[pingequalab]

Summary

Follow-up to #3082 (fefc13b). Fixes silent single-sector / tail-sector write drops (CMD24 / sdWriteSector()) and premature sdWait() return on shared-SPI / pulled-up MISO buses in lib/HAL/sd_card/sd_diskio.cpp.

(#3082 was merged into dev shortly before our 20-round byte-level MD5 single-sector hardware stress test finished; this PR submits the companion busy-wait fix 6beb4ae caught and verified by that stress suite.)

Hardware Tested

• Board: m5stack-sticks3 with external shared-SPI module (CC1101 + NRF24 + 32GB MicroSD sharing SCK=5, MISO=4, MOSI=6), verified under both USB power and standalone internal battery power.
• Applies to all SPI SD targets using lib/HAL/sd_card/sd_diskio.cpp.

Root Cause

• After Fix SPI SD card remount zombie state, MISO release clock, and USBSerial vprintf #3082 added the trailing 0xFF clock in sdDeselectCard() so the SD card properly releases DAT0/MISO to pulled-up 0xFF (1 kOhm pull-up on the shared SPI bus), sdWriteSector() (CMD24) still deasserted CS immediately after reading the data-response token (0x05) without waiting for internal NAND programming to finish.
• Immediately afterward (~1 us), sdTransaction(SEND_STATUS) (CMD13) re-asserted CS and called sdWait(pdrv, 500). Because sdWait() previously exited as soon as a single byte was non-zero (while (resp == 0x00) with signed char resp, returning resp > 0x00), the initial non-zero N_CS transition byte sampled on the pulled-up MISO line before the card drove DO low (0x00 busy) caused sdWait() to return true prematurely.
• Subsequent CMD13 / CMD24 commands collided with active NAND programming, silently dropping <512 B files, tail sectors, and directory-entry updates without logging an error.

Changes (lib/HAL/sd_card/sd_diskio.cpp)

• sdWait(): Use uint8_t resp and require two 0xFF idle bytes (resp == 0xFF && ++ff_count == 2) before returning true, so an initial N_CS transition byte after CS assertion cannot prematurely pass a busy check.
• sdSelectCard(): Call sdDeselectCard(pdrv) instead of raw digitalWrite(card->ssPin, HIGH) on sdWait() timeout so MISO is cleanly clocked to Hi-Z.
• sdWriteSector() (CMD24): Check uint8_t token == 0x05 (deselecting and retrying on 0x0A / 0x0B CRC error), call sdWait(pdrv, 500) while CS is still held LOW before sdDeselectCard(pdrv) (matching multi-sector sdWriteSectors() / CMD25), and initialize unsigned int resp = 0.

Verification

• Build: pio run -e m5stack-sticks3 -> SUCCESS (RAM: 40.2% / 131,748 B, Flash: 56.5% / 4,736,991 B).
• 20-round automated hardware stress suite (OVERALL=PASS | S0=True | S1_PureSD=5/5 | S2_Stress=15/15 | Clean=True):
  • S0 + S1 (5 rounds): Multi-sector (CMD25) + tail single-sector (CMD24) writes (2,880 B & 1,920 B), byte-level MD5 verification, 12-sector golden reference CRC32 readback, and directory-entry remove -> 5/5 PASS, 0 SPI/SD errors.
  • S2 (15 rounds): Continuous shared-SPI switching across CC1101 RSSI Spectrum -> NRF24 80-ch 10MHz Spectrum -> UI Files -> SD Card directory browse -> dynamic multi/single-sector SD write + MD5 + 12-sector CRC32 + remove -> 15/15 PASS, 0 Card Failed / Select Failed / Wait Failed, stable heap (113,219 B -> 111,123 B) and PSRAM (8,370,212 B -> 8,370,212 B).
• Standalone battery stress test: Verified under internal battery operation with concurrent RF + MicroSD stress load -> PASS.