Re: pvfb: Absolute vs relative mouse tracking mystery

[Jeremy Fitzhardinge <jeremy@goop.org>]

On 06/19/2010 05:22 PM, Pasi Kärkkäinen wrote:
> On Sat, Jun 19, 2010 at 04:48:46PM +0100, Jeremy Fitzhardinge wrote:
>   
>> On 06/19/2010 04:41 PM, Pasi Kärkkäinen wrote:
>>     
>>> On Tue, May 11, 2010 at 11:56:38AM +0100, Stefano Stabellini wrote:
>>>   
>>>       
>>>> On Mon, 10 May 2010, Jeremy Fitzhardinge wrote:
>>>>     
>>>>         
>>>>> On 05/10/2010 07:41 AM, Stefano Stabellini wrote:
>>>>>       
>>>>>           
>>>>>> On Fri, 7 May 2010, Jeremy Fitzhardinge wrote:
>>>>>>   
>>>>>>         
>>>>>>             
>>>>>>> On one of my host machines, in PV guests using pvfb I get proper
>>>>>>> absolute mouse tracking.
>>>>>>>
>>>>>>> On another host machine, I get relative tracking in pvfb-using guests.
>>>>>>>
>>>>>>> Both are using identical versions of Fedora 12, identical domain
>>>>>>> configs, and toolstacks and kernels built from the same source.  The log
>>>>>>> files of the X servers are more or less identical.
>>>>>>>
>>>>>>> Help?  Any clues?
>>>>>>>
>>>>>>>     
>>>>>>>           
>>>>>>>               
>>>>>> relative vs absolute depends on a node on xenstore called
>>>>>> "request-abs-pointer" that defaults to 0 and has to be written by the
>>>>>> guest.
>>>>>>   
>>>>>>         
>>>>>>             
>>>>> OK, it looks like it is being set, but the guest is still showing
>>>>> relative behaviour:
>>>>>
>>>>>     vkbd = ""
>>>>>      0 = ""
>>>>>       backend = "/local/domain/0/backend/vkbd/5/0"
>>>>>       backend-id = "0"
>>>>>       state = "4"
>>>>>       page-ref = "1137660"
>>>>>       event-channel = "11"
>>>>>       request-abs-pointer = "1"
>>>>>
>>>>> Does this mean qemu-dm's vnc server is not doing the right thing?
>>>>>       
>>>>>           
>>>>  
>>>> I think is probably xenfb (that is the framebuffer backend in qemu) that
>>>> is not doing the right thing.
>>>> Try adding some debug output in hw/xenfb.c:input_connect.
>>>>
>>>>     
>>>>         
>>> Jeremy: Did you figure this out? I'm again seeing this problem on my Fedora 13 + Xen 4.0.1-rc3-pre box..
>>>   
>>>       
>> Yes, that should be fixed in current xen-unstable - it was a xenbus
>> race-condition.  What guest are you seeing it in?  I've found that I
>> still don't get absolute pointers in Centos guests.
>>
>>     
> Yeah, it's a CentOS 5.5 guest that gets relative pointer on my Fedora 13 dom0.
> On a CentOS5 dom0 it gets absolute though..
>
> Fedora 13 guest gets absolute pointer on my Fedora 13 dom0..
>   

Huh, that is interesting.  Maybe there's still a race or something there...

    J