GPDMA Linked List Circular Buffering: Correct Transfer Event Mode?

[cubett]

I want to capture images larger than 0xFFFF via DCMI using double buffering with two GPDMA1 linked list nodes. As guidance, I'm following the DCMI_ContinuousCap_EmbeddedSynchMode example. However, it seems that the DCMI_DMAXferCplt function is written as if it should be called after every node completes, but the example configures:

  handle_GPDMA1_Channel12.Instance = GPDMA1_Channel12;
  handle_GPDMA1_Channel12.InitLinkedList.Priority = DMA_LOW_PRIORITY_LOW_WEIGHT;
  handle_GPDMA1_Channel12.InitLinkedList.LinkStepMode = DMA_LSM_FULL_EXECUTION;
  handle_GPDMA1_Channel12.InitLinkedList.LinkAllocatedPort = DMA_LINK_ALLOCATED_PORT1;
  handle_GPDMA1_Channel12.InitLinkedList.TransferEventMode = DMA_TCEM_LAST_LL_ITEM_TRANSFER;
  handle_GPDMA1_Channel12.InitLinkedList.LinkedListMode = DMA_LINKEDLIST_CIRCULAR;

Am I misunderstanding TransferEventMode, or why is it not set to DMA_TCEM_EACH_LL_ITEM_TRANSFER?

[Sarra.S]

Hello @cubett, 

In the example, the event is triggered only after the last linked list item transfer is completed, so the completion of the entire linked list transfer is considered as a single event

If TransferEventMode is set to DMA_TCEM_EACH_LL_ITEM_TRANSFER, this will trigger the transfer complete event after each linked list item transfer, so you handle each node's completion individually

[cubett]

That matches my understanding. But in that case, the example DCMI_ContinuousCap_EmbeddedSynchMode must be wrong? I can't see how the DCMI_DOUBLE_BUFFER Mode (data size > 0xFFFF) with two linked list nodes could work if DCMI_DMAXferCplt is only triggered after the last node (second node).

static void DCMI_DMAXferCplt(DMA_HandleTypeDef *hdma)
{

  DCMI_HandleTypeDef *hdcmi = (DCMI_HandleTypeDef *)((DMA_HandleTypeDef *)hdma)->Parent;
  uint32_t tmp1;
  uint32_t tmp2;
  DMA_NodeTypeDef *pnode;
  uint32_t pbuff;
  uint32_t transfernumber;
  uint32_t transfercount;
  uint32_t transfersize ;

  /* Update Nodes destinations */
  if (hdcmi->XferSize != 0U)
  {
    pbuff          = hdcmi->pBuffPtr;
    transfernumber = hdcmi->XferTransferNumber;
    transfercount  = hdcmi->XferCount;
    transfersize   = hdcmi->XferSize;

    tmp1 = hdcmi->DMA_Handle->Instance->CLLR & DMA_CLLR_LA;
    tmp2 = hdcmi->DMA_Handle->Instance->CLBAR & DMA_CLBAR_LBA;
    pnode = (DMA_NodeTypeDef *)(uint32_t)(tmp1 | tmp2);

    if (hdcmi->XferCount > 1U)
    {
      pnode->LinkRegisters[NODE_CDAR_DEFAULT_OFFSET] = pbuff + ((transfernumber - transfercount + 2U) * transfersize);
      hdcmi->XferCount--;
    }

    else if (hdcmi->XferCount == 1U)
    {
      pnode->LinkRegisters[NODE_CDAR_DEFAULT_OFFSET] = hdcmi->pBuffPtr;
      hdcmi->XferCount--;
    }
    else
    {
      pnode->LinkRegisters[NODE_CDAR_DEFAULT_OFFSET] = hdcmi->pBuffPtr + hdcmi->XferSize;

      /* When Continuous mode, re-set dcmi XferCount */
      if ((hdcmi->Instance->CR & DCMI_CR_CM) == DCMI_MODE_CONTINUOUS)
      {
        hdcmi->XferCount = hdcmi->XferTransferNumber ;
      }
      /* When snapshot mode, set dcmi state to ready */
      else
      {
        hdcmi->State = HAL_DCMI_STATE_READY;
      }

      __HAL_DCMI_ENABLE_IT(hdcmi, DCMI_IT_FRAME);
    }
  }
  else  /* Snapshot Mode */
  {
    /* Enable the Frame interrupt */
    __HAL_DCMI_ENABLE_IT(hdcmi, DCMI_IT_FRAME);

    /* When snapshot mode, set dcmi state to ready */
    if ((hdcmi->Instance->CR & DCMI_CR_CM) == DCMI_MODE_SNAPSHOT)
    {
      hdcmi->State = HAL_DCMI_STATE_READY;
    }
  }
}