Re: [Xen-devel] Nested events in 64bit mini-OS

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Sat, 2012-11-10 at 13:52 +0000, Samuel Thibault wrote:
> > +    .byte 0x78,0x78,0x78,0x78,0x78            # jmp    hypercall_page + 
> > (__HYPERVISOR_iret * 32)
> 
> Here we would also need a fixup table for the code at hypercall_page!
> A nicer fix would be to inline the hypercall code here.  That said, I
> have to say I don't know any detail about the NMI flag, I don't see it
> defined in the Intel manual, and I don't see it being set in the Xen
> hypervisor.

It's a "software defined" bit (i.e. stolen out of eflags :-() used in
the classic-Xen Linux guest to indicate that return must go via
HYPERVISOR_iret (rather than direct iret), I think in order to let the
hypervisor implement the correct semantics wrt NMI masking. The
hypervisor itself doesn't actually know anything about the bit and the
guest could implement this some other way if it wanted (doesn't 64 bit
minios *always* use HYPERVISOR_iret?).

>   Unless somebody knows more about it, I would assume that it
> currently never happens,

That's probably a pretty good assumption, you would expect NMIs only in
dom0 (real NMIs getting forwarded) or in a guest which you explicitly
sent an NMI to via the toolstack, which I doubt you would do to an
unsuspecting mini-os domain (not in the expectation of anything good
happening at least).

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel