Re: Xen 4.3 development update

[Fabio Fantoni <fabio.fantoni@m2r.biz>]

Il 13/05/2013 17:08, George Dunlap ha scritto:
> On 13/05/13 15:53, Fabio Fantoni wrote:
>> Il 13/05/2013 12:26, Ian Campbell ha scritto:
>>> On Fri, 2013-05-10 at 15:30 +0100, Fabio Fantoni wrote:
>>>> Il 10/05/2013 13:24, Anthony PERARD ha scritto:
>>>>> On 10/05/13 12:08, Anthony PERARD wrote:
>>>>>> On Fri, May 10, 2013 at 12:00 PM, George Dunlap
>>>>>> <George.Dunlap@eu.citrix.com>  wrote:
>>>>>>> On Fri, May 10, 2013 at 11:26 AM, George Dunlap
>>>>>>> <George.Dunlap@eu.citrix.com>  wrote:
>>>>>>>> * qxl not actually working
>>>>>>>>     > qxl apparently not enabled during compile by default
>>>>>>>>     > Appear to be some other bugs even when it is working:
>>>>>>>> >http://lists.xen.org/archives/html/xen-devel/2012-11/msg00713.html 
>>>>>>>>
>>>>>>>>     owner: ? (Anthony was going to take a look)
>>>>>>>>     We need to either fix both of these problems or disable qxl 
>>>>>>>> for 4.3
>>>>>>> Anthony, you were going to take a look at this; did you make any 
>>>>>>> progress?
>>>>>> Yes, I'm still trying to figure out how to enable qxl in the VM.
>>>>> And now, I can search why QEMU crash when starting Xorg with qxl.
>>>>>
>>>>>
>>>> Hi, thanks for your help.
>>>> I'm trying to have qxl full working on xen since end of 2011.
>>>> Since half of 2012 is at least working as standard vga.
>>> This means with:
>>>          vga = "qxl"
>>>
>>> in your configuration file?
>> Yes
>>> Are you saying that this configuration
>>> works?
>> For now it works only as standard vga
>>>   If not this then which configuration do you believe works?
>> With qxl driver enabled on domU's os, qemu crashes
>>>> The libxl patch should be correct, also with memory size setting.
>>>> I have tried same parameters of libxl qxl patch without xen and are 
>>>> working.
>>>> Also without kvm was working even if performance was really bad but at
>>>> least qemu did not crash on start with qxl driver.
>>>> Based on this test the problem seems to be xen related problem, 
>>>> probably
>>>> about hvmloader.
>>> What is the configuration which fails?
>> See on attachment
>>>
>>> Which libxl patch "should be correct"? libxl has lots of patches!
>> This patch that I refer:
>> http://xenbits.xen.org/gitweb/?p=xen.git;a=commit;h=aab4d1b266ce9891a684704f6abf6a5f6b3f7c24 
>>
>>>
>>> So are you saying that although the libxl patch "should be correct" it
>>> is in fact not sufficient to make qxl work?
>> The qemu parameters used are correct, for now on xen works only as 
>> standard vga but on kvm and qemu-only tests works also with qxl driver.
>> The difference with xen appares to be only on hvmloader.
>> I tried to find the exactly problem without result, based on seabios 
>> debug I saw that when it detects xen hypervisor, seabios relies on 
>> tables passed by hvmloader instead generate them.
>> Is the only effective difference viewed with seabios debug with and 
>> without xen.
>
> OK -- so you're saying that qxl actually *does* work, but just in a 
> limited fashion: as long as the guest only access the standard vga 
> interface, then you can connect over spice; but if the guest uses the 
> special qxl interface, then it crashes.
>
Yes
> We don't really use qxl, so that wasn't clear at first; we thought 
> that you were saying it didn't work at all.
>
> If it didn't work at all, then there's no point exposing functionality 
> that can't actually work.  But since there is new functionality 
> (albeit limited) exposed by the "qxl" option, we should keep it in.
>
> But we do need to make sure to set people's expectations properly -- 
> our release notes will have to advertize "partial qxl support".
I think it is good point to maintain it and advice that is 
experimental/partial so can be tested/debugged by users.
Without that libxl qxl patch users can't test it with simple 
device_model_args on xl configuration file but they must change libxl 
code for disable other emulated vga (and also setting proper videoram size).
>
>  -George