Hi,
The only fix I can think of is allocating a receive buffer inside
the UHCI code for read transfers, and copy the result over to the
guest memory when we are re-scanning the schedule, encounter the
completed td and are going to signal its completion to the guest.
I could take a shot at writing a patch with the proposed fix, but I
was sort of waiting on a reaction on the proposal first.
Gerd what do you think of my proposed fix ?
That should fix it indeed. Yes, for uhci we can live with the overhead,
it's low bandwidth anyway. No, I don't have a better idea. Patch is
very welcome.