How do I use HAL_SPI_Receive_DMA in conjunction with a receive buffer on the stack and caching enabled?

[arnold_w]

I am working with the STM32F760I-EVAL board and caching (SCB_EnableICache() and SCB_EnableDCache() are called) and MPU (MPU_Config() is called) are enabled. I am using DMA with SPI and if I transmit data, HAL_SPI_Transmit_DMA, I need to call SCB_CleanDCache() to see any MOSI-data on the oscilloscope. What if want to receive data using HAL_SPI_TransmitReceive_DMA, what function do I need to call and at what point? I tried calling SCB_InvalidateDCache() when my receive is finished, but I then got stuck in an infinite loop inside SCB_InvalidateDCache():

  do {

   ways = (uint32_t)(CCSIDR_WAYS(ccsidr));

   do {

    SCB->DCISW = (((sets << SCB_DCISW_SET_Pos) & SCB_DCISW_SET_Msk) |

           ((ways << SCB_DCISW_WAY_Pos) & SCB_DCISW_WAY_Msk) );

    #if defined ( __CC_ARM )

     __schedule_barrier();

    #endif

   } while (ways--);

I'm aware that I could use fixed buffers allocated within TCMRAM to avoid the caching problems, but I would like to use buffers on the stack for the SPI and I don't see how I can make my stack buffers appear in TCMRAM. Can someone please help me?

[Tesla DeLorean · Accepted solution]

SCB_InvalidateDCache() has a lot of nasty side-effects, use the ByAddr variant and limit it's scope to the buffer in question. It works on 32-byte aligned addresses.

[David Littell]

Make sure your receive buffer is aligned and sized to a 32-byte boundary and call SCB_InvalidateDCache_by_Addr() before accessing the receive buffer after the DMA completes.

[arnold_w]

Thanks, it works great with SCB_InvalidateDCache_by_Addr().

Just as a side note, strangely I had to call SCB_CleanDCache (line 2 below) before my call to HAL_SPI_Receive_DMA. If I remove line 2, then line 32 would be called forever as if someone had put an invisible while(TRUE)-loop around line 32!?!?! By the way, the code (you may have to click "Show More" to see it) is for the built-in STM bootloader SPI protocol.

static void SPI_BL_Rx(uint8_t *pRxData32bitAligned, uint16_t Size) {
    SCB_CleanDCache();
    setNssLow();
    SPI_DMA_InProgress = TRUE;
    HAL_SPI_Receive_DMA(hspi, pRxData32bitAligned, Size);
    while (SPI_DMA_InProgress) {}
    setNssHigh();
    SCB_InvalidateDCache_by_Addr((uint32_t*)pRxData32bitAligned, Size);
}
 
Bool_t BL_GetIDCmd(uint16_t* idCmd) {
    uint8_t cmdSeq[3];
    uint32_t rxNumberOfBytes32bitAligned = 0;
    uint8_t* rxNumberOfBytes = (uint8_t*)&rxNumberOfBytes32bitAligned;
    uint8_t ***** = 0x00;
    uint32_t rxBuffer32bitAligned = 0;
    uint8_t* rxBuffer = (uint8_t*)&rxBuffer32bitAligned;
 
    /* Send start of frame (0x5A) + Get ID command frame (0x02 0xFD) */
    cmdSeq[0] = BL_SPI_SOF;
    cmdSeq[1] = GET_ID_COMMAND;
    cmdSeq[2] = GET_ID_COMMAND ^ 0xFF;
    SPI_BL_Tx(cmdSeq, 0x3);
 
    /* Wait for ACK or NACK frame */
    if (waitForAck(DEFAULT_WAIT_FOR_ACK_TIMEOUT_MS) == FALSE) {
        return FALSE;
    }
 
    /* Receive data frame */
    SPI_BL_Tx(&*****, 0x1); /*!< Send ***** */
    SPI_BL_Rx(rxNumberOfBytes, 0x1);
    if (*rxNumberOfBytes > 2) {
        return FALSE;
    }
    SPI_BL_Rx(rxBuffer,(uint16_t) *rxNumberOfBytes + 0x1);
 
    /* received Byte #0 is ID MSB and received Byte #1 is ID LSB */
    *idCmd = (((uint16_t) rxBuffer[0]) << 8) | (rxBuffer[1]);
 
    /* Wait for ACK or NACK frame */
    return waitForAck(DEFAULT_WAIT_FOR_ACK_TIMEOUT_MS);
}

[Amel NASRI]

If this may bring some help: https://community.st.com/s/article/FAQ-DMA-is-not-working-on-STM32H7-devices. It concerns STM32H7, but should be valid for STM32F7 as well.