Re: Problem since commit c73bbaa4ec3e [rc-core: don't lock device at rc_register_device()]

[Heiner Kallweit <hkallweit1@gmail.com>]

Am 27.02.2016 um 20:50 schrieb Mauro Carvalho Chehab:
> Em Sat, 27 Feb 2016 19:39:16 +0100
> Heiner Kallweit <hkallweit1@gmail.com> escreveu:
> 
>> Am 27.02.2016 um 19:05 schrieb Mauro Carvalho Chehab:
>>> Em Sat, 27 Feb 2016 17:10:41 +0100
>>> Heiner Kallweit <hkallweit1@gmail.com> escreveu:
>>>   
>>>> Am 27.02.2016 um 13:14 schrieb Heiner Kallweit:  
>>>>> Since this commit I see the following error when the Nuvoton RC driver is loaded:
>>>>>
>>>>> input: failed to attach handler kbd to device input3, error: -22
>>>>>
>>>>> Error 22 (EINVAL) comes from the new check in rc_open().
>>>>>     
>>>>
>>>> Complete call chain seems to be:
>>>>   rc_register_device
>>>>   input_register_device
>>>>   input_attach_handler
>>>>   kbd_connect
>>>>   input_open_device
>>>>   ir_open
>>>>   rc_open
>>>>
>>>> rc_register_device calls input_register_device before dev->initialized = true,
>>>> therefore the new check in rc_open fails. At a first glance I'd say that we have
>>>> to remove this check from rc_open.  
>>>
>>> Hmm... maybe we could, instead, do:
>>>
>>> 	if (!rdev->initialized) {
>>> 		rval = -ERESTARTSYS;
>>> 		goto unlock;
>>> 	}
>>>   
>> Looking at the source code of the functions in the call chain I see no special
>> handling of ERESTARTSYS. It's treated like any other error, therefore I don't
>> think this helps.
> 
> The expected behavior is that the Kernel syscall code to handle ERESTARTSYS
> internally, and either return EAGAIN to userspace, or try again until
> it succeeds, depending on the open mode.
> 
I tested it and returning ERESTARTSYS instead of EINVAL doesn't help.
The behavior is the same.
As far as I can see no syscall code is involved in this call chain.

> So, it seems a worth trial.
> 
> Thanks,
> Mauro
>