Re: [patch] voluntary-preempt-2.6.8-rc2-L2 PS2 keyboard gone south

[Lee Revell <rlrevell@joe-job.com>]

On Sat, 2004-07-31 at 20:36, Felipe Alfaro Solana wrote:
> On Sat, 2004-07-31 at 13:56 -0400, Lee Revell wrote:
> > On Sat, 2004-07-31 at 05:48, Felipe Alfaro Solana wrote:
> > > On Sat, 2004-07-31 at 00:03 -0400, Lee Revell wrote:
> > > > Shane & Felipe,
> > > > 
> > > > I am removing LKML and Ingo from the cc: list so we can compare
> > > > hardware, as I definitely am not seeing this problem.  Maybe we can
> > > > track it down to a difference there.  What hardware are you guys using?
> > 
> > > I'm running 2.6.8-rc2-bk9 plus a bunch of other patches, like Con's
> > > Staircase Scheduler and Ingo's voluntary-preempt (of course). Kernel is
> > > compiled with ACPI + APIC.
> > 
> > Hmm.  Maybe it's an ACPI/APIC problem?  I do not have these enabled.  It
> > could also be an interaction between some other patch you have.
> > 
> > Have you tried with 2.6.8-rc2-M5, without the other patches?  Booting
> > with acpi=off?
> 
> OK, OK :-) I've been running 2.6.8-rc2-mm1-M5 with ACPI but without APIC
> for more than ten minutes), compiling kernels, sending mails with
> Evolution and it hasn't locked up yet. Crossfingers. Should we report
> this to Ingo and Andrew?
> 
> Anyways, I'll keep on running this puppy to see if this behavior is
> consistent.
> 
> # grep . /proc/irq/*/*/threaded
> /proc/irq/11/eth0/threaded:1
> /proc/irq/12/Intel 82801BA-ICH2/threaded:1
> /proc/irq/14/ide0/threaded:1
> /proc/irq/15/ide1/threaded:1
> /proc/irq/1/i8042/threaded:1
> /proc/irq/5/uhci_hcd/threaded:1
> /proc/irq/8/rtc/threaded:1
> /proc/irq/9/acpi/threaded:1
> /proc/irq/9/uhci_hcd/threaded:!
> 
> # grep . /proc/sys/kernel/*_preemption
> /proc/sys/kernel/voluntary_preemption:3
> /proc/sys/kernel/preemption:1

The next thing I was going to suggest was the software RAID.  You appear
to have a RAID 0 or 1 with one disk on irq14 and one on irq15.  I am not
sure how interrupts are handled by Linux IDE RAID, but it seems like
this would be tricky.  I bet the threading is screwing up the
synchronization between the devices, and you end up with one waiting
forever for the other - the possibilities for lockup are endless.

However, I have no idea what you would do about this.  You could try
making those irqs non-threaded, but you'd have to do it one at a time,
and I would *definitely* expect weird stuff with one threaded and one
not.

I will have to defer to someone with more kernel expertise.

Lee