Hello, After a complicated investigation, it turns out that c/s 2529c850ea48 broke xc_vcpu_getinfo().
The bug looks as if it is in vcpu_runstate_get(), which doesn't account for XEN_RUNSTATE_UPDATE and calculating a wildly inappropriate delta. Ultimately, the result of XEN_DOMCTL_getvcpuinfo ends up very occasionally with op->u.getvcpuinfo.cpu_time being wrong by 1 << 63. Given some of the callers of vcpu_runstate_get(), I don't think it is reasonable to pause the VCPU while reading the runstate info. However, it is also unclear whether waiting for XEN_RUNSTATE_UPDATE to drop in vcpu_runstate_get() is safe either. Thoughts? ~Andrew _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
