bump MSC buffer for RP2040 boards exposing SD-over-USB

[mikeysklar]

Workaround for macOS bulk-IN endpoint stalls on RP2040 during multi-sector SD READ(10) probes with the default 1024-byte MSC buffer. SD card never mounts on macOS; Linux works. Bumping the buffer to 16 KB clears the stall.

Applied only when CHIP_VARIANT=RP2040 and CIRCUITPY_SDCARDIO=1. No RAM cost for RP2350 boards or non-SD RP2040 boards.

Tested (CP 10.1.4 stable)

Board	Chip	Default 1024 B	This PR (16 KB on RP2040+SD)
Feather RP2040 Adalogger	RP2040	macOS stalls, gives up after ~5 min	Mounts first try
Metro RP2040	RP2040	LUN 1 never probed (separate bug — #10965)	LUN 1 still never probed (separate bug — #10965)
Metro RP2350	RP2350	Works	Not applied (unchanged)
Fruit Jam (3-LUN)	RP2350	Works	Not applied (unchanged)

Host OSes: macOS 26.x on Apple Silicon M2 (both Mac mini and MacBook Air), Ubuntu 24.04 (kernel 6.17). Linux is unaffected in every cell.

This is a stopgap

Metro RP2350 and Fruit Jam handle the same macOS traffic fine at 1024. The root cause is almost certainly in tinyusb's rp2 device driver or the RP2040 USB peripheral itself. Root-cause ticket: hathach/tinyusb#3607. Once fixed upstream, this conditional can be reverted.

Related

• RP2040: MSC bulk-IN stalls on macOS with default 1024-byte buffer during SD multi-sector reads hathach/tinyusb#3607 — root-cause ticket for the RP2040 bulk-IN stall
• Metro RP2040: SD card LUN never probed by macOS via USB MSC (works on RP2350) #10965 — separate Metro RP2040 LUN 1 probe bug (not addressed by this PR)
• Mass Storage Controller microSD on Metro RP2040 Adafruit_TinyUSB_Arduino#455 — prior Arduino-side report
• shared/supervisor/usb_msc_flash.c: check for valid vfs from get_vfs() #10457 — complementary no-card NULL-check
• Mount SD card at filesystem init so macOS can see it via USB MSC #10963 — companion PR, moves SD automount to filesystem init

[bablokb]

No RAM cost for RP2350 boards or non-SD RP2040 boards.

In py/circuitpy_mpconfig.mk you have: CIRCUITPY_SDCARDIO ?= $(CIRCUITPY_FULL_BUILD), so this will by default turn on CIRCUITPY_SDCARDIO unless it was explicitly disabled. And currently I see no RPxxxx board explicitly disabling this.

[mikeysklar]

You're right — my claim of "no RAM cost for non-SD RP2040 boards" was wrong. With CIRCUITPY_SDCARDIO ?= $(CIRCUITPY_FULL_BUILD) defaulting on, my previous Makefile gate would have bumped the buffer by 15 KB on basically every RP2040 board including ones with no SD hardware (Pico, QT Py, KB2040, etc.). Not OK.

I've pivoted to a per-board opt-in: boards that actually expose an SD card over USB MSC set CIRCUITPY_USB_MSC_LARGE_BUFFER = 1 in their mpconfigboard.mk, and the port Makefile gates the bump on that variable. Default stays at 1024 — zero RAM cost for every existing board, since I haven't opted in any of them upstream (no RP2040 boards in upstream have DEFAULT_SD_* defined anyway). Anyone with custom firmware or a future board with SD hardware can opt in with one line.

Verified locally on a Feather RP2040 Adalogger (with the opt-in added for testing): with CIRCUITPY_USB_MSC_LARGE_BUFFER = 1 → buffer compiles to 16384, SD mounts on macOS 26.x first try. Without it → buffer compiles back to 1024, confirmed via make -n.

Updated in the latest commit. Thanks for the catch.