J6000 hangs with 2.6.14-rc1-pa1-32smp
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 11 lines · Antonio Cardoso Martins
Hello,
could anybody help with tracing this bug deeper?
Machine: HP J6000 workstation with serial console, 2048 Mb
Linux version 2.6.14-rc1-pa1-32smp (root@moonlight) (gcc version 3.3.5 (Debian 1:3.3.5-13)) #1 SMP Sun Sep 18 20:24:10 WEST 2005
...
SMP CALL FUNCTION TIMED OUT! (cpu=0), try 1
Antonio,
I still occasionally see the above bug. I think we deadlock when interrupts are disabled and we try to send an IPI. Ergo we can never harvest the completion. I haven't had time to dig more into the one good TOC dump that I do have.
I have to wonder if we are loosing nested IPI or deadlock if both CPUs happen to want to send an IPI to each other at the same time.
I've parked a kernel I've built that boots on my j6k at:
ftp://gsyprf10.external.hp.com/kernels/c3000/2.6.14-rc1-pa1.tgz
If you want to give that a try and then compare the .config files, that would be helpful.
Quoted 1 line
Would someone please help me?
I think the next step is to add a bug catcher that will panic if/when one cpu fails to get the IPI spinlock or pending_ipi is already set in ipi_send() (only works on 2-way). Or something like that. I have to think about this lot more.
hth,
grant
_______________________________________________
parisc-linux mailing list
[email protected]
http://lists.parisc-linux.org/mailman/listinfo/parisc-linux