machinamentum / ctrx Goto Github PK

I have an original 3ds xl 4.5 eu with rxtools.

???

Oddly, the success rate of booting a kernel through the debugger is somewhat low. Often CTRX will just soft-lock to the point where even removing Cubic Ninja from the game card slot won't cause the 3DS to give an error screen.

Test case: current CTRX BIOS performs 3 nops followed by a branch immediately followed by a branch back to 0xBFC00000. After running this loop several times, the branches stop executing and the program counter continues to increase.

Seems to affect all instructions. The instruction never gets decoded. May not even have been fetched or MachineCode in main.cpp is being overwritten at a poor timing.

It's lame.

mtc0 (and likely related instructions) doesn't write C0 registers and, consequently, can't update exception state.

mtc0 does get decoded and goes through execution properly, but write back is broken.

Some sets of instructions are very commonly generated by MIPS compilers. These can be optimized to be executed as a single instruction together instead of suffering from the overhead of executing two or more emulated instructions.

Example

lui t0, 0x1FB0
ori t0, t0, 0xF560

These instructions together effectively move a 32-bit immediate into register t0 and can be optimized into a single "move 32-bit immediate" instruction. Since this requires already decoding sets of instructions together, the interpreter requires a mechanism to pool together pre-decoded instructions that can later be executed without re-decoding, speeding up small loops. This also would enable a quick way to discard nops (commonly implemented as sll zero, zero, 0 in most compilers).