Hi,
> > It's not different from other input devices. No buffer space -> drop
> > event. What else do you think should happen? We could signal "you lost
> > events" to the guest, but I suspect that buys us nothing. Other input
> > devices don't have that capability, so guests are likely not prepared to
> > handle the situation.
>
> For assigned device input events, how about we don't read events off the
> input device file if there's nowhere to put them?
That'll just offload the problem to the kernel.
> For things like sync that qemu generates, I suspect it's a good idea
> to buffer them in QEMU otherwise guest will get out of sync, right?
Grouping things makes sense indeed. Mouse movements for example are
reported as three events: one for the x axis, one for the y axis, and
the sync. We should report either all three or none of them to the
guest to avoid confusion. I'll think about how to do that best.
> I'm also pretty sure whoever's running the hypervisor does not
> want to see the fprintf.
I'll drop it. Or maybe better make it a tracepoint.
cheers,
Gerd
