large ncpus / memory support going in, HEADS UP - master may have some instability

[Matthew Dillon]

    As part of the 48-core support some significant scheduler changes
    have gone into master, so again HEADS UP there may be some instability.

    The new scheduler changes how contended tokens are dealt with.
    Instead of just randomly spinning trying to acquire the token the LWKT
    scheduler now uses a cache-friendly FIFO index queue using atomic
    instructions which time-slots the contending cpus so they only
    spin against the currenet lock holder.  The old algorithm did not scale
    well to 48 cores while the new one scales quite nicely.

    The new algorithm is still a work in progress but is far more flexible
    than the old one and will soon be able to accommodate e.g. MWAIT and
    other hardware features.

						-Matt