SDMMC Data End Interrupt sometimes does not occur

[by-gnome]

Describe the set-up
The device based on STM32L496VG with high load application. ICCARM v8.32, STM32Cube_FW_L4_V1.18.1.

Describe the bug
Writing to an SD memory card using SDMMC+DMA sometimes times out in the SD_write() function. This happens because the DATAEND interrupt does not occur.
After some debugging I found that the SD_DMATransmitCplt() function enables the DATAEND interrupt by setting the DATAENDIE bit, but interrupt does not occur, and after the timeout in the SD_write() function the DATAEND bit is set and the DATAENDIE bit is not set.

How To Reproduce
SDMMC parameters:

• Clock source - HSI48 RC
• Clock transition - Rising
• Clock divider bypass - Enable
• Clock power save - Enable
• HW flow control - Enable
• Wide bus mode - 4 bit
• DMA - DMA2 channel 5

The application writes data to the SD card for 12-24 hours, with a data rate of 128 kB/s. After writing several gigabytes of data, a timeout occurs in the SD_write() function.

Additional context
Let's consider the SD_DMATransmitCplt() function:

static void SD_DMATransmitCplt(DMA_HandleTypeDef *hdma)
{
  SD_HandleTypeDef* hsd = (SD_HandleTypeDef* )(hdma->Parent);

  /* Enable DATAEND Interrupt */
  __HAL_SD_ENABLE_IT(hsd, (SDMMC_IT_DATAEND));
}

In case of abnormal operation, before setting DATAENDIE, the SDMMC_STA register has the state 0x00105400:
TXDAVL = 1, TXFIFOHE = 1, TXACT = 1, DBCKEND = 1 or 0 (doesn't matter)
and SDMMC_MASK register has the state 0x00000000 !!!

After setting DATAENDIE, the SDMMC_STA register has the state 0x00105400:
TXDAVL = 1, TXFIFOHE = 1, TXACT = 1, DBCKEND = 1 or 0 (doesn't matter)
and SDMMC_MASK register has the state 0x00000100:
DATAENDIE = 1

After timeout in SD_write the SDMMC_STA register has the state 0x00000500:
DBCKEND = 1, DATAEND = 1
and SDMMC_MASK register has the state 0x0000001A:
TXUNDERRIE = 1, DTIMEOUTIE = 1, DCRCFAILIE = 1

The correct values ​​in the SDMMC_MASK register are 0x0000001A and 0x0000011A. Therefore, it can be assumed that in the SD_DMATransmitCplt() function, reading/writing the SDMMC_MASK register is sometimes disrupted.

Possible solution
If I set the DATAENDIE bit in the HAL_SD_WriteBlocks_DMA() function, the problem does not occur.

--- a/Drivers/STM32L4xx_HAL_Driver/Src/stm32l4xx_hal_sd.c
+++ b/Drivers/STM32L4xx_HAL_Driver/Src/stm32l4xx_hal_sd.c
@@ -1566,7 +1566,7 @@ HAL_StatusTypeDef HAL_SD_WriteBlocks_DMA(SD_HandleTypeDef *hsd, uint8_t *pData,
       (void)SDMMC_ConfigData(hsd->Instance, &config);
 
       /* Enable SD Error interrupts */
-      __HAL_SD_ENABLE_IT(hsd, (SDMMC_IT_DCRCFAIL | SDMMC_IT_DTIMEOUT | SDMMC_IT_TXUNDERR));
+      __HAL_SD_ENABLE_IT(hsd, (SDMMC_IT_DCRCFAIL | SDMMC_IT_DTIMEOUT | SDMMC_IT_TXUNDERR | SDMMC_IT_DATAEND));
 #else
       /* Enable transfer interrupts */
       __HAL_SD_ENABLE_IT(hsd, (SDMMC_IT_DCRCFAIL | SDMMC_IT_DTIMEOUT | SDMMC_IT_TXUNDERR | SDMMC_IT_DATAEND));
@@ -3225,9 +3225,6 @@ HAL_StatusTypeDef HAL_SD_Abort_IT(SD_HandleTypeDef *hsd)
 static void SD_DMATransmitCplt(DMA_HandleTypeDef *hdma)
 {
   SD_HandleTypeDef* hsd = (SD_HandleTypeDef* )(hdma->Parent);
-
-  /* Enable DATAEND Interrupt */
-  __HAL_SD_ENABLE_IT(hsd, (SDMMC_IT_DATAEND));
 }

[by-gnome]

Project configuration

Clock configuration

[KRASTM]

Hello @by-gnome,

Thank you for the report.

Regarding the issue, I tested an application with STM32L496G-Disco and work just fine, it's under Projects\32L496GDISCOVERY\Applications\FatFs\FatFs_uSD_DMA_RTOS

However, I will share your proposal to our team in order to analyze it. In the meanwhile, I will try to investigate more and test an example using your config, especially as you mentioned The application writes data to the SD card for 12-24 hours, with a data rate of 128 kB/s. After writing several gigabytes of data, a timeout occurs in the SD_write() function, it's a lot of time and Data.

Just a few questions, if you may:

• Did you do the test several times, and each time you have this error or timeout?
• You project with a custom board or ST board?

With regards,

[by-gnome]

The project runs on a custom board.
I tested for several weeks. The problem does not always occur, but quite often. More often with the FAT32 file system than with the exFAT. It might also depend on interrupt priorities and system load.
This problem might be similar to STM32CubeL4/pull/35

[by-gnome]

To localize the problem I modified the SD_DMATransmitCplt() and SD_write() as follows:

static void SD_DMATransmitCplt(DMA_HandleTypeDef *hdma)
{
  SD_HandleTypeDef* hsd = (SD_HandleTypeDef* )(hdma->Parent);
  SdDmaTxCplt1 = SDMMC1->STA | 0x80000000;
  SdDmaTxCplt0 = SDMMC1->MASK | 0x80000000;
  /* Enable DATAEND Interrupt */
  __HAL_SD_ENABLE_IT(hsd, (SDMMC_IT_DATAEND));
  SdDmaTxCplt3 = hsd->Instance->STA | 0x80000000;
  SdDmaTxCplt2 = hsd->Instance->MASK | 0x80000000;
}

DRESULT SD_write(BYTE lun, const BYTE *buff, DWORD sector, UINT count)
{
  uint32_t tickstart;
  WriteStatus = 0;
  if (BSP_SD_WriteBlocks_DMA((uint32_t*)buff, (uint32_t)(sector), count) == MSD_OK)
  {
    tickstart = HAL_GetTick();
    while ((WriteStatus == 0) && ((HAL_GetTick() - tickstart) < SD_TIMEOUT));
    if (WriteStatus != 0) {
      do {
        if (BSP_SD_GetCardState() == SD_TRANSFER_OK) {
          return RES_OK;
        }
      } while ((HAL_GetTick() - tickstart) < SD_TIMEOUT);
    }
  }
  SdmmcTxCplt3 = hsd1.Instance->STA | 0x80000000;
  SdmmcTxCplt2 = hsd1.Instance->MASK | 0x80000000;
  return RES_ERROR;
}

When timeout, the variables contain the following values:

SdDmaTxCplt1 - 0x80105400
SdDmaTxCplt0 - 0x80000000
SdDmaTxCplt3 - 0x80105400
SdDmaTxCplt2 - 0x80000100
SdmmcTxCplt3 - 0x80000500
SdmmcTxCplt2 - 0x8000001a

It might be a hardware issue with SDMMC_MASK register.

[KRASTM]

ST Internal Reference: 196054