K60 RAM allocation issue when adding a thread

[kfranz]

I started with the TWR-K60N512_Quick_Start_Demo. I added a thread to TASK_TEMPLATE_STRUCT MQX_template_list:

  { MC_TIME_TASK,  time_task,        800,     12,      "mcTime",         0,                   0,      0 },

Doing this causes MQX to "blow up" and restart when _mqx() calls:

 _mem_zero((pointer)kernel_data, (_mem_size)sizeof(KERNEL_DATA_STRUCT));

Investigation shows that before I add the thread to MQX_template_list, kernel_data is at address 0x1FFF0960-0x1FFF0C80. After I add the thread, kernel_data is at address 0x2000FEF0-0x20010210. There is no RAM past 0x2000FFFF.

Before adding thread:

MQX_init_struct -> START_OF_KERNEL 0x1FFF0960

MQX_init_struct -> END_OF_KERNEL     0x2000FFF0

After adding thread:

MQX_init_struct -> START_OF_KERNEL 0x2000FEF0

MQX_init_struct -> END_OF_KERNEL     0x00000411

What caused MQX_init_struct to have invalid addresses for the START_OF_KERNEL and END_OF_KERNEL?

thanks,

Kevin

[MarkP_]

Hi Kevin,

one explanation could be that your task uses a lot of RAM.

Could you check the RAM allocations in map-file.

~Mark

[kfranz]

No, it's a small simple task. If I delete one of the tasks in the TWR-K60N512_Quick_Start_Demo which use more stack space than the task I wish to create (for example, sdcard_task) and replace it with my task everything is okay.

I have been doing embedded development for many years on other platforms using an RTOS. So, I have a good understanding of the process. The plan is for us to port all of our products to Kinetis. I just have to figure out Kinetis and MQX, .

My guess is that something is getting placed at an absolute address and when I add my task I am crossing a boundary or exceed an allocation. The question is what boundary or allocation? and, how do I fix it? I looked at the Lightweight Memory. It says I have 16K free at 0x1fffbe10 and 65k free at 0x20000004.

Do you have any other ideas?

thanks,

Kevin

[MarkP_]

Hi,

what values map file shows for these segments:

  1FFF8000 00014C9C 00001210 .main_application_data ram
  1FFF9210 1FFF9210 00001B10 .main_application_bss ram
  1FFFAD20 1FFFAD20 00000000 .kernel_data ram
  20007FE0 20007FE0 00000000 .end_of_kernel_data end_of_kd

(Those values are for K10, it has only 64k RAM)

There is a discontinuity in RAM at address 0x20000000, hidden in init_bsp.c:

uint_32 _bsp_enable_card(void)

...

    /* Memory splitter - prevent accessing both ram banks in one instruction */
    _mem_alloc_at(0, (void*)0x20000000);

~Mark

[kfranz]

I am using the IAR Embedded Workbench compiler version 6.30.

I have traced the problem to the initialization of MQX_init_struct before main() is called.

int main(void)
{
  /* Start MQX */
  _mqx( (MQX_INITIALIZATION_STRUCT_PTR) &MQX_init_struct );
   return 0;
}

__iar_data_init3 is an assembler function which appears to be initializing RAM constants. It is provided by IAR and there is no source code for it. When this function executes, MQX_init_struct gets set to values which are not correct. At this point I think I need to contact IAR.

thanks,

Kevin