Re: Regression: ONE CPU fails bootup at Re: [3.2.0-RC7] BUG: unable to handle kernel NULL pointer dereference at 0000000000000598 1.478005] IP: [<ffffffff8107a6c4>] queue_work_on+0x4/0x30

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Wed, Jan 04, 2012 at 10:36:52AM -0800, John Stultz wrote:
> On Wed, 2012-01-04 at 09:17 +0100, Stefan Bader wrote:
> > Over night I had still be thinking on this and maybe one important fact I had
> > been ignoring. This really has only been observed on paravirt guests on Xen as
> > far as I know. And one thing that I should have pointed out is that
> > 
> > [    0.792634] rtc_cmos rtc_cmos: rtc core: registered rtc_cmos as rtc0
> > [    0.792725] rtc_cmos: probe of rtc_cmos failed with error -38
> > 
> > So first the registration is done and the first line is the last thing printed
> > in the registration function. Then, and that line always comes after, the probe,
> > which looks like being done asynchronously, detects that the rtc is not
> > implemented. I would assume that this causes the rtc to be unregistered again
> > and that is probably the point where, under the right circumstances, the worker
> > triggered by the initialize alarm is trying to set another alarm. Probably while
> > some of the elements of the structure started to be torn down. I need to check
> > on that code path, yet. So right now its more a guess.
> 
> Hrm. Do you see the same probe error with 3.1 kernels as well?  
> 
> Konrad: Is the probe failure a known issue on Xen? Any clues on whats
> going on there?

Hey John,

Stefan kind of summarized it - the paravirtualized guests do not have access
to the CMOS. In fact they have no access to any legacy device (except if one does
PCI passthrough) - so the rtc_core returning -38 is correct.

We have our own timer - which is the Xen hypervisor stamps the the nanosecond
resolution data in a per-cpu field that the timer API uses.