Can't erase flash on STM32407VG Discovery

[Alan Chambers]

I'm a bit lost by this one. I can write to the flash but cannot erase a sector. I can erase sectors with the ST-Link Utility. The result codes at every step indicate success, but the flash is not actually erased. I am using the Standard Peripheral Library as follows (the status checks with while (true) are just part of the debugging effort):

void FlashDriver::erase_sector(uint8_t sector_index)
{
 DisableInterrupts di;
 // Unlock the Flash Program Erase controller 
 if (FLASH_GetStatus() != FLASH_COMPLETE)
 while (true); 
 FLASH_Unlock();
 
 // Clear pending flags (if any)
 FLASH_ClearFlag(
 FLASH_FLAG_EOP |
 FLASH_FLAG_OPERR |
 FLASH_FLAG_WRPERR |
 FLASH_FLAG_PGAERR |
 FLASH_FLAG_PGPERR |
 FLASH_FLAG_PGSERR
 );
 
 // Erase the given logical sector.
 if (FLASH_GetStatus() != FLASH_COMPLETE)
 while (true);
 if (FLASH_EraseSector(sector_index, VoltageRange_3) != FLASH_COMPLETE)
 {
 ...
 }
 
 if (FLASH_GetStatus() != FLASH_COMPLETE)
 while (true);
 FLASH_Lock(); 
}

I'd be grateful for any insights on this one.

Al

[Alan Chambers]

I should add that I converted the Discovery board to work with J-Link (using a utility from SEGGER). I don't know if this might have some impact on fuses or other features which might affect the flash. Kind of clutching at straws.

[Alan Chambers]

Fixed it. The variable sector_index is a logical index in my implementation: 0, 1, 2, ... but FLASH_EraseSector() expects FLASH_Sector_0, FLASH_Sector_1, FLASH_Sector_2 which are completely different values. Silly error that was far too easy to make. Feeling sheepish.

I started writing a custom HAL some time ago that would have made this bug almost impossible - the register field values are typesafe enum classes. Perhaps I should return to that project. I won't use Cube's HAL under any circumstances, but I don't suppose SPL is going to support STM32MP1 (say).

[Tesla DeLorean]

>>but FLASH_EraseSector() expects...

ST seems to favour bit patterns it can apply directly rather than indexes it can shift into place.

For the F2/F4 I'd built most of this in assembler, and had to code routines to compute the variable/non-uniform sector sizes. Extra fun on the 2MB parts, where the 16K/64K/128K patterns were repeated at the beginning of the second bank, and the index was discontinuous.

The code actually turned out reasonably compactly, I was aiming for something I could copy to RAM without a rabbit warren of call dependencies.

Hope for SPL is rather grim, we'll just have to frankenstein the usable parts from HAL...