I can confirm this behavior. The standard of Micro-SD cards demands that initialization is done with 1 bit bus and a slow clock. This is already perfectly implemented by HAL_SD_Init using an own init definition with 1 Bit and slower speed. However, when trying to widen the bus to 4 bit the described error above occurred.
After a lot of tries I found a better strategy: For your own high speed SD_InitTypeDef definition start with a slow variant (1 bit + clock divider) and increase speed before widening the bus. By this the full SD card initialization and the first part of the widen bus configuration is done with the robust slow speed. No worries the fast speed configuration is still applied because the HAL_SD_ConfigWideBusOperation also calls another round of SDIO_Init.
Here is a code example:
uint8_t BSP_SD_Init(void)
{
uint8_t SD_state = MSD_OK;
/* uSD device interface configuration */
uSdHandle.Instance = SDIO;
uSdHandle.Init.ClockEdge = SDIO_CLOCK_EDGE_RISING; /* warning NEG edge does not work. See errata */
uSdHandle.Init.ClockBypass = SDIO_CLOCK_BYPASS_DISABLE;
uSdHandle.Init.ClockPowerSave = SDIO_CLOCK_POWER_SAVE_DISABLE;
uSdHandle.Init.BusWide = SDIO_BUS_WIDE_1B; /* Must be one bit for initial configuration. Widened later */
uSdHandle.Init.HardwareFlowControl = SDIO_HARDWARE_FLOW_CONTROL_DISABLE; /* warning HW flow control does not work. See errata */
uSdHandle.Init.ClockDiv = SDIO_INIT_CLK_DIV; /* bypassed for ~ +20% performance */
/* Check if the SD card is plugged in the slot */
if(BSP_SD_IsDetected() != SD_PRESENT)
{
return MSD_ERROR;
}
/* Msp SD initialization */
BSP_SD_MspInit(&uSdHandle, NULL);
if(HAL_SD_Init(&uSdHandle) != HAL_OK)
{
SD_state = MSD_ERROR;
}
/* Configure SD Bus width */
if(SD_state == MSD_OK)
{
/* This is an essential Delay. Otherwise the following WideBusOperation can get stuck
* in optimized builds. It is unsure and irrelevant if this behavior is caused by software
* or hardware (RC-Delays, 4 bit lines in different lengths,...)
*/
OSAL_SleepMs(1);
uSdHandle.Init.BusWide = SDIO_BUS_WIDE_4B;
uSdHandle.Init.ClockBypass = SDIO_CLOCK_BYPASS_ENABLE; /* Use 48MHz generated by PLL directly for SDIO clock line (instead of PCLK2 /(2+uSdHandle.Init.ClockDiv)) */
uSdHandle.Init.ClockDiv = SDIO_TRANSFER_CLK_DIV; /* bypassed for ~ +20% performance */
/* Enable wide operation */
if(HAL_SD_ConfigWideBusOperation(&uSdHandle, uSdHandle.Init.BusWide) != HAL_OK)
{
SD_state = MSD_ERROR;
}
else
{
SD_state = MSD_OK;
}
}
return SD_state;
}
By this the failure rate of the above described problem is reduced to 0 (before it was about 1 in 3-6 initializations). I am also pretty sure that the 1 millisecond sleep is not needed anymore.
