AnsweredAssumed Answered

K60 DMA ping-pong issue

Question asked by segfault on Oct 30, 2014
Latest reply on Nov 18, 2014 by segfault

Hello everybody!


I'm working on a project for K60 MCU and encountered a weird issue with the DMA controller.


I need to read some data from the ADC. The interface is similar to SAI/I2S, however this ADC is using eight data lines instead of one. That is why I've configured the I2S0 interface to provide SCLK and FSYNC signals and wanted to use DMA for data acquisition. The ADC data outputs are connected to PORT_E[0:7] lines. Whenever we raise FSYNC line the ADC provides us 24 bytes on every raising edge of the SCLK signal (I mean, the time we should read them). Below is an example of saturated output (For clearer visibility FSYNC is raised for 24 SCLK periods).



blue - SCLK, violet - FSYNC (up for 24 periods), green - saturated data (0x7FFFFF), yellow - doesn't matter.


I used SCLK rising edge to trigger the DMA (one byte minor loop, 24 bytes major loop) and FSYNC signal was meant to enable the DMA (which would get disabled after every major loop). Another thing is that I need to do the data postprocessing immediately after I gather the samples and it takes a little bit longer than the time to next FSYNC rise (next data packet). If I used one DMA it would start overwriting the data being currently processed and that's bad. That's why I wanted to use two DMA channels with exactly the same configuration and only one difference: the destination buffer address. They were supposed to run one after another in cycles so I have time to proccess the data.


I tried three different methods to enable the channels on rising FSYNC edge:

  1. I connected IRQ handling routine to FSYNC (way too slow, especially during read/write operations to external SRAM).
  2. I used third DMA that copies one predefined byte to DMA_SERQ register. The byte is being changed by both ISR channels (since I was not able to use PORT_A for the third DMA I had to loopback the FSYNC to PORT_D.15)
  3. It was a variation of the second idea. Third DMA had two bytes in major loop and the source address was a 2-byte array containing DMA_1 and DMA_2 channel numbers. This way I did not have to change anything in the ISR anymore. Below code snippets belong to this version.


Some defines first:

#define I2S0_DMA_SAMPLE_COUNT ( 24 )
#define DMA_ENABLER_CHANNEL   ( 13 )
#define I2S0_DMA_1_CHANNEL    ( 15 )
#define I2S0_DMA_2_CHANNEL    ( 14 )
#define BUF_SIZE              ( 64 )

#define I2S0_DMA_1_VECT_NUM   DMA15_DMA31_IRQn
#define I2S0_DMA_2_VECT_NUM   DMA14_DMA30_IRQn


Data structures:

static volatile uint8_t AdcDataBuffer[2][BUF_SIZE] __attribute__((aligned (BUF_SIZE)));
DMA_Type *DmaPtr = DMA0;

uint16_t DmaI2sChan_1   = I2S0_DMA_1_CHANNEL;
uint16_t DmaI2SChan_2   = I2S0_DMA_2_CHANNEL;
uint16_t DmaEnablerChan = DMA_ENABLER_CHANNEL;

static volatile uint8_t DmaEnableValues[2] = {


First DMA channel configuration (channel 15):

  DMAMUX0->CHCFG[DmaI2sChan_1] = 0x0;

  /* source address is Port E PDIR register */
  DmaPtr->TCD[DmaI2sChan_1].SADDR     = (uint32_t) &PTE->PDIR;
  /* source does not increment */
  DmaPtr->TCD[DmaI2sChan_1].SOFF      = 0;
  /* source does not increment at the end of the major loop */
  DmaPtr->TCD[DmaI2sChan_1].SLAST     = 0;

  DmaPtr->TCD[DmaI2sChan_1].DADDR     = (uint32_t) &AdcDataBuffer [0];
  DmaPtr->TCD[DmaI2sChan_1].DOFF      = 1;
  DmaPtr->TCD[DmaI2sChan_1].DLAST_SGA = (uint32_t) -I2S0_DMA_SAMPLE_COUNT;

  DmaPtr->TCD[DmaI2sChan_1].NBYTES_MLNO   =   1;
  DmaPtr->TCD[DmaI2sChan_1].ATTR          =   DMA_ATTR_SSIZE (0)
                                            | DMA_ATTR_SMOD  (0)
                                            | DMA_ATTR_DSIZE (0)
                                            | DMA_ATTR_DMOD  (0);

                                  | DMAMUX_CHCFG_ENBL_MASK;

                                  | DMA_CSR_DREQ_MASK;
  DmaPtr->CERQ = DmaI2sChan_1;


The only difference for second DMA channel (no. 14) configuration:

  // ...
  DmaPtr->TCD[DmaI2sChan_2].DADDR = (uint32_t) &AdcDataBuffer [1];
  // ...


Here's the enabler DMA configuration:

  DMAMUX0->CHCFG[DmaEnablerChan]            = 0x0;
  DmaPtr->TCD[DmaEnablerChan].SADDR         = (uint32_t) &DmaEnableValues[0];
  DmaPtr->TCD[DmaEnablerChan].SOFF          = 1;
  DmaPtr->TCD[DmaEnablerChan].SLAST         = (uint32_t) -2;
  DmaPtr->TCD[DmaEnablerChan].DADDR         = (uint32_t) 0x4000801b; /* SERQ address */
  DmaPtr->TCD[DmaEnablerChan].DOFF          = 0;
  DmaPtr->TCD[DmaEnablerChan].DLAST_SGA     = 0;
  DmaPtr->TCD[DmaEnablerChan].CITER_ELINKNO = 2;
  DmaPtr->TCD[DmaEnablerChan].BITER_ELINKNO = 2;
  DmaPtr->TCD[DmaEnablerChan].ATTR =   DMA_ATTR_SSIZE(0)
                                     | DMA_ATTR_SMOD (0)
                                     | DMA_ATTR_DSIZE(0)
                                     | DMA_ATTR_DMOD (0);
                                     | DMAMUX_CHCFG_ENBL_MASK;
  DmaPtr->TCD[DmaEnablerChan].CSR  = 0x0;

  DmaPtr->SERQ = DmaEnablerChan;


Port settings for DMA:

  /* SCLK is internally generated on PTA.14 (DMA_1 and DMA_2 on rising edge) */

  /* FSYNC is loopbacked to PTD.15 (DMA enabler on rising edge) */


IRQ priorities:

  /* lower all priorities */
  for (i = -14; i < 105; i++)
    NVIC_SetPriority (i, 0x7);

  /* raise data processing DMA priorities */
    NVIC_SetPriority (I2S0_DMA_1_VECT_NUM, 0x0);
    NVIC_SetPriority (I2S0_DMA_2_VECT_NUM, 0x0);


DMA priorities:

/* channels 0 to 12 have values 0x80 to 0x8c respectively. */
DMA_DCHPRI0  = 0x80; /* can be preempted, can't preempt */
// ...
DMA_DCHPRI12 = 0x8c;

DMA_DCHPRI13 = 0x4f; /* can preempt, cannot be preempted, highest priority */
DMA_DCHPRI14 = 0x8d;
DMA_DCHPRI15 = 0x8e; /* can be preempted, second to highest priority */


(I think I've tried all possible configurations here without luck).


Below are the DMA finished IRQ handling routines:

void DMA_1_ISR (void)
  const uint_16 DmaChannel = I2S0_DMA_1_CHANNEL;

  DmaPtr->CINT = DMA_CINT_CINT (DmaChannel);

  /* data processing */
  process (AdcDataBuffer[0]);

void DMA_2_ISR (void)
  const uint16_t DmaChannel = I2S0_DMA_2_CHANNEL;

  DmaPtr->CINT = DMA_CINT_CINT (DmaChannel);

  /* data processing */
  process (AdcDataBuffer[1]);


Now here's the issue: Always the channel with higher order number reads corrupted results. If I used (as in example above) channels 14 and 15, data from channel 15 was wrong. If I used 12 and 14, the 14 became corrupted. Nothing changed as I modified the priorities and preemption settings so I assume it has something to do with the DMA channel order.


Here's an example image (every second sample is incorrect but in some regular way):





From the image below it looks like the second DMA always kicks off too early. The weird thing is that the data (inspected in the debugger) do look OK actually. I enabled DMA13 HALF and MAJOR interrupts to check if maybe the third DMA was causing the issue but they both were in correct places (right after the FSYNC rising edge).



blue - SCLK, violet - FSYNC (up for 24 periods), yellow - data postprocessing in DMA ISR.


I've already spent too much time on this issue but maybe I'm missing something obvious here? I've already tried many different settings and combinations before giving up this implementation.


For those who are interested in what is my current solution...

I'm using one DMA channel. After it enters the ISR it immediately changes DMA_TCD_DADDR register value to next buffer address and then begins data processing.


The main modifiaction is in the DMA_1 ISR:

static volatile uint_8 AcquisitionBuffer = I2S0_DMA_1_DATA_BUF;

// ...

void DMA_1_ISR (void)
  DMA_Type * const DmaPtr       = DMA0;
  const uint_16    DmaChannel   = I2S0_DMA_1_CHANNEL;
  const uint_8     BufToProcess = AcquisitionBuffer;

  AcquisitionBuffer = (AcquisitionBuffer + 1) & 0x1;
  DmaPtr->TCD[DmaChannel].DADDR = (uint32_t) &AdcDataBuffer [AcquisitionBuffer];

  DmaPtr->CINT = DMA_CINT_CINT (DmaChannel);

  /* data processing */
  process (AdcDataBuffer[BufToProcess]);



If somebody has an idea what could be incorrect in my setup, please let me know. I'd like to make sure that it's still a software issue and not hardware related.


Thank you,