Hi, Folks.

I changed pmap.c a bit again.  No functional difference from the
prevous commit (should be).  I'm trying to make other changes to
improve TLB management, however, things repeatedly tell I don't yet
understand the current implementation of NetBSD/mips pmap
thoughoutly...

BTW, I mistakenly stated a wrong figure.

> I'm now modifying several things of NetBSD/mips VM.  I've just changed
> a tiny code segment of TLB flush operation.  The new code does
> preserve all of kernel TLB entries, which map things KSEG2 space,
> whenever TLBPID (ASID) generation number wraps (happens every ~60
> context switchs). 

TLB flush operation is done when TLBPID generation is incremented as
the result of TLBPID (64 in MIPS1) is exhausted.  The assignment of
new TLBPID not necessarily happens at every context switch.  Then, TLB
flushing will be done less frequently than I said previously.

Tohru Nishimura
Nara Institute of Science and Technology