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

feat: preserve native HD rumble on Switch 2 controllers

Browse source
This commit is contained in:
Joey Yakimowich-Payne 2026-09-07 08:19:53 -06:00
commit 006b573ee8
25 changed files with 1722 additions and 60 deletions
Download patch file Download diff file Expand all files Collapse all files

77
README.md
View file

@ -266,6 +266,12 @@ build services one CYW43 packet per poll and explicitly reschedules remaining
input. Packet-level ring reads and bounded incoming-credit batching reduce
bus work without disabling flow control. `haptics-experiment profile --json`
adds transport timings, clock/voltage settings and packet-size diagnostics.
Switch 2 native output adds separate ingress and output-stage drop counters.
The management response extends from 32 to 40 bytes; the updated host tool
still reads older 32-byte responses and treats their missing counters as
unreported, not zero. These count firmware queue discards/rejections, not
physical actuator-delivery receipts.
### Per-controller profiles
@ -333,15 +339,74 @@ The AIO firmware implements the proprietary BLE protocol for Nintendo `057E:2069
- **Pairing:** fresh SYNC pairing requires the existing bounded pairing window. A directed reconnect must target this adapter's Bluetooth address and match its persistent application-level authorization. These links are unencrypted and are **not authenticated SMP bonds**. No global Bluetooth security downgrade is made; automatic SMP requests for these devices fail closed while other controllers retain their existing policy. Public/static addresses can own profiles; transient private addresses are not promoted to persistent identities.
- **Joy-Con ownership:** an opposite ready half automatically joins a solo half. Either connection order works; the first-ready player slot is retained, with the left controller's identity/profile owning the pair. The right half supplies motion. A disconnected half's inputs and pending effects are removed immediately; the surviving half returns to sideways solo operation and its own profile identity. A lone half has rotated controls and SL/SR shoulders. Two pairs exhaust the four physical Bluetooth connections. Pairing does not reopen discovery outside the existing connection policy.
- **Protocol:** service, characteristic and CCCD UUIDs are discovered rather than trusting fixed ATT handles. Setup requires matching acknowledgements, reads user/factory stick calibration and gyro bias, and rejects malformed/failed transactions. Motion is normalized to the existing SDL-oriented units; sensor clock/range classification and physical axis accuracy still need wider model qualification.
- **Rumble:** strong/weak amplitudes feed fixed low/high carriers, repeated in three-frame HOLD packets on an approximately 13 ms cadence. Keepalives retain active effects instead of silencing them; finite effects expire, XInput held effects persist until replaced/stopped, and teardown cancels output. Pro output mirrors the same two-band effect to both actuators. This does **not** preserve Nintendo HD substeps or independent left/right HD effects and does not use the original Switch-native opt-in backend.
- **Rumble:** Switch HD commands now retain independent left/right frequency/amplitude fields and up to three ordered subframes through the native Switch 2 encoder and bounded queues described below. XInput and local feedback retain their conventional fixed-carrier behavior. This is separate from the original Switch-native opt-in backend.
- **Not implemented:** Joy-Con mouse output, native GameChat signaling, NFC/IR and Switch 2 NSO GameCube support. C and back/rail inputs can instead be remapped to controls the selected USB mode supports.
This is a scoped reimplementation informed by [Bluepad32 PR #219](https://github.com/ricardoquesada/bluepad32/pull/219), reviewed at `9c95e43a87d3bd8a68565da0836d8a758bd8d8af`, not a wholesale fork import. Protocol references: [ndeadly's research](https://github.com/ndeadly/switch2_controller_research), [Nadeflore](https://github.com/Nadeflore/switch2-controllers), [Switch2Connect](https://github.com/TommyWabg/Switch2Connect), and [SDL's Switch 2 sensor implementation](https://github.com/libsdl-org/SDL/blob/main/src/joystick/hidapi/SDL_hidapi_switch2.c).
**Verification:** 287 tests passed; AIO, XInput/feasibility, HD-rumble, haptics and UART firmware variants built. Native protocol tests use the SDK's real BTstack types/accessors and cover discovery, acknowledgement ordering, calibration, persistence, output deadlines and teardown. Lifecycle tests cover both Joy-Con connection orders, multiple pairs, detach/replacement and pairing-policy isolation. The editor's extra/Shift mappings were exercised in Chromium.
**Verification:** 289 tests passed; AIO, XInput/feasibility, HD-rumble, haptics and UART firmware variants built. Native protocol tests use the SDK's real BTstack types/accessors and cover discovery, acknowledgement ordering, calibration, persistence, output deadlines and teardown. Lifecycle tests cover both Joy-Con connection orders, multiple pairs, detach/replacement and pairing-policy isolation. The editor's extra/Shift mappings were exercised in Chromium.
On the flashed Pico, a real Switch 2 Pro (`3C:A9:AB:65:73:12`) completed setup, appeared in persistent pairing/profile inventories, and delivered live sticks, accelerometer, gyro and independent C/GL/GR presses. A 100-report USB rumble exercise retained its connection while 3,033 controller reports arrived. Schema-7 extra mappings were written/read and restored on hardware; all 32 pre-existing profiles, metadata and active selections were compared against a pre-flash backup and preserved, with adapter configuration generation 13 / CRC `3af5ee18` unchanged. Physical rumble feel, Joy-Con 2 pair behavior, long-duration reconnect and mixed-controller transport remain hardware qualification items. Use schema-7-capable firmware after saving expanded profiles.
### Switch 2 native HD rumble
Switch-mode host commands use decoded HD parameters, not the compatibility
strong/weak peak values. Pro output preserves two independent actuators;
paired Joy-Con 2 output routes each source side to its physical half.
A solo Joy-Con uses the louder source independently for each band, retaining
that band's frequency, with left winning ties and shorter sequences holding
their final sample. Profiles scale amplitudes before this conversion.
Physical microphone characterization on the Pro Controller established:
- Each five-byte sample contains two **10-bit frequency + 10-bit amplitude**
fields. The measured frequency model is
`Hz ~= 10 * 2^((code - 1) / 96)`.
Original low/high indices map to `193 + 3*index` / `289 + 3*index`;
index 64 therefore produces codes 385/481 (160/320 Hz).
- Block headers `0x50`, `0x60`, `0x70`, plus the four-bit sequence counter,
select **one, two or three** valid samples. Unused slots are zeroed.
Filling three slots under `0x50` does not play the later slots.
- A randomized 90-packet run using the other actuator as an acoustic timing
reference measured **5.27 ms/frame, ±0.16 ms statistical 95% interval**.
Acoustic/threshold systematic error is not included. The sender uses
conservative **6/11/16 ms** submission guards, not a claim of exact onset.
Linear Q0.15 amplitudes use SDL's conservative `29000/65535` envelope, producing
native codes 0–453. This preserves a linear input curve but is not calibrated
physical-force equivalence; it can feel different from compatibility rumble.
Source frequency indices are bounded to 1–127.
Each logical slot has a 16-command cross-core ingress FIFO; each physical
Switch 2 controller has a 16-command transport FIFO. Commands keep their
original receipt time and connection/output generation. Native commands expire
after 50 ms; a batch that cannot fit its complete playback guard before that
deadline is discarded rather than started halfway stale. Expiring unplayed
history does not interrupt current playback or force a useless HOLD ahead of
fresh work. Consecutive identical one-sample holds may refresh a pending
command; multi-sample sequences are never coalesced.
Stops flush older host work, including under backpressure. Local feedback owns
a separate bounded override while host state advances underneath it; resuming
uses the current valid final sample, not a replay of masked history.
Keepalives likewise send only the final sample with count 1. Pending ATT
write-request buffers remain immutable; late completion cannot resurrect an
old epoch after stop, reconnect or Joy-Con topology change.
Final single-Pro hardware runs:
| Workload | Result |
|---|---|
| Stereo, frequency sweep, and three-subframe patterns (195 USB reports) | Zero ingress/output drops; 1,193 input reports continued |
| 512 changing one-subframe commands at 125.14 Hz | Zero ingress/output drops; clean stop |
| 128 changing three-subframe commands at 125.11 Hz | Zero ingress drops; 69 output-stage commands discarded/superseded; clean stop and empty ingress |
Three-subframe commands at 125 Hz exceed the native playback budget. These
results do **not** establish lossless arbitrary workloads or mixed-controller
radio performance. Joy-Con hardware and perceptual equivalence remain separate
qualification items. All 40 profiles, names, active selections and adapter
configuration were preserved during this upgrade.
### Rumble per controller
Commands remain bound to a USB slot and Bluetooth connection generation. Compatibility output uses a latest-value mailbox; native output keeps a bounded timestamped command history instead of collapsing substeps.
@ -805,15 +870,15 @@ linked binary, not from the larger debug-bearing ELF or UF2 transport file:
| Resource | Used or reserved | Device capacity |
|---|---:|---:|
| Executable flash image | 806,872 bytes | 4 MiB |
| Executable flash image | 815,192 bytes | 4 MiB |
| Indexed profile arenas | 256 KiB | 4 MiB flash |
| Adapter configuration | 8 KiB | 4 MiB flash |
| BTstack bonds and Switch 2 application authorizations | 8 KiB | 4 MiB flash |
| RP2350 terminal sector | 4 KiB | 4 MiB flash |
| Allocated/reserved SRAM, including heap and stacks | 141,984 bytes | 520 KiB |
| Allocated/reserved SRAM, including heap and stacks | 150,824 bytes | 520 KiB |
The executable plus persistent reservations consume 1,089,496 bytes of flash,
leaving 3,104,808 bytes. Allocated SRAM sections leave 390,496 bytes of link-time
The executable plus persistent reservations consume 1,097,816 bytes of flash,
leaving 3,096,488 bytes. Allocated SRAM sections leave 381,656 bytes of link-time
headroom; this is not a runtime heap high-water measurement. Core 0 has a
4 KiB stack, and Core 1 uses a dedicated 16 KiB stack in main SRAM for nested
catalog migration/compaction rather than overflowing its 4 KiB scratch bank.