[ENBD] stability issues, big red neon sign

[Peter T. Breuer]

"A month of sundays ago Arne Wiebalck wrote:"
> On Wed, 27 Mar 2002, Peter T. Breuer wrote:
> > I forget what your results were for "-n 1" ...! Did you say it errors
> > with SMP or not? I'll have to go back and check ...
> 
> with 1 channel under SMP it was not stable under TCP/IP, but it survived
> under SCI ... maybe this was just luck.
> Nonetheless, I'll apply the patch and see what happens.

It'll be quite hard to exclude the kernel thread from the driver while
the client threads are already there. That's because the kernel thread
enters with the global i/o spinlock already held, and so I can't make it
wait (or we starve and die).

The best I can do is make the kernel thread take all the local spinlocks
at the earliest opportunity, as it enters.  Let me know and I'll send
you the diff. Or let me know how I can reschedule the q->request_fn()
for later if I see that I want to delay the kernel thread ... it _must_
run.

Peter