xTaskIncrementTick() could be called when pxCurrentTCB == NULL

dddddppppp wrote on Sunday, July 20, 2014:

I’m running code, generated by STM32Cube and have dropped to HardFault_Handler() on start of execution.
By debugging I have found this issue:
STM32Cube code generator placed xPortSysTickHandler() into SysTick_Handler() for 1 ms tick counting.
xPortSysTickHandler() calls xTaskIncrementTick(), which execute follow code:

if( listCURRENT_LIST_LENGTH( &( pxReadyTasksLists[ pxCurrentTCB->uxPriority ] ) ) > ( UBaseType_t ) 1 )

But, clock system had initialized before any tasks have been created.
And interrupt SysTick_Handler() may occurs when no tasks and even scheduler is not running yet.
So, pointer pxCurrentTCB equal to zero at this moment and [pxCurrentTCB -> uxPriority] indicates some random value.
It caused system exception.

As temporary solution I just have added one of task creation procedure before clock initialization. That is not good coding, but works.

rtel wrote on Sunday, July 20, 2014:

The clock should not be initialised until the scheduler has started. The sequence used by FreeRTOS is as follows:

  1. Ensure interrupts are disabled.
  2. Configure clock to generate tick interrupts.
  3. Start the first task (which automatically re-enables interrupts).

So the clock initialisation code must itself be enabling the systick, which it need not do as FreeRTOS will do that itself at the appropriate time.

One solution for you, which assumes the ST code is not also enabling interrupts, is to simply call taskENTER_CRITICAL (or globally disable interrupts) before the clock initialisation is done. The scheduler will automatically set the global interrupts and critical section state to be correct when it starts.

Richard Barry.

dddddppppp wrote on Wednesday, July 23, 2014:

Thanks Richard!
ST promisses to fix it in the next version.
So, I have decided to develop own project, using Cube as reference only.
Based on examlpes from FreeRTOS package it work fine.

Dmitry Popov