HostAP Stability (fixed??)

[Jouni Malinen]

On Sun, May 11, 2003 at 06:24:43PM +0100, Mike Saywell wrote:

> Bascially I was recompiling the kernel on one of my APs and noticed I
> had SMP enabled, so I disabled it (it's a single processor machine),
> and since doing this hostap hasn't locked up once!

> I'm a bit confused about the problems reported with soekris boards though,
> I wouldn't have thought people would be running an SMP kernel on them...

I think these are two separate issues, although they could be related. I
don't currently have easy access to an SMP host, so I have been waiting
with SMP testing. However, based on your findings, it would be enough to
just use SMP enabled kernel.. I'll do that at some point.

I have now borrowed a Soekris board (thanks Sam!) and can duplicate the
reported hang easily. This sure makes it easier to debug the issue. It
looks like kernel does not really hang, it just happens to be busy
looping in hardware interrupt handler.. I was able to use Prism2 tick
event to detect such hangs and write debugging code for reporting what
went wrong. In addition, I have code that is able to automatically
recover from such a hang.

I haven't committed these changes yet since they are just hiding the
real issue: RX events seem to be unmasked in some cases before they have
been processed. This will end up in a state where the card has RX event
and thus IRQ line asserted all the time. A tasklet is trying to process
the pending RX entries, but it does not get anything done since it is
constantly interrupted by the event that it is actually trying to clear.

-- 
Jouni Malinen                                            PGP id EFC895FA