On Wed, Oct 23, 2002 at 11:51:55AM -0700, Scott Ellis wrote:
> I filed a PR for this (18676).  On my system, using a Uniprocessor
> kernel solved the problem (albeit at the expense of one CPU!).
> The whole MP IRQ handling seems a little..broken.  I had a few
> system hangs when using both CPUs on the system and lots of
> device IO, and of course there's the whole "can't share IRQ's
> at different levels" issue (which is annoying, as it means there's technically
> no way to get the 3 cards in my system I want..network, USB, and sound).

The '..broken' message for sharing an interrupt line with different
IPLs is misleading; it's not actually broken. It'll work, but
it's suboptimal, the lower priority interrupt will also block
the higher priority one on the same line. However, this has
basically been the case for shared irq lines in the i386
intr code forever.

I'm rewriting the i386 interrupt code (getting closer to finishing it),
to alleviate this issue, and make MP interrupt handling faster.

The bge(4) driver is more suspect with regard to causing problems
than the MP code. There's quite a list of errata (a different one
for each revision) of the chip on these cards, and the driver
doesn't address all of them.

If you can reproduce any problems with a different card, let me know.
I have not been able to cause any problems on my main system
(dual Athlon with some SCSI disks, an IDE disk, 2 FE internet
interfaces and 2 gig-e interfaces, one wm(4) and one gsip(4)) under
heavy I/O. I.e. running a full source tree build while hammering
it as an NFS server.

- Frank

-- 
Frank van der Linden                                    fvdl@wasabisystems.com
==============================================================================
Quality NetBSD Development, Support & Service.   http://www.wasabisystems.com/