Re: [Xen-devel] [BUG] XEN crash and double fault when doing cpu online/offline

[Jürgen Groß <jgross@xxxxxxxx>]

On 08.01.20 09:32, Tao Xu wrote:
> On 1/8/20 3:50 PM, Jürgen Groß wrote:
> > On 08.01.20 06:50, Tao Xu wrote:
> > > Hi,
> > >
> > > When I use xen-hptool cpu-offline/cpu-online to let CPU in a socket 
> > > online/offline using the script as follows:
> > > for((j=48;j<=95;j++));
> > > do
> > >    xen-hptool cpu-offline $j
> > > done
> > >
> > > for((j=48;j<=95;j++));
> > > do
> > >    xen-hptool cpu-online $j
> > > done
> > >
> > > Xen crash when cpu re-online. I use the upstream XEN(0dd92688) and 
> > > try many days, it still crash. But if I only do cpu online/offline 
> > > for CPU 48~59, Xen will not crash. The bug can be reproduced when we 
> > > do cpu online/offline for most CPU in a socket. And interesting thing 
> > > is when we use the script as follow:
> > > for((j=48;j<=95;j++));
> > > do
> > >    xen-hptool cpu-offline $j
> > >    xen-hptool cpu-online $j
> > > done
> > >
> > > Xen will not crash too. Is there a bug in sched_credit2?
> > >
> > > The crash message as follows:
> > >
> > > (XEN) Adding cpu 77 to runqueue 1
> > > (XEN) Adding cpu 78 to runqueue 1
> > > (XEN) Adding cpu 79 to runqueue 1
> > > (XEN) Adding cpu 80 to runqueue 1
> > > (X(ENXE) N) *** DOUBLE FAULT ***
> > > (XEN) Assertion 'debug->cpu == smp_processor_id()' failed at 
> > > spinlock.c:88
> > > (XEN) ----[ Xen-4.14-unstable  x86_64  debug=y   Not tainted ]----
> > > (XEN) Debugging connection not set up.
> > > (XEN) CPU:    48
> > > (XEN) ----[ Xen-4.14-unstable  x86_64  debug=y   Not tainted ]----
> > > (XEN) CPU:    0
> > > (XEN) RIP:    e008:[<ffff82d080240bfc>] _spin_unlock+0x40/0x42
> > So the original problem causes a double fault, but spinlock debugging
> > causes a subsequent panic.
> >
> > Can you please retry the tests with the attached patch? It should
> > result in diagnostic data related to the real problem.
> >
> >
> > Juergen
> Hi Juergen,
>
> After apply your patch, spin_lock still assert. And the address 
> ffff82d0bffce880 is not in the xen-syms.
Yes, I had a bug in my modified ASSERT(), but this time the data is
better.

> (XEN) Adding cpu 78 to runqueue 1
> (XEN) *** DOUBLE FAULT ***
> (XEN) ----[ Xen-4.14-unstable  x86_64  debug=y   Not tainted ]----
> (XEN) CPU:    49
> (XEN) RIP:    e008:[<ffff82d0bffce880>] ffff82d0bffce880
This seems to be a crash in the stub page of cpu 48.

I don't think this is related to the scheduler, but to stub page
handling.

Can you please try the attached patch?


Juergen

Attachment: 0001-xen-x86-clear-per-cpu-stub-page-information-in-cpu_s.patch
Description: Text Data

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel