Xen devel - May 2013 - Re: [PATCH-v2] xen: Don't call arch_trigger_all_cpu_backtrace in dom0(pvm)

On Wed, 2013-05-15 at 16:40 +0800, Zhenzhong Duan wrote:
> On 2013-04-10 00:36, Ian Campbell wrote:
> > On Mon, 2013-04-08 at 08:42 +0100, Jan Beulich wrote:
> >>>>> On 07.04.13 at 07:54, Zhenzhong Duan
<zhenzhong.duan@oracle.com> wrote:
> >>> nmi isn''t supported in dom0, fallback to general all
cpu backtrace code.
> >> Since when is sending NMIs not supported, and since when is this
> >> Dom0-specific? If you want to deal with this, you should do so
> >> properly: Special case sending NMIs in the respective Xen specific
> >> code (using VCPUOP_send_nmi), and carry this out in a way not
> >> dependent upon running (un)privileged.
> > You''d also need to implement the upcall support for receiving
NMIs,
> > which IIRC isn''t yet done for pvops.
> Hi Ian,
> 
> Could you give a suggestion on which file to change to support NMI upcall?
> I compare with vMCE code, made similar change.
> Use VCPUOP_send_nmi to send nmi between pvm guest vcpus, but nmi
isn''t
> triggered.
You need to register a callback with CALLBACKOP_register
CALLBACKTYPE_nmi. You also need to write the code in entry.S to receive
that callback. IIRC you also need to arrange that returning from an NMI
is always done with HYPERVISOR_iret and not optimised to a direct iret
as it can be otherwise. This is to allow the hypervisor to implement NMI
masking correctly.

The linux-2.6.18-xen.hg tree implements NMI callbacks so you may find
inspiration there, although how upstreamable that approach is I''m not
sure. In particular stealing a EFLAGS bit to force the HYPERVISOR_iret
is certain to not be acceptable upstream.

Ian.