• gcc ld and bss sections on SDRAM

    From pozz@21:1/5 to All on Fri Mar 18 08:56:14 2022
    My platform is LPC546xx, Cortex-M4 by NXP, and the build environment is
    based on GCC ARM toolchain (MCUXpresso IDE). However the question is
    generic.

    The default linker script instructs linker to put bss sections (zero initialized data) in internal RAM. During startup, before main, bss
    sections, described in a table on Flash, are reset to zero:


    __attribute__ ((section(".after_vectors.init_bss")))
    void bss_init(unsigned int start, unsigned int len) {
    unsigned int *pulDest = (unsigned int*) start;
    unsigned int loop;
    for (loop = 0; loop < len; loop = loop + 4)
    *pulDest++ = 0;
    }

    extern unsigned int __bss_section_table;
    extern unsigned int __bss_section_table_end;

    void ResetISR(void) {
    ...
    // At this point, SectionTableAddr = &__bss_section_table;
    // Zero fill the bss segment
    while (SectionTableAddr < &__bss_section_table_end) {
    ExeAddr = *SectionTableAddr++;
    SectionLen = *SectionTableAddr++;
    bss_init(ExeAddr, SectionLen);
    }

    ...
    main();
    ...
    }


    This works well until I add a new RAM memory section on external bus (it
    is a SDRAM). Suppose I want to allocate a big variable on SDRAM. This
    can be done in different ways (I use some facilities of MCUXpresso IDE).

    Now the startup code is broken, because SDRAM memory can be accessed
    only after it is configured (during main), but the bss initialization is
    done before main.

    I know I can rewrite ResetISR() or bss_init() to avoid accessing
    external bus addresses or I can configure SDRAM during ResetISR() before calling bss_init().

    I suspect there's a better and more cleaner way. Noinit sections?

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From David Brown@21:1/5 to pozz on Fri Mar 18 11:15:11 2022
    On 18/03/2022 08:56, pozz wrote:
    My platform is LPC546xx, Cortex-M4 by NXP, and the build environment is
    based on GCC ARM toolchain (MCUXpresso IDE). However the question is
    generic.

    The default linker script instructs linker to put bss sections (zero initialized data) in internal RAM. During startup, before main, bss
    sections, described in a table on Flash, are reset to zero:


    __attribute__ ((section(".after_vectors.init_bss")))
    void bss_init(unsigned int start, unsigned int len) {
        unsigned int *pulDest = (unsigned int*) start;
        unsigned int loop;
        for (loop = 0; loop < len; loop = loop + 4)
            *pulDest++ = 0;
    }

    extern unsigned int __bss_section_table;
    extern unsigned int __bss_section_table_end;

    void ResetISR(void) {
        ...
        // At this point, SectionTableAddr = &__bss_section_table;
        // Zero fill the bss segment
        while (SectionTableAddr < &__bss_section_table_end) {
            ExeAddr = *SectionTableAddr++;
            SectionLen = *SectionTableAddr++;
            bss_init(ExeAddr, SectionLen);
        }

        ...
        main();
        ...
    }


    This works well until I add a new RAM memory section on external bus (it
    is a SDRAM). Suppose I want to allocate a big variable on SDRAM. This
    can be done in different ways (I use some facilities of MCUXpresso IDE).

    Now the startup code is broken, because SDRAM memory can be accessed
    only after it is configured (during main), but the bss initialization is
    done before main.

    I know I can rewrite ResetISR() or bss_init() to avoid accessing
    external bus addresses or I can configure SDRAM during ResetISR() before calling bss_init().

    I suspect there's a better and more cleaner way. Noinit sections?

    There are several ways to handle this. You can set up the sdram before clearing the bss - I think you'll find a weak function called
    SytemInitHook or similar, that you can override with your own version.
    If you are putting a lot of different stuff into the sdram, this is
    probably the best.

    You can fiddle with either the linker script or the ResetISR function to
    avoid clearing the sdram bss (and copying the sdram initialised data)
    during ResetISR, and do it manually after initialising the ram.

    Often, however, it is a good idea /not/ to use sdram for bss or data.
    Use the internal ram (DTC ram if possible, or OCRAM - depending on the particular chip) for that. Put your heap in sdram, along with other big buffers such as network buffers, FreeRTOS heap, etc. These can all be initialised later once the ram is configured. And you can mark any
    large statically allocated arrays or structures with a section attribute
    for a "noinit" section of the sdram.

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From pozz@21:1/5 to All on Fri Mar 18 11:49:58 2022
    Il 18/03/2022 11:15, David Brown ha scritto:
    On 18/03/2022 08:56, pozz wrote:
    My platform is LPC546xx, Cortex-M4 by NXP, and the build environment is
    based on GCC ARM toolchain (MCUXpresso IDE). However the question is
    generic.

    The default linker script instructs linker to put bss sections (zero
    initialized data) in internal RAM. During startup, before main, bss
    sections, described in a table on Flash, are reset to zero:


    __attribute__ ((section(".after_vectors.init_bss")))
    void bss_init(unsigned int start, unsigned int len) {
        unsigned int *pulDest = (unsigned int*) start;
        unsigned int loop;
        for (loop = 0; loop < len; loop = loop + 4)
            *pulDest++ = 0;
    }

    extern unsigned int __bss_section_table;
    extern unsigned int __bss_section_table_end;

    void ResetISR(void) {
        ...
        // At this point, SectionTableAddr = &__bss_section_table;
        // Zero fill the bss segment
        while (SectionTableAddr < &__bss_section_table_end) {
            ExeAddr = *SectionTableAddr++;
            SectionLen = *SectionTableAddr++;
            bss_init(ExeAddr, SectionLen);
        }

        ...
        main();
        ...
    }


    This works well until I add a new RAM memory section on external bus (it
    is a SDRAM). Suppose I want to allocate a big variable on SDRAM. This
    can be done in different ways (I use some facilities of MCUXpresso IDE).

    Now the startup code is broken, because SDRAM memory can be accessed
    only after it is configured (during main), but the bss initialization is
    done before main.

    I know I can rewrite ResetISR() or bss_init() to avoid accessing
    external bus addresses or I can configure SDRAM during ResetISR() before
    calling bss_init().

    I suspect there's a better and more cleaner way. Noinit sections?

    There are several ways to handle this. You can set up the sdram before clearing the bss - I think you'll find a weak function called
    SytemInitHook or similar, that you can override with your own version.
    If you are putting a lot of different stuff into the sdram, this is
    probably the best.

    ResetISR() is the first call after a reset.


    __attribute__ ((section(".after_vectors")))
    void
    ResetISR(void) {

    //
    // Copy the data sections from flash to SRAM.
    //
    unsigned int LoadAddr, ExeAddr, SectionLen;
    unsigned int *SectionTableAddr;

    // Load base address of Global Section Table
    SectionTableAddr = &__data_section_table;

    // Copy the data sections from flash to SRAM.
    while (SectionTableAddr < &__data_section_table_end) {
    LoadAddr = *SectionTableAddr++;
    ExeAddr = *SectionTableAddr++;
    SectionLen = *SectionTableAddr++;
    data_init(LoadAddr, ExeAddr, SectionLen);
    }
    // At this point, SectionTableAddr = &__bss_section_table;
    // Zero fill the bss segment
    while (SectionTableAddr < &__bss_section_table_end) {
    ExeAddr = *SectionTableAddr++;
    SectionLen = *SectionTableAddr++;
    bss_init(ExeAddr, SectionLen);
    }

    #if defined (__USE_CMSIS) || defined (__USE_LPCOPEN)
    SystemInit();
    #endif

    #if defined (__cplusplus)
    //
    // Call C++ library initialisation
    //
    __libc_init_array();
    #endif

    #if defined (__REDLIB__)
    // Call the Redlib library, which in turn calls main()
    __main() ;
    #else
    main();
    #endif

    //
    // main() shouldn't return, but if it does, we'll just enter an
    infinite loop
    //
    while (1) {
    ;
    }
    }


    There isn't any weak function call, anyway I can add my own.


    You can fiddle with either the linker script or the ResetISR function to avoid clearing the sdram bss (and copying the sdram initialised data)
    during ResetISR, and do it manually after initialising the ram.

    Yes, this is another possibility.

    You didn't mention NOLOAD (noinit). It seems to me another good approach
    for my problem.



    Often, however, it is a good idea /not/ to use sdram for bss or data.
    Use the internal ram (DTC ram if possible, or OCRAM - depending on the particular chip) for that. Put your heap in sdram, along with other big buffers such as network buffers, FreeRTOS heap, etc. These can all be initialised later once the ram is configured. And you can mark any
    large statically allocated arrays or structures with a section attribute
    for a "noinit" section of the sdram.

    I have some images saved in a SDCard and I have to show them on the
    display in some cases. I have plenty of SDRAM for other purposes
    (framebuffer, for example) so I thought, why read images from SDCard
    every time I have to show them?

    At startup, I copy images from SDCard to SDRAM and later I can access
    images directly from SDRAM.
    The maximum size of all images is 2MB and I have this free space on SDRAM.

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Stefan Reuther@21:1/5 to All on Fri Mar 18 17:57:26 2022
    Am 18.03.2022 um 08:56 schrieb pozz:
    I know I can rewrite ResetISR() or bss_init() to avoid accessing
    external bus addresses or I can configure SDRAM during ResetISR() before calling bss_init().

    I suspect there's a better and more cleaner way. Noinit sections?

    When I did bootloaders, I did the SDRAM initialisation there, before the application starts. This also means there's a nice place to deal with
    board varieties, i.e. if I have a board variant with a different SDRAM
    chip, I only have to change the bootloader. And it allows loading parts
    of the application code into SDRAM, if that is relevant for you.

    Otherwise, I'd probably use a noinit section.


    Stefan

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From pozz@21:1/5 to All on Sat Mar 19 16:05:35 2022
    Il 18/03/2022 17:57, Stefan Reuther ha scritto:
    Am 18.03.2022 um 08:56 schrieb pozz:
    I know I can rewrite ResetISR() or bss_init() to avoid accessing
    external bus addresses or I can configure SDRAM during ResetISR() before
    calling bss_init().

    I suspect there's a better and more cleaner way. Noinit sections?

    When I did bootloaders, I did the SDRAM initialisation there, before the application starts. This also means there's a nice place to deal with
    board varieties, i.e. if I have a board variant with a different SDRAM
    chip, I only have to change the bootloader. And it allows loading parts
    of the application code into SDRAM, if that is relevant for you.

    Otherwise, I'd probably use a noinit section.

    Yes, you're right, bootloader is another good place to configure hw and
    next launch application.

    Thank you for your suggestion.

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)