On 09/23/2010 10:11 PM, Timo Juhani Lindfors wrote:
> Ian Campbell <i...@hellion.org.uk> writes:
>> I guess the right thing to do would be to using some existing paravirt
>> interface to send the IPI instead of tweaking the APIC directly, or if
> Is
>
> void xen_send_IPI_one(unsigned int cpu, enum ipi_vector vector);
>
> in xen/events.h perhaps already this?
>
>
The crashing code is actually using the LAPIC to send a NMI to the other
CPUs, not a regular IPI. We don't have any equivalent to that at
present. We could probably do the higher-level "show all cpu's
stacktraces" operation some other way entirely (since we can grab the
other CPU's context via Xen without having to rely on the CPU itself to
do the job). But I haven't looked into what it would take to do that.
For now, it might just be easiest to put a "if (xen_pv_domain())
return;" to nop out the whole thing. We could also set up a dummy lapic
page to absorb the write, but that would just leave it spinning.
Hm, since we don't enable APIC in the CPU capabilities, that code
shouldn't be running anyway; that's probably the cleanest workaround for
now.
J
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
