Xen devel - Nov 2008 - guest state on realmode emulation failure

Due to a not yet determined reason (data corruption elsewhere most likely)
we''re seeing emulation failures. What caught my attention is the
inconsistency of the CS values in this log fragment:

(XEN) realmode.c:132:d4 Failed to emulate insn.
(XEN) realmode.c:174:d4 Real-mode emulation failed @ aa55:0000ffea: cb 80 fa 8f
7e 02
(XEN) domain_crash called from realmode.c:175
(XEN) Domain 4 (vcpu#0) crashed on cpu#0:
(XEN) ----[ Xen-3.3.1_18486_01-2.2  x86_32p  debug=n  Not tainted ]----
(XEN) CPU:    0
(XEN) EIP:    f000:[<0000ffea>]
(XEN) EFLAGS: 00000012   CONTEXT: hvm guest
(XEN) eax: 0000aa55   ebx: 00000000   ecx: 000d0000   edx: 00000080
(XEN) esi: 00007e00   edi: 000005cc   ebp: 00000000   esp: 0000ffda
(XEN) cr0: 00000010   cr4: 00000000   cr3: 00000000   cr2: 00000000
(XEN) ds: 0000   es: 07c0   fs: 0000   gs: 0000   ss: 0000   cs: f000

For the register dump to be meaningful I''m therefore wondering whether,
prior to crashing the domain, there shouldn''t be a call to
hvm_emulate_writeback().

As a secondary thing, I can''t really see how emulation of a real-mode
far
return could ever fail with the register state given.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 17/11/08 15:26, "Jan Beulich" <jbeulich@novell.com> wrote:
> As a secondary thing, I can''t really see how emulation of a
real-mode far
> return could ever fail with the register state given.
Have you tested with xen-unstable changeset 18766?

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>>> Keir Fraser <keir.fraser@eu.citrix.com> 17.11.08 16:32
>>>
>On 17/11/08 15:26, "Jan Beulich" <jbeulich@novell.com>
wrote:
>
>> As a secondary thing, I can''t really see how emulation of a
real-mode far
>> return could ever fail with the register state given.
>
>Have you tested with xen-unstable changeset 18766?
Not yet, but shouldn''t that affect things only when
''mis-''using real mode
(e.g. by leaving a 32-bit default operand size set)?

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 17/11/08 15:55, "Jan Beulich" <jbeulich@novell.com> wrote:
>>> As a secondary thing, I can''t really see how emulation of
a real-mode far
>>> return could ever fail with the register state given.
>> 
>> Have you tested with xen-unstable changeset 18766?
> 
> Not yet, but shouldn''t that affect things only when
''mis-''using real mode
> (e.g. by leaving a 32-bit default operand size set)?
No, realmode_deliver_exception() was testing a potentially uninitialised
field. It''s a nasty little bug that thus occasionally, whether real
mode is
abused or not, causes you to write through ESP rather than SP.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel