PDP-11 C compiler bug

[utzoo!decvax!ucbvax!menlo70!sri-unix!chico!duke!phs!dennis at Berkeley]

If you're on an 11/34 or 11/60 with 11/45-style floating point,
the problem may not be the segmentation registers; there's a deficiency
in the hardware, although this only occurs when you use an
auto-increment or -decrement floating point instruction
as the trapping instruction (ie, the one that causes the fault).
The problem is that these two 11s don't have
SSR2(?), which records how much each register has
changed during an aborted instruction.
It is impossible in this case to backup and restart
the instruction, and the kernel gives you
a memory fault as a reward.

On our 11/60, this only showed up once or twice
in several years, but then we didn't do too much
number-crunching.  Of course, this problem
shows up no matter what system you run.