[PATCH v3 1/4] ARM: davinci: da8xx-dt: Add ti-aemif lookup for clock matching

[kbeldan@baylibre.com (Karl Beldan)]

On Thu, Aug 18, 2016 at 10:04:37AM +0100, Russell King - ARM Linux wrote:
> On Thu, Aug 18, 2016 at 07:33:19AM +0000, Karl Beldan wrote:
> > Checking clk_get:
> > 
> > struct clk *clk_get(struct device *dev, const char *con_id)
> > {       
> > 	[...]
> > 	if (dev) {
> > 		struct clk *__of_clk_get_by_name(struct device_node *np,
> > 						 const char *dev_id,
> > 						 const char *name)
> > 		clk = __of_clk_get_by_name(dev->of_node, dev_id, con_id);
> > 		[...]
> > 	}
> > 	return clk_get_sys(dev_id, con_id);
> > }
> > 
> > In DT case the con_id _is_ the clock name, so the assertion "clk_get()
> > does not look up a clock by name" would be false ?
> 
> Wrong.
> 
> 	clocks = <&provider 0>, <&provider 1>;
> 	clock-names = "fck", "ick";
> 
> 	fck = clk_get(dev, "fck");
> 
> 	ick = clk_get(dev, "ick");
> 
> Works just the same.  The whole point of the string is to identify an
> _individual_ _input_ _to_ _the_ _device_ and not a global clock name.
> 

I hope so.
You are matching con_id with _a_ clock name, that's my point.

Maybe my writing 'the clock name' instead of 'a clock name' made you
deduce I was asserting in some way that the clock tree was addressed
by some kind of a unique root namespace via the clock-names ?? It was
not the case ('the' was contextual if that means anything).

> Think what happens if you specify a clock name.  Where does that name
> come from?  What if the clock is named differently on another SoC?
> With that approach, we need to define a new set of APIs to allow a
> device to determine the global clock name for the clock that it's
> interested in - which is completely absurd when we've already got
> that within clk_get().
> 
> The whole point of the clk API is to abstract those differences.  That's
> why it takes a _connection_ _id_ and not a clock name.
> 
> I really can't understand why people keep getting this wrong in their
> heads.  It makes no sense to me for this to take a global clock name.
> 

Nope, I am fine with that, it is a convenient scheme.

> > Also, numerous commits refer to *clk_get(*, NULL) as getting an unnamed
> > clock, although it only really is accurate to the point in DT cases.
> 
> Table-driven clkdev copes fine with that, and will try to locate an
> appropriate table entry with a NULL connection ID, not matching any
> non-NULL connection ID entry.
> 

Yes it does, my change relies on that and I was happy to be able to so.

It is good to see the maintainers like Sekhar and you seizing the
opportunity to throw light on such things.
 
Karl

> In the DT case, it's always the first specified clock.
> 
> -- 
> RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
> FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
> according to speedtest.net.