Comment # 3 on bug 1218455 from Frederic Weisbecker

Handler 8 is NET_RX_SOFTIRQ. This means that CPU is going idle with a
NET_RX_SOFTIRQ softirq pending. This is not supposed to happen in an online CPU
because at this stage if a softirq is pending, ksoftirqd should be pending,
thus need_resched() should be set, and the tick shouldn't be stoppeable.

One possibility though is that the CPU is going offline after the smpboot
kthreads (including ksoftirqd) haven been parked (past
CPUHP_AP_SMPBOOT_THREADS). In that case need_resched() may not be set even if
softirqs are pending. But isn't networking supposed to deal with that somewhow?