On Thu, Oct 07, 2010 at 06:20:53PM +0200, Avi Kivity wrote:
> On 10/07/2010 06:03 PM, Gleb Natapov wrote:
> >> >
> >> > Isn't SET_USER_MEMORY_REGION so slow that calling it 2^32 times
> >> > isn't really feasible?
> >>
> >> Assuming it takes 1ms, it would take 49 days.
> >>
> >We may fail ioctl when max value is reached. The question is how much slot
> >changes can we expect from real guest during its lifetime.
> >
>
> A normal guest has a 30 Hz timer for reading the vga framebuffer,
> multiple slots. Let's assume 100 Hz frequency, that gives 490 days
> until things stop working.
>
And reading vga framebuffer needs slots changes because of dirty map
tracking?