On 11/06/12 15:08, Hans de Goede wrote: > This patchsets move interrupt handling for usb-redir and usb-hid devices > over from polling to async handling. Together with a patch to allow > async stepdown in ehci when the periodic schedule only contains interrupt > queues, and a patch to allow usb-hid devices to connect to ehci, this > fixes the usb-tablet causing a crazy cpu load for idle guests. > > With this patchset the cpuload for a fully idle vm with usb-tablet > is lowered from 20% to 2-3% (on my laptop).
Not looked at it (yet) in detail, just merged quickly into a test branch, then found it blowing up in an assert with a win7 guest when moving the mouse for the first time after boot. So something is fishy in the hid patches I guess ... cheers, Gerd
