[WoW64] DynaRec translates .data section as code → SIGILL loop on Cortex-A510 (BGI engine game)

[as14725836]

Box64 v0.4.3 e2345d7 (2026-06-15)
DynaRec ARM64: ASIMD AES CRC32 PMULL ATOMICS SHA1 SHA2 USCAT FLAGM FLAGM2 FRINT
CPU: Cortex-A510 ×8, 39-bit address space, pagesize 4096
Wine: wine-9.0proton-vanilla-wow64 (Termux)
Game: sakura2.exe (PE32 i386, BGI engine, imports ChakraCore.dll)

============================================================
CRASH CHAIN:

[1] DynaRec dead loop — SIGILL + Divide by 0, same db hash
x64pc=0x949300 (sakura2.exe .data section, UNINITIALIZED)
opcode=00 00 00 00 42/62 00 00 58 (0F 0B E8 E1 36)
db hash=a2f58006 (cached bad translation)
→ box64 siglongjmp → retry → infinite loop
→ BOX64_DYNAREC=0 fixes this but Wine won't start (too slow)

[2] If survives [1], massive SIGSEGV storm in Wine system DLLs
x64pc=0x79418fd3 (C7 00 00 00 00 00) × dozens
x64pc=0x794395fb (C7 07 00 00 00 00) × many
x64pc=0x79458b3a, 0x794548c3, 0x79456756, 0x79681d26...
→ All writing zero to unmapped addresses (0x11xxxxxx range)
→ Final: SIGSEGV @0x7984dab5 F3 0F 7F 07 (movdqu)

[3] ChakraCore.dll loads successfully but also hits SIGSEGV
x64pc=0x12111082, 0x1211303d, 0x12113223... (base 0x10000000)

[Attempted fix] Patched ChakraCore.dll JsCreateRuntime to force
JsRuntimeAttributeDisableNativeCodeGeneration (OR [esp+4],4)
→ DLL loaded (confirmed by x86opcode=D2 3A 00 00 00 CC in crash)
→ No effect — crash is in sakura2.exe .data, not in ChakraCore

============================================================
ROOT CAUSE:

box64 DynaRec translates sakura2.exe's .data section (non-code)
as x86 instructions → generates invalid ARM64 → SIGILL loop.
This is a DynaRec bug triggered by Cortex-A510 + WoW64 + BGI engine.

trace-26579.txt

[ptitSeb]

How did you get to this conclusion?

The crash

[1] DynaRec dead loop — SIGILL + Divide by 0, same db hash
x64pc=0x949300 (sakura2.exe .data section, UNINITIALIZED)
opcode=00 00 00 00 42/62 00 00 58 (0F 0B E8 E1 36)

Looks normal to me: the x86 code is 0F 0B which is ud2 which will trigger a SIGILL by design. So I would say the behviour is totaly normal at this point, and the game just exit/crash on purpose. The issue, if there is one, is certainly before that point.

[ptitSeb]

The game might be encrypted / obfuscated and might need BOX64_DYNAREC_SAFEFLAGS=2 to operate correctly.

[as14725836]

The situation is still the same, nothing has changed.

[ptitSeb]

You did not answered my questions

[as14725836]

This is the log with BOX64_DYNAREC_SAFEFLAGS=2 enabled.

trace-12025.txt