------------Picnic Paranoia------------ A 4am crack 2019-01-26 -------------------. updated 2020-06-24 |___________________ Name: Picnic Paranoia Genre: action Year: 1982 Publisher: Synapse Software Platform: Apple ][+ or later Media: 5.25-inch disk Sides: 1 OS: custom ~ Chapter 0 In Which Various Automated Tools Fail In Interesting Ways COPYA immediate disk read error Locksmith Fast Disk Backup unable to read any track EDD 4 bit copy (no sync, no count) read errors on tracks $13-$22 copy just hangs on boot Copy ][+ nibble editor Tracks appear to have 4-4 encoded nibbles, no standard sectors or structure Why didn't COPYA work? not a 16-sector disk Why didn't Locksmith FDB work? not a 16-sector disk Why didn't my EDD copy work? I don't know. Half tracks, maybe? The disk sounds like it's seeking to successive tracks very quickly, so maybe it's doing a spiral-y kind of thing. Hard to tell just on sound alone though. On the bright side, the game is single- load; once it's loaded, it never uses the disk again. (There is no high score saving and no reload when a game ends.) I think this will be one of those "capture the game in memory and rebuild it from the ground up" cracks. Which probably means they did everything in their power to prevent that. Next steps: 1. Trace bootloader 2. Capture game code in memory 3. Write game to a standard disk and build a bootloader to load it 4. Declare victory (*) (*) go to the gym ~ Chapter 1 In Which We Brag About Our Humble Beginnings I have two floppy drives, one in slot 6 and the other in slot 5. My "work disk" (in slot 5) runs Diversi-DOS 64K, which is compatible with Apple DOS 3.3 but relocates most of DOS to the language card on boot. This frees up most of main memory (only using a single page at $BF00..$BFFF), which is useful for loading large files or examining code that lives in areas typically reserved for DOS. [S6,D1=original disk] [S5,D1=my work disk] The floppy drive firmware code at $C600 is responsible for aligning the drive head and reading sector 0 of track 0 into main memory at $0800. Because the drive can be connected to any slot, the firmware code can't assume it's loaded at $C600. If the floppy drive card were removed from slot 6 and reinstalled in slot 5, the firmware code would load at $C500 instead. To accommodate this, the firmware does some fancy stack manipulation to detect where it is in memory (which is a neat trick, since the 6502 program counter is not generally accessible). However, due to space constraints, the detection code only cares about the lower 4 bits of the high byte of its own address. Stay with me, this is all about to come together and go boom. $C600 (or $C500, or anywhere in $Cx00) is read-only memory. I can't change it, which means I can't stop it from transferring control to the boot sector of the disk once it's in memory. BUT! The disk firmware code works unmodified at any address. Any address that ends with $x600 will boot slot 6, including $B600, $A600, $9600, &c. ; copy drive firmware to $9600 *9600 Now the entire contents of the stack are preserved at $2100..$21FF. ; reconnect my work disk DOS (it got ; disconnected by the jump to $FF59) *3D0G ; save the copy of the decrypted code- ; on-the-stack ]BSAVE OBJ.0100-01FF,A$2100,L$100 The stack pointer was set to #$9C, then we pushed #$7A bytes onto the stack, which moves the stack pointer down to #$22. The "RTS" at $081A moves the stack pointer up one byte, treats the next two bytes as an address, increments that address by 1, and jumps to it. Thus... *2123.2124 2123- 26 01 ...means we are "returning" to $0127. Piece of cake. ~ Chapter 3 In Which It Is Not, In Fact, A Piece of Cake $0127 is part of the code we just decrypted and pushed onto the stack. (Of *course* you can run code from the stack page. It's just a normal page in normal memory.) That code is in memory now at $2127. *2127L ; show hi-res page 1 2127- 2C 50 C0 BIT $C050 212A- 2C 52 C0 BIT $C052 212D- 2C 57 C0 BIT $C057 ; ($00) points to $0400 (text page, ; so maybe an address to load the next ; stage?) 2130- A9 04 LDA #$04 2132- 85 01 STA $01 2134- A9 00 LDA #$00 2136- 85 00 STA $00 ; $2B holds the boot slot x16 2138- A6 2B LDX $2B ; reset the disk drive latch 213A- BD 8C C0 LDA $C08C,X 213D- BD 8E C0 LDA $C08E,X ; maybe a counter? 2140- A9 03 LDA #$03 2142- 85 03 STA $03 ; look for a non-standard nibble ; prologue "D7 AA 97 97 AA AA" 2144- A0 00 LDY #$00 2146- BD 8C C0 LDA $C08C,X 2149- 10 FB BPL $2146 214B- C9 D7 CMP #$D7 214D- D0 F7 BNE $2146 214F- BD 8C C0 LDA $C08C,X 2152- 10 FB BPL $214F 2154- C9 AA CMP #$AA 2156- D0 F3 BNE $214B 2158- BD 8C C0 LDA $C08C,X 215B- 10 FB BPL $2158 215D- C9 97 CMP #$97 215F- D0 EA BNE $214B 2161- BD 8C C0 LDA $C08C,X 2164- 10 FB BPL $2161 2166- C9 96 CMP #$96 2168- D0 E1 BNE $214B 216A- BD 8C C0 LDA $C08C,X 216D- 10 FB BPL $216A 216F- C9 AA CMP #$AA 2171- D0 D8 BNE $214B 2173- BD 8C C0 LDA $C08C,X 2176- 10 FB BPL $2173 2178- C9 AA CMP #$AA 217A- D0 CF BNE $214B ; read 4-4-encoded data 217C- BD 8C C0 LDA $C08C,X 217F- 10 FB BPL $217C 2181- 2A ROL 2182- 85 02 STA $02 2184- BD 8C C0 LDA $C08C,X 2187- 10 FB BPL $2184 2189- 25 02 AND $02 ; store it in ($00), which is $0400, ; the text page 218B- 91 00 STA ($00),Y 218D- C8 INY 218E- D0 EC BNE $217C ; increment target page 2190- E6 01 INC $01 ; decrement sector count 2192- C6 03 DEC $03 ; branch ahead if done 2194- F0 03 BEQ $2199 ; branch back if more 2196- D0 E4 BNE $217C ; what 2198- 4C 4C 00 JMP $004C 219B- 04 ??? Oh, ha ha, the "BEQ" at $0194 doesn't branch to $0198, it branches to $0199, which is in the "middle" of the listed instruction. This is a standard obfuscation technique to make it more difficult to do the very thing I am now attempting to do -- list the boot code in the monitor and understand it. *2199L 2199- 4C 00 04 JMP $0400 And that's where I get to interrupt the boot. But how? Well, the decrypted code that runs from the stack page is entirely self-contained. After the initial entry (via the stack, using the "RTS" to jump to $0127), the code uses nothing but relative branches until it finally jumps to $0400 (at $0199). So I could copy the entire decrypted code anywhere in memory and it would still work. ; set up boot trace *9600 Poking around in memory, I can confirm that the game loads into $0800..$93FF. Everything above that is still the unusual byte ($FD) I put there myself. ; Save lower part of game code into ; higher memory (my work disk loads DOS ; directly into the language card, so ; this part of main memory is unused ; during boot) *9800<800.1FFFM ; reboot to my work disk *C500G ... ]CALL -151 ; restore the lower part of the game ; code *800<9800.AFFFM ; save it all at once *BSAVE OBJ,A$800,L$8C00 Now let's see what's going on at $8015. *8015L ; copy some stuff into lower memory 8015- A2 00 LDX #$00 8017- BD 00 81 LDA $8100,X 801A- 9D 00 04 STA $0400,X 801D- BD 00 82 LDA $8200,X 8020- 9D 00 05 STA $0500,X 8023- BD 00 83 LDA $8300,X 8026- 9D 00 06 STA $0600,X 8029- BD 00 84 LDA $8400,X 802C- 9D 00 07 STA $0700,X 802F- E8 INX 8030- D0 E5 BNE $8017 8032- A2 00 LDX #$00 8034- BD 00 85 LDA $8500,X 8037- 9D 00 02 STA $0200,X 803A- BD 00 86 LDA $8600,X 803D- 9D 00 03 STA $0300,X 8040- E8 INX 8041- D0 F1 BNE $8034 ; continue elsewhere 8043- 4C 01 90 JMP $9001 *9001L ; trash the previous stage of the boot ; that we just came from (this is all ; to frustrate memory capture cards, ; by the way -- it makes it more ; difficult to know which parts of ; memory contain important data and how ; it all got there) 9001- A2 00 LDX #$00 9003- AD 70 C0 LDA $C070 9006- 5D 00 F8 EOR $F800,X 9009- 9D 00 80 STA $8000,X 900C- 9D 00 81 STA $8100,X 900F- E8 INX 9010- D0 F1 BNE $9003 ; and exit gracefully 9012- 60 RTS Wait, what? Returning to what? Where's the game? This "RTS", like any "RTS", returns to the next thing on the stack. What's on the stack at this point? If you recall (and it's totally cool if you don't), when we first decrypted the bootloader to the stack, we "returned" to $0127. Now we're "returning" to the next address on the stack. Hey, good thing I saved that piece earlier. *BLOAD OBJ.0100-01FF *2125.2126 2125- FA 78 The next address on the stack was part of the decryption-onto-the-stack that happened a loooong time ago. (I timed the boot process in an emulator; it was about 5 million CPU cycles ago.) Now we're finally "returning" to $78FB to start the game. *78FBL 78FB- A9 01 LDA #$01 78FD- 85 B4 STA $B4 78FF- 85 8F STA $8F 7901- A0 00 LDY #$00 7903- 84 7C STY $7C 7905- 84 7B STY $7B 7907- 20 C3 45 JSR $45C3 790A- 20 B0 5E JSR $5EB0 790D- 20 03 0C JSR $0C03 I'm not sure what that is, but at least it's valid code. Let's see if I really understand what's going on. *300:A9 78 48 A9 FA 48 4C 15 80 *300L 0300- A9 78 LDA #$78 0302- 48 PHA 0303- A9 FA LDA #$FA 0305- 48 PHA 0306- 4C 15 80 JMP $8015 ...games loads and runs... Whew. Now, how to reconstruct this game on a standard disk. The entire game fits in main memory, and not even all of it -- there's room for a standard DOS 3.3 to load this game from a file. We could write a little pre-load hook to set up the stack, then jump to $8015. Or we could patch the code at $9001 so it jumps to $78FB instead of doing an "RTS". The whole boot process wouldn't take more than 15 seconds or so. Or we could write a fastloader and load the entire game in 2 seconds. So obviously, we're going to do that. First, I wrote a little write loop that writes all the game code onto tracks $01-$09 (but, like, on real tracks, not consecutive half tracks), in a standard format. It assumes track $01 data is at $0800, track $02 at $1800, &c. It's slow because it's writing sectors in increasing order, but this is not how the bootloader will read them. (It will read them much faster, as we'll see in a minute.) *9800L 9800- A9 90 LDA #$90 9802- 85 FF STA $FF 9804- A9 00 LDA #$00 9806- 85 FE STA $FE 9808- A9 98 LDA #$98 980A- A0 88 LDY #$88 980C- 20 D9 03 JSR $03D9 980F- E6 FE INC $FE 9811- A4 FE LDY $FE 9813- C0 10 CPY #$10 9815- D0 07 BNE $981E 9817- A0 00 LDY #$00 9819- 84 FE STY $FE 981B- EE 8C 98 INC $988C 981E- 98 TYA 981F- 8D 8D 98 STA $988D 9822- EE 91 98 INC $9891 9825- C6 FF DEC $FF 9827- D0 DF BNE $9808 9829- 60 RTS *9888.9897 9888- 01 60 01 00 01 00 FB F7 9890- 00 08 00 00 02 00 00 60 *BSAVE WRITER,A$9800,L$C0 [insert a blank disk into slot 6] *9800G ...write write write... Piece of... well, you know. ~ Chapter 6 0boot Once upon a time, I wrote a little thing called 4boot. It was fast and small and I was more than a little bit proud of it. The boot1 code was a mere 742 bytes and fit in $BD00..$BFFF. Then qkumba did that thing he does, and now it fits in zero page. With his blessing, I present: 0boot v3. 0boot lives on track $00, just like me. Sector $00 (boot0) reuses the disk controller ROM routine to read sector $0E (boot1). Boot0 creates a few data tables, copys boot1 to zero page, modifies it to accomodate booting from any slot, and jumps to it. Boot0 is loaded at $0800 by the disk controller ROM routine. ; tell the ROM to load only this sector ; (we'll do the rest manually) 0800- [01] ; The accumulator is $01 after loading ; sector $00, or $03 after loading ; sector $0E. We don't need to preserve ; the value, so we just shift the bits ; to determine whether this is the ; first or second time we've been here. 0801- 4A LSR ; second run -- we've loaded boot1, so ; skip to boot1 initialization routine 0802- D0 0D BNE $0811 ; first run -- increment the physical ; sector to read (this will be the next ; sector under the drive head, so we'll ; waste as little time as possible ; waiting for the disk to spin) 0804- E6 3D INC $3D ; X holds the boot slot (x16) -- ; munge it into $Cx format (e.g. $C6 ; for slot 6, but we need to accomodate ; booting from any slot) 0806- 8A TXA 0807- 20 7B F8 JSR $F87B 080A- 09 C0 ORA #$C0 ; push address (-1) of the sector read ; routine in the disk controller ROM 080C- 48 PHA 080D- A9 5B LDA #$5B 080F- 48 PHA ; "return" via disk controller ROM, ; which reads boot1 into $0900 and ; exits via $0801 0810- 60 RTS ; Execution continues here (from $0802) ; after boot1 code has been loaded into ; $0900. On real Apple hardware, the Y ; register is always 0 at $0801, but it ; turns out the CFFA 3000 firmware does ; not always match this behavior -- ; which is exactly the sort of bug that ; qkumba enjoys(*) uncovering -- so we ; initialize Y here (to 1, which is the ; value of the accumulator after the ; drive firmware loaded physical sector ; $03 and we performed an LSR). 0811- A8 TAY (*) not guaranteed, actual enjoyment may vary ; munge the boot slot, e.g. $60 -> $EC ; (to be used later) 0812- 8A TXA 0813- 09 8C ORA #$8C ; Copy the boot1 code from $0901..$09FF ; to zero page. ($0900 holds the 0boot ; version number. This is version 3. ; $0000 is initialized later in boot1.) 0815- BE 00 09 LDX $0900,Y 0818- 96 00 STX $00,Y 081A- C8 INY 081B- D0 F8 BNE $0815 ; There are a number of places in boot1 ; that need to hit a slot-specific soft ; switch (read a nibble from disk, turn ; off the drive, &c). Rather than the ; usual form of "LDA $C08C,X", we will ; use "LDA $C0EC" and modify the $EC ; byte in advance, based on the boot ; slot. $00E4 is an array of all the ; places in the boot1 code that need ; this adjustment. 081D- C8 INY 081E- B6 E0 LDX $E0,Y 0820- 95 00 STA $00,X 0822- D0 F9 BNE $081D ; munge $EC -> $E0 (used later to ; advance the drive head to the next ; track) 0824- 29 F0 AND #$F0 0826- 85 CB STA $CB ; munge $E0 -> $E8 (used later to ; turn off the drive motor) 0828- 09 08 ORA #$08 082A- 85 D3 STA $D3 ; push sector interleave array to the ; bottom of the stack (by setting the ; stack pointer to #$0F and pushing ; #$10 bytes, those bytes will end up ; in $0100..$010F) 082C- A2 0F LDX #$0F 082E- 9A TXS 082F- BD 96 08 LDA $0896,X 0832- 48 PHA 0833- CA DEX 0834- 10 F9 BPL $082F For reference, this is the sector interleave array: 0896- .. .. .. .. .. .. 00 07 0898- 0E 06 0D 05 0C 04 0B 03 08A0- 0A 02 09 01 08 0F ; push the final game entry point so ; the "RTS" in the game code (at $9012) ; will return to the proper place 0836- A9 78 LDA #$78 0838- 48 PHA 0839- A9 FA LDA #$FA 083B- 48 PHA ; push the penultimate game entry point ; (we will return to this after our ; own boot code is done, to transfer ; control to the game) 083C- A9 80 LDA #$80 083E- 48 PHA 083F- A9 14 LDA #$14 0841- 48 PHA ; push several addresses to the ; stack (more on this later) 0842- A2 06 LDX #$06 0844- B5 DA LDA $DA,X 0846- 48 PHA 0847- CA DEX 0848- D0 FA BNE $0844 ; number of tracks to load (x2) -- ; this game uses 9 tracks 084A- A0 12 LDY #$12 ; loop starts here 084C- 8A TXA ; the carry was set by the "LSR" at ; $0801, so we won't take this branch ; the first time (but, as we will see ; shortly, the carry gets flipped off ; and on, and we end up taking this ; branch every second time through the ; loop) 084D- 90 03 BCC $0852 ; X is 0 going into this loop, and it ; never changes, so A is always 0 too. ; So this will push $0000 to the stack ; (to "return" to $0001, which reads a ; track into memory) 084F- 48 PHA 0850- 48 PHA ; There's a "SEC" hidden here (because ; it's opcode $38), but it's only ; executed if we take the branch at ; $084D, which lands at $0852, which is ; in the middle of this instruction. ; Otherwise we execute the compare, ; which clears the carry bit because A ; is always #$00 at this point. So the ; carry flip-flops between set and ; clear, so the BCC at $084D is only ; taken every other time. Please clap. 0851- C9 38 CMP #$38 ; Push $00B6 to the stack, to "return" ; to $00B7. This routine advances the ; drive head to the next half track. 0853- 48 PHA 0854- A9 B6 LDA #$B6 0856- 48 PHA ; loop until done 0857- 88 DEY 0858- D0 F2 BNE $084C Because of the carry flip-flop, we will push $00B6 to the stack every time through the loop, but we will only push $0000 every other time. The loop runs for twice the number of tracks we want to read, so the stack ends up looking like this: --top-- $00B6 (move drive 1/2 track) $00B6 (move drive another 1/2 track) $0000 (read track into memory) $00B6 \ $00B6 } second group $0000 / $00B6 \ $00B6 } third group $0000 / . . [repeated for each track] . $00B6 \ $00B6 } final group $0000 / $FE88 (IN#0, pushed at $0846) $FE92 (PR#0, pushed at $0846) $00D1 (turn off drive motor) $8014 (game 1st entry point) $78FA (game 2nd entry point) --bottom-- Boot1 reads the game into memory from tracks $01-$09, but it isn't a loop. It's one routine that reads a track and another routine that advances the drive head. We're essentially unrolling the read loop on the stack, in advance, so that each routine gets called as many times as we need, when we need it. Like dancers in a chorus line, each routine executes then cedes the spotlight. Each seems unaware of the others, but in reality they've all been meticulously choreographed. Because of the carry flip-flop, we will push $00B6 to the stack every time through the loop, but we will only push $0000 every other time. ~ Chapter 7 6 + 2 Before I can explain the next chunk of code, I need to pause and explain a little bit of theory. As you probably know if you're the sort of person who reads this sort of thing, Apple II floppy disks do not contain the actual data that ends up being loaded into memory. Due to hardware limitations of the original Disk II drive, data on disk must be stored in an intermediate format called "nibbles." Bytes in memory are encoded into nibbles before writing to disk, and nibbles that you read from the disk must be decoded back into bytes. The round trip is lossless but requires some bit wrangling. Decoding nibbles-on-disk into bytes-in- memory is a multi-step process. In "6-and-2 encoding" (used by DOS 3.3, ProDOS, and all ".dsk" image files), there are 64 possible values that you may find in the data field (in the range $96..$FF, but not all of those, because some of them have bit patterns that trip up the drive firmware). We'll call these "raw nibbles." Step 1: read $156 raw nibbles from the data field. These values will range from $96 to $FF, but as mentioned earlier, not all values in that range will appear on disk. Now we have $156 raw nibbles. Step 2: decode each of the raw nibbles into a 6-bit byte between 0 and 63 (%00000000 and %00111111 in binary). $96 is the lowest valid raw nibble, so it gets decoded to 0. $97 is the next valid raw nibble, so it's decoded to 1. $98 and $99 are invalid, so we skip them, and $9A gets decoded to 2. And so on, up to $FF (the highest valid raw nibble), which gets decoded to 63. Now we have $156 6-bit bytes. Step 3: split up each of the first $56 6-bit bytes into pairs of bits. In other words, each 6-bit byte becomes three 2-bit bytes. These 2-bit bytes are merged with the next $100 6-bit bytes to create $100 8-bit bytes. Hence the name, "6-and-2" encoding. The exact process of how the bits are split and merged is... complicated. The first $56 6-bit bytes get split up into 2-bit bytes, but those two bits get swapped (so %01 becomes %10 and vice- versa). The other $100 6-bit bytes each get multiplied by 4 (a.k.a. bit-shifted two places left). This leaves a hole in the lower two bits, which is filled by one of the 2-bit bytes from the first group. A diagram might help. "a" through "x" each represent one bit. ------------- 1 decoded 3 decoded nibble in + nibbles in = 3 bytes first $56 other $100 00abcdef 00ghijkl 00mnopqr | 00stuvwx | split | & shifted swapped left x2 | | V V 000000fe + ghijkl00 = ghijklfe 000000dc + mnopqr00 = mnopqrdc 000000ba + stuvwx00 = stuvwxba ------------- Tada! Four 6-bit bytes 00abcdef 00ghijkl 00mnopqr 00stuvwx become three 8-bit bytes ghijklfe mnopqrdc stuvwxba When DOS 3.3 reads a sector, it reads the first $56 raw nibbles, decoded them into 6-bit bytes, and stashes them in a temporary buffer (at $BC00). Then it reads the other $100 raw nibbles, decodes them into 6-bit bytes, and puts them in another temporary buffer (at $BB00). Only then does DOS 3.3 start combining the bits from each group to create the full 8-bit bytes that will end up in the target page in memory. This is why DOS 3.3 "misses" sectors when it's reading, because it's busy twiddling bits while the disk is still spinning. ~ Chapter 8 Shift Happens 0boot also uses "6-and-2" encoding. The first $56 nibbles in the data field are still split into pairs of bits that need to be merged with nibbles that won't come until later. But instead of waiting for all $156 raw nibbles to be read from disk, it "interleaves" the nibble reads with the bit twiddling required to merge the first $56 6-bit bytes and the $100 that follow. By the time 0boot gets to the data field checksum, it has already stored all $100 8-bit bytes in their final resting place in memory. This means that 0boot can read all 16 sectors on a track in one revolution of the disk. That's crazy fast. To make it possible to do all the bit twiddling we need to do and not miss nibbles as the disk spins(*), we do some of the work earlier. We multiply each of the 64 possible decoded values by 4 and store those values. (Since this is accomplished by bit shifting and we're doing it before we start reading the disk, this is called the "pre-shift" table.) We also store all possible 2-bit values in a repeating pattern that will make it easy to look them up later. Then, as we're reading from disk (and timing is tight), we can simulate all the bit math we need to do with a series of table lookups. There is just enough time to convert each raw nibble into its final 8-bit byte before reading the next nibble. (*) The disk spins independently of the CPU, and we only have a limited time to read a nibble and do what we're going to do with it before WHOOPS HERE COMES ANOTHER ONE. So time is of the essence. Also, "As The Disk Spins" would make a great name for a retrocomputing-themed soap opera. The first table, at $0200..$02FF, is three columns wide and 64 rows deep. Astute readers will notice that 3 x 64 is not 256. Only three of the columns are used; the fourth (unused) column exists because multiplying by 3 is hard but multiplying by 4 is easy (in base 2 anyway). The three columns correspond to the three pairs of 2-bit values in those first $56 6-bit bytes. Since the values are only 2 bits wide, each column holds one of four different values (%00, %01, %10, or %11). The second table, at $036C..$03D5, is the "pre-shift" table. This contains all the possible 6-bit bytes, in order, each multiplied by 4 (a.k.a. shifted to the left two places, so the 6 bits that started in columns 0-5 are now in columns 2-7, and columns 0 and 1 are zeroes). Like this: 00ghijkl --> ghijkl00 Astute readers will notice that there are only 64 possible 6-bit bytes, but this second table is larger than 64 bytes. To make lookups easier, the table has empty slots for each of the invalid raw nibbles. In other words, we don't do any math to decode raw nibbles into 6-bit bytes; we just look them up in this table (offset by $96, since that's the lowest valid raw nibble) and get the required bit shifting for free. addr | raw | decoded 6-bit | pre-shift -----+-----+----------------+---------- $36C | $96 | 0 = %00000000 | %00000000 $36D | $97 | 1 = %00000001 | %00000100 $36E | $98 [invalid raw nibble] $36F | $99 [invalid raw nibble] $370 | $9A | 2 = %00000010 | %00001000 $371 | $9B | 3 = %00000011 | %00001100 $372 | $9C [invalid raw nibble] $373 | $9D | 4 = %00000100 | %00010000 . . . $3D4 | $FE | 62 = %00111110 | %11111000 $3D5 | $FF | 63 = %00111111 | %11111100 Each value in this "pre-shift" table also serves as an index into the first table (with all the 2-bit bytes). This wasn't an accident; I mean, that sort of magic doesn't just happen. But the table of 2-bit bytes is arranged in such a way that we take one of the raw nibbles that needs to be decoded and split apart (from the first $56 raw nibbles in the data field), use that raw nibble as an index into the pre- shift table, then use that pre-shifted value as an index into the first table to get the 2-bit value we need. That's a neat trick. ; this loop creates the pre-shift table ; at $36C 085A- A2 6A LDX #$6A 085C- 1E 6B 03 ASL $036B,X 085F- 1E 6B 03 ASL $036B,X 0862- CA DEX 0863- D0 F7 BNE $085C Wait, what? It turns out the drive firmware already creates a table that looks very similar to the pre-shift table we want... it's just not shifted yet! Since we're not calling the drive firmware anymore, we can take full advantage of this table that's guaranteed to be in memory. And this is the result (".." means the address is unused): 036C- 00 04 .. .. 0370- 08 0C .. 10 14 18 .. .. 0378- .. .. .. .. 1C 20 .. .. 0380- .. 24 28 2C 30 34 .. .. 0388- 38 3C 40 44 48 4C .. 50 0390- 54 58 5C 60 64 68 .. .. 0398- .. .. .. .. .. .. .. .. 03A0- .. 6C .. 70 74 78 .. .. 03A8- .. 7C .. .. 80 84 .. 88 03B0- 8C 90 94 98 9C A0 .. .. 03B8- .. .. .. A4 A8 AC .. B0 03C0- B4 B8 BC C0 C4 C8 .. .. 03C8- CC D0 D4 D8 DC E0 .. E4 03D0- E8 EC F0 F4 F8 FC ; this loop creates the table of 2-bit ; values at $200, magically arranged to ; enable easy lookups later 0865- C8 INY 0866- 46 BA LSR $BA 0868- 46 BA LSR $BA 086A- B5 E7 LDA $E7,X 086C- 99 FF 01 STA $01FF,Y 086F- E6 AF INC $AF 0871- A5 AF LDA $AF 0873- 25 BA AND $BA 0875- D0 05 BNE $087C 0877- E8 INX 0878- 8A TXA 0879- 29 03 AND #$03 087B- AA TAX 087C- C8 INY 087D- C8 INY 087E- C8 INY 087F- C8 INY 0880- C0 04 CPY #$04 0882- B0 E6 BCS $086A 0884- C8 INY 0885- C0 04 CPY #$04 0887- 90 DD BCC $0866 And this is the result: 0200- 00 00 00 .. 00 00 02 .. 0208- 00 00 01 .. 00 00 03 .. 0210- 00 02 00 .. 00 02 02 .. 0218- 00 02 01 .. 00 02 03 .. 0220- 00 01 00 .. 00 01 02 .. 0228- 00 01 01 .. 00 01 03 .. 0230- 00 03 00 .. 00 03 02 .. 0238- 00 03 01 .. 00 03 03 .. 0240- 02 00 00 .. 02 00 02 .. 0248- 02 00 01 .. 02 00 03 .. 0250- 02 02 00 .. 02 02 02 .. 0258- 02 02 01 .. 02 02 03 .. 0260- 02 01 00 .. 02 01 02 .. 0268- 02 01 01 .. 02 01 03 .. 0270- 02 03 00 .. 02 03 02 .. 0278- 02 03 01 .. 02 03 03 .. 0280- 01 00 00 .. 01 00 02 .. 0288- 01 00 01 .. 01 00 03 .. 0290- 01 02 00 .. 01 02 02 .. 0298- 01 02 01 .. 01 02 03 .. 02A0- 01 01 00 .. 01 01 02 .. 02A8- 01 01 01 .. 01 01 03 .. 02B0- 01 03 00 .. 01 03 02 .. 02B8- 01 03 01 .. 01 03 03 .. 02C0- 03 00 00 .. 03 00 02 .. 02C8- 03 00 01 .. 03 00 03 .. 02D0- 03 02 00 .. 03 02 02 .. 02D8- 03 02 01 .. 03 02 03 .. 02E0- 03 01 00 .. 03 01 02 .. 02E8- 03 01 01 .. 03 01 03 .. 02F0- 03 03 00 .. 03 03 02 .. 02F8- 03 03 01 .. 03 03 03 .. ; to reproduce the experience of the ; original disk, we switch to hi-res ; page 1 and let the title page load ; progressively during boot 0889- 2C 50 C0 BIT $C050 088C- 2C 54 C0 BIT $C054 088F- 2C 57 C0 BIT $C057 0892- 2C 52 C0 BIT $C052 ; And that's all she wrote. Everything ; else is already lined up on the ; stack. All that's left to do is ; "return" and let the stack guide us ; through the rest of the boot. 0895- 60 RTS [Note to future self: $0889..$08FF is available for game-specific init code, but it can't rely on or disturb zero page in any way. That rules out a lot of built-in ROM routines; be careful.] ~ Chapter 9 0boot boot1 The rest of the boot runs from zero page. It's hard to show you exactly what boot1 will look like, because it relies heavily on self-modifying code. In a standard DOS 3.3 RWTS, the softswitch to read the data latch is "LDA $C08C,X", where X is the boot slot times 16 (to allow disks to boot from any slot). 0boot also supports booting from any slot, but instead of using an index, each fetch instruction is pre- set based on the boot slot. Not only does this free up the X register, it lets us juggle all the registers and put the raw nibble value in whichever one is convenient at the time. (We take full advantage of this freedom.) I've marked each pre-set softswitch with "o_O" to remind you that self-modifying code is awesome. There are several other instances of addresses and constants that get modified while boot1 is running. I've marked these with "/!\" to remind you that self-modifying code is dangerous and you should not try this at home. The first thing popped off the stack is the drive arm move routine at $00B7. It moves the drive exactly one phase (half a track). 00B7- E6 BA INC $BA ; This value was set at $00B7 (above). ; It's incremented monotonically, but ; it's ANDed with $03 later, so its ; exact value isn't relevant. 00B9- A0 3F LDY #$3F /!\ ; short wait for PHASEON 00BB- A9 04 LDA #$04 00BD- 20 C3 00 JSR $00C3 ; fall through 00C0- 88 DEY ; longer wait for PHASEOFF 00C1- 69 41 ADC #$41 00C3- 85 CE STA $CE ; calculate the proper stepper motor to ; access 00C5- 98 TYA 00C6- 29 03 AND #$03 00C8- 2A ROL 00C9- AA TAX ; This address was set at $0826, ; based on the boot slot. 00CA- BD E0 C0 LDA $C0E0,X /!\ ; This value was set at $00C3 so that ; PHASEON and PHASEOFF have optimal ; wait times. 00CD- A9 D1 LDA #$D1 /!\ ; wait exactly the right amount of time ; after accessing the proper stepper ; motor 00CF- 4C A8 FC JMP $FCA8 Since the drive arm routine only moves one phase, it was pushed to the stack twice before each track read. Our game is stored on whole tracks; this half- track trickery is only to save a few bytes of code in boot1. (Hey, we're on zero page; space is tight!) The track read routine starts at $0001, because that let us save 1 byte in the boot0 code when we were pushing addresses to the stack. (We could just push $00 twice.) ; sectors-left-to-read-on-this-track ; counter (incremented to $00) 0001- A2 F0 LDX #$F0 0003- 86 00 STX $00 We initialize an array at $00EB that tracks which sectors we've read from the current track. Astute readers will notice that this part of zero page had real data in it -- some addresses that were pushed to the stack, and some other values that were used to create the 2-bit table at $0200. All true, but all those operations are now complete, and the space is now available for unrelated uses. The array is in logical sector order; we convert physical to logical sectors immediately after reading the address field. Values are the actual pages in memory where that sector should go, and they get zeroed once the sector is read (so we don't waste time decoding the same sector twice). ; starting address (game-specific; ; this one starts loading at $0800) 0005- A9 08 LDA #$08 /!\ 0007- 95 EB STA $EB,X 0009- E6 06 INC $06 000B- E8 INX 000C- D0 F7 BNE $0005 000E- 20 D5 00 JSR $00D5 ; subroutine reads a nibble and ; stores it in the accumulator 00D5- AD EC C0 LDA $C0EC o_O 00D8- 10 FB BPL $00D5 00DA- 60 RTS Continuing from $0011... ; first nibble must be $D5 0011- C9 D5 CMP #$D5 0013- D0 F9 BNE $000E ; read second nibble, must be $AA 0015- 20 D5 00 JSR $00D5 0018- C9 AA CMP #$AA 001A- D0 F5 BNE $0011 ; We actually need the Y register to be ; $AA for unrelated reasons later, so ; let's set that now. (We have time, ; and it saves 1 byte!) 001C- A8 TAY ; read the third nibble 001D- 20 D5 00 JSR $00D5 ; is it $AD? 0020- 49 AD EOR #$AD ; Yes, which means this is the data ; prologue. Branch forward to start ; reading the data field. 0022- F0 22 BEQ $0046 If that third nibble is not $AD, we assume it's the end of the address prologue. ($96 would be the third nibble of a standard address prologue, but we don't actually check.) We fall through and start decoding the 4-4 encoded values in the address field. 0024- A0 02 LDY #$02 The first time through this loop, we'll read the disk volume number. The second time, we'll read the track number. The third time, we'll read the physical sector number. We don't actually care about the disk volume or the track number, and once we get the sector number, we don't verify the address field checksum. YOLO. 0026- 20 D5 00 JSR $00D5 0029- 2A ROL 002A- 85 AF STA $AF 002C- 20 D5 00 JSR $00D5 002F- 25 AF AND $AF 0031- 88 DEY 0032- 10 F2 BPL $0026 ; take physical sector number (in A) ; and use it to look up the logical ; sector number 0034- AA TAX 0035- BC 00 01 LDY $0100,X ; store logical sector number 0038- 84 AF STY $AF ; use logical sector number as an ; index into the sector address array ; to get the target page (where we want ; to store this sector in memory) 003A- B6 DB LDX $DB,Y ; store the target page in several ; places throughout the following code 003C- 86 9E STX $9E 003E- CA DEX 003F- 86 6E STX $6E 0041- 86 86 STX $86 0043- E8 INX ; This is an unconditional branch, ; because the ROL at $0029 will always ; set the carry. We're done processing ; the address field, so we need to loop ; back and wait for the data prologue. 0044- B0 C8 BCS $000E ; execution continues here (from $0022) ; after matching the data prologue 0046- E0 00 CPX #$00 ; If X is still $00, it means we found ; a data prologue before we found an ; address prologue. In that case, we ; have to skip this sector, because we ; don't know which sector it is and we ; wouldn't know where to put it. 0048- F0 C4 BEQ $000E Nibble loop #1 reads nibbles $00..$55, looks up the corresponding offset in the preshift table at $036C, and stores that offset in the temporary buffer at $0300. ; initialize rolling checksum to $00 004A- 85 58 STA $58 004C- AE EC C0 LDX $C0EC o_O 004F- 10 FB BPL $004C ; The nibble value is in the X register ; now. The lowest possible nibble value ; is $96 and the highest is $FF. To ; look up the offset in the table at ; $036C, we need to subtract $96 from ; $036C and add X. 0051- BD D6 02 LDA $02D6,X ; Now the accumulator has the offset ; into the table of individual 2-bit ; combinations ($0200..$02FF). Store ; that offset in the temporary buffer ; at $0300, in the order we read the ; nibbles. But the Y register started ; counting at $AA, so we need to ; subtract $AA from $0300 and add Y. 0054- 99 56 02 STA $0256,Y ; The EOR value is set at $004A ; each time through loop #1. 0057- 49 00 EOR #$00 /!\ 0059- C8 INY 005A- D0 EE BNE $004A Here endeth nibble loop #1. Nibble loop #2 reads nibbles $56..$AB, combines them with bits 0-1 of the appropriate nibble from the first $56, and stores them in bytes $00..$55 of the target page in memory. 005C- A0 AA LDY #$AA 005E- AE EC C0 LDX $C0EC o_O 0061- 10 FB BPL $005E 0063- 5D D6 02 EOR $02D6,X 0066- BE 56 02 LDX $0256,Y 0069- 5D 02 02 EOR $0202,X ; This address was set at $003F ; based on the target page (minus 1 ; so we can add Y from $AA..$FF). 006C- 99 56 D1 STA $D156,Y /!\ 006F- C8 INY 0070- D0 EC BNE $005E Here endeth nibble loop #2. Nibble loop #3 reads nibbles $AC..$101, combines them with bits 2-3 of the appropriate nibble from the first $56, and stores them in bytes $56..$AB of the target page in memory. 0072- 29 FC AND #$FC 0074- A0 AA LDY #$AA 0076- AE EC C0 LDX $C0EC o_O 0079- 10 FB BPL $0076 007B- 5D D6 02 EOR $02D6,X 007E- BE 56 02 LDX $0256,Y 0081- 5D 01 02 EOR $0201,X ; This address was set at $0041 ; based on the target page (minus 1 ; so we can add Y from $AA..$FF). 0084- 99 AC D1 STA $D1AC,Y /!\ 0087- C8 INY 0088- D0 EC BNE $0076 Here endeth nibble loop #3. Loop #4 reads nibbles $102..$155, combines them with bits 4-5 of the appropriate nibble from the first $56, and stores them in bytes $AC..$FF of the target page in memory. 008A- 29 FC AND #$FC 008C- A2 AC LDX #$AC 008E- AC EC C0 LDY $C0EC o_O 0091- 10 FB BPL $008E 0093- 59 D6 02 EOR $02D6,Y 0096- BC 54 02 LDY $0254,X 0099- 59 00 02 EOR $0200,Y ; This address was set at $003C ; based on the target page. 009C- 9D 00 D1 STA $D100,X /!\ 009F- E8 INX 00A0- D0 EC BNE $008E Here endeth nibble loop #4. ; Finally, get the last nibble, ; which is the checksum of all ; the previous nibbles. 00A2- 29 FC AND #$FC 00A4- AC EC C0 LDY $C0EC o_O 00A7- 10 FB BPL $00A4 00A9- 59 D6 02 EOR $02D6,Y ; If checksum fails, start over. ; Note: we really want to branch ; to $000E, but that's too far, ; so we're branching to an earlier ; unrelated "BCS" which branches ; to $000E. The carry is always ; set at this point (it was set ; by the "CPX #$00" all the way ; back at $0046), so the BCS is ; an unconditional jump and we ; end up where we want (at $000E). 00AC- D0 96 BNE $0044 ; This was set to the logical ; sector number (at $0038), so ; this is a index into the 16- ; byte array at $00DB. 00AE- A0 00 LDY #$00 /!\ ; store #$00 at this index in the ; sector array to indicate that ; we've read this sector 00B0- 96 DB STX $DB,Y ; are we done yet? 00B2- E6 00 INC $00 ; nope, loop back to read more sectors 00B4- D0 8E BNE $0044 ; And that's all she read. 00B6- 60 RTS 0boot's track read routine is done when $0000 hits $00, which is astonishingly beautiful. Like, "now I know God" level of beauty. And so it goes: we pop another address off the stack, move the drive arm, read another track, and so on. Eventually we finish moving and reading, moving and reading, and we get to the home stretch and start calling ROM routines. $FE88 (IN#0, pushed at $0846) $FE92 (PR#0, pushed at $0846) Next on the stack: $00D1 (turn off drive motor) 00D2- AD E8 C0 LDA $C0E8 /!\ Note that this routine falls through to the one at $00D5 which reads a nibble from disk, but that's harmless. Next on the stack is the game's first entry point: $8014 The routine at $8015 exits via $9001, which exits via "RTS", which -- just like the original disk -- brings us to the final address on the stack, which we pushed a loooong time ago (way back at $0836): $78FA ...which, as we know, starts the game. The entire boot process takes about two seconds. Quod erat liberandum. ~ Acknowledgements Thanks to qkumba for writing 0boot, for explaining 6-and-2 encoding to me, for reviewing drafts of this write-up, and for being that rare combination of smart and kind. ~ Changelog 2020-06-24 - typo in the 6-and-2 encoding diagram [thanks Andrew R.] 2019-01-26 - initial release --------------------------------------- A 4am crack No. 1945 ------------------EOF------------------