qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH v2] ppc: spapr-rtas - implement os-term rtas cal


From: Alexander Graf
Subject: Re: [Qemu-devel] [PATCH v2] ppc: spapr-rtas - implement os-term rtas call
Date: Thu, 26 Jun 2014 11:22:50 +0200
User-agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.9; rv:24.0) Gecko/20100101 Thunderbird/24.6.0


On 26.06.14 11:05, Nikunj A Dadhania wrote:
Alexander Graf <address@hidden> writes:

Am 26.06.2014 um 09:55 schrieb Nikunj A Dadhania <address@hidden>:

Alexander Graf <address@hidden> writes:

On 25.06.14 13:27, Nikunj A Dadhania wrote:
Alexander Graf <address@hidden> writes:

Let me put down my understanding:

There are two possible way to handle kernel panic:
1) Kdump service running in guest - already working
2) Pass the kernel panic information to hypervisor - not there in Qemu
    pseries

So without kdump service running, if linux kernel hits a panic, its going
to check os-term and extended-os-term, only then its going to call
os-term.
It's checking both for a reason. Find that reason.
Here is what I figured out from PAPR:

rtas ibm,os-term, does not gaurantee a return back to the guest cpu.

If ibm,extended-os-term property is set rtas call return will always
occur.
With your patch it doesn't occur.
Hmm, you are right.

+    monitor_protocol_event(QEVENT_GUEST_PANICKED, data);
+    qobject_decref(data);
+    vm_stop(RUN_STATE_GUEST_PANICKED);

So the issue here is using vm_stop. So if I do not do this I will be
returning back to the guest.

I was trying to enable:

<on_crash> coredump-restart </on_crash>
<on_crash> coredump-destroy </on_crash>

in libvirt. But anyways, when libvirt receives these event, it will
either destroy or restart the guest. So if I remove the vm_stop, there
will be a return to the guest. And depending on the libvirt domain
config a restart/destroy will take effect. Does this look ok?

I think that's closer to what Linux expects, yes. But I'd like to have an ack on that approach from Ben or Anton.


Alex




reply via email to

[Prev in Thread] Current Thread [Next in Thread]