[PATCH 01/40] clkdev: add clkname to struct clk_lookup

[Richard Zhao]

On Tue, Apr 10, 2012 at 06:11:42PM +0200, Sascha Hauer wrote:
> On Tue, Apr 10, 2012 at 03:30:55PM +0100, Russell King - ARM Linux wrote:
> > On Tue, Apr 10, 2012 at 03:45:14PM +0200, Sascha Hauer wrote:
> > > With the generic clock framework we do not necessarily have
> > > a pointer to the struct clk we want to register a lookup
> > > for, so add a const char *clkname field to struct clk_lookup
> > > so that a lookup can be registered with a clock name.
> > 
> > All this silly names all over the place is getting to be utterly
> > rediculous. Why do we need to name clocks and look them up by name
> > when we have a perfectly good way (clkdev) to do this already?  Yes,
> > it takes a struct clk pointer but that's exactly because that's what
> > it has to return.
> > 
> > Why do we want to have another idiotic string to look up a clock which
> > is named using an idiotic scheme to find out its pointer to only then
> > register that with clkdev?
> > 
> > I think this clk stuff has gone totally insane wrt names recently.
> 
> clk names and clk lookups are two different things. The former is a
> unique identifier for a clock whereas the latter associates a clock
> to a device. Multiple lookups can point to the same clock, so they can't
> be used to identify a particular clock. I know I'm not telling anything
> new to you here, so I think your question goes down to why we need a
> unique identifier string for clocks.
> 
> For me the big advantage for clocks having unique names is that I do not
> have to have a struct clk * to pass the (possible) parents of a clock to
> the clock framework during registration. Also it makes us independent of
> the registration order of the clock tree (no need to register the
> parents before the children). The debugfs representation of the clock
> tree also makes use of the names.
I don't like string loopup either. And after DT binding, we can use
phandler to refer clk.

BRs
Richard