Files
psxrecomp/runtime/include/overlay_api.h
T
Matthew Stanley 94a1966444 recursion(fix): universal continuation-passing — FIXES the idle-freeze host-stack leak (BIOS+game)
The complete §25 fix: make BOTH recompiler emitters emit guest calls as
TAIL-TRANSFERS driven by the unified flat trampoline + the guest's own $ra/$sp,
instead of nesting one host C frame per guest call. The host stack can no longer
mirror unbounded guest re-entrancy — the architectural root of the ~14-min idle
freeze. Gated gen-time by PSX_CPS; legacy codegen byte-identical when unset.

VALIDATED (user): full-CPS Tomba boots BIOS -> FMV -> New Game -> village idle and
plays; ce_profile cur_max_kb FLAT at 5 KB from frame ~9.9k through 56,890 (turbo
soak) — blew past the ~50k-frame freeze threshold with the host stack bounded
(legacy climbed +1.17 KB/frame -> 48 MB -> freeze).

KEY CORRECTION to the prior session's §25.1: there are TWO emitters. The BIOS
(full_function_emitter.cpp) was done last session but the §8 leaking chain
(func_8001A954 ...) is GAME code, emitted by code_generator.cpp (via main_psx.cpp)
— never converted. Also, §25.1's "continuations not in the dispatch table"
hypothesis was wrong (they were always added); the real boot regressions were
found by dumping the always-on fntrace ring on the abnormal psx_dispatch return.

Emitter changes (gated PSX_CPS):
- full_function_emitter.cpp (BIOS): BIOS A0/B0/C0 vector runtime-fallback
  tail-transfers instead of nested psx_dispatch_call (was the frame-0 exit).
- code_generator.cpp/.h (game): jal/jalr/jr-$ra/jr-table/branch-split/j-split ->
  tail-transfer (jalr captures rs before the link, alias-safe); each call's
  return point registered as a dispatchable continuation; entry-switch (keyed on
  cpu->pc) at the top of every function AND every overlapping-alias body. 0
  psx_call_contract / call_by_address emitted under CPS.
- main_psx.cpp: psx_dispatch_game_compiled gains continuation cases
  (case cont: cpu->pc=cont; func_owner(cpu)); entries clear cpu->pc.
- strict_translator.cpp: syscall emits `if (psx_syscall(...)) return;` so a void
  syscall falls through to the inline post-syscall jr $ra (real-HW EPC+4).

Runtime changes (the legacy "return via nested dispatch / cpu->pc=0" idioms that
exit the flat CPS trampoline):
- traps.c: psx_syscall returns int (0=void/fall-through, 1=transfer);
  Enter/ExitCriticalSection return 0; case-3 ChangeThread-success returns 0 — the
  fiber scheduler leaves cpu->pc=0 on resume, which was the frame-712 exit.
- cpu_state.h / overlay_api.h: psx_syscall signature void->int; overlay callback
  ABI bumped 4->5 (stale overlay caches rejected).
- dirty_ram_interp.c: gate interp_enter_compiled behind PSX_HAS_GAME_DISPATCH
  (un-breaks the BIOS-only build).
- main.cpp: dump the always-on fntrace ring tail + stack to psx_cps_exit_trace.json
  on an abnormal psx_dispatch return (diagnostic; CLAUDE.md ring-buffer model).

Docs: RECURSION_BUG.md §25.2-§25.4. Remaining: dwarf->overworld playtest;
Stage 3 (sljit/overlay CPS parity); then consider making CPS default.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-18 12:45:02 -07:00

138 lines
7.1 KiB
C

#ifndef OVERLAY_API_H
#define OVERLAY_API_H
/* Shared ABI between psx-runtime.exe and overlay DLLs.
*
* After LoadLibrary, the runtime calls overlay_init() with this struct.
* The DLL stores the pointers and routes all cross-module calls through them.
*/
#include "cpu_state.h"
#include <stdint.h>
/* ABI version exported by every overlay DLL as `overlay_abi()`. The loader
* rejects (and deletes, so autocompile regenerates) any DLL whose version
* doesn't match — including all pre-versioning DLLs, which lack the export.
*
* v2: dispatch call-contract (Bug D family) — DLL call sites carry (ra, sp)
* contract checks and share the runtime's bail state via the appended
* callback pointers below.
* v3: widescreen backdrop screenX squash ([widescreen.backdrop] x_sites) —
* overlay backdrop handlers now emit psx_ws_backdrop_x() at their screenX
* store. This is an EMIT-content change (not a struct-ABI change), but the
* version is the cache discriminator: bumping it forces the loader to
* reject pre-backdrop DLLs so autocompile regenerates them with the squash.
* (The deferred follow-up is to auto-derive this tag from a hash of the
* widescreen config + emit headers; until then it's a manual bump.)
* v4: widescreen backdrop column PRELOAD ([widescreen.cull] auto_backdrop) —
* overlay backdrop generators now emit psx_ws_backdrop_value() at the
* window START/END finalize, and the callback struct grows a
* ws_backdrop_value pointer (appended last). Bumping rejects pre-preload
* DLLs (which lack both the emit and a host that supplies the callback). */
/* v5: psx_syscall callback return type void->int (CPS, RECURSION_BUG.md §25);
* overlays compiled under CPS emit `if (psx_syscall(...)) return;`. */
#define PSX_OVERLAY_ABI_VERSION 5
/* Codegen flavor of the recompiled output the overlays + runtime were built
* against. Overlays are keyed in the cache by guest-bytes CRC, which is
* flavor-BLIND — the same guest overlay yields the same filename whether it was
* compiled by the base recompiler or a variant (e.g. widescreen, which emits
* extra runtime symbols + GTE adjustments). Without a flavor tag a widescreen
* DLL and a base DLL for the same overlay collide. The flavor is folded into
* overlay_abi() (high 16 bits) so the loader's existing ABI gate rejects (and
* regenerates) any DLL whose flavor differs from the running build's, keeping
* caches separate even in a shared directory.
*
* Base/master MUST stay 0 so the tag == PSX_OVERLAY_ABI_VERSION (backward
* compatible with all existing base-flavor DLLs). A variant build overrides it
* (e.g. -DPSX_OVERLAY_FLAVOR=1 for widescreen) for BOTH the runtime and
* compile_overlays.py (its --flavor arg). */
#ifndef PSX_OVERLAY_FLAVOR
#define PSX_OVERLAY_FLAVOR 0
#endif
/* Combined tag exported by overlay_abi() and checked by the loader. */
#define PSX_OVERLAY_ABI_TAG \
((int)((PSX_OVERLAY_ABI_VERSION & 0xFFFF) | ((PSX_OVERLAY_FLAVOR & 0xFFFF) << 16)))
/* Codegen version — bumped whenever the recompiler emits DIFFERENT bytes for the
* same guest overlay WITHOUT an ABI change (the load-time overlay_abi() gate only
* catches ABI/flavor changes; a pure-emitter change produces a different DLL the
* gate would happily accept). The cache is namespaced by this version
* (gcc/<arch-abi>/cg<N>/), so a build with new codegen writes + reads a FRESH
* directory and never reuses a stale DLL; old versions coexist on disk (no
* auto-delete — a user who downgrades still finds their matching cache). Python
* (compile_overlays.py) parses this same constant from this header, so the two
* sides can never drift. BUMP THIS when you change code_generator.cpp emit in a
* way that alters overlay output (e.g. the auto_screen_x widescreen widening).
* 1: baseline + auto_screen_x render-funnel cull via psx_ws_cull_sltiu.
* 2: + auto_backdrop far-backdrop column preload via psx_ws_backdrop_value.
* 3: auto_backdrop switched to camera-tracked window WIDENING
* (psx_ws_backdrop_value gains a window_cols arg). */
#define PSX_OVERLAY_CODEGEN_VER 3
typedef struct {
/* Core dispatch: routes call_by_address() and out-of-overlay jal */
void (*dispatch_call)(CPUState *cpu, uint32_t addr, uint32_t ra);
/* Interrupt check: called after every function return in overlay */
void (*check_interrupts)(CPUState *cpu);
/* GTE coprocessor 2 execution */
void (*gte_execute)(CPUState *cpu, uint32_t cmd);
/* MIPS syscall (break/syscall instructions). Returns 1 if control
* transfers (cpu->pc set), 0 for a directly-handled void syscall. See
* cpu_state.h. The signature changed with CPS (RECURSION_BUG.md §25), so
* PSX_OVERLAY_ABI_VERSION was bumped to reject stale overlay caches. */
int (*psx_syscall)(CPUState *cpu, uint32_t code);
/* Unresolved dispatch target */
void (*psx_unknown_dispatch)(CPUState *cpu, uint32_t addr, uint32_t phys);
/* Debug instrumentation: called at every function entry (may be NULL) */
void (*log_call_entry)(uint32_t func_addr);
/* RestoreState/ReturnFromException longjmp escape (interrupts.c). The
* recompiler emits this at longjmp-return sites in exception-context
* kernel code (the install-slot / kernel-window class). Appended LAST:
* overlay_init copies the struct by value, so older DLLs built against
* the shorter struct simply never read this member. */
void (*psx_restore_state_escape)(void);
/* Call-contract state shared with the runtime (ABI v2; see the contract
* model in cpu_state.h). DLL code reads the bail flag and bumps the
* counters through these pointers. */
int *call_bail_flag;
uint64_t *bail_first;
uint64_t *bail_resolved;
/* Widescreen hooks (ABI v3). The recompiler can emit psx_ws_* calls into
* OVERLAY code (backdrop screenX squash; and, for other games, sprite-tag
* or cull sites that resolve into overlays). These compute against the
* host's live widescreen state (gpu.c), so the DLL forwards to the runtime
* through these. Appended LAST (struct grows back-compatibly); may be NULL
* on a host that predates them — the DLL glue falls back to identity. */
int (*ws_backdrop_x)(int x);
int (*ws_x_margin)(void);
void (*ws_sprite_tag)(CPUState *cpu);
/* Widescreen backdrop column-preload value substitution (ABI v4). Overlay
* backdrop generators call this at a detected window START/END bound; the
* runtime returns `orig` unless native-wide is engaged, in which case it
* widens the camera-tracked window by the 16:9 reveal (START -> orig-margin,
* END -> orig+margin; margin scales with window_cols). Appended LAST (struct
* grows back-compatibly); may be NULL on a host that predates it — the DLL
* glue falls back to orig. */
uint32_t (*ws_backdrop_value)(uint32_t orig, int is_end, int window_cols);
} OverlayCallbacks;
#ifdef __cplusplus
extern "C" {
#endif
/* Exported by every overlay DLL. Call once after LoadLibrary. */
#ifdef _WIN32
__declspec(dllexport)
#else
__attribute__((visibility("default")))
#endif
void overlay_init(const OverlayCallbacks *cbs);
#ifdef __cplusplus
}
#endif
#endif /* OVERLAY_API_H */