vmspace changes to use sysref
dillon at apollo.backplane.com
Mon Apr 30 19:14:24 PDT 2007
Oh, also try with the absolute latest HEAD. I don't think SYSREF
could have had any effect on the interrupt routing, but the other
work recently committed certainly could have.
I have some ideas with regards to the livelock issues. If reducing
the values solves that part of the problem I think we can safely
boot up with a much lower livelock limits, then increase them
as part of the RC.
The only thing that can cause a large number of stray irq 7's that
I know of is when the ICU is used (no APIC_IO), and the BIOS fails
to program the hardware to route the ICU interrupts to just one of
the cpu's. If interrupts go to both cpu's then both cpu's try to
ack the INTA cycle and one of the cpu's will always get a stray
interrupt vector (which is usually irq 7). On AMD boxes that can be
fixed. I'm not sure how to do it on Intel boxes.
What I'd like to know regarding the stray irq 7 is whether you ever
used the ICU with older kernels running on this particular box,
or whether you just always used APIC_IO. It might not be a 'new'
issue for your box with that particular combination (i.e. not created
by recent commits).
More information about the Kernel