common clock framwork: clk_set_rate issue

[Chao Xie]

On Tue, Dec 18, 2012 at 4:19 AM, Sascha Hauer <s.hauer at pengutronix.de> wrote:
> On Thu, Dec 06, 2012 at 10:52:03AM +0800, Chao Xie wrote:
>> hi
>> When develop the clk drivers for SOCs based on common clock framework.
>> I met a issue.
>> For example there is a uart device, it's function clock comes from a
>> divider, and the divider's parent is a mux. It means
>>
>> MUX --> DIV --> UART
>>
>> As we know that UART can work at low baudrate for a terminal, while it
>> can also connect to GPS module which needs a high rate. So the MUX
>> will provide two clock source, a low clock rate and high clock rate.
>>
>> The MUX clk driver clk-mux.c does not implement a ->round_rate callbacks.
>> It means that when uart driver is used for a GPS and it want to change
>> it clock, the driver will call clk_set_rate(); clk_set_rate will loop
>> upward to DIV, and DIV will try to set its divider, and it need loop
>> upward to MUX.
>> In fact the current clk drivers have some issue.
>> MUX clk driver should provide the round_rate callback, it then can
>> provide a new_rate. It means that in clk_calc_subtree MUX can switch
>> the clock source.
>
> It's not that simple. The input clocks to a mux may not only differ in
> their rate but can also have other different properties, like for
> example one input may be always present whereas another input only runs
> when the CPU is in run mode.
>
> It may be a possibility to add a flag to the mux to explicitely
> allow reparenting on a rate change.
>
> Sascha
>
> --
> Pengutronix e.K.                           |                             |
> Industrial Linux Solutions                 | http://www.pengutronix.de/  |
> Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
> Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-555

There is already a flag to do it.
CLK_SET_RATE_PARENT
if the mux does not want to changes the input for clk_set_rate called
by its child, it can clear this flag.
The question is whether we need add the round_rate/recalc_rate for MUX
type of clock? Is there any special issue about it that why current
MUX implementation does not have these callbacks?