pmu: non-sense kernel utilisation metric under smp

[KurtWu10]

Under the SMP configuration, the reported kernel utilisation metric in the networking benchmark can be wrong, resulting in unsigned integer underflows (please check the SMP 4-core udp benchmark under the 11/25 multicore (maaxboard) tab).

I suspect this is due to a data race issue with the global ksEnter variable in the kernel. This variable may be updated out of the big kernel lock, e.g. here.

A potential solution is available here, though I am not sure whether this has solved all data race issues.

EDIT: The scope of the issue is larger and also includes metrics in Total utilisation details.