[Wine] SIGILL/SIGSEGV when running game with SMC (Self-Modifying Code) patch

[as14725836]

Description

I'm running a Windows game through Wine + Box64 on ARM64 Android (Termux). The original game works fine, but after applying a Chinese translation patch (which uses SMC/self-modifying code), Box64 crashes with SIGILL/SIGSEGV.

The error occurs immediately when the patched EXE starts. The original unpatched EXE runs without issues.

Environment

• CPU: Cortex-A510, 8 cores
• OS: Android 15
• Box64 Version: 0.4.5 (dc57589)
• Build Date: Aug 26 2026 09:25:48
• Wine Version: wine-10.0-3proton-vanilla-wow64
• Termux: Latest from F-Droid/GitHub

Box64 Environment

BOX64_DYNAREC=1
BOX64_DYNAREC_BIGBLOCK=0
BOX64_DYNAREC_DIRTY=1
BOX64_DYNAREC_NOHOTPAGE=1
BOX64_DYNAREC_FASTNAN=1
BOX64_DYNAREC_FORWARD=1024
BOX64_DYNAREC_MISSING=1
BOX64_DYNAREC_NATIVEFLAGS=1
BOX64_DYNAREC_WEAKBARRIER=1
BOX64_DYNAREC_CALLRET=0
BOX64_LOG=1
BOX64_SHOWSEGV=1

Error Log

Marked db 0x75cc0e94f0 as dirty, and address 0x7b2089 as creating internal SMC
29783|Emit Signal 11 at IP=0x7b20b9(??? / ?) / addr=0x7b20b9, code=0xbad0
ES=0x002b CS=0x0023 SS=0x002b DS=0x002b FS=0x004b GS=0x0000 FSBASE=0x7ffc2000 GSBASE=0x7ffc0000
RAX=00000001 RCX=00000000 RDX=00869000 RBX=7ffd1000 flags=???????
RSP=0020ff3c RBP=0020ff58 RSI=00000000 RDI=00000000 EIP=007b20b9 Emu Stack: 0x75cc4a0000 0x800000
Signal 11: si_addr=(nil), TRAPNO=13, ERR=0, RIP=0x7b20b9, prot=87, mmapped:3
Context has been changed in Sigactionhanlder, doing siglongjmp to resume emu at 0x4ffff2f82c, RSP=0x20f800 (resume with Dynarec)
... (repeated) ...
0x6fc000: Dynarec stopped because of x86 Opcode 0F 3F 07 0B 64 8F 05 00 00 00 00 83 C4 04 83
29783|Emit Signal 4 at IP=0x6fc000(??? / ?) / addr=0x6fc000, code=0x0
SIGILL: Opcode at ip is 0f 3f 07 0b 64 8f
Signal 4: si_addr=0x6fc000, TRAPNO=6, ERR=0, RIP=0x6fc000, prot=87, mmapped:3

### Key Observations
1. **Original game**: Works fine
2. **Patched game**: Crashes immediately
3. **Error type**: SIGILL (Illegal instruction) at address 0x6fc000
4. **Offending opcode**: `0F 3F 07 0B 64 8F` (unrecognized by Dynarec)
5. **SMC detected**: "Marked db ... as dirty, and address ... as creating internal SMC"
6. **Workaround**: Running with `BOX64_DYNAREC=0` (interpreter mode) works but is slow

### What I've Tried
- Adjusting `BOX64_DYNAREC_BIGBLOCK=0`
- Adjusting `BOX64_DYNAREC_FORWARD=0`
- Adjusting `BOX64_DYNAREC_CALLRET=0`
- Setting `BOX64_DYNAREC_WEAKBARRIER=1`
- All resulted in the same crash

### Expected Behavior
The translated/patched game should run similarly to the original game.

### Additional Context
- The translation patch modifies the game's code at runtime (SMC)
- This is common for Chinese translation patches for visual novels
- The game is a visual novel (RPG Maker/KiriKiri engine)

### Request
Could the Dynarec be improved to handle SMC better, or could the `0F 3F` opcode be added to the emulation?

Thank you for your work on Box64!

[ptitSeb]

Try to use DIRTY=2 and NOHOTPAGE=0 instead maybe

Also, 0F 3F 07 0B doesn't seems like a valid opcode to me.

Is the patch a runtime patch, like a separate exe that patch the game while it's loaded in memory or does it change the file exe on disk?

[as14725836]

A standalone translation patch executable "patch.exe"
This is the log after running with DIRTY=2 and NOHOTPAGE=0.

trace-19132-1.txt

[ptitSeb]

What is BOX64_JMPBUF_STRATEGY=1? are you use a custom build of box64?

According to the logs, the program seems to load fine until it jumps to 0x62f062e0 wich then (repeatedly) fail as non-executable memory.

[as14725836]

Note: BOX64_JMPBUF_STRATEGY=1 is not a valid variable and was mistakenly left in the environment. Please ignore it.

The process hangs indefinitely—it neither loads nor crashes.

[as14725836]

When I set BOX64_DYNAREC=0, the game loads successfully, but performance and speed are very poor.

trace-9085-1.txt