Skip to content

Major-Number Registry

Pick a character major in 51–67 or 69. Everything below 51 is either stock, claimed by a community driver, or a trap; 70 and above does not exist without editing the array bounds. That one sentence is the whole answer — the rest of this page is the evidence, the other three namespaces nobody tracks, and the two collisions that already exist in the wild.

Amix has no authority that allocates device numbers. A driver "registers" by occupying an index in cdevsw[] or bdevsw[] in master.d/kernel.c and rebuilding the kernel ✅. Every community project picks its own slot and hopes; zz9000-amix's own INSTALL.md tells the reader to "Confirm that it is free in your own cdevsw[]" ✅, and va2000-amix does not mention conflicts at all ✅. This page is the shared artifact that gap calls for. It is descriptive, not binding — no kernel enforces it. It is offered so that two drivers built by two people can be linked into one kernel.

For what a major/minor is and how the switch tables work, see the Amix device-driver model. For the SCSI minor-number encoding and the card list, see Device & card reference.

1. Character majors — cdevsw[]

The stock table has 70 entries, indices 0–69 ✅. CDEVSIZE is sizeof(cdevsw)/sizeof(cdevsw[0]), so a major ≥ 70 requires extending the array — no community driver has done this, and doing so changes cdevcnt, which other code reads. Treat 69 as the ceiling.

Transcribed byte-exact from usr/sys/master.d/kernel.c in two independent trees ✅: the published isoriano1968/hydra-amix tree @ dcb7d24, and this project's own relink tree. A full diff between the two shows five hunks, all of them hydra — so both trees agree on every other row ✅.

Major Occupant Class Status
0 console stock occupied ✅
1 prf stock occupied ✅
2 tty (sy) stock occupied ✅
3 mem stock occupied ✅
4 bb stock occupied ✅
5 sl (streamtab slinfo) stock occupied ✅
6 amiga stock occupied ✅
7 /*7=win?*/ — labelled, no functions stock ⚠ do not use — reserved marker ✅
8 xt stock occupied ✅
9 sxt stock occupied ✅
10 screen stock occupied ✅
11 scsi — gsioctl raw-CDB passthrough (/dev/scsi) stock occupied ✅
12 machid stock occupied ✅
13 ql stock occupied ✅
14 pts stock occupied ✅
15 ptmx stock occupied ✅
16 ct — tape stock occupied ✅
17 fd — floppy (char) stock occupied ✅
18 aen — A2065 LANCE STREAMS/DLPI (&aeninfo) stock occupied ✅ — note: unrelated to block 18 (dd)
19 sld — SLIP DLPI stock occupied ✅
20 loop stock occupied ✅
21 par — parallel (/dev/par) stock occupied ✅
22 tiga — A2410 stock occupied ✅
23 timod stock occupied ✅
24 tirdwr stock occupied ✅
25 log stock occupied ✅
26 sp stock occupied ✅
27 clone stock occupied ✅
28 ticots stock occupied ✅
29 ticotsord stock occupied ✅
30 ticlts stock occupied ✅
31 res stock occupied ✅
32 ip stock occupied ✅
33 tcp stock occupied ✅
34 udp stock occupied ✅
35 rawip stock occupied ✅
36 icmp stock occupied ✅
37 arp stock occupied ✅
38 /*38=fd*/ — labelled but all-ND stock ⚠ avoid — reserved-looking, functionally empty ✅
39 /*39=hd*/ — labelled but all-ND stock ⚠ avoid — same ✅
40 dd — SCSI raw/char disk stock occupied ✅
41 ben stock occupied ✅
42 — — free ✅
43 — — free ✅
44 rkd — only under KERNEL_DEBUGGER stock ⚠ avoid — empty in a normal build, occupied in a debugger build ✅
45 kdebug — only under KERNEL_DEBUGGER stock ⚠ avoid — same ✅
46 audio stock occupied ✅
47 mx (stock contrib) ⚔ hydra (community) — ⚔ COLLISION — see §2.1 ✅
48 random (isoriano1968 dev tree, unpublished) — vacated by z3660eth 2026-07-30 — 🟡 dev-tree prose only; collision resolved — see §2.2
49 zz9000 (isoriano1968) community claimed ✅
50 sad — STREAMS administrative driver stock occupied ✅
51 z3660eth/zen0 (this family) — renumbered from 48, 2026-07-30 community claimed ✅ — live in shipped kernels: mknod /dev/zen0 c 51 0 verified on real hardware 2026-09-09 (A4000 + Z3660, kernel 68060-260828-50, zen0 up at 0 % packet loss). The "48 until the next kernel rebuild" caveat is now discharged
52–67 — — ✅ FREE — verified row-by-row in two trees; take a slot here
68 va2000 (asokero) community claimed ✅
69 — — ✅ FREE — the last slot in the array
70+ does not exist — 🚫 requires extending cdevsw[] ✅

Every row from /*51*/ to /*69*/ reads literally ND,ND,ND,ND,ND,ND,ND,ND,ND,ND,notty,nostr,nullflag, in the published community tree ✅ — slots 48, 49 and 68 are the same empty filler there, because those drivers live in other authors' trees.

2. The two live collisions

2.1 Major 47 — mx vs hydra ⚔

Amix 2.1 ships a driver that already claims 47. /usr/sys/amiga/alien/contrib/mx/ is in the stock distribution (present in two independently obtained trees, files dated 1992-03-21) ✅. It is the Palomax MAX-125 IBM-style serial adapter driver — up to 8 extra serial ports — by an author tagged jck. Its kernel.c.mx says, verbatim ✅:

/* placed in struct cdevsw cdevsw[] = {  at major 47 */

ND,ND,ND,ND,ND,ND,ND,ND,ND,ND,notty,&mxinfo,nullflag,   /* 47 jck MAX-125 */

The modern hydra AmigaNet driver also takes 47 ✅.

If you have installed the MAX-125 driver from your own Amix media, adding hydra will silently overwrite it (or fail to link, depending on which extern wins). mx additionally claims an int2_tbl[] ISR (mxtxintr) and an io_poll[] entry (mxpoll) ✅ — so the collision is not only in cdevsw[].

This is the oldest conflict in the ecosystem and predates every modern driver by three decades. Neither side knows about the other.

The other stock contrib item, contrib/viper.sh (Archive Viper 2150S tape support, Frank J. Edwards, 1991, GPL), claims no major — it only adds ctioctl to the existing /*16=ct*/ row ✅. Recorded here so it need not be re-checked.

2.2 Major 48 — z3660eth vs random — resolved 2026-07-30

This family's z3660eth driver occupied char 48, tag zen, interface zen0 ✅ — chosen in 2026-06 because 48 is a free nostr row on the stock and A4091 kernels, and burned into real-hardware-proven kernels and shipped /dev/zen0 nodes.

Against it, zz9000-amix's integration/kernel.c.txt states ✅:

"The tested development tree uses major 47 for Hydra, 48 for random, 49 for ZZ9000, and 50 for sad. Verify your own table before assigning major 49."

That is the only evidence for 48 = random anywhere — a prose line describing an author's unpublished development tree 🟡. The published hydra-amix tree leaves 48 empty ✅.

Resolution (2026-07-30): z3660eth renumbered off 48 to 51 ✅ — the lowest slot of the verified-free 51–67/69 range — because this project's own convention (§5) says community claims win, and a documented claim — even a weak one — is cheaper to respect than to argue with. The renumber is source-side: kernels and /dev/zen0 nodes built before the next kernel rebuild still register and open 48 ✅. 48 now carries only the unpublished-dev-tree random claim 🟡.

3. Block majors — bdevsw[]

32 entries, indices 0–31 ✅ (BDEVSIZE).

Major Occupant Status
0–15 — free ✅
16 fd — floppy occupied ✅
17 hd occupied ✅
18 dd — SCSI disk occupied ✅ — shared, see below
19 dummydisk occupied ✅
20 ramdisk occupied ✅
21–31 — free ✅

No SCSI host adapter consumes a block major. The A4091 and Z3660 piscsi drivers both register inside major 18 by adding a row to sd.c's scsicard[] table, keyed on AutoConfig product ID ✅. A new SCSI controller driver should do the same — do not take a new block major for a host adapter. Note SDCARDS = 2: only two controllers can be live at once (see composing multi-driver kernels).

4. Filesystem types — vfssw[]

The third namespace, and the one the old registry missed entirely. Filesystems are matched by name at mount time (mount -F cdfs …), so index drift is less dangerous than a cdevsw collision — but nfstype is sizeof(vfssw)/sizeof(vfssw[0]) and the row order is part of the kernel's ABI, so it still needs recording.

Index Type Status
0 BADVFS stock ✅
1 s5 stock ✅
2 ufs stock ✅
3 fifofs stock ✅
4 rfs stock ✅ — conditional on INCLUDE_RFS (defined in the shipped tree)
5 xnamfs stock ✅
6 bfs stock ✅
7 proc stock ✅
8 specfs stock ✅
9 namefs stock ✅
10 fdfs stock ✅
11 nfs stock ✅
12 cdfs (this family) — appended last claimed ✅

✅ Index resolved 2026-07-27: cdfs = 12. The staged master.d/filesys.c defines INCLUDE_RFS unconditionally in the file itself, immediately above the #ifdef guarding the rfs row — so rfs (index 4) always compiles, stock rows are 0–11 (nfs at 11), and the appended cdfs row is 12. No build knob ever turns INCLUDE_RFS off (the kerntools fstype staging only appends the row; it never touches the guard). Confirmed at the binary level: the vfssw tables dumped out of two independently linked kernels during the a3091 bounce investigation are identical row for row, cdfs last at 12 (amix-kerntools/docs/a3091-bounce-boot-patch.md) ✅. The competing figure — "index 11, stock 0–10" in amix-cdfs/docs/packet-c-bench-results.md (since corrected) — was a static miscount that treated the #ifdef INCLUDE_RFS row as not compiled; excluding rfs yields exactly 11/0–10, which is how the discrepancy arose. That doc's load-bearing claim (row appended last, no index shift of stock rows) was true all along. Note the index is load-bearing for nothing: cdfsinit() self-locates its row by name at boot and records the index it finds (cdfs_fstype), and mount matches by name — by design, cdfs works at either index ✅.

New filesystems should likewise append, never insert — inserting renumbers every type above it.

5. AutoConfig product IDs

The fourth namespace. This is where hardware identity collides, and nobody tracks it either. The key is (manufacturer << 16) | product, the argument to autocon() ✅.

ID Board Claimed by
0x02020001 A2090 stock ✅
0x02020003 A2091 stock ✅
0x0202F003 A3000 on-board SCSI stock ✅
0x02020054 A4091 (Zorro III, 53C710) A4091 project / this family ✅
0xc0de0001 A4092 this family ✅
0xc0de0002 A4770 (SCSI.ME) this family ✅
0x08490001 Hydra AmigaNet (2121/1) isoriano1968 ✅
0x144B0001 Z3660 accelerator (eth + piscsi, combo base 0x10000000) shanshe / this family ✅
0x6D6E0001 MNT VA2000 (mfr 0x6D6E, product 1) asokero ✅
0x6D6E0003 MNT ZZ9000, Zorro II product 3 isoriano1968 ✅
0x6D6E0004 MNT ZZ9000, Zorro III identity ⚠ detected and rejected — Amix has no Zorro III bus support ✅

No collisions today. Two boards from the same manufacturer (0x6D6E, MNT) coexist because they carry different product numbers.

6. The other thing that must not collide: init tables

A major is not the only slot a driver claims. Four separate boot-time tables exist in master.d/kernel.c, and community drivers use different ones ✅:

Table Purpose Used by
init_tbl[] boot-time driver init hydra (hydrainit) ✅, zz9000 (zz9000init) ✅
io_init[] late I/O init va2000 (va2000init) ✅
int2_tbl[] level-2 autovector ISRs stock (aciaaintr, jbintr, a2091intr, a3091intr, aenintr), hydra (hydraintr) ✅, mx (mxtxintr) ✅
io_poll[] 16.667 ms poll callbacks stock (qlintr, slpoll), mx (mxpoll) ✅

Record which table your driver hooks alongside its major. z3660eth deliberately hooks none — it autoconfigures on first open() and services RX from a timeout() callout, so a kernel built without the hardware still boots cleanly and open() merely returns ENXIO ✅. That is the friendliest pattern for a shared kernel and the one this project recommends.

7. Conventions for adding a driver

  1. Take 51–67 or 69. Never 7, 38, 39, 44, 45 (traps), never ≥ 70 (off the end of the array).
  2. Respect existing community claims — 47, 48, 49, 68 (48 = an unpublished dev-tree claim; this project vacated it anyway, 2026-07-30). When in doubt, the party who has not shipped hardware-proven kernels renumbers.
  3. A SCSI host adapter takes no new major — add a scsicard[] row under block 18 (§3).
  4. A filesystem appends to vfssw[] — never inserts (§4).
  5. Gate the slot mechanically. Match the free row by its /*<major>*/ comment tag and abort if it is not the nostr/ND filler — do not blind-write the row. This project's harness does exactly that (FATAL: cdevsw slot N is not the free 'nostr' row) ✅; at least one community installer patches by literal comment match with no such check, and will silently no-op if the row was retagged ✅.
  6. Prefer probe-on-open to a boot-time init_tbl[] entry where the hardware allows it (§6).
  7. Record the claim here — the driver admission checklist gate 3 requires it.

8. Claims by project

Project Char Block Other Verified on metal
z3660eth (this family) 51 (renumbered from 48, 2026-07-30 source-side; shipped kernels carry 48 until the next rebuild) — — ✅ A4000 + Z3660, 2026-06 (at 48)
z3660scsi (this family) via 40 via 18 (scsicard[], 0x144B0001) — ✅ 2026-06
a4091 (this family) via 40 via 18 (scsicard[], ×3 IDs) — 🟡 emulated; metal pending
cdfs (this family) — — vfssw[] last row ✅ 2026-07
hydra (isoriano1968) 47 ⚔ — init_tbl[], int2_tbl[] ✅ 2026-06
zz9000 (isoriano1968) 49 — init_tbl[] ✅
random (isoriano1968, unpublished) 48 (uncontested since 2026-07-30) — — 🟡 prose-only
va2000 (asokero) 68 — io_init[] ✅
mx MAX-125 (stock contrib, 1992) 47 ⚔ — int2_tbl[], io_poll[] 🔴 no modern report
viper tape (stock contrib, 1991) none — extends 16 — — 🔴 no modern report

See also

Sources

  • usr/sys/master.d/kernel.c — cdevsw[] (rows 0–69) and bdevsw[] (rows 0–31) read byte-exact from two independent trees: isoriano1968/hydra-amix @ dcb7d24, and this project's relink tree. A full diff between them shows five hunks, all hydra.
  • usr/sys/master.d/filesys.c — vfssw[] rows, read from this project's staged tree with cdfs appended.
  • amix-kerntools/docs/a3091-bounce-boot-patch.md — the 2026-07-26 a3091 investigation's binary vfssw table dumps from two independently linked kernels (cdfs last, index 12) — the measurement that resolved the 11-vs-12 index question.
  • /usr/sys/amiga/alien/contrib/mx/{kernel.c.mx,README} — the stock Amix 2.1 MAX-125 contrib driver's major-47 claim, int2_tbl[] and io_poll[] entries (present in two independently obtained copies of the stock tree; files dated 1992-03-21).
  • /usr/sys/amiga/alien/contrib/{README.install,viper.sh} — the Archive Viper tape contrib (Frank J. Edwards, 1991); confirmed to extend char 16 rather than claim a new major.
  • isoriano1968/zz9000-amix @ 75e2449 — integration/kernel.c.txt (major 49 entry, the 47/48/49/50 dev-tree prose line), INSTALL.md, mkdev.sh (mknod /dev/zz9000 c 49 0), integration/c1-console.txt.
  • asokero/va2000-amix @ fec1c68 — KERNEL_CHANGES.md (cdevsw slot 68, io_init[] entry, mknod /dev/va2000 c 68 0), install-va2000-driver.sh (the /*68*/ literal-match patcher), src/va2000.c (VA2000_PRODUCT = 0x6D6E0001).
  • isoriano1968/hydra-amix @ dcb7d24 — README.md (char 47, tag hya, mknod /dev/hya0 c 47 0, AutoConfig 0x08490001).
  • This project's driver manifests — amix-z3660net/driver.conf (net z3660eth … 51 zen (48 until 2026-07-30)), amix-z3660scsi/driver.conf (0x144B0001 z3660queue), amix-a4091/driver.conf (0x02020054 / 0xc0de0001 / 0xc0de0002) — and the build harness's free-row gate (build-cross-kernel.sh, build-net-kernel.sh).
  • Upstream-watch sweeps #1 (2026-07-26) and #2 (2026-07-27) — the cross-tree claim survey behind this page.