FUNCTION-42-RETURN - what SCSI READ FORMAT returns, the 5SCIN state, and whether it skips block 0¶

Byte-verified reverse-engineering of SCSI disk-layer function 42 (READ FORMAT) in SINTRAN III VSX/500 L07: every return value it can hand its caller, the condition for each, the 5SCIN initialization-flag state machine, and whether any of that causes the @ENTER-DIRECTORY ,,DISC-SCSI-1,0 mount to skip the block-0 (page-0 directory) read.

This closes the specific hypothesis that the two companion carves (../SCSI-DRIVER/, ../RCBLO/, ../ENTER-DIRECTORY/) left as "function 42 returns geometry, block-0 is a separate upstream request." Here we prove exactly what function 42 returns and what state it leaves, from the bytes.

Segment: 065-S3SIPIT (load base 32000B; byte-identical twin 066-S3IIPIT).
Source bin: ../../segments/065-S3SIPIT.bin (52224 words).
Disasm: nd100-dis -a -o -b 13312 over the byte-swapped image.
Evidence: VERIFIED = read from L07 bytes (dd offset = (addr-32000B)*2 decimal). INFERRED = from NPL logic / architecture. OPEN = uncarved / runtime. All addresses/values octal.

0. The revision-offset fact that unblocked this (byte-verified)¶

The SCSI disk layer (IP-P2-SCSI-DISK: SCSDI / SCSID / INQUI / function 42) lives in the L07 image at NPL-source-label + 376B:

Symbol	NPL label	L07 (SYMBOL-2-LIST)	delta
SCSID	061621	062217	+376B
SCSDI	-	057215	-

Proof: 062217 - 061621 = 376B. The SCSI driver (IP-P2-SCSI-DRIV: SCLLD=067160, SCINT=067247, SELEC=070165, INITO=070261) is a separate compilation unit and DOES align with its NPL labels - which is why the companion SCSI-DRIVER carve verified the driver at the NPL addresses but its function-42 disassembly (taken at the NPL labels 057074 / 062214 / 062503) decoded to IOF / ION / FAD garbage. Those were the wrong addresses. Adding 376B makes every function-42 instruction below decode coherently and match the NPL.

Byte spot-checks (dd, big-endian words):

L07 addr	byte off	word	mnemonic	meaning
`057472`	22132	`171042`	`SAT 42`	`IF A=42` (function-42 gate)
`057506`	22156	`174075`	`BSET ZRO 70 DA`	CLEAR 5SCIN on entry
`062217`	24862	`006010`	`STA ,X 10`	`SCSID:` entry (confirms +376B)
`063101`	25730	`044423`	`LDA ,B 23`	pre-`BSET ONE 5SCIN` (INQUI done)

1. Function-42 return-value table (VERIFIED)¶

Function 42 hands its caller a return code in T (0 = OK) and, on success, writes geometry into the caller's DMA buffer (SLINK.MEMA1/MEMA2). Internal disk-layer code values are proven at the ERREX->user-error map (062124: IF D=TYPER(1), IF D=ILAOP(4), IF D=BADPA(5)).

`T`	Symbol	Condition (byte site)	Caller-visible effect
0	OK	control record read, HSTAT ok, XOR checksum = 0, `2 < NPART <= NCOPA` (`057716->057747`)	buffer := {data-area size, UHLIM, status word `36`}; op terminates OK
1	TYPER	after inquiry `SUTYP>>8 != 0` = illegal device type (`057552-057555`)	error; mapped to user code `61`
4	ILAOP	illegal operation (`057233`, e.g. non-direct-access)	error; mapped to user code `55`
5	BADPA	bad partition / address out of `UHLIM` bounds (`057401`; partition path `056707+`)	error; mapped to user code `174B` (illegal parameter)
11	NOCRC	control-record XOR `!= 0`, or `NPART<=2`, or `NPART>NCOPA` (`057707-057720`)	error "no control record"
(6 / 13)	UNIT-ATTN / ABORTED	SCSID status `6` or `13`	not returned - retried in-place (`TACOU--`, `GO RETRY`)
PFAIL/SBRST/LIRST	bus reset / power fail	at ERREX (`057240` region)	`MIN TACOU; GO RETRY`

Which one OUR run produces: blockSize=1024, control record valid (NPART=8, XOR =0), UHLIM=121560. Trace: 057472 (A=42) -> preamble -> 057514 RETRY -> automatic INQUIRY + READ CAPACITY (5SCIN was cleared) -> 057557 control-record READ(6) at the last LBA -> 057655 FINEX -> checksum loop passes (057716) -> geometry published (057737-057744) -> 057747 T := 0 (SUCCESS). This is the single READ(6) seen on the wire, and it returns success, not an error.

2. The 5SCIN state machine (VERIFIED)¶

5SCIN = bit 7 of SUTYP (offset 23), "inquiry / geometry initialisation done".

Step	L07 addr	Action on 5SCIN
func-42 entry	`057506` `BSET ZRO 70 DA`	CLEAR (force re-inquiry)
RETRY gate	`057515` `BSKP ZRO 70 DA`	if clear -> run INQUIRY+READ CAPACITY
INQUI success	`063102` `BSET ONE 70 DA`	SET ("INITIALIZATION FINISHED")
after func 42 returns	-	5SCIN = SET
next fn-0 (block-0) read reads it	`063752-063754` `LDA SUTYP; BSKP ONE 70 DA; JMP I 163 -> INQUI`	5SCIN SET -> does NOT divert to INQUI; proceeds to transfer

So: after function 42, 5SCIN is SET. The only consumer of 5SCIN on the next operation is the queue/entry gate at 063752. With 5SCIN SET it takes the fall-through (normal transfer -> CACOB -> EXCOM -> SCLLD -> INITO -> SCWAQ), i.e. it enqueues block 0. It would only divert to INQUI if 5SCIN were CLEAR. 5SCIN cannot cause a block-0 skip - it is left in the state that permits the block-0 read.

3. SCSI vs SMD function 42 - the real divergence¶

	SCSI (`IP-P2-SCSI-DISK`, VERIFIED L07 bytes)	SMD/Winchester (`IP-P2-DISK-START`, INFERRED NPL, uncarved)
Disk I/O	INQUIRY + READ CAPACITY + one control-record `READ(6)`	none
Geometry source	computed from the platter (control record -> `UHLIM`, partitions)	static in-core table: format number `DISPN` from `HTABL` (`054632 *DEPO`)
Return on our run	`T := 0` (OK) + buffer {UHLIM, partitions, status `36`}	`HSTAT = 0` (OK) + buffer[0] = format number
Non-`BDISK` driver	n/a	`054066`: silent no-op, no error
Failure modes	TYPER/ILAOP/BADPA/NOCRC/PFAIL + retries	essentially none (table lookup)

The divergence is NOT the return code (both report success) - it is (a) SCSI does real, failure-prone I/O to learn geometry, and (b) SCSI's output is a disk-derived UHLIM + partition table, whereas SMD's is a static format index. Both then let the mount proceed to the separate block-0 read.

Because SMD's function 42 is a table lookup that cannot fail and needs no follow-up state, the SMD mount proceeds to CHDSI/RXDIR/RCBLO -> block 0 unconditionally. The SCSI path reaches the same block-0 request point; nothing in the SCSI disk layer diverts away from it.

4. The caller branch that decides "issue block-0 vs not"¶

There is no such branch inside the SCSI disk layer or the driver. Verified:

Function 42 (057472-057747) terminates the operation via RETEX -> RETOP with T=0; it neither reads block 0 nor chains a block-0 request.
The block-0 (page-0) directory read is a separate fn-0 request issued by the device-agnostic path: ENDIR -> CHDSI -> RXDIR (037651 sets block 0) -> RCBLO -> JPL I ,B 10 via datafield ,X 14 -> SCSDISK fn 0 (fully carved in ../ENTER-DIRECTORY/ and ../RCBLO/).
When that fn-0 request re-enters the disk layer it hits the 063752 5SCIN gate, which (5SCIN SET) lets it through to SCLLD -> SCWAQ.

So the decision "issue the block-0 read or not" is taken entirely in the uncarved device-agnostic connect/mount overlay (the caller of function 42 and of CHDSI), not in any byte carved here. The exact deciding instruction is therefore OPEN from the SCSI disk-layer bytes - the same boundary the RCBLO/ENTER-DIRECTORY carves reached, now confirmed from the disk-layer side that the layer does not make that decision.

5. THE ANSWER¶

The hypothesis - "SCSI function 42 returns a DIFFERENT value (or leaves a different device state) and the caller BRANCHES on it and skips the block-0 read" - is DISPROVED on both the return-value axis and the 5SCIN-state axis, for our run, from the bytes:

Return value: function 42 returns T = 0 (SUCCESS) - byte-verified at 057747, reached because the control record is valid (XOR =0, NPART=8, UHLIM set). This is the same success sense SMD reports (HSTAT=0). There is no distinct "SCSI error return" for a caller to branch on, because function 42 succeeded.
5SCIN state: function 42 leaves 5SCIN SET (063102). The sole consumer, the fn-0 re-entry gate at 063752, reads 5SCIN and - because it is SET - does not divert or block; it lets the block-0 read through to SCLLD/SCWAQ. A SET 5SCIN enables block 0, it cannot skip it. (A CLEAR 5SCIN would only cause an extra INQUIRY, still not a skip.)

Therefore the block-0 skip is not caused by function 42's return value or its 5SCIN state. The fault is upstream, in the uncarved device-agnostic connect/mount overlay that consumes function 42's output (the disk-derived UHLIM / partition table / status 36) and is responsible for then issuing the block-0 request. Consistent with the ground truth (SCWAQ empty; no block-0 enqueue; silence after the control-record read), that caller never advances to, or is rejected before, the block-0 request. This matches the RCBLO / ENTER-DIRECTORY conclusion and now rules out the disk layer itself as the brancher.

The ONE settling runtime check¶

After the function-42 control-record read completes (T=0), with DAP breakpoints:

Caller advance: break CHDSI 037763B. If never hit after func-42 completion -> the connect/mount overlay never handed off to the page-0 read (it consumed UHLIM/status-36 and stopped or errored). Carve that overlay next - not the SCSI disk layer.
Disk-layer re-entry: break SCSID 062217B and the 5SCIN gate 063752B. If a second SCSID entry with (ABFUN & 077) == 0 (a READ) and block-0 address arrives -> read SUTYP@23 (expect 5SCIN SET) and watch whether it reaches SCLLD 067160B. If it errors first (e.g. BADPA, T=5, from the UHLIM bounds check at 056760), the block-0 address is mis-scaled against the SCSI UHLIM (the caller built a bad LBA). If SCSID is never re-entered, the caller never issued block 0.

The distinguishing read: at each break, dump the func-42 output buffer (SLINK.MEMA1/MEMA2 = UHLIM + status 36) to see whether the caller received sane geometry it could turn into a valid block-0 LBA.

6. VERIFIED / INFERRED / OPEN¶

#	Claim	Verdict
1	Disk layer is at L07 = NPL + 376B (`SCSID 062217` = NPL `061621`+376)	VERIFIED (symbol + dd)
2	func-42 gate `IF A=42` at `057472` (`SAT 42`/`SKP IF DA EQL ST`)	VERIFIED (dd `171042`)
3	func-42 entry CLEARS 5SCIN: `057506 BSET ZRO 70 DA` -> `STA ,X 23`	VERIFIED (dd `174075`)
4	RETRY gate `057515 BSKP ZRO 70 DA` runs INQUIRY when 5SCIN clear	VERIFIED
5	INQUI SETS 5SCIN: `063102 BSET ONE 70 DA` ("INITIALIZATION FINISHED")	VERIFIED (dd `044423` @063101)
6	Single control-record `READ(6)` path `057557-057612` (`X.ABP32=1`, one block)	VERIFIED
7	Success return `T:=0` at `057747`; publishes UHLIM + status `36` (`057737-057744`)	VERIFIED
8	Return-code values OK=0/TYPER=1/ILAOP=4/BADPA=5/NOCRC=11 (ERREX map `062124`)	VERIFIED
9	Our run -> `T=0` (ctrl record valid, XOR=0, NPART=8, UHLIM set)	VERIFIED (bytes) + ground-truth trace
10	After func 42 -> 5SCIN SET; next fn-0 gate `063752` lets block 0 through	VERIFIED
11	SMD func 42 = static format number, no I/O, HSTAT=0	INFERRED (NPL `IP-P2-DISK-START` 054066/054620; different uncarved segment)
12	No branch in the SCSI disk layer decides issue-block-0 vs not	VERIFIED (no such branch in the carved body)
13	Hypothesis (func-42 return/state causes block-0 skip)	DISPROVED for our run (return=0, 5SCIN=SET)
14	Exact deciding instruction (upstream never-issue vs mis-addressed LBA)	OPEN - settled by the section-5 DAP check

Provenance. Carved 065-S3SIPIT L07 bytes (load base 32000B); symbols SINTRAN/NPL-SOURCE/SYMBOLS/L07/SYMBOL-2-LIST.SYMB.TXT (SCSID/SCSDI/SCLLD...) and FILSYS-SYMBOLS/SYMBOL-1-LIST (SUTYP=23, 5SCIN=7, SCDFA, SCWAQ); logic (different revision, INFERRED) SINTRAN/NPL-SOURCE/NPL/IP-P2-SCSI-DISK.NPL (function 42 / INQUI / FINEX / RETEX) and IP-P2-DISK-START.NPL (SMD function 42). Ground-truth trace: INQUIRY + READ CAPACITY + one control-record READ(6), then silence; SCWAQ empty.