Entry 3

One executable holds all the code, and a Rust toolbox to read it

Files6 in 3 directories
crates/hwtr-psx/src/mips.rsexe.rsanalysis.rscrates/hwtr-re/src/main.rsdocs.rssymbols/cccpsx.txt

CCCPSX.EXE is the whole game: a PS-X EXE whose 796672-byte image loads at 0x80010000 and enters at 0x8009fc10. The BIG archive's .OVL members are sprite sheets, not code (2), and no string names a code overlay, so there is one image to reverse.

The decision: the RE tools are Rust

Earlier projects kept their reverse engineering tools in Python under tools/ and the port in Rust. This one writes the tools in Rust too, in the same workspace as the port, so a reader of a format exists once and the port uses the same code the investigation proved. Python stays available for throwaway probes (the BIG checksum in 2 was found with one), but anything kept is Rust. The crates so far:

hwtr-disc   CUE sheets, raw Mode 2 sectors, ISO 9660 with CD-XA     (port uses it)
hwtr-data   BIG archives; later every asset format                    (port uses it)
hwtr-psx    PS-X EXE, R3000A + GTE disassembler, static analysis
hwtr-re     the command line: disc, big, exe, disasm, funcs, strings, docs

hwtr-re docs index|check replaces piney_apples' tools/docs.py.

The disassembler

MIPS I with COP0 and the GTE: every R3000A opcode, the GTE commands by function field with their sf, lm and MVMVA fields, lwc2/swc2 and the GTE data and control register names. It prints the usual pseudo-ops (nop, move, li, b, beqz, negu) and annotates lui/addiu and lui/load-store pairs with the string or name they form. Unit tests pin encodings, including RTPS (0x4a180001) and RTPT (0x4a280030).

crt0 and the memory map

The entry point is PsyQ's crt0, unchanged in shape:

0x8009fc10  clear 0x800d23cc .. 0x80143b44   (bss, 464760 bytes)
            sp  = (word at 0x800c8690) - 8 | 0x80000000
            gp  = 0x800d0b48
            InitHeap(0x80143b44 + 4, ...)   (BIOS A0:39)
            jal 0x80010a5c                   main

The executable carries gp0 = 0 in its header, so $gp is only known from this code. Game globals are read gp-relative (lw v0, 6364(gp) in 0x800113f8), which the reference scan does not yet resolve.

Function discovery

hwtr-re funcs finds 1540 starts: 834 jal targets, 668 frames opened right after a jr ra, 36 BIOS stubs (li t2, 0xA0|0xB0|0xC0; jr t2; li t1, N, 31 named), the entry, and one data pointer. The PsyQ libraries sit from about 0x8009fc10 up, by their $Id strings (sys.c 1.135 1997/09/02, intr.c 1.76, bios.c 1.86); the game below.

Names live in symbols/cccpsx.txt, read by disasm, funcs and strings, each with a confidence note.

Still unknown

Which of the ~1540 found starts are PsyQ library functions; the prologue heuristic's false positives; the jalr and jump-table targets the scan does not see.