IOP: Reset recompiler and map RAM mirrors on IOP soft reset

[fjtrujy]

Summary

Fixes IOP soft reset (SifIopReset) crashing with stale recompiler blocks.

When SifIopReset triggers (SBUS_F240 bit 19), psxReset() reloads the IOP BIOS kernel into RAM but does not flush the IOP recompiler cache. The recompiler keeps stale compiled x86 blocks for IOP RAM addresses that now contain different code. When the fresh kernel executes, the recompiler serves old translations, causing the IOP to crash on garbage instructions (psxUNK).

Changes

1. psxCpu->Reset() after psxReset() in the SBUS_F240 handler — flushes all cached recompiler blocks, forcing fresh recompilation from the new RAM contents. This is the critical fix.

2. Map IOP RAM mirrors in recLUT (kuseg pages 0x0000-0x1DFF) — the PS2 IOP mirrors its 2MB RAM across the entire kuseg address space. The BIOS uses mirrored addresses during soft resets (e.g. 0x0d000100 → physical 0x00000100). Without mirrored pages in recLUT, code at mirrored addresses hits unmapped pages.

How it was found

Tested with ps2link performing SifIopReset via ps2client reset. Without the fix:

• psxReset() reloads the IOP kernel
• IOP starts executing (DEV9 hardware detection runs)
• Recompiler serves stale blocks → psxUNK: f0000102 crash
• The "Soft reboot" log message never appears after the reset

With the fix, the recompiler cache is flushed and the IOP reboots cleanly.

Test plan

• ps2client reset performs a clean IOP soft reboot
• Consecutive ps2client reset commands work
• ps2client execee + ps2client reset cycle works
• No regression in normal game boot (which also uses IOP soft resets during BIOS init)

[Ziemas]

Hm this doesn't sound quite right to me, if I make a simple executable that just calls SifResetIop() it doesn't crash.

Reading 0x0d000100 on a real PS2 seems to result in a bus error as well.

dsidb R> iw 0x0d000100
*** Bus Error