On Mar 31, 2006, at 6:43 PM, jonathan@dsg.stanford.edu wrote:

> Close enough, seeing we
> get all (2^64)-1) values.

...and we have precedent in the source tree for ignoring such 64-bit  
(or 63-bit) counter wraps.  From the Alpha pmap:

                 if (cpma->pma_asngen == 0) {
                         /*
                          * The generation number has wrapped.  We could
                          * handle this scenario by traversing all of
                          * the pmaps, and invalidating the generation
                          * number on those which are not currently
                          * in use by this processor.
                          *
                          * However... considering that we're using
                          * an unsigned 64-bit integer for generation
                          * numbers, on non-ASN CPUs, we won't wrap
                          * for approx. 585 million years, or 75 billion
                          * years on a 128-ASN CPU (assuming 1000 switch
                          * operations per second).
                          *
                          * So, we don't bother.
                          */
                         panic("pmap_asn_alloc: too much uptime");
                 }

(Yay, Friday!)

-- thorpej