Re: [Xen-devel] Regression introduced by a3db233 x86emul: use DstEax also for {, I}{MUL, DIV

2016-09-29 Thread Jan Beulich 
>>> On 29.09.16 at 11:34,  wrote:
 On 29.09.16 at 11:11,  wrote:
>> The commit a3db233 x86emul: use DstEax also for {,I}{MUL,DIV}
>> introduces an regression when doing SR-IOV passthrough.
> 
> I'll see if I can repro,

Actually, I can see some variant of this (and without any SR-IOV), as
soon as I use "unrestricted_guest=0" on the command line. Debugging
now ...

Jan


___
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

Re: [Xen-devel] Regression introduced by a3db233 x86emul: use DstEax also for {, I}{MUL, DIV

2016-09-29 Thread Jan Beulich 
>>> On 29.09.16 at 11:34,  wrote:
 On 29.09.16 at 11:11,  wrote:
>> The commit a3db233 x86emul: use DstEax also for {,I}{MUL,DIV}
>> introduces an regression when doing SR-IOV passthrough.
> 
> I'll see if I can repro, but I can't see anything wrong with the
> commit (yet).

And sadly I can't. Also - is SR-IOV pass-through really related here?
At the point the issue occurs it shouldn't matter yet.

For investigation, could you make the vgabios.bin available
somewhere (might even be small enough to attach to mail)? I
ask because the binary I have doesn't even come close to have
the instruction byte sequence you ran into the issue with.

Jan


___
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

Re: [Xen-devel] Regression introduced by a3db233 x86emul: use DstEax also for {, I}{MUL, DIV

2016-09-29 Thread Jan Beulich 
>>> On 29.09.16 at 11:11,  wrote:
> The commit a3db233 x86emul: use DstEax also for {,I}{MUL,DIV}
> introduces an regression when doing SR-IOV passthrough.

I'll see if I can repro, but I can't see anything wrong with the
commit (yet). The more that ...

> Specifically I get:
> ..
> (d1) Invoking ROMBIOS ...
> (XEN) stdvga.c:174:d1v0 entering stdvga mode
> (d1) VGABios $Id: vgabios.c,v 1.67 2008/01/27 09:44:12 vruppert Exp $
> (XEN) realmode.c:119:d1v0 Failed to emulate insn.
> (XEN) Real-mode emulation failed: d1v0 Real @ c000:0154 -> 0f 75 05 e8 b0 
> 5f eb 7c 80 fc 1a 75 05 e8 d3 6a

... we don't emulate opcode 0f 75 at all, and looking at the following
bytes it feels like we're in the middle of an instruction, yet that patch
clearly doesn't alter what size instructions end up being.

Jan


___
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel