Re: [PATCH net-next v3 1/2] net: phy: replace PHY_HAS_INTERRUPT with a check for config_intr and ack_interrupt

[Heiner Kallweit]

On 09.11.2018 21:33, Florian Fainelli wrote:
> On 11/9/18 12:22 PM, Heiner Kallweit wrote:
>> On 09.11.2018 21:13, Andrew Lunn wrote:
>>> Hi Heiner
>>>
>>>> +static bool phy_drv_supports_irq(struct phy_driver *phydrv)
>>>> +{
>>>> +	return phydrv->config_intr || phydrv->ack_interrupt;
>>>> +}
>>>
>>> Should this be && not || ? I thought both needed to be provided for
>>> interrupts to work. 
>>>
>>> 	   Andrew
>>>
>> I've seen at least one driver which configures interrupts in
>> config_init and doesn't define a config_intr callback
>> (ack_interrupt callback is there)
> 
> That driver should probably be fixed, while it most likely does not make
> any significant difference during probe/connect, since config_init() and
> config_intr() are virtually happening at the same time, this is not
> necessarily true when disconnecting from the PHY where we really want
> config_intr() to effectively disable the interrupts and not leaving
> something enabled that would now become unmaskable, because no more
> driver attached.
> 
Found the driver: It's the IP101A/G in icplus.c
It should be easy to fix the behavior and move the interrupt config
to a config_intr callback. But the last real changes to the driver
have been done 6 years ago, so I'm not sure there's anybody out
there who can test.

>> Intention of this check is not to ensure that the driver defines
>> everything to make interrupts work. All it states:
>> If at least one of the irq-related callbacks is defined, then
>> we interpret this as indicator that the PHY supports interrupts.
> 
> I agree with Andrew here, that this should be an AND here, both
> callbacks must be implemented for interrupts to work correctly.
>