Description of problem:

When running offline virtualized container in a long time, the host kernel may cause divide zero crash occasionally when running __cpuacct_get_usage_result.


Version-Release number of selected component (if applicable):


How reproducible:

Running offline virtualized workload, and run `cat cpuacct.proc_stat_show` for a longtime, it may cause the panic or not.

Steps to Reproduce:


Actual results:
Host panic if the offline virtualized container run a long time.

Expected results:
Work normally

Additional info:
This bug is caused by race condition when read per_cpu `kcpustats` variable, and the non-consistent tick_user and tick_guest, make the `tick_user - tick_guest` be negative.