Potential DMA IRQ problem.

[Anubirux]

I've run into something disturbing in DMA driver. STM32F1 XL and HD devices have shared DMA2 4th and 5th channel. At dmaStreamRelease there is no check whether when disabling 4th channel, 5th is allocated, or opposite.

Here you have fix proposal (2.6.x):

Code: Select all

void dmaStreamRelease(const stm32_dma_stream_t *dmastp) {

  chDbgCheck(dmastp != NULL, "dmaStreamRelease");

  /* Check if the streams is not taken.*/
  chDbgAssert((dma_streams_mask & (1 << dmastp->selfindex)) != 0,
              "dmaStreamRelease(), #1", "not allocated");


  /* Marks the stream as not allocated.*/
  dma_streams_mask &= ~(1 << dmastp->selfindex);

  /* Disables the associated IRQ vector.*/
#if !(STM32_HAS_DMA2 && !defined(STM32F10X_CL)) || defined(__DOXYGEN__)
  nvicDisableVector(dmastp->vector);
#else
  /* Check unless it is 10 or 11 stream. If yes, make additional check before
     disabling IRQ.*/
  if (dmastp->selfindex < 10) {
    nvicDisableVector(dmastp->vector);
  }
  else {
    if (dma_streams_mask & (3 << 10) == 0)
      nvicDisableVector(dmastp->vector);
  }
#endif/* STM32_HAS_DMA2 && !STM32F10X_CL */

  /* Shutting down clocks that are no more required, if any.*/
  if ((dma_streams_mask & STM32_DMA1_STREAMS_MASK) == 0)
    rccDisableDMA1(FALSE);
#if STM32_HAS_DMA2
  if ((dma_streams_mask & STM32_DMA2_STREAMS_MASK) == 0)
    rccDisableDMA2(FALSE);
#endif
}

[Giovanni]

Hi,

Thanks, fixed as bug #510.

Giovanni

[Anubirux]

Hello,

small misunderstanding

This line:

Code: Select all

 /* Marks the stream as not allocated.*/
dma_streams_mask &= ~(1 << dmastp->selfindex); 

has to be before disabling IRQ. Its because check whether 4th and 5th channel is free will always return false:

Code: Select all

    if (dma_streams_mask & (3 << 10) == 0)
      nvicDisableVector(dmastp->vector);

[Giovanni]

Oops, missed that, fixed now.

Giovanni