My R3000 is suffered from "kernel touch FPA" symptom repeatedly.

> R3000 processor posts INT5 interrupt requesting the intervension of
> kernel emulation for FP insn(s).  EPC always points the 2nd insn of
> ast() routine which does not make sense to me.  I'm confused about
> what't the exact condition(s) produces FP interrupt on R3000.  A
> possible cause would be a glitch in fp.S which will repeatedly
> generate (it's what I'm watching) INT5 interrupts once it dives into
> an infinite loop.  Can someone out there throw a towel to me? 

I found MIPS kernel runs differently than used to be.  I have been
using an old NetBSD/pmax kernel made last March 3rd (very stable), and
it never posts FPA interrupts when two small FPA exciser processes rush
to grab a single FPA.  On the other hand, recent NetBSD/pmax kernel
posts FPA interrupts in moderate pace, a few per second given CPU
speed.  While my recent change of delayed lazy FPA context switch might
be blamed, why R3000 FPA interrupt behaves differently after all?

== a kernel built this afternoon ==
$ vmstat -i
interrupt         total     rate
softclock         14939        9
softnet            1860        1
serial0            1467        0
ether              3048        1
scsi              13142        8
clock            394946      256
fpu                 114        0
Total            429516      279
$ uptime
12:02AM  up 26 mins, 2 users, load averages: 1.05, 0.54, 0.32

=== old kernel built 3/3 ===
$ vmstat -i
interrupt         total     rate
softclock       6501964        7
softnet         2152899        2
ether           2369416        2
scsi             766167        0
clock          225912335      256
optslot0       63671043       72
optslot1         791943        0
Total          302165767      342
$ uptime
12:03AM  up 10 days,  5:08, 4 users, load averages: 0.23, 0.25, 0.40

Tohru Nishimura