GNU gcc 4.9.1: undefined symbol _end

Attilio Dona'

I'm porting code written from mspgcc 4.6.3 into a CCS 6.0.1 project with TI mspgcc 4.9.1.

MSP430FR59xx is the MCU type.

When linking the symbol _end (as I can read from the code it is the start of RAM + sizeof(data) + sizeof(bss) ) is undefined:

./cpu/msp430-common/msp430-main.o:(.data+0x0): undefined reference to `_end'
collect2.exe: error: ld returned 1 exit status

What symbol name instead of _end I have to use for gcc 4.9.1?

below the code snippet:

extern int _end;        /* Not in sys/unistd.h */
static char *cur_break = (char *) &_end;

#define asmv(arg) __asm__ __volatile__(arg)

#define STACK_EXTRA 32


/*
 * Allocate memory from the heap. Check that we don't collide with the
 * stack right now (some other routine might later). A watchdog might
 * be used to check if cur_break and the stack pointer meet during
 * runtime.
 */
void *sbrk(int incr)
{
    char *stack_pointer;

    asmv("mov r1, %0" : "=r"(stack_pointer));
    stack_pointer -= STACK_EXTRA;

    if (incr > (stack_pointer - cur_break)) {
        return (void *) - 1;    /* ENOMEM */
    }

    void *old_break = cur_break;
    cur_break += incr;
    /*
     * If the stack was never here then [old_break .. cur_break] should
     * be filled with zeros.
    */
    return old_break;
}

Greetings

Attilio

over 11 years ago

0 George Mock over 11 years ago

TI__Guru**** 245180 points

I cannot find where this detail is explained in the documentation. I think the symbol name is "end". That is because I looked at some executables with the names utility. The command could look like ...

% msp430-elf-nm -n executable_file | less

Search for "end". I expect you will only find "end" and never "_end". The -n option sorts the output by address, so you can see, based on the names of nearby symbols, whether the address makes sense for your system.

Thanks and regards,

-George

0 Attilio Dona' over 11 years ago in reply to George Mock

Intellectual 415 points

Thanks George!

It make sense and seems reasonable that the end symbol is the right address, but ...

for confirmation I made this simple case:

#include <msp430.h>

/*
 * main.c
 */
int main(void) {
    WDTCTL = WDTPW | WDTHOLD;	// Stop watchdog timer

    return 0;
}

then:

$ msp430-elf-size.exe e2e-post.out
   text    data     bss     dec     hex filename
    582     146      18     746     2ea e2e-post.out

So I expect end = 0x1c00 + 146 + 18 = 0x1ca4

but instead

$ msp430-elf-nm.exe -n e2e-post.out
00000012 A __bsssize
0000007e A __romdatacopysize
00001c00 D __datastart
00001c00 d __JCR_END__
00001c00 d __JCR_LIST__
00001c02 D __dso_handle
00001c04 D _impure_ptr
00001c06 d impure_data
00001c7e B __bssstart
00001c7e D _edata
00001c7e b completed.2761
00001c80 b dtor_idx.2763
00001c82 b object.2776
00001c90 N end

from nm outuput end is 0x1c90, not 0x1ca4.

What I miss?

regards

Attilio

0 George Mock over 11 years ago in reply to Attilio Dona'

TI__Guru**** 245180 points

That's because "end" marks the end of the .bss section. 0x1c7e + 0x12 = 0x1c90.

Thanks and regards,

-George

0 Archaeologist over 11 years ago in reply to Attilio Dona'

TI__Guru* 84225 points

The "data" column from the "size" program apparently includes more than what you're considering data. I played around with some GCC executables, and it seems that (_edata-__datastart) is the size of the .data section only; there are other data sections.

Try using the "-A" argument to msp430-elf-size.exe for a more detailed section listing.

Looking at the output of "nm", we see that (end-__bssstart) is 18, as expected, and (_edata-__datastart) is 126. This leaves 20 extra bytes unaccounted for in the "data" column of the "size" program. Possibly const data?

0 Archaeologist over 11 years ago in reply to Archaeologist

TI__Guru* 84225 points

It looks like the "data" column is the total size of all writable data sections including .data, but excepting .bss.

0 Attilio Dona' over 11 years ago in reply to Archaeologist

Intellectual 415 points

Ok, data section reported by msp430-elf-size is the sum of .data (RAM) and .rodata (ROM) data:

$ msp430-elf-size.exe -A e2e-post.out
e2e-post.out  :
section                     size    addr
__reset_vector                 2   65534
.rodata                       20   17408
.text                        580   17428
.data                        126    7168
.bss                          18    7294
.MP430.attributes            322       0
.comment                     123       0
.debug_aranges               400       0
.debug_info                 6233       0
.debug_abbrev               1149       0
.debug_line                 2370       0
.debug_frame                 156       0
.debug_str                  1823       0
.debug_loc                   563       0
.debug_ranges                120       0
.debug_line.text.__errno      22       0
.debug_line.text.memmove      64       0
.debug_line.text.memset       34       0
Total                      14125

Thanks again!

Attilio

Code Composer Studio™︎

Code Composer Studio forum

GNU gcc 4.9.1: undefined symbol _end