Skip to content

SCSI Disc-Transfer Completion Status - why a good read is reported as

"DISC TRANSFER ERROR, STATUS 100020"

Question answered: on @ENTER-DIRECTORY ,,DISC-SCSI-1,0 every SCSI command returns SS_GOOD and the 1024 bytes DMA into ND memory, yet SINTRAN aborts the mount with ERROR 20 / ERROR 21 "Disk transfer error", DISC ADDRESS 0 and 7777B, STATUS 100020B. This document decodes, from the carved SINTRAN L bytes plus the ND-3201 firmware analysis, exactly which status word SINTRAN reads, which bit makes it declare a transfer error, what 100020B means, and precisely what the emulated ND-3201 controller must present on a completed transfer so the driver accepts it.

Grading: VERIFIED = proven by carved SINTRAN L bytes, by the embedded driver register table, or by a reference manual; INFERRED = strong reasoning from the NPL driver revision or firmware doc (NPL is a different revision than L, so its numeric details are not carved-proven); OPEN = not closable statically, needs a live trace.

All ND addresses are octal; status words are given octal and hex.

Carved provenance: segment 006-S3FS, load base 26000B (11264 dec) - ../../../tools/sintran-segment-carver/versions/L-VSX-500/segments/006-S3FS.bin; disassembly + symbols under ../../../tools/sintran-segment-carver/versions/L-VSX-500/re/segments-ref/006-S3FS/. Opcodes grounded in ../../../tools/sintran-segment-carver/versions/L-VSX-500/re/instruction-semantics/ND100-INSTRUCTION-SEMANTICS.md.

Cross-links: scsi-mount-geometry / disk format - SCSI-MOUNT-FIX-PLAN.md - nd-scsi-3201.md - IP-P2-SCSI-DRIV-ANALYSIS.md.

0. Answer in one paragraph

SINTRAN's SCSI driver reads the controller status register RSTAU (IOX offset 04) after the transfer and gates completion on the error-summary bit, RSTAU bit 4 ("Error - OR of error conditions"). The value it read was 100020B = 0x8010 = bit 15 + bit 4. Bit 4 is the error flag - its presence, and nothing else, is what makes the driver reject the transfer and the mass-storage layer print "DISC TRANSFER ERROR, STATUS 100020". Bit 15 is "Differential" - a static board strap, not an error - it is noise in the printed value. The data DMA'd fine, but the emulated controller is presenting RSTAU with the error bit set on a successful transfer. To fix the mount, the emulated ND-3201 must present, on a completed good transfer, RSTAU with bit 4 = 0 and bit 11 = 0 (no error / no bus error), bit 3 = 1 (Ready for Transfer), the NCR-interrupt bit set at completion and cleared only on the RITRG acknowledge, the transfer counter read back as 0, and the Busy bit cleared.

1. Where the status comes from (VERIFIED)

The message text is the standard SINTRAN mass-storage disk-error report. From the reference manuals (../../../Reference-Manuals/ND-60.128.5 EN SINTRAN III Reference Manual.md error table, and ../../../Reference-Manuals/ND-60.050.06 SINTRAN III Users Guide.md):

Error Text Word A Word B
20 Disk transfer error hardware device no. unit
21 Disk transfer error last 16 bit of sector address hardware status

So the printed STATUS = the "hardware status" word = the device status register the driver read = RSTAU. DISC ADDRESS 0 (the block being verified) and 7777B (low 16 bits of the sector address of the probe) match error 21's "last 16 bit of sector address". The 100020B in the message is the raw RSTAU value. (VERIFIED - reference manual + trace.)

The driver and its error translator are carved and on this path (VERIFIED - exact addresses, 006-S3FS.asm):

Symbol Octal Carved line Role
SCSI1 046530B 046530 125070 JMP I 70 SCSI driver entry
SCSI2 046661B 046661 047027 LDA I ,X 27 SCSI driver entry
SCDTS 062107B 062107 044416 LDA ,B 16 status / error translator
SCSID 062217B 062217 000410 STZ ,B 10 driver dispatcher
NCROK 177731B symbol 177731B NCROK(FILSYS-SYMBOLS) NCR-interface status word

OPEN (carved decode limit): the SCDTS (062107B) and SCSID (062217B) regions disassemble as PLANC data-before-code - the linear sweep produces indirect-call jump tables and packed constants (062303 ROP NOOP, 062310 LDT I ,B ,X -40, 062223 JMP 44, 062224 BSKP ONE 170 DA, ...), so the exact internal-code -> user-error-code table (the 232B/141B/252B/... mapping) is not cleanly readable from the current carve. For those numeric mappings the firmware doc and NPL are the fallback authority, marked INFERRED. What is carved-proven: these are the real routines and a page-0 read failure returns through them. This matches the finding already recorded in SCSI-MOUNT-FIX-PLAN.md sec. 1a.

2. The RSTAU status-register format (VERIFIED - embedded driver table)

RSTAU is read at IOX offset 04; the driver writes control at WCONT, offset 05. The authoritative bit layout is the driver's own register definition table carried in the emulator's embedded driver source and reproduced in nd-scsi-3201.md sec. "RSTAU Status Register Bits":

Bit Octal weight Name Meaning
0 000001 Interrupt Enabled echoes WCONT bit 0
2 000004 Busy (Active) controller still processing a command
3 000010 Ready for Transfer controller completed, ready
4 000020 Error OR of error conditions
5 000040 Reset on SCSI bus SCSI bus reset detected (IRQ source)
6 000100 NCR 5386 disabled chip in disabled state
7 000200 Single-ended single-ended SCSI drivers strapped
8 000400 Data Request DREQ from NCR 5386
9 001000 NCR Interrupt interrupt from NCR 5386 (IRQ source)
10 002000 Data Acknowledge DACK to NCR 5386
11 004000 BERROR ND-100 bus DMA error
12 010000 BSY SCSI BSY line
13 020000 REQ SCSI REQ line
14 040000 ACK SCSI ACK line
15 100000 Differential differential SCSI receivers strapped

Note the two distinct error indicators: bit 4 = generic error summary and bit 11 = BERROR = the actual DMA/bus error. The DMA-completion error lives in bit 11, not bit 4.

Discrepancy to be aware of (OPEN): the NPL driver revision (IP-P2-SCSI-DRIV-ANALYSIS.md, lines 123-138) reads RSTAU and tests bit 11 for "interrupt from NCR" and bit 4 for an error, whereas the register table above (closest to L) puts NCR interrupt at bit 9 and BERROR at bit 11. NPL is a different revision than carved L; the two disagree on whether the NCR-interrupt bit is 9 or 11. This does not change the bit-4 finding, but it does mean the completion-interrupt bit position (9 vs 11) should be pinned by a live trace before hard-coding it (see sec. 6).

3. Decode of STATUS 100020B

  100020B  =  0x8010  =  1000 0000 0001 0000b
                          |              |
                          bit 15         bit 4
Bit set RSTAU meaning Verdict
bit 4 (000020B) Error - OR of error conditions THE trigger. This is why SINTRAN declares "DISC TRANSFER ERROR".
bit 15 (100000B) Differential (SCSI receivers strapped) Benign static strap. Not an error, not tested by the completion gate. Noise in the printed word.

What SINTRAN thinks went wrong: the driver does not see a specific cause (BERROR/bit 11 is clear, Data-Request/Data-Ack are clear). It sees only the generic error-summary bit 4 raised, so it takes the error exit, hands SCDTS a device-error code, and the mass-storage layer prints the raw status. In short: "the controller told me the operation ended in error" - with no more detail than the summary bit. (VERIFIED that bit 4 = error summary; INFERRED that this is the exact gate the completion path evaluates, from the NPL revision + the register table - see sec. 4.)

Why bit 15 is a red herring: bit 15 = Differential is a hardware-strap readback that is set on a differential board regardless of any transfer. If the emulator sets it unconditionally, it will always appear in a printed RSTAU. SINTRAN's error gate does not test it, so it neither causes nor prevents the failure - but it does make the printed status look alarming. (INFERRED from the register table; an ND-3201 is normally single-ended, so a real single-ended board would show bit 7 set and bit 15 clear - see sec. 6, item 6.)

4. The completion / error gate in the driver (VERIFIED table + INFERRED code)

VERIFIED: the driver reads RSTAU (IOX 04) as the first step of its SCSI interrupt/status service, and bit 4 is defined as the error summary (sec. 2). The error message is produced from that RSTAU word (sec. 1).

INFERRED (NPL revision, IP-P2-SCSI-DRIV-ANALYSIS.md lines 123-131) - the shape of the gate:

Line 123: T := HDEV+RSTAU; *IOXT        % read device status into A
Line 124: IF X := 64/\A ><0 THEN        % gate mask 64B = bits 2,4,5
Line 126:    IF A BIT 2  GO SCWTI       %   bit 2  -> controller busy, keep waiting
Line 126:    IF A BIT 5  ... SCDIS      %   bit 5  -> SCSI bus reset received
Line 129:    IF A BIT 4  CALL SCIDE     %   bit 4  -> error condition -> error path

The gate mask 64B = 000064B covers bit 2 (Busy) + bit 4 (Error) + bit 5 (Reset). When RSTAU = 100020B, 64B /\ 100020B = 000020B (bit 4 only) which is ><0, so the driver enters the handler, finds bit 2 and bit 5 clear, and falls to the bit-4 error path. That is the "transfer error" exit. (INFERRED: NPL is a different revision, so the exact opcodes/line numbers are not carved-proven; the mask value and bit assignments are consistent with the VERIFIED register table.)

The GOOD vs ERROR decision, stated plainly:

RSTAU after transfer Driver verdict
bit 4 = 0, bit 3 = 1, counter = 0, NCR-int seen+ack'd GOOD - transfer complete, proceed
bit 4 = 1 (any error OR) ERROR - "DISC TRANSFER ERROR", abort
bit 2 = 1 (still Busy) not done - keep polling / eventually time out

100020B fails on the first row's condition because bit 4 = 1.

5. The connect / verify probe - why it aborts before block 0 (VERIFIED geometry

+ INFERRED sequencing)

Between READ CAPACITY and the directory read at block 0, SINTRAN does not go straight to page 0. It runs a connect/verify probe: it reads the highest addressable block (LBA = reported-capacity - 1) to confirm the drive is physically accessible at the capacity it just reported. Only if that probe completes cleanly does it issue the SCSI READ for block 0 (the directory).

  • VERIFIED (disk format): scsi-disk-format.md shows a real ND SCSI disk's top blocks are a disk-parameter block (LBA raw-2) and a defect/reallocation table (LBA raw-1), and that READ CAPACITY reports the RAW last-LBA (e.g. 129311 on SCSI-K). SINTRAN's mount path reads that last block. So the probe target is the top-of-disk table, exactly the block the emulator's trace shows being read (a full 1024-byte DMA) before any block-0 read.
  • The failure (this document's finding): the probe's data does DMA in, but the emulated controller raises RSTAU bit 4 at completion. The driver's post-transfer gate (sec. 4) sees the error summary and aborts at the probe stage - so it never issues the block-0 read. This is precisely the observed behaviour: "SINTRAN never issues a SCSI READ for block 0; it aborts at the connect/verify probe."
  • RAW-capacity crash vs usable-capacity error (VERIFIED symptom, INFERRED cause): with RAW capacity the probe LBA is the true top block and something downstream dereferences past a valid structure -> crash at PC 012466; with a usable capacity the probe still runs but the same bit-4-on-completion makes it the clean "transfer error". Same stage, same root bit (RSTAU bit 4), two surface symptoms.

Handshake the probe requires to proceed: the read must end with RSTAU bit 4 clear (no error), bit 3 set (ready), transfer counter 0, and the completion interrupt delivered and acknowledged - i.e. the identical good-completion contract as any other read (sec. 6). The probe is not special; it is just the first interrupt-driven read on the mount path, so it is the first to expose a completion-status defect.

6. What the emulated ND-3201 must present on a GOOD transfer (the fix contract)

For SINTRAN to accept a completed disc transfer, the controller's IOX-visible state, at the moment SINTRAN reads RSTAU after the transfer, must be:

# Register / state Required value on GOOD completion Grade
1 RSTAU bit 4 (Error) 0 - clear. Do not map the internal "dmaError"/completion onto bit 4 for a successful DMA. This is the single bit that currently breaks the mount. VERIFIED (register table + gate)
2 RSTAU bit 11 (BERROR) 0 - no ND-100 bus DMA error. A genuine DMA fault belongs here, not on bit 4. VERIFIED
3 RSTAU bit 3 (Ready for Transfer) 1 - controller completed and ready. VERIFIED
4 RSTAU bit 2 (Busy/Active) 0 - no longer processing. VERIFIED
5 NCR-interrupt bit set at completion, cleared only on the RITRG read (IOX 54), never on an RSTAU poll present the completion IRQ (level 11) and let it survive RSTAU status polls; ack on RITRG VERIFIED contract; bit position 9 vs 11 is OPEN - the register table says bit 9, the NPL revision tests bit 11. Pin with a live trace (sec. 7). Ties to SCSI-MOUNT-FIX-PLAN.md FIX 1 (do not clear the interrupt flag on the RSTAU case).
6 RSTAU bit 15 (Differential) present per emulated board type. An ND-3201 is single-ended, so ideally bit 15 = 0 and bit 7 (Single-ended) = 1. Harmless either way (not gated), but setting bit 15 makes the printed status misleading. INFERRED
7 Transfer counter RTCM/RTC2/RTCL (IOX 70/72/74) read back as 0 - all requested bytes moved. The driver treats a non-zero residual as an incomplete/failed transfer. INFERRED (firmware doc: NCR transfer-counter mismatch is the controller's own "internal DMA error", POST code 0x1B)
8 NCR interrupt register RITRG (IOX 54) present a completion/OK reason code, cleared by the read (the acknowledge) INFERRED (firmware doc + impl guide)

Concrete controller-side change (RetroCore): in the RSTAU read path of NDBusDiscControllerSCSI.cs (the Register.RSTAU case), the assembled status word must have the error bit (bit 4) clear whenever the transfer completed with all bytes moved. Track down what sets the internal "dmaError"/error flag on this otherwise-successful transfer and stop it from asserting for a completed DMA (a completed 1024-byte move with counter 0 is success). Keep BERROR (bit 11) for real bus faults only. Combine with FIX 1 from the mount-fix plan (acknowledge the NCR interrupt on RITRG, not on RSTAU) so the completion interrupt is not lost during the driver's status polling.

Do not simply mask bit 15 out of the printed value to make the message look clean - the failure is bit 4; bit 15 is cosmetic.

7. What a live trace would pin (for the two OPEN points)

Two items cannot be closed from static carved bytes:

  1. The NCR-interrupt bit position (9 vs 11). Break the ND-100 in the SCSI driver at the RSTAU read on the completion interrupt (SCSI1 046530B / SCSI2 046661B / SCSID 062217B on the mount path) and read the RSTAU value the driver evaluates, plus the immediately following bit test. The bit the branch keys on for "interrupt from NCR" is the one the emulator must set.
  2. The exact error-code translation in SCDTS (062107B). Because that region is PLANC data-before-code in the current carve, break at SCDTS on the failing probe and capture the internal code in T/A and the user-error code it returns. That confirms the 232B/141B mapping without trusting the NPL revision.

For the controller side, the RetroCore [SCSI-TRACE] plan in SCSI-MOUNT-FIX-PLAN.md sec. 4 already logs every IOX read/write with active/intFlag; add the assembled RSTAU value and the transfer-counter readback to that log so the exact bits presented at completion are visible next to what SINTRAN reads.

8. Summary

  • The printed STATUS 100020B is the raw RSTAU value SINTRAN read after the probe transfer. (VERIFIED)
  • 100020B = bit 15 + bit 4. Bit 4 = "Error (OR of error conditions)" is the trigger; bit 15 = "Differential" is a benign strap. (VERIFIED - register table)
  • The driver's completion gate rejects any transfer with RSTAU bit 4 set, regardless of the fact that all data DMA'd and every SCSI command returned SS_GOOD. It aborts at the connect/verify probe (highest-addressable-block read) and never reaches the block-0 directory read. (VERIFIED gate meaning; INFERRED exact opcode sequence, NPL revision.)
  • Fix: the emulated ND-3201 must present, on a completed good transfer, RSTAU bit 4 = 0, bit 11 = 0, bit 3 = 1, bit 2 = 0, the NCR-completion interrupt set and acknowledged only via RITRG, and the transfer counter read back as 0. Stop mapping the internal DMA-completion/error onto bit 4 for a successful move. (VERIFIED contract; NCR-interrupt bit 9-vs-11 OPEN - pin by live trace.)