2017-06-27 14:28+0200, Paolo Bonzini: > On 27/06/2017 14:23, Wanpeng Li wrote: >>>>> I have considered single_task_running() before. But since there is no >>>>> such paravirtual interface currently and i am not sure whether it is a >>>>> information leak from host if introducing such interface, so i didn't do >>>>> it. Do you mean vcpu_is_preempted can do the same thing? I check the >>>>> code and seems it only tells whether the VCPU is scheduled out or not >>>>> which cannot satisfy the needs. >>>> Can you help to answer my confusion? I have double checked the code, but >>>> still not get your point. Do you think it is necessary to introduce an >>>> paravirtual interface to expose single_task_running() to guest? >> >> I think vcpu_is_preempted is a good enough replacement. >> For example, vcpu->arch.st.steal.preempted is 0 when the vCPU is sched >> in and vmentry, then several tasks are enqueued on the same pCPU and >> waiting on cfs red-black tree, the guest should avoid to poll in this >> scenario, however, vcpu_is_preempted returns false and guest decides >> to poll. > > ... which is not necessarily _wrong_. It's just a different heuristic. Right, it's just harder to use than host's single_task_running() -- the VCPU calling vcpu_is_preempted() is never preempted, so we have to look at other VCPUs that are not halted, but still preempted. If we see some ratio of preempted VCPUs (> 0?), then we stop polling and yield to the host. Working under the assumption that there is work for this PCPU if other VCPUs have stuff to do. The downside is that it misses information about host's topology, so it would be hard to make it work well.
