OMAP3530 Dsplink based application MMU Error

There are some articles about the OMAP3 DSP-side MMU configuration here:

   * http://tiexpressdsp.com/index.php/DSP_MMU_Faults

   * http://tiexpressdsp.com/index.php/OMAP3_DSP_MMU_Configuration

Chris

Hi,

It is not clear from the code you have pasted, how the interaction with DSPLink changes when you add the convolution code. For example, it's not clear whether you are copying the contents of the received buffer from the GPP into your local buffer before performing the convolution operation on it. Also whether you are copying the results of your convulation buffer into the DSPLink CHNL buffer allocated from the POOL. Can you show the exact changes you made to the application to send the updated buffer back to the GPP?

The MMU fault indicates that you're trying to access a memory address that is not mapped to the DSP. This can happen if you have some incorrect memory access on the DSP, or if the DSP is going into the weeds.

Regards,
Mugdha

Here's the code from GPP side

// populate outgoing buffer with data

    if (DSP_SUCCEEDED (status)) {
        temp = LOOP_Buffers [0] ;
    for (i = 0 ; i < ARRAY_SIZE ; i++) {
        fillVar=gr_inst.rfDataInt[i];
         memcpy(temp, &fillVar, sizeof(fillVar));
        temp = temp + stepSize;
        }

}

convolution(gr_inst.rfDataInt, ARRAY_SIZE, gr_inst.rfDataInt, ARRAY_SIZE, gr_inst.rfDataOut);

// the rest of the GPP code is the same for the LOOP example example the verify function checks the solution against the GPP convolution result

DSP Side

all the modifications I did was in the loop execute function, so I copied and pasted the whole function here

Int TSKLOOP_execute(TSKLOOP_TransferInfo * info)
{
    Int         status  = SYS_OK ;
    Char *      buffer  = info->buffers [0] ;
    Arg         arg     = 0 ;
    Uint32      i=0,k=0 ;
    Int         nmadus ;
    bufferType buffLocal[MAX_SIZE];
    bufferType buffLocal2[MAX_SIZE];
    /* Execute the loop for configured number of transfers
     * A value of 0 in numTransfers implies infinite iterations
     */

    //BCACHE_setMode(BCACHE_L1D, BCACHE_FREEZE);


    for (i = 0 ;
         (   ((info->numTransfers == 0) || (i < info->numTransfers))
          && (status == SYS_OK)) ;
         i++) {
        /* Receive a data buffer from GPP */
        status = SIO_issue(info->inputStream,
                           buffer,
                           info->bufferSize,
                           arg) ;
        if (status == SYS_OK) {
            nmadus = SIO_reclaim (info->inputStream,
                                  (Ptr *) &buffer,
                                  &arg) ;
            if (nmadus < 0) {
                status = -nmadus ;
                SET_FAILURE_REASON (status) ;
            }
            else {
                info->receivedSize = nmadus ;
            }
        }
        else {
            SET_FAILURE_REASON(status);
        }
        /* Do processing on this buffer */
    
    if (status == SYS_OK) {


         memset(buffLocal, 0, sizeof(buffLocal));
         memcpy(buffLocal2, buffer, sizeof(bufferType)*ARRAY_SIZE);

        // to make sure that the convolution result is written back to external memory and is not in cache
         HAL_CACHE_WBALL;
         //HAL_CACHE_WB(buffLocal, sizeof(bufferType)*ARRAY_SIZE);
         memcpy(buffer, buffLocal, sizeof(bufferType)*ARRAY_SIZE);

            /* Add code to process the buffer here*/
        }
    else {
            SET_FAILURE_REASON(status);
        }
        /* Send the processed buffer back to GPP */
        if (status == SYS_OK) {
            status = SIO_issue(info->outputStream,
                               buffer,
                               info->receivedSize,
                               arg);

            if (status == SYS_OK) {
                nmadus = SIO_reclaim (info->outputStream,
                                      (Ptr *) &(buffer),
                                      &arg) ;
                if (nmadus < 0) {
                    status = -nmadus ;
                    SET_FAILURE_REASON (status) ;
                }
            }
            else {
                SET_FAILURE_REASON (status) ;
            }
        }
    }
    return status ;
}

Thanks for your reply, in my case would I be looking at expanding the "DSPLINK (MEM)" segment out of curiosity?  and do I need to worry about the data and program segments seperately? since I'm assuming data and instructions are in two seperate memory banks

One more things I forgot to mention in my code ... sorry for the multiple posts :(

typedef short bufferType

I'm passing the channel size as 2048 in the same manner as the TI LOOP app

Hi,

I suspect stack overrun. Can you try making the local buffer arrays global instead of on the stack, or increase the task stack size when you are creating it?

The buffer sizes you are using are small enough to not need increasing the DSPLINKMEM size. If you had to increase it, you would have got failures from either POOL_open or PROC_attach:

http://wiki.davincidsp.com/index.php/DSPLink_POOL_FAQs

http://wiki.davincidsp.com/index.php/Changing_DSPLink_Memory_Map

Regards,
Mugdha

Mugdhak, I finally got around testing your suggestion of making the buffers global and that fixed it ... you're a genius thanks !!! out of curiosity how can I increase the task Stack size?  I'm assuming it's through modifying the pool_alloc call?

POOL_alloc( SAMPLE_POOL_ID, (Ptr*)transInfo->buffer[i], transInfo->bufferSize );

would that be by changing "transInfo->bufferSize" into "transInfo->bufferSize+BufferSize1+BufferSize2"  thanks again you're a life saver :)

Hi,

No, this is the DSP/BIOS TSK I'm talking about. That's created (in the loop sample) in file main.c, main ().

    tskLoopTask = TSK_create(tskLoop, NULL, 0);

If you want to increase the stack size, you should use non-NULL params and set the stackSize in it. You can check out DSP/BIOS documentation for TSK module for the correct parameters to be used.

Regards,
Mugdha