OpenBLT and Keil RTX

The application is started via pProgResetHandler read out form the application vector table (CPU_USER_PROGRAM_VECTABLE_OFFSET + 4).

However the stack pointer is not configured based on the application (first entry of the application vector table) but rather inherited from the bootloader. Maybe this is causing issues.

Also is the core left in handler mode and using MSP (as default out of reset)? Otherwise RTX initialization and startup might fail.

The application is started via pProgResetHandler read out form the application vector table (CPU_USER_PROGRAM_VECTABLE_OFFSET + 4).

However the stack pointer is not configured based on the application (first entry of the application vector table) but rather inherited from the bootloader. Maybe this is causing issues.

Also is the core left in handler mode and using MSP (as default out of reset)? Otherwise RTX initialization and startup might fail.

Hi Robert,

thanks for your feedback.

The stack pointer should be set to the applications stack pointer on the SystemInit at the RTX application:

; Reset handler routine
Reset_Handler    PROC
                 EXPORT  Reset_Handler             [WEAK]
     IMPORT  __main
     IMPORT  SystemInit
                 LDR     R1, =__initial_sp      ; Load position of SP
                 MOV     SP, R1                 ; set SP to position
                 LDR     R0, =SystemInit
                 BLX     R0
                 LDR     R0, =__main
                 BX      R0
                 ENDP

While stepping through, the SP is located at the following addresses:
0x20000D68 on reset
0x20000D50 when entering this Reset_Handler and
0x200024D8 after moving R1 into SP

About the core modes: I'm not so sure about that, but as far that I know, the core is started in thread mode with privileged access.

Have you tried the RTX application without the bootloader?

If yes, you could check with the debugger what behaves differently.

Hi Robert,

so I've debugged the last days to check various things.

What I've found out (and will be most obvious to you I think) is, that the _platform_post_lib_init() function is used to create a main thread to call the user code main() function.

After creating the thread, the osKernelStart is called and should execute the main thread and thus should never "return" to the _platform_post_lib_init() function where the next instruction is the exit = hard fault, if semihosting is enabled...

Anyhow, the problem seems to be in the creation of the task. So the function svcThreadCreate() could not create the main task and so no task is running when the kernel starts.

// Thread Service Calls

/// Create a thread and add it to Active Threads and set it to state READY
osThreadId svcThreadCreate (const osThreadDef_t *thread_def, void *argument) {
  P_TCB  ptcb;
  OS_TID tsk;
  void  *stk;

  if ((thread_def == NULL) ||
      (thread_def->pthread == NULL) ||
      (thread_def->tpriority < osPriorityIdle) ||
      (thread_def->tpriority > osPriorityRealtime)) {
    sysThreadError(osErrorParameter);
    return NULL;
  }

  if (thread_def->stacksize != 0U) {            // Custom stack size
    stk = rt_alloc_mem(                         // Allocate stack
      os_stack_mem,
      thread_def->stacksize
    );
    if (stk == NULL) {
      sysThreadError(osErrorNoMemory);          // Out of memory
      return NULL;
    }
  } else {                                      // Default stack size
    stk = NULL;
  }

  tsk = rt_tsk_create(                          // Create task
    (FUNCP)thread_def->pthread,                 // Task function pointer
    (uint32_t)
    (thread_def->tpriority-osPriorityIdle+1) |  // Task priority
    (thread_def->stacksize << 8),               // Task stack size in bytes
    stk,                                        // Pointer to task's stack
    argument                                    // Argument to the task
  );

  if (tsk == 0U) {                              // Invalid task ID
    if (stk != NULL) {
      rt_free_mem(os_stack_mem, stk);           // Free allocated stack
    }
    sysThreadError(osErrorNoMemory);            // Create task failed (Out of memory)
    return NULL;
  }

  ptcb = (P_TCB)os_active_TCB[tsk - 1U];        // TCB pointer

  *((uint32_t *)ptcb->tsk_stack + 13) = (uint32_t)osThreadExit;

  return ptcb;
}

Where exactly does the svcThreadCreate() function fail?

Hi Robert,

I think it's a problem inside the rt_tsk_create() function where a memory box should be allocated:

OS_TID rt_tsk_create (FUNCP task, U32 prio_stksz, void *stk, void *argv) {
  /* Start a new task declared with "task". */
  P_TCB task_context;
  U32 i;

  /* Priority 0 is reserved for idle task! */
  if ((prio_stksz & 0xFFU) == 0U) {
    prio_stksz += 1U;
  }
  task_context = rt_alloc_box (mp_tcb);
  if (task_context == NULL) {
    return (0U);
  }
  /* If "size != 0" use a private user provided stack. */
  task_context->stack      = stk;
  task_context->priv_stack = (U16)(prio_stksz >> 8);
  /* Pass parameter 'argv' to 'rt_init_context' */
  task_context->msg = argv;
  /* For 'size == 0' system allocates the user stack from the memory pool. */
  rt_init_context (task_context, (U8)(prio_stksz & 0xFFU), task);

  /* Find a free entry in 'os_active_TCB' table. */
  i = rt_get_TID ();
  if (i == 0U) {
    return (0U);
  }
  os_active_TCB[i-1U] = task_context;
  task_context->task_id = (U8)i;
  DBG_TASK_NOTIFY(task_context, __TRUE);
  rt_dispatch (task_context);
  return ((OS_TID)i);
}

on the line where rt_alloc_box() gets called, the task_conext is Null because the function call returns after this line. It's a bit tricky because this code is only as assembler available and some of it can't get accessed with the debugger. For example I can't get the content of the tast_context but jumping to the return statement let's me guess, that there is a problem with this.

Here is the called function where the __USE_EXCLUSIVE_ACCESS is defined (so the #else part is running):

void *rt_alloc_box (void *box_mem) {
  /* Allocate a memory block and return start address. */
  void **free;
#ifndef __USE_EXCLUSIVE_ACCESS
  U32  irq_mask;

  irq_mask = (U32)__disable_irq ();
  free = ((P_BM) box_mem)->free;
  if (free) {
    ((P_BM) box_mem)->free = *free;
  }
  if (irq_mask == 0U) { __enable_irq (); }
#else
  do {
    if ((free = (void **)__ldrex(&((P_BM) box_mem)->free)) == 0U) {
      __clrex();
      break;
    }
  } while (__strex((U32)*free, &((P_BM) box_mem)->free));
#endif
  return (free);
}

It looks as if the memory for task control block could not be allocated. This should normally not occur since the memory is provided statically by the kernel.

Are you sure that rt_tsk_create() fails due to rt_alloc_box() returning NULL?

Even if the main thread could not be created, the function osKernelStart() should not return but switch to the Idle task which is created implicitly during osKernelInitialize().

I suggest you debug further and check first osKernelInitialize() and then osKernelStart().

Or you could try out a simple RTX based example without using the bootloader and examine what is different. Maybe you have a tool or configuration issue.

You could take a look at existing examples (Blinky) found in the Device Family Pack for STM32F1 (http://www.keil.com/dd2/pack/).

It should be also very easy to create a test project with empty main that uses RTX and verify that is starts properly. It takes less than a minute (a few clicks) to create an RTX project with MDK and using Run-Time Environment.
It can be also configured for the simulator.

You could also consider the newer RTX5 (CMSIS RTOS2) with many new features (available as source code also - so debugging should be eaiser).

Try these in conjunction with what Robert #1 suggested.

1) See that your RTX_Config.c file is configured properly. Make sure you are looking at the EXACT one that is being used.

2) Look in the map file and see how much space is allocated for mp_tcb, mp_stk, os_stack_mem and make sure it matches what your RTX_Config.c files thinks they should be.

3) For a test, build a scatter file for the APP that loads at 0x08000000, rebuild and overwrite the bootloader and see if this runs properly or not.

An item of interest is that the Bootloader seems to set the stack to the highest RAM address as opposed to something within the range of 0x20000000 and 0x20000C00. Probably not your issue, but if you are really intending to keep these 2 separate, they are not.

It looks as if the memory for task control block could not be allocated. This should normally not occur since the memory is provided statically by the kernel.

Yes it is not something that occurs on its own, but it is only "semi" statically allocated. It can be set to 0 (i.e. it is possible that this is "empty" or "has no items" in it, but the compiler will not tell you that (so not quite statically allocated). As it seems that the rt_alloc_box for the STACKS may also be failing, maybe it is also zero or something else is wrong but related what makes the rt_alloc fail for both the TCB and Stack area's.

Even if the main thread could not be created, the function osKernelStart() should not return but switch to the Idle task which is created implicitly during osKernelInitialize().

Yes and I am not seeing any reason to believe that it not being scheduled and dispatched. It would seem that it is probably the only thread the system thinks it has. The TCB for the idle thread is statically allocated (it is REALLY statically allocated such that if it compiles space for it will exist). My guess is the rt_alloc for the stack for the Idle thread is failing, and this causes it to be set to 0. When the initial stack frame is generated and the stack adjusted, it will not actually be pointing to or set to anything valid. I believe the hard fault comes when the system "returns" to the idle thread.

I suggest you debug further and check first osKernelInitialize() and then osKernelStart().

And remember (or note) you can add the source code to the os to your application and it will allow you to do source level debugging of your app including the os. (just make sure that after you add the files, you select do not include in build)

You could take a look at existing examples (Blinky) found in the Device Family Pack for STM32F1 (http://www.keil.com/dd2/pack/).

Yes, the blinky examples are always great.

Hi Robert,

thanks for your answer. Unfortunately changing/upgrading the rtos seems to much for me, due to licensing and affected products. But I've created a simple blinky project (who matches my hardware) and the problem occures too. So I've debugged a lot and found out, that the variables os_initialized and os_running are not set to zero when I start the application and thus the memory box (who is used to allocate memory for everything in the rtx) is not created. If I add the following code into the startup reset handler:

; Reset handler routine
Reset_Handler    PROC
                 EXPORT  Reset_Handler             [WEAK]
         IMPORT  __main
         IMPORT  SystemInit
         IMPORT  ClearSpecial
                 LDR     R1, =__initial_sp                      ; Added for OpenBLT support
                 MOV     SP, R1

                 LDR     R0, =ClearSpecial
                 BLX     R0

                 LDR     R0, =SystemInit
                 BLX     R0
                 LDR     R0, =__main
                 BX      R0
                 ENDP

and then in my c-code:

void ClearSpecial(void)
{
        os_initialized = 0;
        os_running = 0;
}

the RTX code works (!), but only on the "simple" blinky project.

When I went further to use this on my application, who I want to use with the OpenBLT, the kernel starts and the threads were initialized, but after the threads are running I get again a hardfault.

This time the hardfault is fired "somewhere". By somewhere I mean that while stepping through the code, that I think not the shown code in the debugger is causing the issue, but more some IRQ related part...

As I've looked into the *.map file I found, that the both variables are data (for me = not zero initialized). So they must be initialized somewhere in the startup / arm library, because the RTX is running when I don't use the bootloader part.

So I've set an access-breakpoint on the os_initialized (write) and "found" the function __scatterload. So now I'm observing this (but I wanted to inform you about it), because I think my problems are caused by some not-initialized variables...

Hello together,

sorry for you Robert #2, I haven't seen your reply but finally I got things running. But I'm a bit confused, if I've found the "best practise" solution for it. So to summarize what I've done (and as a possible ToDo-List for future projects):

- Prepare the OpenBLT depending on your needs (in my case only UART is needed)

- Specify on your own, which memory addresses are used for the bootloader and the rtx application you would like to run, depending on used microcontroller in my case a STM32F103C6: + OpenBLT (only with UART), ROM from 0x08000000 - 0x080017FF, RAM from 0x2000000 - 0x20000BFF + RTX App, ROM from 0x080001800 - 0x08008000, RAM from 0x20000C00 - 0x20002800

- Apply the OpenBLT specific settings (vector table offset, checksum placeholder) on the bootloader application

- To run the RTX application with the OpenBLT, the semihosting must be disabled with the following pragma in the c-code, in my case placed in the main.c file:

#pragma import(__use_no_semihosting)

- After compiling, implement the needed semihosting functions (like in retarget.c, the compiler should tell you the missing functions) like the _sys_exit:

void _sys_exit(int code)
{
        // in my case: just reset to re-run the application, replace with more specific solution
        NVIC_SystemReset();
}

- Implement an "RTX Reset" function, who cleares the os_initialized and os_running variables, call it at the startup of the RTX application, in my case the function is named "ClearSpecial" in startup_stm32f10x_ld.s:

; Reset handler routine
Reset_Handler    PROC
                 EXPORT  Reset_Handler             [WEAK]
     IMPORT  __main
     IMPORT  SystemInit
     IMPORT  ClearSpecial
                 LDR     R1, =__initial_sp                      ; Added for OpenBLT support to get the correct stack pointer location
                 MOV     SP, R1
                 LDR     R0, =ClearSpecial                      ; Added for RTX support
                 BLX     R0
                 LDR     R0, =SystemInit
                 BLX     R0
                 LDR     R0, =__main
                 BX      R0
                 ENDP

void ClearSpecial(void)
{
        os_initialized = 0;
        os_running = 0;
}

- Use a scatter file for relocating the RTX application:

; *************************************************************
; *** Scatter-Loading Description File for OpenBLT          ***
; *************************************************************

LR_IROM1 0x08001800 0x00005800  {    ; load region size_region
  ER_IROM1 0x08001800 0x00005800  {  ; load address = execution address
   *.o (RESET, +First)
   *(InRoot$$Sections)
   .ANY (+RO)
  }
  RW_IRAM1 0x20000C00 0x00001C00  {  ; RW data
   .ANY (+RW +ZI)
  }
}

- Edit the core_cm3.h file definitions (line 250 and 256 in original) to place constant variables in ROM and not in RAM. To do this, explore to the file and remove the write-protection and edit the following:

/* IO definitions (access restrictions to peripheral registers) */
/**
    \defgroup CMSIS_glob_defs CMSIS Global Defines

    <strong>IO Type Qualifiers</strong> are used
    \li to specify the access to peripheral variables.
    \li for automatic generation of peripheral register debug information.
*/
#ifdef __cplusplus
  #define   __I     volatile             /*!< Defines 'read only' permissions */
#else
  //#define   __I     volatile const       /*!< Defines 'read only' permissions */
  #define   __I     const       /*!< Defines 'read only' permissions, special version for compiler V5.05, update 2, build 169 */
#endif
#define     __O     volatile             /*!< Defines 'write only' permissions */
#define     __IO    volatile             /*!< Defines 'read / write' permissions */

/* following defines should be used for structure members */
//#define     __IM     volatile const      /*! Defines 'read only' structure member permissions */
#define     __IM     const      /*! Defines 'read only' structure member permissions, special version for compiler V5.05, update 2, build 169 */
#define     __OM     volatile            /*! Defines 'write only' structure member permissions */
#define     __IOM    volatile            /*! Defines 'read / write' structure member permissions */

/*@} end of group Cortex_M3 */

- Add the after build command in the RTX application to generate a *.srec file, used by the OpenBLT MicroBoot application (Options for Target->User->Run #1), and don't forget to check the checkbox!

fromelf --m32 --output=OutputDirectory\RTX_App.srec OutputDirectory\RTX_App.axf --cpu Cortex-M3

To debug the RTX application, do the following: + Compile the OpenBLT application and flash the microcontroller + Compile the RTX application and flash the microcontroller (start a debug-session) + Flash the controller again with the MicroBoot application (microcontroller isn't leaving the bootloader, because the checksum is incorrect until the RTX application is flashed within the bootloader)

- Be happy!

There is a character limit of 7000 per post ;-)

My final thoughts on this one is the editing of the CMSIS library. I think this isn't a good way to do this, but I'm unaware, where I could change this. When I got the 2nd hardfault I was wondering about the clock speeds of the microcontroller, read by the function RCC_GetClocksFreq(): PCLK1 and PCLK2 were set to 72MHz (SYSCLK), what was a bit surprising. Then when running the RTX application without the bootloader, the clock speeds were PCLK1 32MHz, PCLK2 72MHz. 32MHz on the PCLK1 is the maximum and should not exceed! So I've found out, that the function is using a table to determine the frequencies:

(snippet from RCC_GetClocksFreq())
/* Get PCLK1 prescaler */
  tmp = RCC->CFGR & CFGR_PPRE1_Set_Mask;
  tmp = tmp >> 8;
  presc = APBAHBPrescTable[tmp];
  /* PCLK1 clock frequency */
  RCC_Clocks->PCLK1_Frequency = RCC_Clocks->HCLK_Frequency >> presc;
  /* Get PCLK2 prescaler */
  tmp = RCC->CFGR & CFGR_PPRE2_Set_Mask;
  tmp = tmp >> 11;
  presc = APBAHBPrescTable[tmp];
  /* PCLK2 clock frequency */
  RCC_Clocks->PCLK2_Frequency = RCC_Clocks->HCLK_Frequency >> presc;

where the table is located in stm32f10x_rcc.c:

/** @defgroup RCC_Private_Variables
  * @{
  */

static __I uint8_t APBAHBPrescTable[16] = {0, 0, 0, 0, 1, 2, 3, 4, 1, 2, 3, 4, 6, 7, 8, 9};
static __I uint8_t ADCPrescTable[4] = {2, 4, 6, 8};

and the macro __I is defined as (in core_cm3.h):

#define   __I     volatile const       /*!< Defines 'read only' permissions */

So normally I've expected to locate the const array in the RO section but the map-file tells me, that this table is located in RAM?!?:

    APBAHBPrescTable                         0x20000cd8   Data          16  stm32f10x_rcc.o(.data)
    ADCPrescTable                            0x20000ce8   Data           4  stm32f10x_rcc.o(.data)

With the debugger I could see, that the RAM is cleared "somewhere" and thus not filled up with the const data! (I think the initializing function is erazing the wrong sectors...). And by changing the __I macro to

#define   __I     const       /*!< Defines 'read only' permissions, special version for compiler V5.05, update 2, build 169 */

the map-file looks like expected (in ROM):

    APBAHBPrescTable                         0x08006e98   Data          16  stm32f10x_rcc.o(.constdata)

So I guess this could be very specific, so here are my versions:

Toolchain:          MDK-ARM Standard Cortex-M only: 1 user(s)  Version: 5.15.0
Toolchain Path:     C:\Keil_v5\ARM\ARMCC\Bin
C Compiler:         Armcc.exe                                  V5.05 update 2 (build 169)
Assembler:          Armasm.exe                                 V5.05 update 2 (build 169)
Linker/Locator:     ArmLink.exe                                V5.05 update 2 (build 169)
Library Manager:    ArmAr.exe                                  V5.05 update 2 (build 169)
Hex Converter:      FromElf.exe                                V5.05 update 2 (build 169)
CPU DLL:            SARMCM3.DLL                                V5.15.0
Dialog DLL:         DCM.DLL                                    V1.13.2.0
Target DLL:         UL2CM3.DLL                                 V1.155.8.0
Dialog DLL:         TCM.DLL                                    V1.14.5.0
ARM::CMSIS          5.0.0-Beta4
STM32F1xx_DFP       1.1.0

So I would like to get your thoughts about this, I think I found a way around the problem but not the source of it...

It seems that RW data (including zero initialized variables) is not properly initialized in your application. This is most likely causing the issues that you have seen and tried to bypass with different fixes.

"ClearSpecial" should not be needed since both os_xxx variables should be initialized to zero by the scatter loader.

Modifying the CMSIS core_cm3.h header file should also not be needed to overcome the RCCGetClocksFreq function implementation provided by ST.

It is true that the table APBAHBPrescTable and ADCPrescTable could be made const and would be located in ROM (not use any RAM). However even if they are located in RAM they should be initialized with the specified values by the scatter loader.

You should check what happens with the mentioned RW data initialization during bootloader and when the actual application is being started.

Yes to all Robert #1 said.

Your MDK version is a little old, but probably not your issue. BUT, I would not count it out. Upgrading to something more recent (I still use 5.16a on a few active projects and I don't have issues running an RTX application from a bootloader)

So I would like to get your thoughts about this, I think I found a way around the problem but not the source of it...

IF your scatter loading for your RTX application is not executing properly I cannot imagine that you will not find something else that is not initialized and causes an issue. I don't think having variables not initialized and zero-init data not being zeroed is going to be OK for anything.

Bottom line is you are correct, there is a problem. Scatterloading not happening is a serious issue. Calling main() instead of __main() will cause Scatterloading to not happen, but also _platform_post_lib_init() would not get called, and it is.

I would step into __main() in the dissassembly window and then step into the __scatterload() function. It is the first function call after entering __main(). you need to determine if / way RW variables are not being initialized and 0 init variables are not zeroed.

If scatterloading is not happening properly you will likely have many problems as your application grows.

Maybe another test would be see if you can create a minimal bootloader that does nothing more then jump to the RTX application.

Hello again,

thanks for your answers. Truly there must be something wrong with this scatterload function, but I've never seen/used it (and also questionned this basic functionality because it has always worked)...

Just to get this correctly and understand the functionality: the function call to the __scatterload is made to initialize variables with pre-defined values including zeroing some variables. As far as I understand, there is a __scatterload_copy and a __scatterload_zeroinit function to do exactly this.

The copy function is loading the variables with the pre-defined values and the zeroinit is filling other variables with zeroes.

Unfortunately I haven't found the "source code" of the __scatterload functions but with the *.map file I could set breakpoints where those functions are called. Or do you have a Idea of how to get to the source of this?

So I've changed back the __I definition and removed the call to the "ClearSpecial" function to get to the "basic", failing rtx application.

So here is the assembler code from the disassembly view:

                 __scatterload_copy:
0x0800192C 3A10      SUBS     r2,r2,#0x10
0x0800192E BF24      ITT      CS
0x08001930 371A      ADDCS    r7,r7,#0x1A
0x08001932 AFFF      ADDCS    r7,sp,#0x3FC
0x08001934 D8FA      BHI      __scatterload_copy (0x0800192C)
0x08001936 0752      LSLS     r2,r2,#29
0x08001938 BF24      ITT      CS
0x0800193A C830      LDMCS    r0!,{r4-r5}
0x0800193C C130      STMCS    r1!,{r4-r5}
0x0800193E BF44      ITT      MI
0x08001940 6804      LDRMI    r4,[r0,#0x00]
0x08001942 600C      STRMI    r4,[r1,#0x00]
0x08001944 4770      BX       lr
0x08001946 0000      MOVS     r0,r0
                 __scatterload_zeroinit:
0x08001948 2300      MOVS     r3,#0x00
0x0800194A 2400      MOVS     r4,#0x00
0x0800194C 2500      MOVS     r5,#0x00
0x0800194E 2600      MOVS     r6,#0x00
0x08001950 3A10      SUBS     r2,r2,#0x10
0x08001952 BF28      IT       CS
0x08001954 C178      STMCS    r1!,{r3-r6}
0x08001956 D8FB      BHI      0x08001950
0x08001958 0752      LSLS     r2,r2,#29
0x0800195A BF28      IT       CS
0x0800195C C130      STMCS    r1!,{r4-r5}
0x0800195E BF48      IT       MI
0x08001960 600B      STRMI    r3,[r1,#0x00]
0x08001962 4770      BX       lr

I've added a breakpoint on the line 0x0800192C and 0x08001948. So when entering the __scatterload_copy I have:

R0 =      0x08006E9C
R1 =      0x20000C00
R2 =      0x00000114
R3 =      0x0800192D
R4 =      0x08001AE1
R5 =      0x20000504
R6 =      0x00000000
R7 =      0x08006E7B
R8 =      0x00000000
R9 =      0x20000160
R10 =     0x08006E8C
R11 =     0x08006E9C
R12 =     0x00000008
R13(SP) = 0x20002750
R14(LR) = 0x08001907
R15(PC) = 0c0800192C

and when entering the __scatterload_zeroinit:

R0 =      0x08006FB0
R1 =      0x20000D14
R2 =      0x00001A3C
R3 =      0x08001949
R4 =      0x00000100
R5 =      0x20000504
R6 =      0x00000000
R7 =      0x20002B4C
R8 =      0x00000000
R9 =      0x20000160
R10 =     0x08006E9C
R11 =     0x08006E9C
R12 =     0x00000008
R13(SP) = 0x20002750
R14(LR) = 0x08001907
R15(PC) = 0c08001948

I'm a bit overwhelmed with the assembler code, but I saw there two addresses: 0x20000C00 in the copy function (R1) and 0x20000D14 in the zeroinit. Considering the *.map file I found that from 0x20000C00 to 0x20000D14 are .data variables located and from 0x20000D14 upwards .bss variables. For the note: the APBAHBPrescTable is located at 0x20000CD8 and the os_initialized and os_running at 0x20000CEC / 0x20000CED = in the .data section.

However, I think that the __scatterload_copy is using the wrong address to copy the initialization values and thus creating those problems, so I'm starting now to research those assembler commands to get the logic of the copy function and to get the address of the initialization values who should be stored in the ROM section I think...

What you have shown looks very correct. A basic project would have 1 copy section and 1 zero init section. The compiler will compress the copy data stored in flash if the total size of the compressed data and the code to decompress it is less then the total size of the uncompressed init data. (my guess is your data is compressed)

what values does os_initialized and os_running have immediately after calling the __scatterload_copy? If these are zero, are either of them non-zero at the osKernelInitialize call?

You should be able to tell if __scatterload_copy was sucessful as after it is run, ANY variable that is initialized you should be able to find it's address in the map file and then do a view memory in the debugger. If you think scatterloading is happening to/from a wrong address, provide more information on that.

There is not yet any indication that __scatterload_copy or __scatter_load_zeroinit is bad so no need to get the source for it. (the zero_init is extreamly simple to analyze for correctness in assembly language as well as copy for sections that are not compressed.)

It might be helpful if you can extract a few 32-bit values fro the Application code.

You should have a variable in the map file called Region$$Table$$Base.
The next 8 32-bit values may be helpful, though it really appears as if scatterload is setup properly. This should tell us if a wrong address is being used to copy from scatter loading wise.

Hi Robert,

thanks for your message. So I've stepped through the __scatterload_copy and the os_initialized and os_running are 1 when I get to the __scatterload_copy, after it, they were also 1 and even after the __scatterload_zeroinit they were 1. For fun, I've set them to a value of 120 at the beginning of the __main call and they were untouched until the osKernelInitialize call, where they were set to 1, but because of the "non zero" value the memory don't gets allocated and so on...
On the other hand, when I set a variable who is in the zeroinit section to non-zero values at the beginning of the __main, the variable get's correctly set to zero, so I think we're on the right track here.

The variable Region$$Table$$Base is in my case located at 0x08006E7C and the Region$$Table$$Limit is located at 0x08006E9C.

The memory values are (hex from the memory window):

Address    = Hexhexhexhex = converted -> my guess, what this number means
0x08006E7C = 9C 6E 00 08 = 0x08006E9C -> end of this array
0x08006E80 = 00 0C 00 20 = 0x20000C00 -> scatterload starting address for initializing
0x08006E84 = 14 01 00 00 = 0x00000114 -> size for initializing
0x08006E88 = 2C 19 00 08 = 0x0800192C -> address of the copy function __scatterload_copy
0x08006E8C = B0 6F 00 08 = 0x08006FB0 -> special memory address?!?
0x08006E90 = 14 0D 00 20 = 0x20000D14 -> scatterload starting address for zeroing
0x08006E94 = 3C 1A 00 00 = 0x00001A3C -> size for zeroing
0x08006E98 = 48 19 00 08 = 0x08001948 -> address of the zero function __scatterload_zeroinit
0x08006E9C = 00 01 00 00 = 0x00000100 -> special size?!?

Can someone see some hint in the numbers? The "special memory address" seems suspicious to me but maybe that is just the "decompress" function...