Re: [Qemu-devel] [Xen-devel] [PATCH 3/3 v4] xenfb: Add [feature|request]-raw-pointer

[Owen Smith]

> -----Original Message-----
> From: Stefano Stabellini [mailto:address@hidden
> Sent: 12 October 2017 18:27
> To: Paul Durrant <address@hidden>
> Cc: 'Gerd Hoffmann' <address@hidden>; 'Stefano Stabellini'
> <address@hidden>; Anthony Perard <address@hidden>;
> address@hidden; address@hidden; Owen Smith
> <address@hidden>
> Subject: RE: [Xen-devel] [PATCH 3/3 v4] xenfb: Add [feature|request]-raw-
> pointer
> 
> On Thu, 12 Oct 2017, Paul Durrant wrote:
> > > -----Original Message-----
> > > From: Gerd Hoffmann [mailto:address@hidden
> > > Sent: 12 October 2017 10:26
> > > To: Paul Durrant <address@hidden>; 'Stefano Stabellini'
> > > <address@hidden>; Anthony Perard <address@hidden>
> > > Cc: address@hidden; address@hidden; Owen
> > > Smith <address@hidden>
> > > Subject: Re: [Xen-devel] [PATCH 3/3 v4] xenfb: Add
> > > [feature|request]-raw- pointer
> > >
> > >   Hi,
> > >
> > > > It's probably OS specific though. I guess the behaviour changed
> > > > because the OS favours absolute pointing devices over relative
> > > > ones and how it has two absolute ones to choose from. How it
> > > > reconciles those, who knows?
> > >
> > > Typically hid emulation calls qemu_input_handler_activate() when the
> > > guest initializes the device, which moves the device to the top of
> > > the priority list.
> > >
> > > Visible effect on a typical guest with ps/2 mouse and usb-tablet is
> > > that qemu switches from relative mode (mouse) to absolute mode
> > > (tablet)  when the guest loads the usb hid driver.
> > >
> > > I suspect pvmouse is doing the same thing.  So it may simply depend
> > > on guest driver load order whenever pvmouse or usb-tablet is used.
> > >
> > > Simplest fix is probably to only attach the device you plan to use
> > > to the guest.  If you can't turn off pvmouse for xen guests then you
> > > might want drop the qemu_input_handler_activate() call, so it
> > > behaves simliar to the ps/2 mouse (is used in case no other pointer
> device is present).
> >
> > Avoiding the activate call sounds reasonable and should avoid the
> behavioural change.
> 
> +1
> 
> Owen, are you up for resubmitting the series with this small change?

Having just rebuilt my xen / linux host, and checked qemu, it looks like the 
keycodemapdb patches have gone in, so I'll rebase and re-submit this series

Owen