On 12/05/2012 05:09 AM, Rusty Russell wrote:

"Srivatsa S. Bhat" [off-list ref] writes:

quoted

From: Michael Wang <redacted>

With stop_machine() gone from the CPU offline path, we can't depend on
preempt_disable() to prevent CPUs from going offline from under us.

Minor gripe: I'd prefer this paragraph to use the future rather than
past tense, like: "Once stop_machine() is gone ... we won't be able to
depend".

Fixed in the new version.

Since you're not supposed to use the _stable() accessors without calling
get_online_cpus_stable_atomic(), why not have
get_online_cpus_stable_atomic() return a pointer to the stable cpumask?
(Which is otherwise static, at least for debug).

We don't have to worry about this now because the new version doesn't
use stable cpumask.
 

Might make the patches messier though...

Oh, and I'd love to see actual benchmarks to make sure we've actually
fixed a problem with this ;)

Of course :-) They will follow, once we have a proper implementation
that we are happy with :-)

Regards,
Srivatsa S. Bhat