Project Logs

Collapse

• Picking template code: linker scripts

  Christoph • 06/05/2023 at 07:05 • 0 comments

  Adding a new target to BF shouldn't be too hard if the new target is similar to an already existing target - right? Let's see what we have

  • STM32H743: a bit older
  • STM32H730: somewhat special, because:
    • This chip uses external memory to store the application
    • but it shares a reference manual with H723, H725, H733, and H735
  • STM32H723: as far as I know, this was just a draft and never actually worked.

  Let's take a closer look at the H743.

  Memory areas

  BF's H743 target uses a slightly customized layout that reserves flash for the reset handler and an emulated eeprom. Both BF and CubeIDE define areas in ITCM and DTCM for critical functions and data. Here's a table that compares BF's H743 memory areas with CubeIDE's H725/H735 standard:

  BF H743	CubeIDE H725/H735
  FLASH (reset handler)	FLASH
  FLASH_CONFIG (emulated EEPROM)
  FLASH1 (application and constants)
  ITCM_RAM	ITCMRAM
  DTCM_RAM	DTCMRAM
  RAM	RAM_D1
  D2_RAM	RAM_D2
  MEMORY_B1 (*)	RAM_D3 (**)

  *) external memory, can only be used if there's actually some external memory connected to the MCU

  **) not used by BF so we can hopefully remove that. Or not, since simply having a memory area in a linker script doesn't hurt.

  The BF linker script also defines two aliases:

  REGION_ALIAS("STACKRAM", DTCM_RAM)
  REGION_ALIAS("FASTRAM", DTCM_RAM) 

  The memory area names are a bit different between BF and CubeIDE but that's not a problem because these are only used within the linker script. A little understanding surely doesn't hurt so let's also have a look at the system architecture diagram for the H725. I've already marked the memories used:

  • Top left: ITCM and DTCM are tightly coupled with the core for fast access.
  • Center: Flash and SRAM in D1 domain via AXI
  • Right: SRAM1 and SRAM2 in D2 domain via AXI and D1-to-D2 AHB
  • Bottom: SRAM4 in D3 domain via AXI and D1-to-D3 AHB; not used by BF

  The two DMA blocks in D2 domain (DMA1 and DMA2) allow for fast transfers between memory and peripherals within that domain, and we see something similar for BDMA (basic DMA) in D3 domain. The MDMA controller also has access to blocks in D2 domain, but that's somewhat convoluted. So we do see why it make sense to deliberately place certain I/O buffers in D2 memory ("D2_RAM" or "RAM_D2" in the table above).

  Now we'd like to bring BF's memory areas over to the CubeIDE project. Turns out we can pretty much copy and paste them, as long as everything remains consistent within the linker script. A quick test reveals that the dev board still runs blinky after doing that. Great!

  Memory Sections

  The linker script further defines a number of memory sections. For plain old C we typically see

  • ".text" (instructions),
  • ".data" (initialized variables) and
  • ".bss" (uninitialized variables)

  But there's a lot more, because we need to place certain instructions in certain locations: reset handler and other ISRs need to be placed correctly in flash, and there has to be a section in ITCM that we can used to properly place functions that are supposed to be executed faster than others. Similarly, certain variables go into DTCM and I/O buffers into SRAM in D2 domain. And thus we end up with something like this:
  

  BF H743	CubeIDE H735
  .isr_vector >FLASH	.isr_vector >FLASH
  .text >FLASH1	.text >FLASH
  	.rodata >FLASH
  .tcm_code >ITCM_RAM AT >FLASH1
  .ARM.extab >FLASH1	.ARM.extab >FLASH
  .ARM >FLASH1	.ARM >FLASH
  .pg_registry >FLASH1
  .pg_resetdata >FLASH1
  	.preinit_array >FLASH
  	.init_array >FLASH
  	.fini_array >FLASH
  .data >RAM AT >FLASH1	.data >RAM_D1 AT >FLASH
  .bss >RAM	.bss > RAM_D1
  .sram2 >RAM
  .fastram_data >FASTRAM AT >FLASH1
  .fastram_bss >FASTRAM
  .dmaram_data >RAM AT >FLASH1
  .dmaram_bss >RAM
  .DMA_RAM >RAM
  .DMA_RW_D2 >D2_RAM
  .DMA_RW_AXI >RAM
  .persistent_data >RAM
  ._user_heap_stack >STACKRAM = 0xa5	._user_heap_stack >RAM_D1
  .memory_b1_text >MEMORY_B1

  Let's take a look at the similarities first: All "basic" sections have identical names (.isr_vector, .text,...

  Read more »

• Working Hardware

  Christoph • 06/02/2023 at 21:55 • 0 comments

  We're working with the "Chonker H735" board which was developed solely for the purpose of bringing up the BF target.

  Repo: https://github.com/crteensy/yolo-chonker

  Several of these were built, some with just the base necessities like

  • MCU and optional 8 MHz resonator
  • USB connector
  • 3V3 regulator
  • two LEDs on GPIOs

  Writing basic blinky code in STM32CubeIDE won't get us the desired files we need for the BF target, but it's easy and provides us with:

  • some valid linker script
  • some valid startup code
  • a valid (limited) clock tree setup

  The main() function then might look like this:

  int main(void)
  {
    HAL_Init();
    SystemClock_Config();
    MX_GPIO_Init();
    while (1)
    {
  	HAL_Delay(500);
  	HAL_GPIO_TogglePin(led_blue_A10_GPIO_Port, led_blue_A10_Pin);
    }
  }

  There's not a lot we could remove from this. Some things to note though:

  • HAL_Init() is called first,
  • even before SystemClock_Config(),
  • and then we initialize the actual peripherals.

  However, there's a lot of stuff happening before main() is even entered: that's the startup code which includes the reset handler. It's written in assembler, but the first thing it does after setting up the stack pointer is to call SystemInit(), which is written in C again:

  startup_stm32h735vghx.s:
  

  Reset_Handler:
    ldr   sp, =_estack      /* set stack pointer */
  
  /* Call the clock system initialization function.*/
    bl  SystemInit
  ...

   system_stm32h7xx.c:

  void SystemInit (void)
  {
  #if defined (DATA_IN_D2_SRAM)
   __IO uint32_t tmpreg;
  #endif /* DATA_IN_D2_SRAM */
  ...

  SystemInit() configures things like FPU, Flash timing, available oscillators (there are a few of them, both internal and external), resets the PLL, and configures the interrupt vector table.

  After that, we're back in the asm startup code (startup_stm32h735vghx.s) with a bunch of snippets like this one:

  /* Copy the data segment initializers from flash to SRAM */
    ldr r0, =_sdata
    ldr r1, =_edata
    ldr r2, =_sidata
    movs r3, #0
    b LoopCopyDataInit
  
  CopyDataInit:
    ldr r4, [r2, r3]
    str r4, [r0, r3]
    adds r3, r3, #4
  
  LoopCopyDataInit:
    adds r4, r0, r3
    cmp r4, r1
    bcc CopyDataInit
  

  _sdata, _edata and _sidata are defined in the linker script. The snippet shown above copies data from flash to RAM. This data is used to initialize variables that were defined an initialized somewhere in the code, for example:

  static uint8_t foo = 3;

  The initialization value can only be stored in flash to survive a reset, but the variable itself resides in RAM during runtime - so the init code has to copy the init value to that RAM location during startup to prepare everything for the code that later relies on that variable to be properly initialized.

  There are other sections for variables that are initialized to zero, or not initialized at all. What we need to know now is what sections are generated by STM32CubeIDE, so that we can compare those sections with the sections in a linker script used in a working BF target. 

  Each of these sections has to be placed in some memory area (like flash, or RAM, and some sub-areas of those). These must be known and defined.

  CubeIDE's generated linker script (STM32H735VGHX_FLASH.ld) creates these memory areas:

  MEMORY
  {
    ITCMRAM (xrw)    : ORIGIN = 0x00000000,   LENGTH = 64K
    DTCMRAM (xrw)    : ORIGIN = 0x20000000,   LENGTH = 128K
    FLASH    (rx)    : ORIGIN = 0x08000000,   LENGTH = 1024K
    RAM_D1  (xrw)    : ORIGIN = 0x24000000,   LENGTH = 320K
    RAM_D2  (xrw)    : ORIGIN = 0x30000000,   LENGTH = 32K
    RAM_D3  (xrw)    : ORIGIN = 0x38000000,   LENGTH = 16K
  }

  and these sections:

  .isr_vector
  .text
  .rodata
  .ARM.extab
  .ARM
  .preinit_array
  .init_array
  .fini_array
  .data
  .bss
  ._user_heap_stack
  

   As written above, _sdata, _edata and _sidata are defined in the linker script. Here's the part that does that:

    /* used by the startup to initialize data */
    _sidata = LOADADDR(.data);
  
    /* Initialized data sections goes into RAM, load LMA copy after code */
    .data :
    {
      . = ALIGN(4);
      _sdata = .;        /* create a global symbol at data start */
      *(.data)           /* .data sections */
      *(.data*)          /* .data* sections */
   *(.RamFunc) /* .RamFunc...

  Read more »

View all 2 project logs