more compatibility mode problems - (nf)

[utzoo!decvax!harpo!npoiv!npois!houxm!ihnp4!ixn5c!inuxc!pur-ee!uiucdcs!emrath]

#N:uiucdcs:8200009:000:910
uiucdcs!emrath    Dec 31 06:41:00 1982

I'm trying to bring up some pdp-11 code under unix in compatibility mode.
Debugging can be difficult because of the following problem.
If you do an "rts pc" when the sp points at an odd value, the stack is
not popped into the pc (which would normally be very helpful!),
but the pc is corrupted so that it is difficult to tell what happened.
It appears that what happens is:
	if (pc & 0340) pc |= 0177400; else pc &= 0377;

This is based on the following observed (octal) values:
	    pc before trap    pc after trap
		   052		177452
		   132		177532
		   206		177606
		177422		000022
where the pc before the trap points at an "rts pc" instruction and 35 is on
top of the stack. I also tried about a dozen other values, all of which
behave in the same way. (It took a few to figure out the pattern)
Does anybody know what's going on here?  Is it the vax?
Can anybody tell me where to look in the code for unix?