To: khh pace wer
Subject: lambda crash
--Text Follows This Line--
This has only happened once, but I am starting a policy of trying to
track down every error.  We should have a mailing list and an archive
for this sort of thing.

While executing the following stream (I've put the instructions in the
order they were executed so you don't have to worry about call and
return addresses.)

  ;this read works OK
	((VMA-START-READ M-2) ADD M-2 A-E)
	(call-if-page-fault ...)
  ;now it modifies the MD
	((IMOD-LOW) A-ALUF)
	((M-TEM) SETZ MD A-BITBLT-TEM)
	((M-TEM1) MD)
	((IMOD-LOW) M-K)
  ;now it writes back the modified MD
	((MD-START-WRITE)
		SELECTIVE-DEPOSIT M-TEM (BYTE-FIELD 0 0) A-TEM1)
	(call-if-page-fault ...)
  ;this jump happens
	(JUMP-GREATER-THAN-XCT-NEXT M-4 (A-CONSTANT 1) BITBLT-INNER-3)
       ((M-4) SUB M-4 (A-CONSTANT 1))

BITBLT-INNER-3
  ;ERROR:  M-1 gets the right thing, but the VMA keeps the same value
  ;        that was written on the first instruction above, namely, the
  ;        value that is still in M-2
	((VMA-START-READ M-1) ADD M-1 A-B)
	(call-if-page-fault ...)
	((IMOD-LOW) DPB M-I OAL-MROT A-ZERO)
  ;Machine hangs here waiting for the MD.  The CSM is in the
  ;READ-REQUEST-ERROR loop.  The VMA, as I said, still contains the
  ;value for the first read, and the location that maps to (on a VCMEM
  ;video buffer) can still be read from the other lambda of the 2x2.
	((A-BITBLT-TEM) (BYTE-FIELD 32. 0) MD)

Ken K. says the bit that says whether the destination field of the
micro-instruction is an A memory address or a combined M/Functional
address might have been been too slow for what ever it is that enables a
write on the VMA.  I tried to determine if any other M function
destination was written, but I can't be sure if I checked everything.
I'm reasonablly sure none of the following were clobbered:

	pdl-pointer, or c-pdl-pointer-[push or pop]
	pdl-index, or c-pdl-index-[push or pop]
	location-counter
	rg-mode
	dp-mode
	stat-counter
	stat-counter-aux
	usp, or micro-stack push or pop
	L1 or L2 maps

The lam program has clobbered the macro-ir on me, so I don't know if it
might have been bad before; however, I have no reason to suspect it.

Call me if you have any questions.

Pace