Re: [Qemu-devel] [5056] add DisplayState->idle (Samuel Thibault)

[Anthony Liguori]

Paul Brook wrote:
> On Thursday 21 August 2008, Anthony Liguori wrote:
>   
> > Revision: 5056
> >           http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=5056
> > Author:   aliguori
> > Date:     2008-08-21 20:08:55 +0000 (Thu, 21 Aug 2008)
> >
> > Log Message:
> > -----------
> > add DisplayState->idle (Samuel Thibault)
> >     
>
> This really needs a comment saying exactly what this field means.

Fair enough.  Gerd or Samuel, can you send a patch?

>  I'm guessing 
> it's set by the GUI routines to tell the emulated device that it doesn't need 
> to update the display.
>   

Yes, that's what it does AFAIUI.

> Currently the GUI has to poll the emulated device to get updated output, which 
> means this flag is the exact opposite of what I'd intuitively expect.
>   

The Xen PV frame buffer is a little different than the other devices we 
emulate.  The Linux driver provides a linear framebuffer to userspace 
via the fbdev interface.  There is no update interface for fbdev though 
so it periodically zaps the page table mappings in order to do dirty 
tracking which it then uses to send the host update regions.  
Presumably, this idle flag is actually communicated to the guest to let 
it know that it does not need to do dirty tracking for the time being.  
Presumably, this has an observable performance advantage.  I believe the 
refresh rate is pretty high within the guest so I can imagine this being 
significant when using guests with CONFIG_NO_HZ.

Regards,

Anthony Liguori

> Paul
>