On Fri, Feb 24, 2012 at 08:57:53AM +0100, Andrew Lunn wrote:

Hi Jason

Great to see this work going forward. I hope we can cooperating in the
future. I have an out of tree kirkwood board i would like to supported
via DT, and i also have a QNAP T119+ i can use for testing.

Awesome, I don't have one of those.

I have some patches under development which will help. I've been
porting the different Orion machines to use Mike Turquette generic CLK
and clkdev...

It appears that changing core platform code across an entire
architecture is quite difficult.  ;-)  I hope it gets merged, but I'm
not going to hold my breath.

...so that the device drivers can determine for themselves
what the clock is. tclk is then removed from the platform data
structures, making DT a little bit simpler. This work should also
solve the kirkwood clock gating which is going to be an issue for
DT. For example, your board file is probably currently calling
kirkwood_spi_init(), in order to use SPI. This does:

kirkwood_clk_ctrl |= CGC_RUNIT;

and so ensures that the RUNIT clock does not get turned off in a
lateinit call. Once you have DT, this flag is not going to be set, and
so you have the danger of the clock to the SPI core gets turned
off. The clk framework should solve this, since the SPI device driver
will claim the clock and ensure it keeps running.

Why not convert kirkwood_clock_gate() (mach-kirkwood/common.c:480) to
make use of of_device_is_available() ?  That would eliminate one global
variable.  That wouldn't even require a change to the devicetrees, since
it returns 1 if 'status' doesn't exist (or is set to "okay" or "ok").

I don't even see where CGC_RUNIT is used in kirkwood_clock_gate(),
except at line 528.  Maybe I'm missing something, a lot of this is new
to me.


thx,

Jason.