STM32H753xI SDMMC data corruption with IDMA and CPU DCACHE enabled

[XHach.1]

Reading blocks from a SD card often result in corrupted data if IDMA and CPU Data Cache are enabled. I am not sure if this falls under erratum 2.1.1 of STM32H753xI. Attached are the STM32Cube project and modified main I used to run tests.

The code in the main reads alternatively the first and second blocks of the SD card and resets the read buffer to 0x55 as a sanity check to verify that internal memory is written. In my test, the connected SD card had a valid FAT32 file system so I knew what data should be read.

Using HAL_SD_ReadBlocks_DMA() to read a block always results in corrupted data while CPU DCACHE is enabled. There is however no corruption if either reading blocks using HAL_SD_ReadBlocks_DMA() when CPU DCACHE is disabled (commenting SCB_EnableDCache() in main.c) or if using HAL_SD_ReadBlocks_IT() / HAL_SD_ReadBlocks() when CPU DCACHE is enabled.

I first encountered this issue in another code base I am working on where it is desired to have CPU DCACHE enabled and use SDMMC IDMA. Is this a real issue or am I missing something?

[Tesla DeLorean]

>>Is this a real issue or am I missing something?

Cache coherency is a real problem, that *you* have to manage.

You can do it from the MPU configuration.

You can do it with DCache Clean / Invalidate by Address functions.

Perhaps some examples in the FatFs DISKIO DMA code, see

STM32Cube_FW_F7_V1.16.0\Middlewares\Third_Party\FatFs\src\drivers\sd_diskio_dma_template_bspv1.c

#define ENABLE_SD_DMA_CACHE_MAINTENANCE 1

[mƎALLEm]

Hello,

Indeed, this is a data cache coherency issue.

We suggest you to read the AN4839 "Level 1 cache on STM32F7 Series and STM32H7 Series"

[XHach.1]

Thanks for the info.