx> > > > Signed-off-by: Tony Lindgren <tony@xxxxxxxxxxx> > > > --- > > > Documentation/devicetree/bindings/clock/ti-clkctrl.txt | 6 +++++- > > > drivers/clk/ti/clkctrl.c | 10 ++++++++-- > > > 2 files changed, 13 insertions(+), 3 deletions(-) > > > > > > diff --git a/Documentation/devicetree/bindings/clock/ti-clkctrl.txt b/Documentation/devicetree/bindings/clock/ti-clkctrl.txt > > > --- a/Documentation/devicetree/bindings/clock/ti-clkctrl.txt > > > +++ b/Documentation/devicetree/bindings/clock/ti-clkctrl.txt > > > @@ -20,15 +20,19 @@ Required properties : > > > - #clock-cells : shall contain 2 with the first entry being the instance > > > offset from the clock domain base and the second being the > > > clock index > > > +- reg : clock registers > > > +- reg-names : clock register names for the clock, should be same as the > > > + domain name > > > > Is this necessary? I'd rather see that the names of the clks don't > > actually matter by means of connecting the clk tree through the "clocks" > > property when the parent clks are provided by external nodes and through > > direct pointers when they're within a controller. If that works then it > > should be possible to ignore this name in general? > > This is not for names of the clocks :) This is to name the clock > provider register range. The name of the register range is the > same as the clockdomain that this range of clocks belongs to. > This property is used by the clock provider on init to initialize the > clock provider, not when a clock is requested. > > In this case a clkctrl clock provider instance has one to some tens > clocks where they all belong to the same domain. If some similar clock > would have multiple domains, then it would simply just have multiple > reg ranges and multiple reg-names properties. > > Or do you have some better ideas on how to name a clock controller > in the device tree? > Why does the name of the clock controller or clkdm_name matter? Using a string from reg-names smells like a workaround to describe some sort of linkage between things that isn't being described in DT today.
