[PATCH] clk: add si5351 i2c common clock driver

[Sebastian Hesselbarth]

On 02/11/2013 06:46 AM, Mike Turquette wrote:
> Quoting Sebastian Hesselbarth (2013-02-09 04:59:32)
>> This patch adds a common clock driver for Silicon Labs Si5351a/b/c
>> i2c programmable clock generators. Currently, the driver supports
>> DT kernels only and VXCO feature of si5351b is not implemented. DT
>> bindings selectively allow to overwrite stored Si5351 configuration
>> which is very helpful for clock generators with empty eeprom
>> configuration. Corresponding device tree binding documentation is
>> also added.
>>
>> Signed-off-by: Sebastian Hesselbarth<sebastian.hesselbarth at gmail.com>
>> ---
>> Notes:
>> - During development I used a debugfs clock consumer that I can also
>>    post if there is interest in it.
>
> Please do.  I have a set of patches that implement a fake clock subtree
> for testing the core framework.  I've been thinking of pushing this to
> the list once it is more presentable and your work might fit into that
> nicely.

Mike,

then I will clean the debugfs driver and post it together with this
patch for 3.9-rc1 as an individual patch.

>> - With current (3.8-rc6) common clock framework there is two (minor)
>>    issues:
>>    * although clocks are registered with devm_clk_register they are not
>>      removed from the clock tree on unloading. That makes reloading of
>>      clk-si5351 as module impossible.
>
> This is a known issue.  clk_unregister is a NOP and defining it has
> always been deferred until the day that someone needed it.  Care to
> take a crack at it?

Ok. I can have a look at it and propose a patch but that will take a
while as other stuff came in between. But IMHO, preparing/enabling
clocks by clock consumers should increase reference count so referenced
modules cannot be unloaded.. but that I have never had a look at, yet ;)

>>    * potentially there could be more than one different external si5351
>>      generators but clocks are registered with names that do not refer
>>      to e.g. the device name. Maybe common clock framework should
>>      prepend the device name for each registered clock, i.e. 0-0060.clk0.
>>      That would also avoid name collisions with same clock names from
>>      different drivers (clk0 is likely to be used by others ;))
>
> More unfinished work, just like clk_unregister above.  I'm sure you are
> aware that clk_register takes struct device *dev as input, but does
> nothing with it.  It wouldn't take much to concatenate the device name
> and clock name if dev is present.  However a complication here is that
> the registration code takes a parent string name to match parents up for
> discrete subtrees; how could statically defined data know about the
> device name ahead of time?

I see. Wrt the above comment about spare time, would prepending DT
clocks be sufficient? Or/And use a fallback mechanism that first tries
a full match, full match with own device name, and relaxed match for
clock name as it is now?

> The above design decision took place before the big DT push we have
> today and was short-sighted.  It would be better to change the framework
> to rely less on string name lookups and DT is one way out of that.
>
> 3.8-rc7 is already out and I don't plan to take anything that hasn't
> already been submitted for 3.9 now.  Can you resubmit this after 3.9-rc1
> comes out?

Sure, but I'll be not available next 2 weeks or so. If 3.8 falls
within that time, I will re-post it later. It is ok for me, if it has
to go in after 3.9 also.

Sebastian