TinyUSB re-pin fixes ESP32-S2 and S3 REPL race

[mikeysklar]

What

Moves the CircuitPython TinyUSB pin to current master (441 commits).

Adds one line so nRF52 boards still build.

Why

Pasting more than a few lines into the ESP32-S2 or S3 REPL drops and scrambles characters. TinyUSB fixed it upstream. Might as well re-pin to get the fix.

Link	What it is
hathach/tinyusb#3937	the fix, merged 2026-09-21
hathach/tinyusb#1292	the bug report, now closed
#11354	the CircuitPython-side attempt, closed in favour of fixing TinyUSB

CircuitPython changes

Two lines in supervisor/shared/usb/tusb_config.h. Every TinyUSB source file and option CircuitPython uses still exists.

Line	Why
+#define nrf52_errata_199 errata_199	newer dcd_nrf5x.c calls it; nordic fails with implicit declaration without it. Same mapping as the three already there. First commit, so every commit builds.
-#define CFG_TUH_VENDOR 0	upstream deleted the vendor host class and the option

Flash cost

All nine boards build clean on both sides with GCC 15.2.1, no new warnings. Bytes free, en_US:

Board	main 9701f33a8f	this PR 1e55bbdcff	delta
metro_m0_express	4236	3964	-272
metro_m4_airlift_lite	16232	15940	-292
feather_nrf52840_express	140648	140040	-608
feather_stm32f405_express	374100	373876	-224
adafruit_metro_rp2040	84532	83740	-792
adafruit_metro_rp2350	124900	124092	-808
raspberry_pi_pico2_w	56928	56136	-792
adafruit_metro_esp32s2	1299536	1298848	-688
adafruit_metro_esp32s3	157392	156688	-704

Hardware tested

All eight farm boards on 10.4.0-alpha.2-26-g1e55bbdcff, Ubuntu 24.04 host.

Board	2000 REPL pastes, corrupted	5 x 64 KB CDC, lost or foreign bytes	KB/s	MSC write and read-back
Metro ESP32-S2	0	0	64.1	OK, 8 KB
Metro ESP32-S3	0	0	64.0-64.3	OK
Metro M0 Express	n/a	0	40.0-40.4	OK
Metro M4 AirLift Lite	n/a	0	185.7-187.8	OK
Feather nRF52840 Express	n/a	0	92.0-93.5	OK
Feather STM32F405 Express	n/a	0	254.0	OK
Metro RP2040	n/a	0	117.7-118.5	OK
Metro RP2350	n/a	0	200.4-201.6	OK

How I tested it

Metro ESP32-S3. 40 test files, 50 times each, through the raw REPL:

### flash repin 2026-09-21T07:43:35-07:00
after: mnt=/media/sklarm/CIRCUITPY2 Adafruit CircuitPython 10.4.0-alpha.2-26-g1e55bbdcff on 2026-09-21; Adafruit Metro ESP32S3 with ESP32S3
### burn repin 2026-09-21T07:46:39-07:00
ESP32-S2: 0 failed iterations of 50, 2000 pastes
ESP32-S3: 0 failed iterations of 50, 2000 pastes

[s3 stress] /media/sklarm/CIRCUITPY2
  before:  Adafruit CircuitPython 10.4.0-alpha.2-26-g1e55bbdcff on 2026-09-21; Adafruit Metro ESP32S3 with ESP32S3
    rep 0: n=65536/65536 bad=0 first_bad=-1 1019 ms first->last = 64.3 KB/s, host write 1019 ms, intact=True
    rep 1: n=65536/65536 bad=0 first_bad=-1 1023 ms first->last = 64.1 KB/s, host write 1023 ms, intact=True
    rep 2: n=65536/65536 bad=0 first_bad=-1 1024 ms first->last = 64.0 KB/s, host write 1024 ms, intact=True
    rep 3: n=65536/65536 bad=0 first_bad=-1 1023 ms first->last = 64.1 KB/s, host write 1023 ms, intact=True
    rep 4: n=65536/65536 bad=0 first_bad=-1 1023 ms first->last = 64.1 KB/s, host write 1023 ms, intact=True
    64k write+read-back OK
### STRESS DONE s3
[s3 restore] /media/sklarm/CIRCUITPY2
  firmware line matches pre-test
  files match backup (23 files)

The 64 KB test, sent over the REPL:

import sys,time,supervisor,os
def run(N):
    c=sys.stdin.read(1)
    t=time.monotonic_ns(); last=t; n=1
    bad=0 if c=='x' else 1
    fb=-1 if c=='x' else 0
    while n<N:
        k=supervisor.runtime.serial_bytes_available
        if k:
            d=sys.stdin.read(min(k,N-n))
            x=d.count('x')
            if x!=len(d) and fb<0:
                fb=n+[i for i in range(len(d)) if d[i]!='x'][0]
            bad+=len(d)-x; n+=len(d); last=time.monotonic_ns()
        elif time.monotonic_ns()-last>2000000000:
            break
    print('GOT',n,(last-t)//1000000,bad,fb)

AI assistance

Claude Code was used. It made the change and ran the builds and farm. I checked the results.

[tannewt]

Code changes looks fine. I poked the CI