Re: pvfb: Absolute vs relative mouse tracking mystery

[Joshua West <jwest@brandeis.edu>]

  On 08/03/2010 12:23 PM, John Haxby wrote:
> On 2 Aug 2010, at 20:40, Jeremy Fitzhardinge wrote:
>
>> On 08/02/2010 12:37 PM, Joshua West wrote:
>>> On 08/02/10 15:00, Jeremy Fitzhardinge wrote:
>>>> On 08/01/2010 12:11 PM, Joshua West wrote:
>>>>> Jeremy,
>>>>>
>>>>> Do you know if this will be fixed in the Xen 3.4.x series as well?  In the 3.4.x ioemu-remote qemu code and the 2.6.18.8 xen kernel?
>>>>>
>>>>> I ask because this is an issue for me on RHEL5 domU's which are paravirtualized.  Using Xen 3.4.3 w/ Xen kernel 2.6.18.8 and I'm not yet ready to switch production clusters to 4.0.x.
>>>> RHEL/CentOS5 deliberately avoids using the absolute pointer mode for reasons I don't understand, so it is broken as expected.
>>>>
>>>>     J
>>> Ahh yes, but thats if you're using the RHEL/CentOS Xen kernel.  I'm not using a RHEL kernel... instead I'm using linux-2.6.18-xen.hg, but still the issue persists on RHEL5 domU's.
>> OK, in that case it would be worth looking at backporting the changes to older Xens.
>
> I'm not sure if it will make much difference.
>
> The 2.6.18.x X server uses the PS/2 mouse driver which is strictly relative, in fact that driver basically maps on to the kernel's mousedev driver which carefully converts the absolute pointer events from the xen driver to relative ones and most of the problems with tracking arise from a built-in assumption that the absolute coordinates are on a 1024x768 screen (this is why the local mouse and the guest mouse appear to be connected by a pantograph, albeit a poorly functioning pantograph).
>
> One solution is to use xorg-x11-drv-evdev which _does_ take the absolute pointer events and passes them directly to the X server.   Unfortunately while this is possible it is difficult to configure because the 2.6.18.x kernel has a combined xen mouse and keyboard device and the X server wants to distinct devices (if linux-2.6.18-xen.hg has separated the drivers (and you can see distinct keyboard and mouse devices in /proc/bus/input/devices) then just configure a evdev driver as a pointer for X.
>
> For RHEL/CentOS 5 I appear to have a solution: an X input driver that takes only the mouse events from the kernel's evdev driver.   I have something that works and that I believe I can publish but I need to check both of those when I get back to work (I'm not quite on the beach at the moment, but not far off it).  Of course, for 5.5 you need to re-enable request-abs-pointer on the kernel command line, but that's not too onerous.
>
> jch
Ahh... So then if we get the PVFB going with 1024x768 resolution, 
instead of wonderful 1990's 800x600, we're all set ;-)

I've been testing with evdev driver as well (Option "Device" 
"/dev/input/event1" for me), but that has not worked.  I see what you're 
saying though - I can see two entries in /proc/bus/input/devices, and my 
mouse handlers are "mouse0 event1 ts0", but both the keyboard and mouse 
physical device is the same: "xen/device/vkbd/0".  And this is all with 
linux-2.6.18-xen.hg.

I'm looking forward to your custom RHEL/CentOS driver to get this going.

One last question -- even if using linux-2.6.18-xen.hg, are you saying 
we still need request-abs-pointer kernel argument?  Wasn't that just to 
enable absolute pointer mode in the RHEL mouse driver if using RHEL 5.5 
and their Xen kernel (looked that way from their patch)?  Or is RHEL's 
Xorg somehow also looking for this directive?

Thanks.

-- 
Joshua West
Senior Systems Engineer
Brandeis University
http://www.brandeis.edu