> I'd like to confirm that theo's fix, while it appears to make the machine
> somewhat faster (perhaps because of an overactive imagination on my part) 
> dhrystone test does not show any improvment, I still show dhry-danamic 
> of 8800 on a sparcstation IPC. While IPC's aren't excatly the fastest 
> machines on the planet, I'm fairly sure they're faster than 3/60's. 

Right. Our testing happened on a 4/300, and that changed things a
bit. My patch changes the hint that mmap passes deep into the gory
guts of the vm system, but it turns out that the gory guts ussually
ignore that hint. Therefore once again we end up with mappings which
are un-aligned. And libc runs uncached. Bit performance hit.

When I get more time I'll take another shot at it again. I know what
needs to be done, it's just that I don't know how to do it simply.