Re: [Qemu-ppc] spapr DRC and awaiting_allocation

[Bharata B Rao]

On Fri, Jun 09, 2017 at 12:32:46AM +1000, David Gibson wrote:
> Hi Bharata,
> 
> I'm trying to understand the awaiting_allocation flag in the DRC code,
> and it's not making much sense to me.  It was introduced in aab9913
> "spapr_drc: Prevent detach racing against attach for CPU DR", which
> claims that a hot unplug racing with a not-completed hot plug of a CPU
> can crash the guest.
> 
> The change made only appears to affect the case where this happens:
>    host                               guest
>    attach() CPU
>    notify guest
>    detach() CPU
>                               RTAS set_allocation_state USABLE
> 
> But in that case, I'd expect the set_allocation_state to simply fail
> (becuase the DRC is now awaiting_release) and for the guest to cope
> with that failure.
> 
> Can you remember the mechanism by which the guest was crashed?

I dug up the old bugzilla (for PowerKVM-3.1) which lists the following
as the reproducer.

4. Create & Run the following script
# cat cpu_hotplug_stress.sh
#!/bin/bash
        for i in {0..5}
        do 
                sed "s/X/$i/g" cpu_hp.xml>temp_cpu_hp.xml
                virsh attach-device sles12 temp_cpu_hp.xml --live
        done
        for i in {5..0}
        do
                sed "s/X/$i/g" cpu_hp.xml>temp_cpu_hp.xml
                virsh detach-device sles12 temp_cpu_hp.xml --live
        done

5. Guest will be crashed, check the dmesg on host machine
[ 8489.089952] qemu-system-ppc[21593]: unhandled signal 11 at 0000000000000014 
nip 00003fff8a891a4c lr 00003fff8a891a44 code 30001

Remember that this was when in-kernel CPU hotplug wasn't yet available.

I will spend some more time on this and come back later.

Regards,
Bharata.