lkml.org 
[lkml]   [2013]   [Feb]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [PATCH] clk: add si5351 i2c common clock driver
Date
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@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.

> - 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?

> * 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?

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?

Thanks,
Mike


\
 
 \ /
  Last update: 2013-02-11 07:21    [W:0.289 / U:0.156 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site