On 13 May 2015 at 00:31, Stephen Boyd [off-list ref] wrote:

On 04/27, Joachim Eastwood wrote:

quoted

This patch set adds support for the two main clock peripherals in the
LPC18xx/43xx MCU family.

The Clock Generation Unit (CGU) is the base source of all clocks. It has
five external inputs and contains PLLs, dividers and muxes. The outputs
from the CGU are then routed to a Clock Control Unit (CCU) and a few
peripherals directly. There are two CCUs in the MCU. The CCU is a
collection of gates and a few dividers that sits between the CGU and
most of the peripherals.

Which clocks that are available depends on the specific device and it's
peripherals. It's possible in DT to setup the routing between the CGU
and the CCUs.

Is there any reason why this uses CLK_OF_DECLARE over the
platform device model? Typically we have CLK_OF_DECLARE for cases
where a provider needs to be up and running early during boot for
hardware that can't probe defer (i.e. timers, irq controllers).

One of the CCUs provided clock for two of the timers that are used as
clock sources.
Specifically this timer:
http://marc.info/?l=linux-arm-kernel&m=143138193704763&w=2

Since the CGU provides the clock to the CCU which then again provides
the clock to the timer both must use CLK_OF_DECLARE for clk enable/get
rate to work in the timer driver.

Otherwise we should be able to use the standard linux device
model to probe clk providers. It looks like the clocksource is
using something from CCU, but perhaps CGU isn't "special" and
could be using device model.

It's is the CCU that uses the clocks from the CGU. So I think both
drivers needs to use CLK_OF_DECLARE.

regards,
Joachim Eastwood