> If that is the case, the tolerance level might be the better approach > after all...
We should also take a look at mce_start():
int cpus = num_online_cpus();
...
/*
* Wait for everyone.
*/
while (atomic_read(&mce_callin) != cpus) {
since offline cpus will still show up to rendezvous ... perhaps
"num_present_cpus()"
is the right number??
-Tony
