Can portEND_SWITCHING_ISR() cause problems ..

vespaman2 wrote on Monday, August 20, 2012:

Can portEND_SWITCHING_ISR() cause problems if not called with the result from e.g. xQueueSendToBackFromISR()?

I have a preemptive coldfire target where i have issues with getting caught in vListInsert() sometimes.  (a few times a week). Stacks are ok, and all looks to be fine. I end up in vListInsert() when a particular task is checking its message queue.

I have pinpointed that it has something to do with a particular interrupt posting to this message queue.

One thing that I have found, is that the interrupt, sometimes does not take care of the return code from xQueueSendToBackFromISR(), hence no task switch.

I’m going to fix my code, but I wonder if this can cause this kind of error? Gut feeling says no…

davedoors wrote on Monday, August 20, 2012:

Does the ColdFire port implement interrupt nesting? I think it does, and if so, are you sure the interrupt posting to the queue is running with a priority lower than or equal to whatever configMAX_SYSCALL_INTERRUPT_PRIORITY is set to?

vespaman2 wrote on Monday, August 20, 2012:

Yes, I have actually set configMAX_SYSCALL_INTERRUPT_PRIORITY to 7, the highest (non maskable) interrupt on the Coldfire. All other interrupts are lower. I have made this table (it is the UART interrupts that can cause this to happen):

/*---------------------------2012-06-27 12:21----------------------------
 * Interrupt levels.
 * Level    Priority    Function
 * -----------------------------
 *   1         0        Kernel context switch
 *   1         1        Kernel PIT
 *   2         1        DMA0 UART
 *   2         2        DMA1 UART
 *   2         5        CAN Error
 *   2         6        CAN Bus off
 *   2         7        FEC Babbling rx
 *   3         0        FEC Babbling tx
 *   3         1        FEC Bus error
 *   3         2        FEC Heart beat error
 *   3         3        FEC Late Collision
 *   3         4        FEC Collision retry limit
 *   3         5        FEC FIFO underrun
 *   3         6        FEC Rx buffer
 *   3         7        FEC Rx Frame
 *   4         0        CAN buffer 0
 *   4         1        CAN buffer 1
 *   4         2        CAN buffer 2
 *   4         3        CAN buffer 3
 *   4         4        CAN buffer 4
 *   4         5        CAN buffer 5
 *   4         6        CAN buffer 6
 *   4         7        CAN buffer 7
 *   5         0        CAN buffer 8
 *   5         1        CAN buffer 9
 *   5         2        CAN buffer 10
 *   5         3        CAN buffer 11
 *   5         4        CAN buffer 12
 *   5         5        CAN buffer 13
 *   5         6        CAN buffer 14
 *   5         7        CAN buffer 15
 *   6         1        RTC
 *   6         6        UART0
 *   6         7        UART1
 *   7                  Kernel syscall level 

vespaman2 wrote on Wednesday, August 22, 2012:

OK, so now  I had it happen again. I knew it was too good to be true…
This time it was not on my development system, so no new leads to what might have cause this. What makes me confused is that the interrupt isn’t very complicated, so I guess I’m overlooking something else.

vespaman2 wrote on Thursday, August 23, 2012:

Caught it in the debugger this morning.

vListInsert() at /home/micke/svn/trunk/controller/FreeRTOS_701/Source/list.c:161 0x7a0	
vTaskPlaceOnEventList() at /home/micke/svn/trunk/controller/FreeRTOS_701/Source/tasks.c:1,756 0x154e	
xQueueGenericReceive() at /home/micke/svn/trunk/controller/FreeRTOS_701/Source/queue.c:974 0x1dec	
os_msg_receive() at /home/micke/svn/trunk/controller/src/os_functions.c:407 0x3764	
dsu_thread_main() at /home/micke/svn/trunk/controller/src/control/door_supervisor_task.c:1,709 0xd7e4	

Basically what happens is that when arriving to vListInser(), there’s one item in the list, and both pxNext and pxPrevious point back to itself, so there’s no way of advancements.

pxIterator	volatile xListItem *	0x2000720b	
	xItemValue	portTickType	3	
	pxNext	volatile struct xLIST_ITEM *	0x2000720b	
	pxPrevious	volatile struct xLIST_ITEM *	0x2000720b	
	pvOwner	void *	0x200071f3	
	pvContainer	void *	0x0	
xValueOfInsertion	portTickType	3	

What is actually the correct pointer values when queue contains one item?

The queue is xTasksWaitingToReceive queue.
pvOwner is actually pointing to the same thread that is now trying to insert itself into the queue again, which I guess is compeletely wrong.  Is this the reason for ending up in this endless loop, or am I misunderstanding this queue?

senetti wrote on Thursday, September 20, 2012:

Hi vespaman2,
I´m having the same issue using an LPC1768, have you find a way to solve this issue?
I´m using a binary semaphore  to synchronize the  reception of CAN messages and the function that parse those messages.
One of the test i did was not to use any more the semaphore and use a simple flag to allow the task to parse or not the messages, in this test i did not have the issue any more, but the problem is that the parsing task runs always and is not good for my application.
Please let me know if find the problem, i will continue doing tests.

vespaman2 wrote on Thursday, September 20, 2012:

My issue turned out to be not using binary semaphore, but you might want to have a look into the follow up thread