On Thu, Apr 14, 2022 at 09:30:21PM -0700, Davidlohr Bueso wrote: > On Thu, 14 Apr 2022, Paul E. McKenney wrote: > > > On Thu, Apr 14, 2022 at 06:09:45PM -0700, Davidlohr Bueso wrote: > > > No idea, however, why this would happen when using qemu as opposed to > > > nbd. > > > > This one is a bit strange, no two ways about it. > > > > In theory, vCPU preemption is a possibility. In practice, if the > > RCU grace-period kthread's vCPU was preempted for so long, I would > > have expected the RCU CPU stall warning to complain about starvation. > > But it still might be worth tracing context switches on the underlying > > hypervisor. > > Right, but both cases are VMs which is what throws me off, regardless > of the zns drive. Or is this not the case, Luis? The nvme controller timeout and rcu splat happens on an 8 core KVM guest. The hypervisor has no issues reported. Does that answer the question? Luis
