Article 171315 of comp.os.vms:
Andy Goldstein <Goldstein@star.zko.dec.youknowwhere> writes:

> The memory management code was written by Peter Lipman. Peter was well
> aware of how performance-critical the page fault code would be, so it's
> devilishly tight. It's still there on the VAX. One of the reasons it's
> held up so well is very few people had the courage to go near it. 

Yeah, that's exactly the impression I got. Did he alos write the parts of the
linker that finally got revamped (more like rewritten, I'd guess) for the
Alpha port?

> If you reported the PFN map bug to us it oughta be fixed. If you give me
> some idea of what it was about I can check.

It definitely was reported, but I actually doubt it was fixed, because it only
showed up in arcane circumstances whose likelihood was rapidly dropping at the
time. To boot: If you PFN map a section, the MMG code forgot to increment the
reference counter on the page table. The machine this happened on was very
tight for memory, and the process was subjected to a dead page table scan,
which of course decided to remove that page table from the working set. Then,
that interface device to the physics experiment the thing was driving came
alive (I think they were using the CONNINT driver), the programm wanted to
look at the PFN mapped page...and the page fault code looked at the page
tables and PFN database and didn't much like what it saw. (If you're looking
for an old SPR: Physics Institute, University of Bonn.)

That was _fun_.

	Jan