Hi, On 10/08/2012 06:11 PM, Johannes Stezenbach wrote:
<mega snip> > I also wonder if this is not a generic problem
for all emulated hw if the driver uses some timeout?
Yes it likely is, because the emulated timer interrupt will try to make up for time lost when the vm was not running (in either hypervisor or guest mode), while as hw emulation code which has been pre-empted for what ever reason my simply need some more actual running time before it can complete the task as hand, at which point the timeout will trigger sooner then the hw emulation can handle the task ... Note that with the EHCI emulation we also can have the 2 (guest and hw-emulation time) get out of sync in the other way. When we got behind in processing frames on the periodic schedule compared to "real-time" we would catch up too fast, causing the guest to see large jumps in the frindex register, which it does not like. So all of this really is a precarious balancing act, and to get back to the subject of the thread, the latency spikes some of the kernels debugging options can cause, can upset the balance... Regards, Hans
