Re: PowerMac G5 (Late 2005) Quad [dual dual] , SMP not working

[Michael <macallan%netbsd.org@localhost>]

Hello,

On Sun, 20 Oct 2019 15:13:26 +0200
Romain Dolbeau <romain%dolbeau.org@localhost> wrote:

> (I have to load the kernel directly instead of going through ofwboot,
> but that's not a problem, and I got that to work with both
> netboot/NFS root and local disk boot/root).

That's what I use for development, but ofwboot.xcf loaded from a HFS+
partition should be able to load a kernel from a netbsd partition.

> However, I soon as I try a multiprocessor-enabled kernel, it won't
> boot. The assert at sys/arch/powerpc/pic/intr.c:561: "KASSERT(virq !=
> 0);" fails, and the kernel drops me in a debugger at least for the
> 11_2 variants.

How far into the boot process does that happen?
Or rather, what's the kernel output right before the KASSERT fires?
Do all CPUs attach? If so, are there error messages between / around
their output?

> I've also tried only defining cpu0 & cpu1 (with little hope) instead
> of cpu*, but I get the same result.
> 
> From previous posts to the mailing list, this machine might not have
> been tried before - I see reference to only the single-cpu dual-core
> version for the 970MP. Is there anything that I can try/report to help
> with getting better support for this particular beast? As the most
> poweful PowerMac ever, it would be nice to get full SMP support :-)

The SMP code has never been tested with more than two CPUs, for lack of
hardware in the right hands ( I have a dual core 11,2 with working SMP,
so the 970MP is supported but we screw up when there are two of them )
This shouldn't be hard to fix, powerpcs really don't need a lot of OS
help for SMP to work, it's mostly IPI delivery.

I'll look at the SMP code on my next weekend ( which is wednesday
through friday ) - most of it probably assumes two CPUs and no more.

Meanwhile, please send me the dmesg and ofctl -p output.

have fun
Michael