[comp.sys.mips] MIPS floating-exception/illegal instruction bug?

slevy@poincare.geom.umn.edu (Stuart Levy) (11/30/90)

A few months back someone posted (on comp.sys.mips) a 3-instruction program
to cause a MIPS kernel panic.  I tried it on an Iris; it paniced their
kernel too.  It was something about a branch whose delay-slot was filled
by an illegal floating-point-type instruction, and whose target was
another such instruction, or one causing a floating exception.
I can't remember exactly, and after trying a number of combinations last night
couldn't get anything worse than an "Illegal instruction" core dump.

Anyway, I now need a way to get a crash dump from one of our Irises,
which occasionally misbehaves.  SGI customer support can't tell me any way
to do this, but I think that program might save the day.  Can anyone out
there tell me the magic recipe?

Thanks in advance,

    Stuart Levy, Geometry Group, University of Minnesota
    slevy@geom.umn.edu, (612) 624-1867

wiltse@sgi.com (Wiltse Carpenter) (12/16/90)

In article <1990Nov29.182320.24295@cs.umn.edu> slevy@poincare.geom.umn.edu (Stuart Levy) writes:
>A few months back someone posted (on comp.sys.mips) a 3-instruction program
>to cause a MIPS kernel panic.  I tried it on an Iris; it paniced their
>kernel too.  It was something about a branch whose delay-slot was filled
>by an illegal floating-point-type instruction, and whose target was
>another such instruction, or one causing a floating exception.

Actually, the panic happens when you have a floating point instruction
in the branch delay slot of an illegal branch instruction and the fp
instruction causes an interrupt.  The kernel tries to emulate the
branch and upon seeing that it is not really a legal branch, panics.

>Anyway, I now need a way to get a crash dump from one of our Irises,
>which occasionally misbehaves.  SGI customer support can't tell me any
way >to do this, but I think that program might save the day.  Can
anyone out >there tell me the magic recipe?

This will be fixed in the next major release of Irix, no need to open
another call on it.  Processes exhibiting such behaviour will be sent
the appropriate signal.

	-Wiltse Carpenter