[RFC AR6000 hif patch] Clean up code and do_irq on a deferrable worker

[Werner Almesberger]

Michael Trimarchi wrote:
> So first of all I clean up the code
> of the hif2 and move to a deferrable lock to avoid the release of the
> lock. The bug remain [...]

Good :-) This is a part of the code that I'm fairly confident about,
so I was surprised that you had to make such a big change.

Did you try if the locking change along makes the panic go away ?
If yes, then we probably don't have to worry about rearranging
hif2.

> and decide to take a look to the wext_ioctl.c and
> I don't find any locking during remove of the device, so I change it too.

This one looks better. I only wonder if we could accomplish the same
with fewer changes. I'm still a little reluctant to make sweeping
changes to that driver, because there's still the distant possibility
that a new code drop may come along from Atheros, in which case, each
of the changes we've made would have to be painstakingly copied over.

I'll give your code a closer look in a couple of days. Right now, my
brain is still too wrecked from a 12 hours jet-lag and 45 hours of
travel that I wouldn't even trust myself changing hello.c :-)

Thanks,
- Werner