Re: [PATCH] x86, kdump: No need to disable ioapic in crash path

[Don Zickus <dzickus@redhat.com>]

On Tue, Feb 07, 2012 at 03:35:59PM -0800, Eric W. Biederman wrote:
> Vivek Goyal <vgoyal@redhat.com> writes:
> 
> > On Tue, Feb 07, 2012 at 04:57:41PM -0500, Don Zickus wrote:
> >> On Thu, Feb 02, 2012 at 03:24:46PM -0800, Eric W. Biederman wrote:
> >> > > Eric, brought up a point that because the boot code was restructured we may
> >> > > not need to disable the io apic any more in the crash path.  The original
> >> > > concern that led to the development of disable_IO_APIC, was that the TSC
> >> > > calibration on boot up relied on the PIT timer for reference.  Access
> >> > > to the PIT required 8259 interrupts to be working.  This wouldn't work
> >> > > if the ioapic needed to be configured.  So on panic path, the ioapic was
> >> > > reconfigured to use virtual wire mode to allow the 8259 to passthrough.
> >> > 
> >> > A small clarification originally it was the jiffies calibration that
> >> > would fail if we could cause the PIT to generate interrupts through the
> >> > 8259.  The boot would then hang at calibrating jiffies.
> >> 
> >> Ok.  Thanks!
> >
> > So now what has changed? Do we setup LAPIC and IOAPIC early enough to
> > receive PIT interrupts in regular mode (non-virtual wire mode) or
> > something else?
> 
> Yes.  Part of the Moorstown work required that this be done because
> moorsetown did not support legacy mode.  Last I looked the code hadn't
> been generalized beyond Moorsetown but empirically it works now.
> 
> Don as to what to test the only case I can think of that might be spooky
> is a screaming interrupt during the handover.  You might want to try
> playing with lkcdtm to try some of the more exotic crash scenarios.  But
> all I expect further testing might reveal are places where we are not
> as robust in initializing the hardware as we might be.  Things that
> might have been papered over by the ioapic shutdown.

I ran lkdtm by panic'ing in the interrupt handle thus leaving device
interrupt un-ack'd and the apic might have been un-ack'd too (jprobes
hooked in at do_IRQ).  3 out 3 times the second kernel came up on my core2
quad.

Cheers,
Don