Skip to contentSkip to conversation

Some puzzles about u-boot.lds

U-Boot Peter Chen 1 participant 1 message Jul 11, 2009

lore
Share & tools
lore

Peter Chen

author
#0 lore raw

Dear lists,

I met some problems about integrate our bootloader to u-boot, my working CPU is ARM1136EJ-S.
Our boot routine likes below:

1. Rom code, and copy the first 4KB u-boot to ITCM (0x70000000), switch to u-boot. ( cpu/arm1136/start.S)

2. u-boot running at ITCM, and config PLL, and memory, and switch to romcode. (board/xxx/lowlevel_init.S)

3. If it is cold boot, copy the whole u-boot to Ram (the address ram address is 0xc1000000)

4. Back from the rom code, and running first part of u-boot at ITCM. (board/xxx/lowlevel_init.S, and cpu/arm1136/start.S)

5. After finishing setup stack and clear bss, jump to the start_armboot (cpu/arm1136/start.S), the VMA of start_armboot is ram address (0xc1000xxx)

I would like the first part is at 0x70000000( both VMA and LMA), and the second part is at 0xc1000000 + sizeof(first boot), so I design the u-boot.lds like
below:

OUTPUT_FORMAT("elf32-littlearm", "elf32-littlearm", "elf32-littlearm") OUTPUT_ARCH(arm)
ENTRY(_start)
SECTIONS

{
          
        . = ALIGN(4);
        . = 0x70000000; 
            
        .textboot :
        {
          cpu/arm1136/start.o   (.text)
          board/atlas5cb/lowlevel_init.o        (.text)
        }   

        . = ALIGN(4);
        _textram_start = 0xC1000000 + SIZEOF(.textboot);
            
        .textram 0xC1000000 + SIZEOF(.textboot):
        AT ( _textram_start )
        {
          *(.text)
        }   

        . = ALIGN(4);
        .rodata : { *(.rodata) }
        .ARM.extab   : { *(.ARM.extab* .gnu.linkonce.armextab.*) }
        __exidx_start = .;
        .ARM.exidx   : { *(.ARM.exidx* .gnu.linkonce.armexidx.*) }
        __exidx_end = .;
       . = .;
        __u_boot_cmd_start = .;
        .u_boot_cmd : { *(.u_boot_cmd) }
        __u_boot_cmd_end = .;

        . = ALIGN(4);
        __bss_start = .;
        .bss : { *(.bss) }
        _end = .;

}

But it can't finish the link, it stops at:
arm-none-linux-gnueabi-objcopy --gap-fill=0xff -O
srec /home/nchen/work/L27EVB/Dev/bootloader/uboot/u-boot /home/nchen/work/L27EVB/Dev/out/bootloader/uboot/u-boot.srec

Why I can't set two LMA for two different text sections?

It can work after I change the second text section's LMA just near to the first one

     ...        
    . = 0x70000000;

        .textboot :
        {
          cpu/arm1136/start.o   (.text)
          board/xxx/lowlevel_init.o        (.text)
        }

        . = ALIGN(4);
        .textram 0xC1000000 + SIZEOF(.textboot):
      
        AT ( ADDR(.textboot) + SIZEOF(.textboot) )
        {
          *(.text)
        }

After that my output layout like to be:

Idx Name          Size      VMA       LMA       File off  Algn
  0 .textboot     00000950  70000000  70000000  00008000  2**5

CONTENTS, ALLOC, LOAD, READONLY, CODE
1 .textram 000130e0 c1000950 70000950 00008950 2**5
CONTENTS, ALLOC, LOAD, READONLY, CODE
2 .rodata 00001858 c1013a30 70013a30 0001ba30 2**2
CONTENTS, ALLOC, LOAD, READONLY, DATA
3 .rodata.str1.1 00003442 c1015288 70015288 0001d288 2**0
CONTENTS, ALLOC, LOAD, READONLY, DATA
4 .data 000407a4 c10186cc 700186cc 000206cc 2**2
CONTENTS, ALLOC, LOAD, DATA
5 .u_boot_cmd 00000348 c1058e70 70058e70 00060e70 2**2
CONTENTS, ALLOC, LOAD, DATA
6 .bss 00004cb4 c10591b8 700591b8 000611b8 2**2
ALLOC

It should be ok If I set pc to VMA of start_armboot, as all parts of u-boot is at correct ram regions now, although I did not do relocate. After executing:
ldr pc, [r0] /* r0 stores VMA of start_armboot)
The pc goes to the undefined address, but I can set pc to start_armboot using realview debugger.
Do you have any clues for such problems, Is it possible that my u-boot.lds is not correct?

Although I can set pc to start_armboot, and let it run, but the system's behavior is incorrect, and can't output anything.

By the way, the old structure ( our bootloader + u-boot + Linux ) is work well, so the second part of u-boot should be ok.

Thank you

Best Regards,
Peter Chen