Re: [PATCH] rfkill: always init poll delayed work

[Larry Finger <Larry.Finger@lwfinger.net>]

Johannes Berg wrote:
> On Wed, 2009-06-03 at 09:50 -0500, Larry Finger wrote:
>> 2. Much more serious - When the radio kill switch is turned off, the
>> radio is killed just as expected, but it is not restored when the
>> switch is turned on. The only way to restore the radio is to
>> rmmod/insmod b43. Similarly, if the module is loaded with the switch
>> off, it is not possible to turn the radio on. An unload/load resquence
>> is then needed.
> 
> I suspected that much. And you can't recover that since you can't set
> the interface UP. This is because polling doesn't work while the
> interface is set down. As I said previously, I think that's previously
> been buggy too, if you did
>  1) hard kill
>  2) set interface down
>  3) hard unkill
> 
> then step 3) would not trigger an event to userspace until you set the
> interface up again, afaict.
> 
> We probably need to bring up the core to poll it, if possible?

I have not made much progress in sorting this out. When I turn the
switch off, I see RFKILL_STATE in /sys/class/rfkill/rfkill0/uevent go
from unblocked to hw_blocked. It does not change when the switch is
turned on.

I have verified that b43_rfkill_poll(), the polling callback routine
is being executed, but that the hardware bit in the interface is never
being set again. Whichever part of the old rfkill_input code that made
that change seems not to be functioning.

What diagnostics would be helpful?

Larry