> > > While the parent is indeed selected at boot time only, and only one parent
> > > is thus needed, parent selection could be performed by a DIP switch
> > > connected to MD_CLK on the board for instance. In that case both parents
> > > should be available in DT, as selection will be done by the kernel at
> > > boot time, not at DT compile time.
> > 
> > OK, I understand the case. I still wonder about specifying two parents,
> > though. If a board uses USB_X1, it then has to spefify a dummy EXTAL
> > clock (or an empty one), just because USB_X1 is enumerated as second
> > entry?
> 
> That's a good question. Mike, would it be possible to support "holes" in the 
> DT clocks lists, like the GPIO DT bindings do ?

I am currently playing with the status property for clocks, so we can
"disable" clocks. Opinions?

From: Wolfram Sang <wsa@sang-engineering.com>
Subject: [PATCH] clk: allow clocks to use the 'status' property

Similar to platform devices, this allows us to set up the clock
hierarchy in dtsi files and enable the used clocks in the board file
later.

Signed-off-by: Wolfram Sang <wsa@sang-engineering.com>
---
 drivers/clk/clk.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
index 5517944..0045f55 100644
--- a/drivers/clk/clk.c
+++ b/drivers/clk/clk.c
@@ -2542,6 +2542,9 @@ void __init of_clk_init(const struct of_device_id *matches)
 		matches = &__clk_of_table;
 
 	for_each_matching_node_and_match(np, matches, &match) {
+		if (!of_device_is_available(np))
+			continue;
+
 		of_clk_init_cb_t clk_init_cb = match->data;
 		clk_init_cb(np);
 	}