qmss termination/reinitialization

Hi Alex,
What is the DSP part number used? I did not understand your query, please elaborate in detail. What you mean by "same code"?

After disabling and closing all opened channels and Cppi_close without errors, Cppi_exit return CPPI_CPDMA_NOT_CLOSED.
How can I fix it?

Hi,

Have you ensured that the software recycles host packet/buffer descriptors recycled appropriately and frees the payload?

How did you stop the CPPI subsystem and is your CPPI transmit or receive channel freed or closed appropriately? Could you please ensure this for proper CPPI termination which is more significant?

Thanks & regards,

Sivaraj K

-------------------------------------------------------------------------------------------------------

Please click the Verify Answer button on this post if it answers your question.

-------------------------------------------------------------------------------------------------------

Sivaraj Kuppuraj said:

Also, there is no automatic deletion of payload buffers by a pktDMA, because there's no way for the hardware to know how the buffer was obtained.  And to my knowledge, the CPPI LLD does not do this either.  For Rx descriptor handling, it is generally up to the host software to recycle the descriptor to the appropriate queue and to free the payload buffer if necessary.

Sivaraj Kuppuraj said:

Have you ensured that the software recycles host packet/buffer descriptors recycled appropriately and frees the payload?

Sivaraj Kuppuraj said:

How did you stop the CPPI subsystem and is your CPPI transmit or receive channel freed or closed appropriately? Could you please ensure this for proper CPPI termination which is more significant?

Int32 DeInit_Cppi (Void)
{
    if (Cppi_closeRxFlow(gRxFlowHnd) < 0)
        {
            System_printf ("Error Cppi_closeRxFlow \n");
            return -1;
        }

    int i, res;
    for (i = 0; i < NUM_PA_TX_QUEUES; i ++)
    {
        res = Cppi_channelDisable (gCpdmaTxChanHnd[i]);
        if (res<0)
        {
            printf ("Error Cppi_channelDisable: %d \n", res);
            return -1;
        }
        res = Cppi_channelClose   (gCpdmaTxChanHnd[i]);
        if (res<0)
        {
            printf ("Error Cppi_channelClose: %d \n", res);
            return -1;
        }
    }

    for (i = 0; i < NUM_PA_RX_CHANNELS; i++)
    {
        res = Cppi_channelDisable (gCpdmaRxChanHnd[i]);
        if (res<0)
        {
            printf ("Error Cppi_channelDisable: %d \n", res);
            return -1;
        }
        res = Cppi_channelClose   (gCpdmaRxChanHnd[i]);
        if (res<0)
        {
            printf ("Error Cppi_channelDisable: %d \n", res);
            return -1;
        }
    }

    if ((res = Cppi_close (gCpdmaHnd)) < 0)
    {
        printf ("Error Cppi_close %d \n", res);
        return -1;
    }

    if ((res = Cppi_exit ()) < 0) // return CPPI_CPDMA_NOT_CLOSED
    {
        printf ("Error Cppi_exit %d \n", res);
        return -1;
    }

    return 0;
}

Hi,

To my opinion, Cppi_close() and Cppi_exit () are mapped to unique functionalities which does the desired task and either of the routines doesn't relate to each other.  By this, we could say there is no relevancy between Cppi_close without errors and Cppi_exit return CPPI_CPDMA_NOT_CLOSED

Also, if Qmss_queueClose function doesn't return any error which infers, it closes & destroys the queue successfully and thereby, it doesn't return any error. This is what it does and it mean to be.

Thanks & regards,

Sivaraj K

-------------------------------------------------------------------------------------------------------

Please click the Verify Answer button on this post if it answers your question.

-------------------------------------------------------------------------------------------------------

The first problem in your crappy code:
srio_drv doesn't contain pair function for Srio_init. Internals unaccessible. Without package code modification user can't deinit srio. And is not obvious that deinitialization are necessary.

The second in your crappy code too:
srio_drv.c:1180

/* Enable all the SRIO Receive Channels */    
{
    uint16_t            index;
    Cppi_RxChInitCfg    rxCfg;
    Cppi_ChHnd          chHnd;
    uint8_t             isAllocated;

    for (index = 0; index < 16; index++)
    {
        /* Open the SRIO Receive Channel */
        rxCfg.channelNum = CPPI_PARAM_NOT_SPECIFIED;
        rxCfg.rxEnable   = Cppi_ChState_CHANNEL_DISABLE;
        chHnd = Cppi_rxChannelOpen (gSRIODriverMCB.cppiHnd, &rxCfg, &isAllocated);
        if (chHnd == NULL)
        {
            Srio_osalLog ("Error: Opening SRIO Rx channel %d failed\n", rxCfg.channelNum);
            return -1;
        }

        /* Enable the channel */
        if (Cppi_channelEnable (chHnd) < 0)
        {
            Srio_osalLog ("Error: Enabling SRIO Rx Channel %d failed\n", index);
            return -1;
        }
    }
}

Index is not used. All channels pointers get lost.
I fixed CPPI_CPDMA_NOT_CLOSED when fixed that.

The third is question:
Thought I have closed all qmss queues, PA, cppi flows and cppi itself,
"Qmss_queuePop () function in second program returns Cppi_HostDesc->buffLen=0" still yet.

Sivaraj Kuppuraj said:

Have you ensured that the software recycles host packet/buffer descriptors recycled appropriately and frees the payload?

In both programs I have
pCppiDesc = Qmss_queuePop (gGlobalFreeQHnd)
and then, after descriptors initialization,
they have addresses like
 c1c4700
 c1c4750
 c1c47a0
 c1c47f0
...
end push via Qmss_queuePushDescSize (gTxFreeQHnd, pCppiDesc, SIZE_HOST_DESC);

First call pHostDesc = Qmss_queuePop (gTxFreeQHnd)
in the first program return c1c4700 descriptor pointer.
After all I close all qmss queues, cppi flows and cppi itself.
But in the second program first call pHostDesc = Qmss_queuePop (gTxFreeQHnd) return c1c4750 or even greater address (c1c47f0 e.g). So it isn't first descriptor address c1c4700.

How could this be if I've closed all before new initialization?

Bump!
Is support going to support?!