[Cortex-M3] Execution code from RAM during Flash Programming

Do you mean your bootloader has flash programming feature, and
want to copy the boot loader from flash to SRAM when the startup code is executed?
regards,
Joseph

In that case, it would be easier to copy the whole boot loader (rather than a portion of it) from flash to SRAM in the assembly startup code.
The boot loader, assumed to be mostly written in C, will have to linked to use the SRAM address as program memory range.

In the assembly startup code you will have a reset handler, and normally you can see something like:

Reset_Handler   PROC

             EXPORT  Reset_Handler    [WEAK]
             IMPORT  __main
             LDR  R0,=__main
             BX     R0   ; this branch to C startup code
             ENDP

You can insert an assembly code to copy the memory from flash to SRAM, before jumping to the C startup code which will be in SRAM.

Reset_Handler   PROC

             EXPORT  Reset_Handler    [WEAK]
             LDR  R4,=0x00000000 ; Source starting address
             LDR  R5,=0x20000000 ; Destination starting address
             LDR  R6,=0x4000  ; 16K program size to copy

CopyFlashToSRAMLoop

             LDMIA R4!,{R0-R3}
             STMIA R5!,{R0-R3}   
             LDMIA R4!,{R0-R3}
             STMIA R5!,{R0-R3}   
             LDMIA R4!,{R0-R3}
             STMIA R5!,{R0-R3}   
             LDMIA R4!,{R0-R3}
             STMIA R5!,{R0-R3}
             SUBS R6,#64 ; Each time 64 bytes are copied
             BNE CopyFlashToSRAMLoop
             IMPORT  __main
             LDR  R0,=__main
             BX     R0   ; this branch to C startup code  
             ENDP

In this way, the whole boot loader will be running in SRAM.  You need to be careful to make sure that the data use by the boot loader is not in the same address as the instruction memory.

regards,
Joseph

Hi Rohit,

In this case you could develop the bootloader and flash programming code as two separate projects.
The flash programming project is compiled and link using SRAM as program address.
Then you can generate hex image, and insert this into your boot loader project as a constant array.
Inside your boot loader, you can copy this constant array into SRAM.

To jump from boot loader to flash programming code, personally I got used to use embedded assembly:

/* ARM RVDS or Keil MDK */
__asm void jump_to_sram(void)
{
   LDR   R0,=0x20000000  ; starting address of program in SRAM.
   LDR   R1,[R0]  ; Get initial MSP value
   MOV   SP, R1  ; Set SP value
   LDR   R1,[R0, #4] ; Get initial PC value
   BX R1
}

You can use function pointer to do the same.
regards,
Joseph

For Cortex-M3 the low bit of the address needs to be set to indicate Thumb state, so if the starting address in RAM was say 0x20000100 (i.e. 0x20000101 is stored at 0x20000004 in the asm above).

In C it would be

typedef void RAM_fn_t(void);
RAM_fn_t* fp = (RAM_fn_t*)0x20000101;
(*fp)();


It doesn't set SP the way the asm above does.  The cast is technically undefined behavior in C, but pragmatically it does what it looks like.

Equivalently, if you prefer terseness/parentesis:

(*((void (*)(void))0x20000101))();


In that case, it would be easier to copy the whole boot loader (rather than a portion of it) from flash to SRAM in the assembly startup code.
The boot loader, assumed to be mostly written in C, will have to linked to use the SRAM address as program memory range.

In the assembly startup code you will have a reset handler, and normally you can see something like:

Reset_Handler   PROC

                EXPORT  Reset_Handler             [WEAK]
                IMPORT  __main
                LDR     R0,=__main
                BX        R0   ; this branch to C startup code
                ENDP

You can insert an assembly code to copy the memory from flash to SRAM, before jumping to the C startup code which will be in SRAM.

Reset_Handler   PROC

                EXPORT  Reset_Handler             [WEAK]
                LDR     R4,=0x00000000 ; Source starting address
                LDR     R5,=0x20000000 ; Destination starting address
                LDR     R6,=0x4000     ; 16K program size to copy

CopyFlashToSRAMLoop

                LDMIA    R4!,{R0-R3}
                STMIA    R5!,{R0-R3}               
                LDMIA    R4!,{R0-R3}
                STMIA    R5!,{R0-R3}               
                LDMIA    R4!,{R0-R3}
                STMIA    R5!,{R0-R3}               
                LDMIA    R4!,{R0-R3}
                STMIA    R5!,{R0-R3}
                SUBS    R6,#64 ; Each time 64 bytes are copied
                BNE    CopyFlashToSRAMLoop
                IMPORT  __main
                LDR     R0,=__main
                BX        R0   ; this branch to C startup code        
                ENDP

In this way, the whole boot loader will be running in SRAM.  You need to be careful to make sure that the data use by the boot loader is not in the same address as the instruction memory.

regards,
Joseph

Hi Kent,

It depends on how you want the project to look like.

If the program code in RAM is created as a completely independent project, and then merged into your other project running in flash using a hex file editor, then you don't have to use scatter files. In this way you will have two separated images in your running system. This can make sure you don't accidentally call a function in flash or read it when you are programming the flash.

If you are trying to create a project that contains code running in flash as well as code running in SRAM, then you need to define the code running in SRAM with a section name, and then use linker script to specify this section is to be copied to and executed from SRAM.

regards,
Joseph

You can use function pointer to do the same.