Jump to content
ajxpk

[GEN 3] Mystery Event/Gift Research

Recommended Posts

I've been lurking for a while now, but I think I'm finally ready to add what information I know about the distribution ROM. I apologize if I repeat something that was already known, but I checked the last 5 or so pages, and I didn't find the information I have.

  1. The actual script data for the Aurora Ticket distribution starts at 0x12728. It is the raw data - unencrypted and uncompressed. Noteworthy is the fact that the checksum and header are not stored here. I edited the script to give me both the Aurora Ticket and Mystic Ticket, and it worked! This means the checksum is calculated dynamically, rather than stored elsewhere on the ROM.
  2. The Wonder Card starts at 0x14FB8. At least, that is where the checksum would be. I haven't confirmed if the checksum is calculated dynamically, but I can't find it stored anywhere in the ROM, so I'm assuming it's dynamic.
  3. I tried my hand at disassembly. Here's what I've learned:
    • The disassemblers designed for generic ARM7-TDMI code don't like disassembling THUMB code.
    • The disassemblers designed specifically for the GBA (e.g. luvdis) don't like disassembling ARM code.
    • Overall, I'm bad at reading assembly code.
    • Something (maybe) useful: the Wireless Adapter uses the 32-bit "normal" serial communication mode. (It sends data by writing to register 0x04000120.)
  4. Not strictly about the distribution ROM, but the "virtual" script functions (opcodes B8 through BF) can access data in the Wonder Card and Wonder News sections. Thus, a properly crafted Wonder Card could add an additional 320 bytes of script data. Finding a way to distribute Wonder News would add another 440 bytes of usable script data.
  5. I've been doing some digging into the decompilation projects, and I've found two functions that might be responsible for Emerald rejecting the distribution:
  6. Given that "ValidateMEventClientHeader" appears in FRLG and Emerald, but not RS, I suspect it is this function that is used for the Wireless Adapter, while "CheckCompatibility" is used for the eReader. Regardless, as long as it is one of these two functions, the remedy is the same. Replace a 0x1 somewhere with 0x5, and replace a 0xF with 0x20F or 0x38F. Unfortunately, like the checksums, these values do not appear to be stored directly on the ROM. They are likely caused by MOVS or ADDS instructions, instead.

I hope this helps, and let me know if there's anything else I can do to help.

  • Like 6

Share this post


Link to post
Share on other sites

Long-winded story in the spoiler tag.

Spoiler

For the past week I've been working on finding out how the distribution ROM was sending the compatibility header to the game. To cut a very long story very short, that adventure was fruitless.

First, I tried making a custom disassembler. My issue with reading assembly was that it looked weird, so I thought that if I translated every THUMB function to a "C" look-a-like, I would understand it better. I got pretty far, but then I ran into the horror that are function pointers. This also caused my trials with luvdis's configuration mode to end as well. Simply put, without an extremely good guessing algorithm, function pointers make it near-impossible to analyze a ROM.

I turned back to the internet, and researched if there were any other GBA disassemblers other than luvdis. I found a Medium article that recommended ISA Pro, but at ~$2600, I said no way. In desperation, I stumbled across a blog post (https://cturt.github.io/pinball.html) that recommended a program called Ghidra. Ghidra is a multi-platform, free and open source decompiler, that is compatible with ARMv4T. The one major issue that I have with it is that it is developed by the NSA. Yes, that NSA.

The tool took some getting used to, but I found that it turned the ROM from a messy mix of THUMB and ARM functions into a nice C-style program (save for the absolutely useless variable names like "bStack11"). Even with this, though, I failed to find how the program was writing the script to the SIODATA32 register. (I still have not found how it does this. My best guess is that it actually sends a command to the Wireless Adapter to DMA from ROM, or from WRAM around 0x03002500.)

What Ghidra did help me find was similarities between the distribution ROM and FireRed/Emerald. In particular, I found some low-level functions in the FireRed/Emerald disassemblies (called STWI), that were also present in the distribution ROM. This reminded me that FireRed & Emerald can send Wonder Cards, and thus act as a server as well as a client.

Even this was a wild goose chase. I was unable to find out how the server commands linked to the low-level STWI commands. Thus, I turned back to my original strategy of looking for the hex values associated with the compatibility header, in particularly 0x101. To my surprise, Ghidra found a function that relied upon it. A very familiar function...

It was a copy of FireRed's ValidateMEventClientHeader function! Nothing called it, of course; stupid function pointers ruining everything. But it made me wonder, why was it here? As I wondered, I returned to common_mainseq_4 (https://github.com/pret/pokefirered/blob/2880cf2a51ea36fa36f00d9ecf07177e5955c882/src/mevent_server.c#L118), which had been a common thread in my investigation. Indeed, I had been puzzling over this function for weeks. But, I had been looking at Emerald's version. The one thing the FireRed disassembly has over Emerald's is that more functions and constants have names. This ended up being the key factor in the solution.

In a fit, I looked at FireRed's gMEventSrvScript_SendCard. Emerald's is an absolutely atrocious mess (https://github.com/pret/pokeemerald/blob/5183cdb35722549d6b465ccaf8c4a21422ecb254/src/mevent_scripts.c#L178). But FireRed's is much cleaner (https://github.com/pret/pokefirered/blob/a0b2fa5d33c43bc4f42fc4f132f73e960d771d5b/src/mevent_scripts.c#L179). In particular, it has the line "SRV_SEND(0x20, gMEventClientScript_Send1442CC),". I knew that 1442CC was associated with the compatibility header, so I went down the rabbit hole. The first instruction sent was by gMEventClientScript_Send1442CC was "CLI_SNDHEAD".

I realized that I had everything all wrong. The distribution ROM doesn't send the compatibility header to the games, the games send the compatibility header to the distribution ROM! The only reason the games have the ValidateMEventClientHeader is because they can act as the server. I checked the definition of CLI_SNDHEAD, and saw it was client instruction 8. I looked at the client's version of common_mainseq_4, case 8, and found a function called "BuildMEventClientHeader" (https://github.com/pret/pokefirered/blob/2880cf2a51ea36fa36f00d9ecf07177e5955c882/src/mevent.c#L753). The function, as expected, builds the header in question.

Emerald's counterpart is sub_801B580 (https://github.com/pret/pokeemerald/blob/ebade7affb31d5bcdc17cdcd3895758010ee6f66/src/mevent2.c#L338), which also build the header as expected. At least, if "a1" is 0. If a1 is not 0, then it should already be compatible. One more trace later, and I discovered that a1 is 1 for Wonder News, and 0 for Wonder Cards. Finally, the mystery had been solved.

TL;DR: Rather than the games checking the compatibility header, the distribution ROM does. Patch instructions below. (If anyone wants to make an IPS/UPS/BPS patch, go ahead! I just don't feel like downloading one more program for this project to make one.)

Due to space constraints, it is impossible to patch a fix in to the same location. However, it is possible to patch out the check completely. Replacing 0x714C through 0x7173 (inclusive) with 0x00 patches out the region, game, and language checks. It leaves the sanity check, however. (Also, see below regarding the language check.)

Why does this matter? Using direct injection, I already knew that it is possible to use "comparefarbytetobyte 0x80000AE 'E'" (ASCII 'E', not the proprietary encoding 'E') to check for whether a game is Emerald or not. "comparefarbytetobyte 0x80000AF" can also be used to check for language. Thus, while scripts for both versions - and texts for all languages - would need to fit into 1000 bytes, it is now possible to make a multi-version multi-language distribution ROM.

One last thing. I found a glitchcity forum post (Reply #35 at https://forums.glitchcity.info/index.php?topic=7114.30) that confirmed my suspicions regarding the CheckCompatibility function. FireRed, LeafGreen, and Emerald act as Japanese cartridges, regardless of their actual language. This carries over to ValidateMEventClientHeader. Thus, all games can receive the Mystery Gift, regardless of the language of the distribution. This does leave a question as to how the European distribution ROM knew which version to send. It may be the case that only English carts act as Japanese carts. However, without a European decompilation, and without a European cart of my own, I'm left without means to test this hypothesis.

  • Like 3

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×
×
  • Create New...