feat(hle): parameterize SoundDriver RPC service SID + subcommands per game

[smmathews]

Summary

Moves the SoundDriver RPC HLE's service SID and subcommand (fno) numbers out of hardcoded constants into the per-game compat layout. Upstream already parameterized the per-game addresses via PS2SoundDriverCompatLayout; the service SID and subcommand (fno) numbers were still hardcoded placeholder constants in ps2_iop.h. This change moves the SID(s) and subcommand numbers into the compat layout so a title whose sound driver registers a different SID or different subcommand numbers is served without editing deps.

What changed

• PS2SoundDriverCompatLayout (ps2_runtime.h) gains commandSid/stateSid + a servesSid() helper, the subcommand fno fields (submitFno, getStatusFno, getAddrTableFno, streamOpenFno, channelConfigFno, stopFno), benignStatusValue (default 0xffffff9b), and the extra address fields (streamStateAddr, streamReadyValue, channelAllocFlagTableAddr, stopCompletionFlagAddr).
• handleSoundDriverRpcServiceImpl (RPC.cpp) now:
  • snapshots the layout under g_rpc_mutex (no fiber/scheduler interaction);
  • returns false immediately without touching guest memory when the layout is unconfigured (commandSid == 0 && stateSid == 0);
  • gates on servesSid(sid), then dispatches by matching rpcNum against the layout's fno fields (each match nonzero-guarded) — submit / getStatus / getAddrTable (existing semantics, re-keyed) plus the new streamOpen / channelConfig / stop writes;
  • for an unknown fno on a served SID, writes benignStatusValue to recv[0] and returns false (falls through so LIBSD / game handlers still run).
• Removed the IOP_SID_SNDDRV_* / IOP_RPC_SNDDRV_* placeholder constants from ps2_iop.h and their references in ps2_debug_panel.cpp. IOP_SID_LIBSD is kept (still used by the LIBSD fast path).
• RE:CVX (slus_201.84) game override (game_overrides.cpp) now carries the SID/fno numbers it relies on: it sets stateSid = 1, getStatusFno = 0x12, getAddrTableFno = 0x13 (the values of the deleted IOP_SID_SNDDRV_STATE / IOP_RPC_SNDDRV_GET_STATUS_ADDR / _GET_ADDR_TABLE constants). RE:CVX's getStatus RPC (sid=1, fno=0x12) provisions g_soundDriverRpcState.statusAddr, and the sceSifGetOtherData checksum backfill fires only when srcAddr == statusAddr; without these values the layout would be unconfigured (commandSid == stateSid == 0), the handler's unconfigured guard would return false, statusAddr would never be provisioned, and the checksum backfill would silently stop. This is the only production override that speaks the sound-driver RPC service; the LotR override uses only completionCallbacks (a separate, non-SID-gated path) and correctly needs no SID/fno configuration. (RE:CVX's old submit path used placeholder SID 0 / fno 0 — a non-real service that never matched a live call — and is intentionally left unconfigured.)
• Tests (ps2_sif_rpc_tests.cpp, ps2_sif_dma_tests.cpp): migrated the pre-existing snddrv-state / sound-status unit tests (which relied on the deleted placeholder SID 1 and fnos 0x12/0x13) to register a layout that maps those numbers. Added regression tests: one exercising every subcommand semantic + the benign unknown-fno fall-through; one proving the inert unconfigured layout plus two games routing independently through the single global layout slot; and a real-path test that provisions the layout solely through ps2_game_overrides::applyMatching("slus_201.84") and drives the actual SifCallRpc(getStatus) → sceSifGetOtherData path, asserting the check-array → status backfill ran and that getStatus returns a nonzero status address.

Known limitations

• 0-sentinel (SID 0 / fno 0 inexpressible). 0 is the "unused" sentinel for every SID and fno field, so a real service whose SID is literally 0 or a subcommand whose fno is literally 0 cannot be expressed. This is a deliberate tradeoff and adequate in practice — real SIF-RPC services are nonzero, and the deleted placeholder constants that were 0 (IOP_SID_SNDDRV_COMMAND / _SUBMIT) were never live services. Documented in a code comment on the struct; if a title ever needs SID 0 / fno 0, add explicit has-value flags rather than overloading the sentinel.
• benignStatusValue is handler-local, not guest-visible on the normal path. For an unknown fno on a served SID the handler writes benignStatusValue to recv[0] and returns false; in the real SifCallRpc path the outer recv finalization then overwrites recv[0] (copy-from-send / zero) before the guest observes it. The unknown-fno unit test asserts the handler-local write directly and is annotated to say so — it locks the handler's documented contract, not guest-observable state.

Verification

• Full ps2x_tests suite: 299 passed / 0 failed. Rebuilt clean (GCC 16) and re-run.
• Byte-level assertions confirm: streamOpen writes streamReadyValue to streamStateAddr and signals nowait; channelConfig sets channelAllocFlagTableAddr[channel]=1 (channel from send word 0, <16); stop writes 1 to stopCompletionFlagAddr; getStatus/getAddrTable return distinct nonzero addresses; unknown fno returns false with handler-local recv[0]==0xffffff9b; unconfigured layout leaves guest memory untouched and returns false.
• Revert check: reverting only the game_overrides.cpp override migration makes the new real-path regression test fail (getStatus returns 0, backfill does not run); restoring it returns the suite to green.

Scope notes

• Additional files touched: game_overrides.cpp (the RE:CVX override migration, above) and ps2_sif_dma_tests.cpp (two existing tests issued SifCallRpc with sid=1, rpcNum=0x12 relying on the old hardcoded constants and never configured a layout; updated to set stateSid=1/getStatusFno=0x12, plus the new real-path regression test). Necessary collateral of the layout-gated dispatch. The debug-panel edits are likewise required collateral of deleting the constants.
• DQ8 end-to-end is not fully served by this struct yet (the DQ8 case is referenced here only as an illustrative example). The SPU2 transfer-handle value pairs (0x410b64 / 0x410b78) and the stream descriptor (0x41b550) are not implemented — they have no enumerated *Fno semantic slot, no defined trigger, and no test coverage. A real DQ8 port would need those added.

Known interactions

• An unknown fno on a served SID writes benignStatusValue (0xffffff9b, the documented "no audio work required" discard sentinel) to recv[0] and then falls through. For a game that muxes its driver onto the LIBSD SID, this means every unknown-fno LIBSD call transiently gets 0xffffff9b in recv[0] before the LIBSD backend runs and overwrites it. This is the intended fall-through design; flagged here for whoever later wires a real game onto that SID.
• After servesSid passes, rpcNum is matched against all *Fno fields regardless of whether the call arrived on commandSid vs stateSid. This is by design (servesSid passes, then rpcNum is compared to the *Fno fields); distinct fno numbers disambiguate in real drivers, and the single-SID muxed case relies on it.

🤖 Generated with Claude Code

[smmathews]

Superseded by the upstream draft PR against ran-j/PS2Recomp: ran-j#154