On 03/21/2013 12:46 PM, Martin Fuzzey wrote:
On 21/03/13 17:11, Timur Tabi wrote:
On Thu, Mar 21, 2013 at 3:39 AM, Martin Fuzzey mfuzzey@parkeon.com wrote:
With this patch and that DT example the frequency of clock 162 will be _set_ to 20MHz
Does that mean that it ignores the ' clocks' parameter?
No, in this case the clocks parameter must contain the phandle of a clock whose rate is configurable.
If clock-frequency is omitted the binding is still correct (hence the optional) but the frequency of clock 162 would not be modified.
In the documentation I wrote "If both a clock and clock-frequency are provided the clock's rate will be set. " maybe this is not clear enough? It's not clear what it will be set *to*.
It will be set to the frequency specified by the clock-frquency attribute.
Would "If both a clock and clock-frequency are provided clock must support the set_rate operation and its frequency will be set to the value specified by clock-frequency" be better?
Yes, that's much better.
The possible configurations and their use cases are:
- only 'clocks' specified
clock points to a clock specified in the DT which already has an appropriate frequency and is configured by some other means external to the sgtl5000 driver (bootloader, board setup code, or just a fixed rate clock)
- only 'clock-frequency' specified
The chip is assumed to be clocked by a signal having the given frequency, which may even be generated by a clock unknown to linux. This could actually be represented as a special case of 1) by defining a fixed-rate clock in the DT.
- Both 'clocks' and 'clock-frequency' specified
The chip is assumed to be clocked by a rate programmable clock defined in the clock tree. clk_set_rate() will be called for this clock to set its rate to that specified by clock-frequency
This should all be in the binding document.
Cases 1 and 2 exist in the current code, this patch adds support for case 3.
Prior to commit 81e8e4926167ab32593bbb915b45a42024ca1020 "ASoC: fsl: add sgtl5000 clock support for imx-sgtl5000" only case 2 was supported This explains why 2 is not implemented as a special case of 1) plain 'clock-frequency' was supported before the driver learnt how to get a clock from the DT
And this should be in the patch description.