[Qemu-devel] WinXP mouse patch

[Qemu-devel] WinXP mouse patch

[Mark Jonckheere]

[-- Attachment #1: Type: text/plain, Size: 366 bytes --]

The following patch against the latest CVS (13 april) resolved
for me the slow mouse movements in WinXP. It removes from the
datastream dummy mouse-events that contain no movement or
button-change. This is probably the same problem as in Win2K.

to apply the patch:
save attachment
cd qemu/hw
patch -p0 < /path/to/pckbd.diff
cd ..
make

Happy testing,
Mark.
-- 
:wq

[-- Attachment #2: pckbd.diff --]
[-- Type: text/plain, Size: 546 bytes --]

--- pckbd.c.orig	Sat Apr 10 21:04:48 2004
+++ pckbd.c	Fri Apr 16 00:24:27 2004
@@ -442,9 +442,14 @@
     if (!(s->mouse_status & MOUSE_STATUS_ENABLED))
         return;
 
+#ifdef DEBUG_MOUSE
+    printf("MOUSE %d %d %d %d\n", dx, dy, dz, buttons_state);
+#endif
     s->mouse_dx += dx;
     s->mouse_dy -= dy;
     s->mouse_dz += dz;
+    if (s->mouse_dx == 0 && s->mouse_dy == 0 && s->mouse_dz == 0 && s->mouse_buttons == buttons_state)
+	return;
     s->mouse_buttons = buttons_state;
     
     if (!(s->mouse_status & MOUSE_STATUS_REMOTE) &&

Re: [Qemu-devel] WinXP mouse patch

[Gregory Alexander]

I made this same change in bochs to make the mouse usable there.  At 
least under linux I think it has to do with X sending the events even 
when they're not really needed.  From Mike Nordell's post, it sounds 
like SDL just blindly forwards those events on without filtering.

Also, (although I don't know much about the qemu event queue) it might 
make sense to add the capability to merge multiple outstanding mouse 
events into a single event.  This made a BIG difference on bochs, and 
could help in QEMU as well, if for no other reason than saving CPU cycles.

GREG



Mark Jonckheere wrote:
> The following patch against the latest CVS (13 april) resolved
> for me the slow mouse movements in WinXP. It removes from the
> datastream dummy mouse-events that contain no movement or
> button-change. This is probably the same problem as in Win2K.
> 
> to apply the patch:
> save attachment
> cd qemu/hw
> patch -p0 < /path/to/pckbd.diff
> cd ..
> make
> 
> Happy testing,
> Mark.
> 
> 
> ------------------------------------------------------------------------
> 
> --- pckbd.c.orig	Sat Apr 10 21:04:48 2004
> +++ pckbd.c	Fri Apr 16 00:24:27 2004
> @@ -442,9 +442,14 @@
>      if (!(s->mouse_status & MOUSE_STATUS_ENABLED))
>          return;
>  
> +#ifdef DEBUG_MOUSE
> +    printf("MOUSE %d %d %d %d\n", dx, dy, dz, buttons_state);
> +#endif
>      s->mouse_dx += dx;
>      s->mouse_dy -= dy;
>      s->mouse_dz += dz;
> +    if (s->mouse_dx == 0 && s->mouse_dy == 0 && s->mouse_dz == 0 && s->mouse_buttons == buttons_state)
> +	return;
>      s->mouse_buttons = buttons_state;
>      
>      if (!(s->mouse_status & MOUSE_STATUS_REMOTE) &&
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://mail.nongnu.org/mailman/listinfo/qemu-devel

Re: [Qemu-devel] WinXP mouse patch

[Mike Nordell]

Mark Jonckheere wrote:

> The following patch against the latest CVS (13 april) resolved
> for me the slow mouse movements in WinXP.

Looking at it, I see difference between hosts and do one observation.

The difference is that QEMU on Win32 isn't as far as I have observed or
experienced plagued by this problem. The observation is that if SDL did its
job, this patch would be "dead code" - a code path never taken, why it seems
to me to be an SDL problem/bug related to your specific host OS. Could you
verify this?

I base that observation on the fact that kbd_mouse_event() should only be
called from sdl_send_mouse_event() in sdl.c, which in turn should only be
called when an SDL event of type SDL_MOUSEMOTION, SDL_MOUSEBUTTONDOWN or
SDL_MOUSEBUTTONUP is recieved. If you get SDL_MOUSEMOTION with with dx&dy =
0 from SDL_GetRelativeMouseState, I can't see any other explanation than an
SDL problem.


That said, since it apparently solves a real problem I too think the patch
should be included, with the addition of a comment explaining why that code
is there.


/Mike