Re: [Xen-devel] [PATCH] libxl: Remove qxl support for the 4.3 release

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 18.09.13 at 14:29, Fabio Fantoni <fabio.fantoni@xxxxxxx> wrote:
> Il 18/09/2013 10:07, Jan Beulich ha scritto:
>>>>> On 16.09.13 at 16:10, Fabio Fantoni <fabio.fantoni@xxxxxxx> wrote:
>>> Il 05/07/2013 18:59, George Dunlap ha scritto:
>>>> On Wed, May 29, 2013 at 11:25 PM, Andrew Cooper
>>>> <andrew.cooper3@xxxxxxxxxx> wrote:
>>>>> On 29/05/2013 08:43, Ian Campbell wrote:
>>>>>> On Tue, 2013-05-28 at 19:09 +0100, Keir Fraser wrote:
>>>>>>> On 28/05/2013 17:51, "Ian Jackson" <Ian.Jackson@xxxxxxxxxxxxx> wrote:
>>>>>>>
>>>>>>>> George Dunlap writes ("[PATCH] libxl: Remove qxl support for the 4.3
>>>>>>>> release"):
>>>>>>>>> The qxl drivers for Windows and Linux end up calling instructions
>>>>>>>>> that cannot be used for MMIO at the moment.  Just for the 4.3 release,
>>>>>>>>> remove qxl support.
>>>>>>>>>
>>>>>>>>> This patch should be reverted as soon as the 4.4 development window 
>>>>>>>>> opens.
>>>>>>>>>
>>>>>>>>> The issue in question:
>>>>>>>>>
>>>>>>>>> (XEN) emulate.c:88:d18 bad mmio size 16
>>>>>>>>> (XEN) io.c:201:d18 MMIO emulation failed @ 0033:7fd2de390430: f3 0f 6f
>>>>>>>>> 19 41 83 e8 403
>>>>>>>>>
>>>>>>>>> The instruction in question is "movdqu (%rcx),%xmm3".  Xen knows how
>>>>>>>>> to emulate it, but unfortunately %xmm3 is 16 bytes long, and the 
>>>>>>>>> interface
>>>>>>>>> between Xen and qemu at the moment would appear to only allow MMIO 
>>>>>>>>> accesses
>>>>>>>>> of 8 bytes.
>>>>>>>>>
>>>>>>>>> It's too late in the release cycle to find a fix or a workaround.
>>>>>>>> Acked-by: Ian Jackson <ian.jackson@xxxxxxxxxxxxx>
>>>>>>> It could be plumbed through hvmemul_do_io's multi-cycle read/write 
>>>>>>> logic,
>>>>>>> and done as two 8-byte cycles to qemu. This would avoid bloating the 
>>>>>>> ioreq
>>>>>>> structure that communicates to qemu.
>>>>>> Are you proposing we do this for 4.3? I'm not sure how big that change
>>>>>> would be in terms of impact (just that one instruction, any 16 byte
>>>>>> operand?).
>>>>>>
>>>>>> Of course even if we did this for 4.3 we don't know what the next issue
>>>>>> will be with QXL.
>>>>>>
>>>>>> Ian.
>>>>> Furthermore, AVX instruction emulation would require support for 32byte
>>>>> operands.  I don't see the multi-cycle logic scaling sensibly.
>>>> Andrew, Keir, Jan, does any one of you fancy taking this on for 4.4?
>>> Is there someone that can add full support for SSE on hvm domUs?
>>> Thanks for any reply.
>> Mind giving the attached patch a try?
> 
> Thanks, I have tested your patch but qxl seems to have the same problem 
> (qemu crash on domU xorg start).
> I not see the MMIO error on logs, what must I do to see it or any 
> similar debug information?
> Tell me if I must do other specific test or you need other details.

Without the patch you ought to have seen "bad mmio size ..."
warning in the hypervisor log. Those should be gone with the
patch I handed you. If you didn't see such warnings, then your
problem wasn't with the emulation of SSE instructions in the first
place.

As to qemu crashing - I'll have to defer to those knowing qemu
much better than I do. I merely wanted to provide the hypervisor
side change that I deemed necessary based on the information I
had (and I think that change ought to go in anyway, regardless
of whether it helps in your specific case).

Jan


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