Hi,

A customer of mine has some high-end (8-way) machines and wants to use MSI interrupts to reduce latencies (specifically 
by not sharing interrupts any more). He has a bunch of GigE NICs to service at high packet rates.

He has M8A and is using startup-apic and pci-bios-v2 (and an experimental version of pci-bios-v2).

He had limited success, but provided feedback that is hopefully of value to us. He is available to test more code...

To make anything that booted at all, he mixed some 6.4.1 and 6.5.0 components...

Here's what he says: plus I attached his results
"
After a bunch of experiments I have mixed (some positive and some
negative results).

I manage to create operational APCI IFS, which consisted of
combination of 6.4.1 and 6.5.0A components. Specifically, I "borrow"
from 6.5.0: startup-apic, pci-bios-v2, libc.so, and libc.so.3.

Unfortunately, this IFS works only on one out of 5 types of machines
we have. Other four -- they crash during startup. I attached the
snapshots of several outputs from different PCs. Hopefully, it
contains useful information to QNX BSP and kernel developers. Again,
it is not clean setup (mixture of 6.4.1 and 6.5.0). Thought, it works
on one of the System. I didn't conduct any experiments with this
runtime environment, but, at least, two boxes can communicate over the
network.

Yesterday, I mentioned to you that pci-bios-v2 hangs the System. When
I took this components from 6.5.0 and tried to run the IFS, ldd
complained about lack of "__slog_once" symbol in pci-bios-v2. So, I
have to take libc.so.3 from 6.5.0 as well.
"

Is this any help ?

Any suggestions ? I've asked him to see if reducing the cores to 8 helps...

Regards

Dave

ave