(XS) RFR: 8056183: os::is_MP() always reports true when NMT is enabled

[Aleksey Shipilev]

On 09/09/2014 09:50 AM, David Holmes wrote:
> http://cr.openjdk.java.net/~dholmes/8056183/webrev/

I would think this approach sets us up for the subtle bugs by accepting
"false" as the legitimate result in the multi-threaded system, even if
it is transient. I think we need to special-case NMT (e.g. make it use
the non-isMP-predicated Atomics) and leave the general case alone and
protected by the assert.

Or, can we initialize _processor_count more eagerly?

Thanks,
-Aleksey.