* Michael Turquette [off-list ref] [160217 12:53]:

Quoting Tony Lindgren (2016-02-17 09:39:49)

quoted

* Michael Turquette [off-list ref] [160216 17:32]:

quoted

We have a shiny new series that provides a standard way to do this:

http://lkml.kernel.org/r/[off-list ref]

OK nice, so tagging the MPU and DDR clocks with CLK_IS_CRITICAL or
"clock-critical" should allow removing this code. I think in this
case I still need to set CLK_IS_CRITICAL as the clock is output 1
of the dts defined clock and does not have a separate dts node.

In fact I hate clock-critical as a DT property and it's only there to
support legacy DT bindings that store all clk data in DT and have zero
clk data in C. So please use the CLK_IS_CRITICAL or CLK_ENABLE_HAND_OFF
flags in C.

Do you think you will ever have a driver that wants to gate these
clocks? Probably not for MPU/DDR, but the HAND_OFF flag is a better fit
if you do. It's the same as CRITICAL but transfers the prepare/enable
reference counting to the consumer driver after that driver calls
clk_get() and clk_prepare_enable() on the HAND_OFF clk.

OK I'll use CLK_ENABLE_HAND_OFF then. Who knows maybe somebody
somewhere has code running on one of the coprocessors in SRAM that
actually allows gating these :)

quoted

I can update when those patches hit Linux next, or I can do a
follow-up patch later on if we want to avoid the dependency
here. Which do you prefer?

Well, testing is always welcome :-)

If you rebase onto those patches then I'll add your patches to that
series for testing and merge it that way.

OK will do and repost after some testing.

Regards,

Tony