On Mon, Dec 09, 2002 at 09:36:14AM -0500, Perry E. Metzger wrote:
> I'm sort of making a bigger point. We need to start having a team
> actually begin breaking up Big Lock, and soon.

Frank and I made an interesting observation recently. For a short
while between commits a week or two ago, there was a bug in the
i386 intr code that meant softint's might not take the kernel lock.

What was most interesting about this bug, in the context of the
present discussion, is that both of us got through several full
MP rebuilds while running with the bug.  It was only noticed when
I started trying to rsync the resulting build off to another machine
and heavy network traffic caused mbuf corruption.

Highly unscientific, but I for one choose to see this as an
encouraging sign that at least some chinks can be taken out from
under the Big Lock fairly readily.  :-)

--
Dan.