Skip to content

MON 124B PRSRV (ForceReserve) - byte-verified carve + the block-0 reserve flow

Carve of MON 124B PRSRV (ForceReserve) in SINTRAN III VSX/500 L07 (running system L-VSX-500, symbols L07), and of the read-modify-write of block 0 that @ENTER-DIRECTORY uses to stamp a directory as "entered", with the exact point where the SCSI path diverges from SMD/Winchester/floppy.

Evidence grades. VERIFIED = re-read from the carved .bin with dd / nd100-dis (byte offset in the .bin = (addr_oct - loadbase) * 2, decimal). INFERRED = reasoned from bytes + manual/NPL. OPEN = crosses into an uncarved low-resident overlay or depends on runtime state. All addresses/values octal.

Deliverables: - PRSRV-124B.ASM - commented disassembly of the PRSRV/RESRV/ EXECC/PRLS/RELES family (from 071-S3SM) + the block-0 WXDIR write-back excerpt. - PRSRV-124B.pseudo.c - readable pseudo-C for PRSRV and the block-0 read -> stamp -> write-back reserve flow.

Cross-linked carves: ../ENTER-DIRECTORY/ (device-agnostic mount path), ../RCBLO/ (cache-block dispatcher + "no page-0 read" analysis), ../SCSI-DRIVER/ (SCSDISK / SCLLD), and narrative ../../../../../../SINTRAN/Filesystem/code-logic/enter-directory.md.

(a) Which overlay holds PRSRV - and the sibling-coherence proof

Dispatch is byte-verified. MON N goes ENT14 -> GOTAB[N]=MFELL -> CALLP -> MCTAB[N] -> worker (the skill's MON model). The monitor-call table is MCTAB @ 005620B in 044-S3IDPIT (load base 4000B). I re-dd'd the four reserve/release slots:

Slot byteoff bytes = addr Symbol (L07 SYMBOL-1-LIST)
MCTAB[122B] 1988 3e 43 037103B RESRV (ReserveResource)
MCTAB[123B] 1990 3e 6e 037156B RELES (ReleaseResource)
MCTAB[124B] 1992 3e 3e 037076B PRSRV (ForceReserve)
MCTAB[125B] 1994 3e 67 037147B PRLS (ForceRelease)

All four land exactly on the L07 SYMBOL-1-LIST addresses (PRSRV=037076, RESRV=037103, PRLS=037147, RELES=037156). VERIFIED.

Which overlay's bytes are these? SYMBOL-1-LIST is the resident-monitor symbol world. I scanned all 60 carved segments at the four sibling addresses (re/segments-ref/); the only segments with byte-identical, non-zero code there are, all with load base 30000B and symbol file SYMBOL-1-LIST:

  • 071-S3SM (System Monitor) <- authoritative resident home
  • 070-S3SSM (System Monitor, save) identical bytes
  • 003-S3CP (Command segment) identical bytes

At 037076..037170 these three are word-for-word equal: PRSRV=044322 RESRV=036776 EXECC=045053 PRLS=142200 RELES=146147 (VERIFIED cross-segment scan). Since the monitor level runs with the System Monitor mapped, 071-S3SM is the overlay; 003-S3CP carries the same bytes, so a carve read out of 003-S3CP is reading the correct bytes. No other overlay (006-S3FS, commoncode, the ERWD/NKSE/U110 segments, ...) produces a coherent reserve family at these addresses - they land mid-loop or on zeros.

Sibling coherence: PRSRV/RESRV/PRLS/RELES sit as a contiguous family (037076, 037103, 037147, 037156) with the shared executor EXECC=037110 between them - a parallel entry cluster, exactly the discriminator the skill requires. VERIFIED.

Did the existing 122B-ReserveResource folder use the wrong overlay? No - it used 003-S3CP, whose bytes are identical to the resident System Monitor 071-S3SM, so the BYTES are correct. Two corrections to its interpretation, though:

  1. It described the whole 037103..037155 block as "the RESRV worker = a command-builder that calls MON 70 twice." That block is really EXECC (037110B), a shared executor; RESRV is the 5-word entry stub in front of it. Minor mislabel, not a wrong overlay.
  2. PRSRV (124B) is NOT the same shape as RESRV. PRSRV@037076 is a 2-word trampoline that does not enter EXECC at all (see (c) below). So the command-builder story does not describe MON 124 PRSRV.

(b) Full dependency tree of MON 124B (routines + addresses + further MONs)

MON 124B  ForceReserve
  -> ENT14 072167B  -> GOTAB[124]=MFELL 072114B -> CALLP 032201B     (MON model)
  -> MCTAB[124B] = 037076B = PRSRV                                    VERIFIED (dd)
       PRSRV @037076B  [071-S3SM, resident System Monitor]
         037076 LDA -56    ; A := MEM[037020] = 000003  (selector const) VERIFIED
         037077 JMP I 10   ; PC := MEM[037107] = 027417B = LEAV2         VERIFIED
       -> LEAV2 027417B    monitor "leave/return" routine  (SYMBOL-1-LIST)
                           BELOW the 30000B monitor window -> body OPEN

  Sibling reserve/release family (co-located; NOT on PRSRV's path, listed so the
  worker set is unambiguous):
    RESRV @037103B (MON 122B)  \
    PRLS  @037147B (MON 125B)   >  fall into the shared executor EXECC @037110B
    RELES @037156B (MON 123B)  /
       EXECC @037110B (== OTRAN)  builds a command line, scans it, and issues:
         037124 MON 70B  --.
         037154 MON 70B  --+->  MCTAB[070B] = 050673B = COMSB            VERIFIED (dd)
                                COMSB = the @-command interpreter worker
                                (family: UECOM 050701, UELOG 050726) [003-S3CP]
       EXECC helpers (low-resident, OPEN):
         037120 JPL I 45 -> [037165] = 004177B  (UCACH/DALTO)
         037157 JPL I 10 -> [037167] = 004116B  (ALTOF/SINAL)

MON calls issued by the reserve family: only MON 70B (COMMND/CallCommand -> COMSB), from the shared executor used by RESRV/PRLS/RELES. PRSRV itself issues no MON call - it loads 3 and leaves via LEAV2. VERIFIED.

Consequence that matters for the mount: in this L07 build MON 124 PRSRV is effectively a pass-through (constant-load + monitor leave); it does not run the command executor and does not raise error 147B. What the 3 encodes and any work inside LEAV2 is OPEN, but the practical result - PRSRV returns cleanly on every device type, SCSI included - is what the failing-mount console confirms (no 147B printed; the mount proceeds past MON 124).

(c) The block-0 read -> stamp -> write-back reserve flow (exact addresses)

Important: this "directory reserve" is a different reservation from MON 124. MON 124 reserves the device unit; the block-0 stamp marks the directory as entered/owned. Bodies are in 006-S3FS (base 26000B); fully carved in ../ENTER-DIRECTORY/ and ../RCBLO/. The flow (all VERIFIED there; WXDIR re-checked here):

ENDIR 0140176   0140252: MON 124 PRSRV (device reserve; passes)
                0140402: JPL I 33 -> [0140435]=037763  CHDSI
CHDSI 037763    0040000: JPL I 143-> [040143]=037643   RXDIR   (READ block 0)
                0040023: JPL I 121-> [040144]=037101   GSIZE   (in-core size)
                stamp + 0040127: JPL I 30 -> [040157]=037702  WXDIR (WRITE block 0)
RXDIR 037643    037651: RADD CLD 0 DD          block number := 0
                037652: JPL I 23 -> [037675]=035766  RCBLO
RCBLO 035766    036101: LDA ,X 14              transfer fn ptr (SCSI=SCSDISK)
                036135: JPL I ,B 10            *** device READ of block 0 ***

Step by step:

  1. READ block 0. RXDIR sets the block number to 0 (037651 RADD CLD 0 DD, VERIFIED) and calls RCBLO, which on a cache miss dispatches the per-device transfer via datafield word ,X 14 (036101 LDA ,X 14 ; 036135 JPL I ,B 10, VERIFIED). This is the read of the directory master block.
  2. Checksum + self-heal (CHDSI, VERIFIED in ENTER-DIRECTORY sec 4): additive sum of ext-info words 1..7 (ADD ,X 0 loop). A bad/zero checksum does not reject - CHDSI zero-fills the 8 words and writes the geometry-derived capacity, then falls through to the stamp.
  3. Owner interlock (040110-040117, VERIFIED): reject only when flag bit15 set AND owner != 0 AND owner != entering system (error 32B/34B, code INFERRED).
  4. STAMP (040121-040127, VERIFIED): owner word := entering system, flag word bit15 := 1. This is the actual "reserve/enter" mark on the directory.
  5. WRITE block 0 back. WXDIR 037702B recomputes the checksum (037714- 037721 ADD ,X 0 loop, 037722-037723 store, VERIFIED here by dd) and writes block 0 via WCBLO (037727 JPL I 25 -> [037754]). Failure raises error 35B "Master block transfer error" (037747 SAA 35, VERIFIED).

(d) THE DIVERGENCE: where SCSI fails vs SMD/Winchester/floppy

MON 124 PRSRV is NOT the divergence. It is a pass-through in this build (section (b)); it raises no 147B and the failing SCSI console never shows 147B. The mount reaches CHDSI.

The divergence is in the block-0 READ (step 1), the same seam the RCBLO carve pinned: on the failing SCSI run the per-device transfer for block 0 is never enqueued - SCLLD (the sole entry that puts work on SCWAQ) is never called for block 0, and SCWAQ stays empty. So RCBLO's 036135 JPL I ,B 10 for block 0 never executed: no page-0 READ and no page-0 WRITE ever hit the wire (the write-back in step 5 is therefore never reached either). On SMD/Winchester/ floppy the same JPL I ,B 10 reaches a driver that DOES enqueue the page-0 read, so the read-modify-write completes and the directory is stamped. VERIFIED (ground-truth SCWAQ-empty) / the exact fault instruction is OPEN from static bytes.

The one runtime check that settles the exact instruction (from ../RCBLO/README.md sec 4): live nd100x/DAP breakpoints at CHDSI 037763B, RXDIR 037643B, RCBLO 035766B. - CHDSI never hit -> the resident mass-storage connect/init overlay (which did INQUIRY + READ CAPACITY + the function-42 control-record READ) never advances to the directory read. Carve that overlay next, not 006-S3FS. - RCBLO hit with block 0 -> single-step from 035766 and see which branch (GSIZE prologue at 035773, geometry check 036103-036117, or a spurious cache HIT 036043-036053) returns before 036135 JPL I ,B 10. That instruction is THE answer.

The factor-2 page(2048B)/block(1024B) shift is REFUTED as the cause here: it would corrupt a read that is issued, not remove it; no read is issued (RCBLO README sec 5).

(e) The "APPROACHING END OF ACCOUNTING FILE" message

Not on the ENTER-DIRECTORY error path. VERIFIED: it is SINTRAN error 243B (163 decimal), set in exactly one place - SINTRAN/NPL-SOURCE/NPL/RP-P2-ACCRT.NPL (243=:ERFL when accounting records pass the warning threshold), inside the RT-accounting collector that runs at logout. ENTER-DIRECTORY never enters the accounting collector, never opens ACCOUNTS:DATA, and never touches ERFL; there is no call edge from the mount path to RP-P2-ACCRT. So the message is either a coincidental logout event or (more likely) the genuine mass-storage read-error code (e.g. 141B transfer error via SCDTS) mis-rendered by the emulator's error-to-string table. Full analysis: ../../../../../../SINTRAN/Filesystem/DEBUG-scsi-enter-directory.md. Verdict: red herring - not raised by the mount itself.

(f) VERIFIED / INFERRED / OPEN

# Claim Verdict
1 MCTAB[124B]=037076=PRSRV, [122B]=037103=RESRV, [123B]=037156=RELES, [125B]=037147=PRLS VERIFIED (dd)
2 Worker bodies resident in System Monitor 071-S3SM (== 070-S3SSM == 003-S3CP bytes), base 30000B VERIFIED (cross-seg scan)
3 PRSRV@037076 = 2-word trampoline: A:=MEM[037020]=3; JMP I 10 -> [037107]=027417B VERIFIED (dd of all 3 words)
4 027417B = LEAV2, a monitor leave/return routine; body below the 30000B window VERIFIED (symbol) / body OPEN
5 PRSRV issues no MON call; is a pass-through (no 147B) on every device incl. SCSI VERIFIED (bytes + no-147B console) / meaning of 3 INFERRED
6 Siblings RESRV/PRLS/RELES fall into shared executor EXECC 037110 which issues MON 70B x2 VERIFIED (dd)
7 MON 70B -> MCTAB[070]=050673=COMSB (@-command interpreter) VERIFIED (dd)
8 EXECC helpers [037165]=004177B (UCACH), [037167]=004116B (ALTOF) VERIFIED (ptrs) / bodies OPEN
9 Block-0 RMW: RXDIR reads blk0 (blk:=0 @037651), stamp @040121-040127, WXDIR writes back @037702 (err 35B @037747) VERIFIED (dd; ../ENTER-DIRECTORY)
10 Directory reserve (block-0 stamp) is SEPARATE from device reserve (MON 124) VERIFIED (distinct code paths)
11 SCSI divergence = block-0 device READ never enqueued (SCWAQ empty); MON 124 is not the fault VERIFIED (ground truth) / exact instr OPEN (DAP settles)
12 Accounting message 243B is not on the mount path (RT-accounting, logout only) VERIFIED

Provenance. 071-S3SM.bin (PRSRV family, base 30000B), 044-S3IDPIT.bin (MCTAB, base 4000B), 006-S3FS.bin (CHDSI/RXDIR/WXDIR, base 26000B), all under tools/sintran-segment-carver/versions/L-VSX-500/segments/; symbols SINTRAN/NPL-SOURCE/SYMBOLS/L07/SYMBOL-1-LIST.SYMB.TXT; manual Reference-Manuals/ND-860228-2-EN SINTRAN III Monitor Calls.md (ForceReserve 124B / ReserveResource 122B); accounting SINTRAN/NPL-SOURCE/NPL/RP-P2-ACCRT.NPL. Disassembly via nd100-dis (WSL, little-endian input).

See also