In message <E1F7bCp-0001sY-00@smeg.dsg.stanford.edu>,
Jonathan Stone writes:
>
>For the use I'd intended, losing nForce4 SATA makes the machine a non-starter.
>I needed a 1.5TB raidframe set , and I can't sustain the required throughput
>with 32-bit PCI add-in cards (not even considnering cost).
>
>I'll have to try a dual-core CPU with the acpi interrupt fixup.

I tried the ACPI_PCIFIXUP (i386 kernel so far, its a long story), this
time on a different machine: a Tyan S2985 with two single-core CPUs,
and two nforce4-family chips, one per CPU socket. The ACPI kernel
works better, or at least differently, than GENERIC.MPBIOS.

On that machine, GENERIC.MPBIOS (with or without various MPBIOS
fixups) would exhibit lost viaide interrupts, and would hang hard
within 120 seconds of logging in via VGA. (Sometimes with bogus
autorepeated keys, sometimes without; usually requiring a hard AC
powercycle to recover).

However, the ACPI kernel no longer configures any of the add-in PCI-X
cards I have in the amd8313 PCI-X slots. Nor does it configure the
PCI-Express NICs for which I'm trying to develop a driver. So it's
still a total bust for what I'm trying to accomplish.

(In contrast, the MPBIOS kernel would find those devices, but doesn't
stay up long enough to compile a driver, let alone a kernel.)