On Mon, Jul 14, 2003 at 10:50:08PM +0100, Richard Earnshaw wrote:
> I'm posting this here because I believe this problem may affect (at least 
> in theory) all platforms based on Jason's generic soft-interrupts 
> implementation for the ARM.

This sounds just like the problem that I saw with the evbarm/adi_brh 
port, too.  In that case, I re-enable and pending and unmasked
interrupts in the exit path.  It occurred readily here because the
softint interrupt is triggerred by software but actually comes in
via a hardware interrupt, instead of via softclock().

After reading your analysis, I think I like your solution a bit
better.

-allen

-- 
 Allen Briggs                     briggs@wasabisystems.com
 Wasabi Systems, Inc.             http://www.wasabisystems.com/