USBD rewrite (b1f7ff96) breaks Crash Bandicoot: Wrath of Cortex over USB in OPL

[horizonends]

Summary

After the 2024-09-04 USBD feature update (b1f7ff96), Open PS2 Loader builds that embed the current usbd_mini / usbmass_bd IRX fail to run Crash Bandicoot: The Wrath of Cortex from USB on real hardware.

Rebuilding the same modern OPL with USB IRX from immediately before that rewrite (2dc6b32f) restores normal behaviour.

This looks like a ps2sdk USBD (and/or usbmass_bd pairing) regression, not an OPL game-code bug.

Symptoms (hardware)

On broken (post-rewrite) USB stack:

• Load to game menu can take up to ~5 minutes (good case is ~10 seconds)
• Starting play then goes to a black screen

On pre-rewrite USB stack (same OPL otherwise):

• ~10s to menu, playable

Reproduction

• Consoles: PAL slim PS2s (SCPH-70003 / 75003 / 77003)
• Device: USB ISO
• Games: NTSC SLUS_202.38 and PAL SLES_503.86 (both fail the same way)
• Host loader: Open PS2 Loader (embeds SDK IRX at build time)

OPL stock bisect (same USB/ISOs):

• Last good OPL Beta build era: Beta-2125
• First bad: Beta-2127
• That window lines up with OPL builds that baked in post-rewrite USBD

Confirmed isolation

Build	Result
Modern OPL + current SDK usbd_mini / usbmass_bd	Broken (slow load → black screen)
Modern OPL (Beta-2245 tree) + USB IRX from ps2sdk @2dc6b32f (pre-b1f7ff96)	Works (~10s, playable)
Stock OPL Beta-2245	Broken

So only swapping the embedded USB IRX back to pre-rewrite fixes it.

Suspected commit

• Introduced: b1f7ff96 — USBD feature update (2024-09-04)
  (“usbd 1.2 API now exposed, module unloading, isochronous scheduling”)
• Last known-good tree for this use case: 2dc6b32f (parent side / immediately before that update)

Related follow-ups in the same era (imports etc.) may also matter; happy to help narrow further if useful.

Related reports

• OPL: [ISSUE]: Crash Bandicoot: The Wrath of Cortex — very slow load + black screen on play (USB); last good Beta-2125 ps2homebrew/Open-PS2-Loader#1751
• wOPL (Double OPL): [ISSUE]: Crash Bandicoot: The Wrath of Cortex — very slow load + black screen on play (USB); upstream last good Beta-2125 ps2homebrew/wOPL#375
  (related symptoms; wOPL CI is not floating :main, but still embeds post-rewrite usbd_mini. A naïve old-IRX pin that works on upstream OPL did not apply cleanly there.)

Notes

• Temporary OPL-side workaround (vendor/pin pre-rewrite IRX) is confirmed on hardware for upstream OPL, but maintainers correctly prefer a real fix here rather than permanently shipping 2024 USB binaries.
• I can provide the pre/post IRX hashes / sizes and help test SDK patches on hardware if needed.

Request

Please investigate USBD (and usbmass_bd if involved) behaviour changes since b1f7ff96 that could cause severe USB read slowdown / failure for this game’s access pattern under OPL’s BDM USB path.

[horizonends]

Opened a candidate fix PR: will link once created.

Changes in device.c:

1. Zero m_transferedBytes on attach (regression vs FreeUsbd)
2. Remap ClearFeature endpoint index from pipe id → bEndpointAddress, reset host toggle

Hardware test ELF with patched usbd_mini + modern usbmass is ready for confirmation.

[horizonends]

Update: the ClearFeature pipe-id remap candidate in #909 was tested on hardware (OPL + wOPL). It did not reproduce the original WoC symptom — instead USB and memory card read/write were completely broken. That PR is closed.

Known-good workaround remains: pin pre-b1f7ff96 usbd_mini + usbmass_bd_mini. Looking at other FreeUsbd vs rewrite deltas next (open-endpoint MPS/align, interrupt bandwidth scheduling, etc.).

[horizonends]

Correction to the previous hardware note (more precise results for niall-sdk908-both):

• OPNPS2LD sdk908 — Fail, black screen (same WoC failure mode; candidate did not fix USBD rewrite (b1f7ff96) breaks Crash Bandicoot: Wrath of Cortex over USB in OPL #908). USB mass was still usable enough to reach that path.
• WOPNPS2LD sdk908 — Does not recognise the USB stick; fails to save OPL settings (likely memory card not seen either). Different failure from OPL.

So #909 is still a dead end for WoC; the earlier “USB and MC completely broken on both” wording was too strong / wrong for OPL. PR remains closed.

[horizonends]

Hardware A/B update

Further candidates on modern USBD alone failed (same WoC: slow load → menu → black screen):

• open-endpoint MPS/align restore
• FreeUsbd interrupt interval map / disable SCE bandwidth scheduling
• ClearFeature pipe remap (usbd: fix ClearFeature pipe-id remap and IoReq byte count (#908) #909) — already closed

New split that works on hardware:

Build	usbd	usbmass	WoC USB
stock modern	new	new	fail
latest-usbfix	old (pre-b1f7ff96)	old	works
hybrid oldusbd-newmass	old	new	works

So the regression is in modern USBD alone — modern usbmass_bd is fine when paired with pre-rewrite usbd. Looking at remaining FreeUsbd vs rewrite deltas next (HCD bring-up, GenTD done-queue / halt recovery, bulk completion paths).