Re: [Qemu-devel] [PATCH v8] kvm: notify host when the guest is panicked

[Anthony Liguori <anthony@codemonkey.ws>]

Marcelo Tosatti <mtosatti@redhat.com> writes:

> On Tue, Aug 14, 2012 at 02:35:34PM -0500, Anthony Liguori wrote:
>> Marcelo Tosatti <mtosatti@redhat.com> writes:
>> 
>> > On Tue, Aug 14, 2012 at 01:53:01PM -0500, Anthony Liguori wrote:
>> >> Marcelo Tosatti <mtosatti@redhat.com> writes:
>> >> 
>> >> > On Tue, Aug 14, 2012 at 05:55:54PM +0300, Yan Vugenfirer wrote:
>> >> >> 
>> >> >> On Aug 14, 2012, at 1:42 PM, Jan Kiszka wrote:
>> >> >> 
>> >> >> > On 2012-08-14 10:56, Daniel P. Berrange wrote:
>> >> >> >> On Mon, Aug 13, 2012 at 03:21:32PM -0300, Marcelo Tosatti wrote:
>> >> >> >>> On Wed, Aug 08, 2012 at 10:43:01AM +0800, Wen Congyang wrote:
>> >> >> >>>> We can know the guest is panicked when the guest runs on xen.
>> >> >> >>>> But we do not have such feature on kvm.
>> >> >> >>>> 
>> >> >> >>>> Another purpose of this feature is: management app(for example:
>> >> >> >>>> libvirt) can do auto dump when the guest is panicked. If management
>> >> >> >>>> app does not do auto dump, the guest's user can do dump by hand if
>> >> >> >>>> he sees the guest is panicked.
>> >> >> >>>> 
>> >> >> >>>> We have three solutions to implement this feature:
>> >> >> >>>> 1. use vmcall
>> >> >> >>>> 2. use I/O port
>> >> >> >>>> 3. use virtio-serial.
>> >> >> >>>> 
>> >> >> >>>> We have decided to avoid touching hypervisor. The reason why I choose
>> >> >> >>>> choose the I/O port is:
>> >> >> >>>> 1. it is easier to implememt
>> >> >> >>>> 2. it does not depend any virtual device
>> >> >> >>>> 3. it can work when starting the kernel
>> >> >> >>> 
>> >> >> >>> How about searching for the "Kernel panic - not syncing" string 
>> >> >> >>> in the guests serial output? Say libvirtd could take an action upon
>> >> >> >>> that?
>> >> >> >> 
>> >> >> >> No, this is not satisfactory. It depends on the guest OS being
>> >> >> >> configured to use the serial port for console output which we
>> >> >> >> cannot mandate, since it may well be required for other purposes.
>> >> >> > 
>> >> >> Please don't forget Windows guests, there is no console and no "Kernel Panic" string ;)
>> >> >> 
>> >> >> What I used for debugging purposes on Windows guest is to register a bugcheck callback in virtio-net driver and write 1 to VIRTIO_PCI_ISR register.
>> >> >> 
>> >> >> Yan. 
>> >> >
>> >> > Considering whether a "panic-device" should cover other OSes is also \
>> >
>> >> > something to consider. Even for Linux, is "panic" the only case which
>> >> > should be reported via the mechanism? What about oopses without panic? 
>> >> >
>> >> > Is the mechanism general enough for supporting new events, etc.
>> >> 
>> >> Hi,
>> >> 
>> >> I think this discussion is gone of the deep end.
>> >> 
>> >> Forget about !x86 platforms.  They have their own way to do this sort of
>> >> thing.  
>> >
>> > The panic function in kernel/panic.c has the following options, which
>> > appear to be arch independent, on panic:
>> >
>> > - reboot 
>> > - blink
>> 
>> Not sure the semantics of blink but that might be a good place for a
>> pvops hook.
>> 
>> >
>> > None are paravirtual interfaces however.
>> >
>> >> Think of this feature like a status LED on a motherboard.  These
>> >> are very common and usually controlled by IO ports.
>> >> 
>> >> We're simply reserving a "status LED" for the guest to indicate that it
>> >> has paniced.  Let's not over engineer this.
>> >
>> > My concern is that you end up with state that is dependant on x86.
>> >
>> > Subject: [PATCH v8 3/6] add a new runstate: RUN_STATE_GUEST_PANICKED
>> >
>> > Having the ability to stop/restart the guest (and even introducing a 
>> > new VM runstate) is more than a status LED analogy.
>> 
>> I must admit, I don't know why a new runstate is necessary/useful.  The
>> kernel shouldn't have to care about the difference between a halted guest
>> and a panicked guest.  That level of information belongs in userspace IMHO.
>> 
>> > Can this new infrastructure be used by other architectures?
>> 
>> I guess I don't understand why the kernel side of this isn't anything
>> more than a paravirt op hook that does a single outb() with the
>> remaining logic handled 100% in QEMU.
>
> From the patch description:
>
> "Another purpose of this feature is: management app(for example:
> libvirt) can do auto dump when the guest is panicked. If management
> app does not do auto dump, the guest's user can do dump by hand if
> he sees the guest is panicked."

Why does this mandated another runstate?  QEMU can simply mark the VCPUs
as stopped and raise a QMP event.  The kernel doesn't care if the VCPUs
are stopped or panicked.

> Wen, auto dump means dump of guest memory?
>
> In that case, the notification should obviously stop the guest 
> otherwise the guest might be reset by the time memdump from QEMU 
> monitor runs.
>
> But kexec supports dumping of memory already (i suppose it can 
> do automatic dump+{reboot,shutdown}).
>
>> > Do you consider allowing support for Windows as overengineering?
>> 
>> I don't think there is a way to hook BSOD on Windows so attempting to
>> engineer something that works with Windows seems odd, no?
>
> Unsure about hooking at BSOD time. But Windows has configurable 
> memory dump/reset/reboot, so yes it should not necessary.

Do you mean it's not necessary to hook BSOD?

I've very often gotten asked: We know 1 person is experiencing this
crash condition, can we figure out from the host how many other VMs are
experiencing this crash too instead of waiting for a user to complain?

That's the primary use-case for this notification IMHO.  Just a simple
status LED from the guest to indicate that it's in a bad state.

Regards,

Anthony Liguori

>
>> 
>> Regards,
>> 
>> Anthony Liguori
>> 
>> >
>> >> Regards,
>> >> 
>> >> Anthony Liguori
>> >> 
>> >> >
>> >> >> 
>> >> >> > Well, we have more than a single serial port, even when leaving
>> >> >> > virtio-serial aside...
>> >> >> > 
>> >> >> > Jan
>> >> >> > 
>> >> >> > -- 
>> >> >> > Siemens AG, Corporate Technology, CT RTC ITP SDP-DE
>> >> >> > Corporate Competence Center Embedded Linux
>> >> >> > --
>> >> >> > To unsubscribe from this list: send the line "unsubscribe kvm" in
>> >> >> > the body of a message to majordomo@vger.kernel.org
>> >> >> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
>> --
>> To unsubscribe from this list: send the line "unsubscribe kvm" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html