128 KB with the Konami mapper WITHOUT SCC (Konami4): sixteen 8 KB banks. There is no register for 0x4000-0x5FFF —bank 0 is fixed there— and the other three windows are chosen by writing the bank number to 0x6000, 0x8000 and 0xA000.
tools/reconocimiento.py measures this on the bytes themselves: not a single write to 0x5000, 0x7000, 0x9000 or 0xB000 —the registers of the mapper with SCC— and all 512 writes to the Konami4 registers carry a bank that obeys the rule, without one exception.
Always the same dance, and every step has its reason:
di while the memory map is half done the interrupt, which
reads from the cartridge, must not come in
ld hl,0xF0F1 the three RAM mirrors, one per window and consecutive
ld a,N the first bank of the trio
ld (0x6000),a to the mapper register
ld (hl),a and noted in its mirror
inc a / inc hl the next bank and the next mirror
ei interrupts can come backThe mirrors at 0xF0F1, 0xF0F2 and 0xF0F3 exist because the mapper cannot be read: writing to 0x6000 sets the bank, but there is no way to ask which one is there. Anyone who needs to put it back has to have written it down.
| banks | what they hold |
|---|---|
| 0 | the core: interrupt, decompressor, painters, the bridges to the data |
| 1, 2, 3, 9, 14 | code |
| 4 to 8, 10 to 13, 15 | data: characters, backdrops, stage scripts, sound |
The data banks never execute an instruction: they are mapped, read and put back. p00:479D is the clean example —it maps banks 12 and 13 only to read four bytes of the table at 0xACBA— and from 0x47E2 onwards there is a long row of bridges doing exactly that. They live in the fixed bank because it is the only place from which you can switch banks without pulling the floor out from under yourself.
The header at 0x4000 is AB with INIT at 0x406A, and that routine does four things before never returning —it ends in a jr to itself—:
RSLREG and keeps the bits of page 1, which is where the cartridge is; with ENASLT it puts that same slot into page 2, so the cartridge spans 0x4000 to 0xBFFF;H.KEYI (0xFD9A) with a jp;At 0x4010 there is another header, CD, which is the one the Konami Game Master reads. It declares thirteen stages —which are not the twenty-four the game has— and gives it the addresses of the things the Game Master knows how to touch: the game data, the score and the routine to jump into.
That routine is arranca_en_la_fase_pedida (p00:40F7), and not one instruction of this cartridge calls it. Its first instruction says whose it is: it reads 0xD31A, which is Game Master RAM —where the stage the user asked for in its MODIFY MODE is written—, and a value of this game would have no business there.
The eight bytes p00:44AE writes from register 7 down to 0 give a layout that is not the one almost any MSX1 game uses:
| where | what |
|---|---|
| 0x0000-0x17FF | colour table (R3 = 0x7F) |
| 0x1800-0x1FFF | sprite patterns (R6 = 0x03) |
| 0x2000-0x37FF | pattern table (R4 = 0x07) |
| 0x3800-0x3AFF | name table (R2 = 0x0E) |
| 0x3B00-0x3B7F | sprite attributes (R5 = 0x76) |
Colours below patterns. R3 and R4 are not addresses but base and mask, and reading them the wrong way round gives correct shapes with striped colours.
At the end of the ROM is the mark Manuel Pazos documented: the catalogue number —RC-743— and the title in katakana, written with a syllabary of its own. tools/marca_konami.py extracts it and checks it. The finding is his; here it is only verified on this cartridge.