[PATCH v6 1/9] drivers: phy: add generic PHY framework

Sylwester Nawrocki s.nawrocki at samsung.com
Tue Jun 4 23:43:18 EST 2013


Hi,

On 06/04/2013 02:26 PM, Kishon Vijay Abraham I wrote:
>>> +static inline int phy_init(struct phy *phy)
>>> +{
>>> +	pm_runtime_get_sync(&phy->dev);
>>
>> Hmm, no need to check return value here ? Also it looks a bit unexpected to
> 
> I purposely dint check the return values in order to support platforms 
> that don’t enable pm_runtime.

Then I guess this should be called conditionally and any errors returned
if runtime PM is enabled ? Not sure if pm_runtime_enabled() would be
helpful such situation.

>> possibly have runtime_resume callback of a PHY device called before ops->init()
>> call ? It seems a bit unclear what the purpose of init() callback is.
> 
> Not really. Anything that is used to initialize the PHY (internal 
> configuration) can go in phy_init. Usually in runtime_resume callback, 
> optional functional clocks are enabled and also in some cases context 
> restore is done. So it really makes sense to enable clocks/module 
> (pm_runtime_get_sync) before doing a PHY configuration (phy_init).

OK, that makes sense. All PHY device resources must be prepared anyway
before a PHY object is registered with the PHY core.

>>> +	if (phy->ops->init)
>>> +		return phy->ops->init(phy);
>>> +
>>> +	return -EINVAL;
>>> +}
>>> +
>>> +static inline int phy_exit(struct phy *phy)
>>> +{
>>> +	int ret = -EINVAL;
>>> +
>>> +	if (phy->ops->exit)
>>> +		ret = phy->ops->exit(phy);
>>> +
>>> +	pm_runtime_put_sync(&phy->dev);
>>> +
>>> +	return ret;
>>> +}
>>
>> Do phy_init/phy_exit need to be mandatory ? What if there is really
> 
> No. phy_init/phy_exit is not mandatory at all.
>> nothing to do in those callbacks ? Perhaps -ENOIOCTLCMD should be
>> returned if a callback is not implemented, so PHY users can recognize
>> such situation and proceed ?
> 
> So currently these APIs return -EINVAL if these callbacks are not 
> populated which is good enough IMHO.

But -EINVAL could be well returned from the callback function. Perhaps
ENOTSUPP could be used instead ?


Thanks,
Sylwester


More information about the devicetree-discuss mailing list