jyapayne/switch-pico
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix(native-usb): prevent transmit-bank starvation and retain battery telemetry

Browse source
This commit is contained in:
Joey Yakimowich-Payne 2026-09-13 21:00:17 -06:00
commit a699a6920b
16 changed files with 342 additions and 82 deletions
Download patch file Download diff file Expand all files Collapse all files

51
README.md
View file

@ -1002,10 +1002,53 @@ the old standalone Wii nominal-charge fallback is removed.
Power-field regressions fail before the correction and pass afterward. Eight
focused regressions and both firmware builds pass; a PC capture of real
motion-bearing input verified Power Info `0x01` on both halves. The
`0.72-native-usb-power` trial is installed with persistent storage verified
unchanged. A sustained Switch/idle test is still required before attributing
the timeout to this flag or claiming recovery. Additional trace-only logs
include raw USB interrupts, frame count and device-watchdog state.
`0.72-native-usb-power` trial was installed with persistent storage verified
unchanged. Later `0.76` passed a ten-minute PC mixed input/control/bulk soak but
still lost input on Switch. Captured failures end after a completed root hub
endpoint-halt clear, with addressed-token and root NAK counters stopped while
SOF continues. This also occurs without USB receive-error flags; the disconnect
cause is not established. Trace-only logs include raw USB interrupts, frame
count and device-watchdog state.
**Wii battery telemetry (0.77):** the Wii parser previously ignored the battery
byte in status report `0x20` and cleared the battery field on every input report.
It now retains the latest measurement between reports, rejects truncated status
updates and clears the old measurement on parser setup. The voltage-derived
capacity uses the same 5/20/70/100-percent bands as
[SDL's Wii driver](https://github.com/libsdl-org/SDL/blob/main/src/joystick/hidapi/SDL_hidapi_wii.c).
The USB bridge forwards that measured level with external power set and charging
clear, rather than forcing full battery. Updates occur when a status report
arrives; this change adds no periodic status requests. Parser regressions cover
retention, low-battery updates and reconnection.
The installed `0.77` PC smoke test read source battery `179/255` (the 70-percent
band) and Power Info `0x19` from both native children, with live IMU and mixed
control/bulk checks passing. Saved flash data was verified unchanged. The
connected-controller JSON percentage also uses the full `1..255` battery range,
so full charge is 100 percent rather than 102. The subsequent Switch trial still
disconnected around 240 seconds while delivering Power Info `0x19`; correcting
false-empty battery reporting alone did not resolve the failure.
**Private transmit-bank publication (0.78):** the earlier
PC checker submitted R/L reads sequentially. USBmon confirmed zero overlapping
child reads. With one reader per child and 500-ms application timeouts, `0.77`
stopped delivering fresh input within about three seconds. Short 10-ms timeouts
had repeatedly cancelled those pending reads and obscured this starvation.
Bank selection had disabled every IN buffer until Core 0 restored the bank,
including HID and bulk payloads already stored in private DPRAM. `0.78` publishes
those prepared private buffers during selection; only the shared EP0 IN image
still waits for Core 0 copying. A regression polls alternating child HID/bulk
endpoints without a foreground task and checks payload isolation, completion,
and newly queued data while EP0 restoration is pending. It fails before the fix
and passes afterward; 14 focused tests and both native firmware builds pass.
The image is installed with persistent storage verified unchanged. With the Wii
reconnected, the same concurrent 500-ms-read test passed 600 seconds: 145,746 R
packets and 149,567 L packets, including 113,316 and 114,276 fresh IMU packets.
USBmon measured overlapping child reads during 94.4% of the steady-state window
with no failed read completions there and no capture drops. A subsequent
15-second mixed control/bulk/input check passed 57 rounds with no errors.
This confirms the PC starvation reproduction is corrected. The user's subsequent
Switch gameplay trial also stopped reproducing the L+R disconnect.
For sensorless hardware, the checker supports `--input-only`: press real buttons
and keep changing controls on both halves during the run. Neutral fallback