If it takes more than the proposed 5 secs, why bother -- your
system is dead anyway ;-)

Hmm.. ok.. I am being parnoid, I just don't trust the FW guys to leave a processor off line that may suck :)
How SLOF make sure the processor is good? does it?

Only the service processor can completely shut down a CPU, on
our blades anyway.  So the best that SLOF can do is hope for
the best.  The secondary-CPU startup code has been engineered
to be pretty much 100% failure proof though.  And a dead CPU
won't be shown in the device tree.

I guess I'm looking for a way to decide that things are bad.

There's one way we are bad right now: we require all secondary
cpu's to be started, they are not idle but running in a spinloop
somewhere.  Cache invalidate those addresses and hell breaks
loose ;-)

Not a real problem in practice, which is why I still didn't fix it.


Xen-ppc-devel mailing list

Reply via email to