IPC/MessageQ problem with ARM & DSP

YES! YES! YES! Found the problem!

It is in MessageQ_get() function (Linux impl.): https://git.ti.com/ipc/ipcdev/blobs/master/linux/src/api/MessageQ.c

There is missing a break statement after a message is found. If all cores have sent a message when get is called, it returns only the last one.

            for (rprocId = 0; rprocId < MultiProc_getNumProcessors();
                 rprocId++) {
                if (rprocId == MultiProc_self()) {
                    continue;
                }
                if (FD_ISSET(obj->fd[rprocId], &rfds)) {
                    /* Our transport's fd was signalled: Get the message: */
                    tmpStatus = transportGet(obj->fd[rprocId], msg);
                    if (tmpStatus < 0) {
                        printf ("MessageQ_get: tranposrtshm_get failed.");
                        status = MessageQ_E_FAIL;
                    }
                }
            }

Does someone know what is the formal way to get the IPC development team aware of this bug & fix?

Vesa,

Excellent work finding that problem, it does indeed look like the cause of your issue.

Vesa Peltonen said:

There is missing a break statement after a message is found. If all cores have sent a message when get is called, it returns only the last one.

We will fix this as you said, shown here (new code in red):

 for (rprocId = 0; rprocId < MultiProc_getNumProcessors(); rprocId++) { if (rprocId == MultiProc_self()) { continue; } if (FD_ISSET(obj->fd[rprocId], &rfds)) { /* Our transport's fd was signalled: Get the message: */ tmpStatus = transportGet(obj->fd[rprocId], msg); if (tmpStatus < 0) { printf ("MessageQ_get: tranposrtshm_get failed."); status = MessageQ_E_FAIL;                     }
                    break;

 } }

Vesa Peltonen said:

Does someone know what is the formal way to get the IPC development team aware of this bug & fix?

This forum is the correct way to alert the IPC development team.  You could also report it to the 'ipc-developers' list to which you are subscribed, but the forum is preferred as it benefits the community in a wider fashion.

Thankyou very much for taking the time to diagnose the bug.

Regards,

- Rob

Vesa Peltonen said:

- when I call  MessageQ_count(g_ddrMsgHandle); it returns -1, which indicates of some error (on DSP side, the count returns zero when there are no messages on the Queue).

FYI, the MessageQ_count() API is disabled on Linux, for the moment.  It simply returns -1 regardless of the number of messages on the queue.

Also, the fix we'll be doing is slightly different than simply putting in that single "break;" statement (which I tried to show in my previous post but the formatting got all messed up).  In order to not favor the lowest numbered remote processor, the "next" call to MessageQ_get() should start looking at the next processor cyclically in the ordinal list (i.e., if it found a message on procId 2, it should start the next search at procId 3, with index wraparound until it gets back to the "last" procId for which it found a message).

Regards,

- Rob