FNK0104B: SD slot is SDMMC 4-bit and SdProfile describes SPI

[iamankushpandit]

SdProfile describes an SPI-attached card: cs / mosi / miso / sclk / spiHz, mounted through SD.begin() on a SPIClass. The Freenove FNK0104B's slot is wired for SDMMC in 4-bit mode, which has no chip select and is a different peripheral entirely. The board profile therefore sets every field to PIN_NONE.

This is a supported configuration, not a fault: sdReady() returns false, optional SD content is not loaded, and everything that reads it already has defaults. See docs/SD_CONTENT_SPEC.md for what is actually optional.

What a user notices

Very little today — SD content is optional across the board. The gap matters because the FNK0104B has 16 MB of flash and 8 MB of PSRAM, so it is the first board where large optional content is genuinely practical.

The hardware

SDMMC 4-bit, from Tutorial_With_Touch/Sketches/Sketch_06.1_SDMMC_Test:

Signal	GPIO
CLK	38
CMD	40
D0	39
D1	41
D2	48
D3	47

Vendor's comment on every one of them is "Please do not modify it" — these are fixed-function pins for the SDMMC host on this layout.

The 3.5-inch FNK0104N variant uses CLK 5 / CMD 4 / D0 6 / D1 7 / D2 2 / D3 3. Do not assume one map across the family.

Arduino core 2.0.17 (pinned via espressif32@7.0.1) exposes this as SD_MMC.setPins(clk, cmd, d0, d1, d2, d3) followed by SD_MMC.begin().

What needs doing

1. Extend SdProfile to describe which kind of slot the board has — the discriminated-record pattern TouchProfile now uses. Flat and /* fieldName */-labelled, because tools/check_boards.py reads those labels; unused lines PIN_NONE.
2. Fill the new fields in for all four boards in the same commit, or the checker fails and a board boots with a silent zero.
3. Implement the SDMMC mount in Board::mountSd() (src/hal/Board.cpp) behind the existing sdReady() / mountSd() API, so nothing above the HAL changes.
4. Guard with #if, not if constexpr — the GUME_TOUCH_CAPACITIVE precedent in include/boards/*.h. Both arms of an if constexpr compile in a non-template function; the touch equivalent linked Wire into every resistive board for +4,476 bytes of dead code. SD_MMC and SD are both non-trivial and flash is at ~75%.
5. Note that Board::Board() constructs sdSpi_(VSPI) unconditionally. On a board with no SPI card that member is pure waste and should be conditional too.
6. Update README.md (the FNK0104B limitations table), src/hal/CLAUDE.md, docs/PORTING.md's optional-peripheral table, and the flash/RAM figures in README.md and CLAUDE.md from your own pio run.

Worth deciding at the same time

The board has 16 MB of flash and currently uses huge_app.csv, a 4 MB table — about 12 MB is stranded. If SD content is the motivation for this work, a larger partition table with a data partition may serve better than a card. That is a separate decision but the same conversation; raise it before building an SDMMC backend nobody needs.

Verifying

Add a mount-and-list step to pio run -e s3diag first: it is standalone, so it confirms the six pins and the 4-bit mode before any HAL work depends on them.