updated nos on thumper

[karn@thumper.bellcore.com (Phil R. Karn)]

I've put a new version of NOS on thumper under /pub/ka9q/nos. This
version fixes the problem with the time of day clock sometimes not
increasing monotonically.

It turns out that the command to sample the 8254 timer chip's count
and output state isn't reliable; if you happen to hit the chip just
after the count hits zero, you might catch it before the output has
toggled. I fixed this by simply checking for a zero count and
re-reading the count and output state when this happens.

I also incorporated full assembler implementations of the long
division and multiplication routines used for converting ticks into
milliseconds; they are quite a bit faster than the C routines I had
written initially.

There's a "test" command in this version that calls the msclock()
function 40,000 times and makes sure that the output always increases.
I'd appreciate it if people would try this out on their machines and
let me know of any failures. If the code functions properly, you
should see no output, just a pause of a few seconds. Be sure you have
said "isat on" before running this test.

This version also includes an optimization in the interrupt buffer pool
code. If an interrupt handler frees a buffer that seems to have come
from the interrupt pool and the interrupt pool isn't full, it goes right
back on the pool.

Phil