How to inform the linker not to produce any data for a .bss section?

How to inform the linker not to produce any data for a .bss section?

[PRC]

Here is my link script:

--------------------------------------------------------------------
OUTPUT_ARCH(mips)

SECTIONS
{
	.text 0x811ed000  :
	{
		_dstart = ABSOLUTE(.) ;	/* Start of code and data	*/
		*(.libc)
		*(.text*)
		*(.rodata*)
	}

	.data :
	{
		*(.data)
	}

	.sbss :
 	{
		*(.sbss)
		*(.scommon)
	}

 	.bss :
	{
 	 	*(.bss)
  	 	*(COMMON)
	}
 	
 	_dend = ABSOLUTE(.) ;	/* end of code and data	*/
}
--------------------------------------------------------------------
and mips-elf-objdump -h a.out shows:

a.out:     file format elf32-littlemips

Sections:
Idx Name          Size      VMA       LMA       File off  Algn
  0 .text         00001ab4  811ed000  811ed000  00001000  2**2
                  CONTENTS, ALLOC, LOAD, READONLY, CODE
  1 .data         0000004c  811eeab4  811eeab4  00002ab4  2**2
                  CONTENTS, ALLOC, LOAD, DATA
  2 .bss          00000620  811eeb00  811eeb00  00002b00  2**3
                  ALLOC
  3 .reginfo      00000018  00000000  00000000  00002b00  2**2
                  CONTENTS, READONLY, LINK_ONCE_SAME_SIZE
  4 .pdr          000003c0  00000000  00000000  00002b18  2**2
                  CONTENTS, READONLY
  5 .mdebug.abi32 00000000  00000000  00000000  00002ed8  2**0
                  CONTENTS, READONLY

The linker create a real '.bss' section in the output ELF file and
store some data inside the section, which greatly increase the size
of the ELF file. 

How can I do to inform the linker don't produce any data for the .bss section?

My linker is:
mips-elf-ld -v
GNU ld version 2.15

BTW, I'd like to know how to get a minimal ELF file. As you see, the offset of the
first section is 0x1000, which means the size of the ELF header is 4K in bytes. 
Is there a way to cut off the size of the header?

Re: How to inform the linker not to produce any data for a .bss section?

[Daniel Jacobowitz]

On Sun, Dec 09, 2007 at 10:50:50AM +0800, PRC wrote:
>   2 .bss          00000620  811eeb00  811eeb00  00002b00  2**3
>                   ALLOC

> The linker create a real '.bss' section in the output ELF file and
> store some data inside the section, which greatly increase the size
> of the ELF file. 

No, it didn't.  It takes up no space in the file because it does not
have the CONTENTS flag.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Re: How to inform the linker not to produce any data for a .bsssection?

[PRC]

There is a global static array in my code:
static char cmd_buffer[1024*64] __attribute__(( section(".mydata") ));

And inform the linker to put it in section ".bss":
.bss :
	{
 	 	*(.bss)
  	 	*(COMMON)
  	 	*(.mydata)
	}

mips-elf-objdump -h a.out 
-------------------------------------------------------------------
  3 .bss          00010620  81413898  81413898  00003898  2**3
                  ALLOC
  4 .reginfo      00000018  00000000  00000000  00013eb8  2**2
                  CONTENTS, READONLY, LINK_ONCE_SAME_SIZE
------------------------------------------------------------------
The difference between sections ".bss" and ".reginfo" is almost 64K,
which is just the size of the buffer `cmd_buffer`. And I check the
output file, and find there are lots of zero starting from the offset
0x3898. It seems the linker puts 64K zeros in the section ".bss".


 
On Sun, Dec 09, 2007 at 10:50:50AM +0800, PRC wrote:
>   2 .bss          00000620  811eeb00  811eeb00  00002b00  2**3
>                   ALLOC

> The linker create a real '.bss' section in the output ELF file and
> store some data inside the section, which greatly increase the size
> of the ELF file. 

No, it didn't.  It takes up no space in the file because it does not
have the CONTENTS flag.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Re: How to inform the linker not to produce any data for a .bsssection?

[Daniel Jacobowitz]

On Tue, Dec 11, 2007 at 09:19:12AM +0800, PRC wrote:
> mips-elf-objdump -h a.out 
> -------------------------------------------------------------------
>   3 .bss          00010620  81413898  81413898  00003898  2**3
>                   ALLOC
>   4 .reginfo      00000018  00000000  00000000  00013eb8  2**2
>                   CONTENTS, READONLY, LINK_ONCE_SAME_SIZE
> ------------------------------------------------------------------
> The difference between sections ".bss" and ".reginfo" is almost 64K,
> which is just the size of the buffer `cmd_buffer`. And I check the
> output file, and find there are lots of zero starting from the offset
> 0x3898. It seems the linker puts 64K zeros in the section ".bss".

This is necessary alignment.  If you don't want .bss to take up space
in the file, do not put allocated sections after it.

-- 
Daniel Jacobowitz
CodeSourcery

Re: How to inform the linker not to produce any data for a .bsssection?

[Dave Murphy]

Daniel Jacobowitz wrote:
> This is necessary alignment.  If you don't want .bss to take up space
> in the file, do not put allocated sections after it.
>   

Is there any way to change that behavior without causing problems?

I'm trying to regain a little extra memory for the heap on a newlib 
based toolchain where the device binary starts in RAM. It's an arm946 
with 48k of tcm.

I've tried placing .itcm & .dtcm sections after the .bss section which 
causes it to consume space in the binary so I tried setting the VMA for 
the bss to the LMA for the first preceeding section, like this

    __dtcm_lma = . ;

    .dtcm __dtcm_start : AT (__dtcm_lma)
    {
        *(.dtcm)
        *(.dtcm.*)
        . = ALIGN(4);
        __dtcm_end = ABSOLUTE(.);
    } >dtcm = 0xff


    __itcm_lma = __dtcm_lma + SIZEOF(.dtcm);

    .itcm __itcm_start : AT (__itcm_lma)
    {
        *(.itcm)
        *itcm.*(.text)
        . = ALIGN(4);
        __itcm_end = ABSOLUTE(.);
    } >itcm = 0xff
   
    .sbss __dtcm_end :
    {
        __sbss_start = ABSOLUTE(.);
        __sbss_start__ = ABSOLUTE(.);
        *(.sbss)
        . = ALIGN(4);    /* REQUIRED. LD is flaky without it. */
        __sbss_end = ABSOLUTE(.);
    } >dtcm
   

   
    __bss_lma = __itcm_lma + SIZEOF(.itcm) ;
    __appended_data = __itcm_lma + SIZEOF(.itcm) ;
    .bss __bss_lma : AT (__dtcm_lma)
    {
        __bss_start = ABSOLUTE(.);
        __bss_start__ = ABSOLUTE(.);
        *(.dynbss)
        *(.gnu.linkonce.b*)
        *(.bss*)
        *(COMMON)
        . = ALIGN(4);    /* REQUIRED. LD is flaky without it. */
        __bss_end = ABSOLUTE(.) ;
        __bss_end__ = __bss_end ;
    } >ewram


    _end = . ;
    __end__ = . ;
    PROVIDE (end = _end);

ld errored with "section .bss [0200c5b8 -> 0200c647] overlaps section 
.itcm [0200c5b8 -> 0200c747]" so I tried placing .bss first & setting 
the LMA of the following section to the bss segment. That got me the 
opposite error.

    __dtcm_lma = . ;
  __bss_start = .;
  __bss_start__ = .;
  .bss            :
  {
   *(.dynbss)
   *(.bss .bss.* .gnu.linkonce.b.*)
   *(COMMON)
   /* Align here to ensure that the .bss section occupies space up to
      _end.  Align after .bss to ensure correct alignment even if the
      .bss section disappears because there are no input sections.
      FIXME: Why do we need it? When there is no .bss section, we don't
      pad the .data section.  */
   . = ALIGN(. != 0 ? 32 / 8 : 1);
  } >ewram
  __bss_end = . ; __bss_end__ = . ;
    _end = . ;
    __end__ = . ;


    .dtcm __dtcm_start : AT (__dtcm_lma)
    {
        *(.dtcm)
        *(.dtcm.*)
        . = ALIGN(4);
        __dtcm_end = ABSOLUTE(.);
    } >dtcm


    __itcm_lma = __dtcm_lma + SIZEOF(.dtcm);

    .itcm __itcm_start : AT (__itcm_lma)
    {
        *(.itcm)
        *itcm.*(.text)
        . = ALIGN(4);
        __itcm_end = ABSOLUTE(.);
    } >itcm

Dave

Re: How to inform the linker not to produce any data for a .bsssection?

[Daniel Jacobowitz]

On Wed, Dec 12, 2007 at 09:08:15AM +0000, Dave Murphy wrote:
> Daniel Jacobowitz wrote:
> > This is necessary alignment.  If you don't want .bss to take up space
> > in the file, do not put allocated sections after it.
> >   
> 
> Is there any way to change that behavior without causing problems?

Try manually defining PHDRs?  If the bss section is the last thing in
its segment, it won't take extra load space.


-- 
Daniel Jacobowitz
CodeSourcery

Re: Re: Re: How to inform the linker not to produce any data for a.bsssection?

[PRC]

Hi, Jacobowitz

Can you tell me how to modify the link script to prevent .bss taking up space in file?
I didn't grap your meanings and have no idea what to do.


 
On Tue, Dec 11, 2007 at 09:19:12AM +0800, PRC wrote:
> mips-elf-objdump -h a.out 
> -------------------------------------------------------------------
>   3 .bss          00010620  81413898  81413898  00003898  2**3
>                   ALLOC
>   4 .reginfo      00000018  00000000  00000000  00013eb8  2**2
>                   CONTENTS, READONLY, LINK_ONCE_SAME_SIZE
> ------------------------------------------------------------------
> The difference between sections ".bss" and ".reginfo" is almost 64K,
> which is just the size of the buffer `cmd_buffer`. And I check the
> output file, and find there are lots of zero starting from the offset
> 0x3898. It seems the linker puts 64K zeros in the section ".bss".

This is necessary alignment.  If you don't want .bss to take up space
in the file, do not put allocated sections after it.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Re: Re: How to inform the linker not to produce any data for a.bsssection?

[Ivan Pulleyn]

On Dec 18, 2007 2:18 AM, PRC <panruochen@gmail.com> wrote:
> Hi, Jacobowitz
>
> Can you tell me how to modify the link script to prevent .bss taking up space in file?
> I didn't grap your meanings and have no idea what to do.

If you define something at the end of the BSS, the linker is forced to
create it. I believe that if you (a) make sure BSS is defined at the
end of your linker script and (b) make sure nothing is inside that
section (get rid of the mydata thing) then BSS will not take up any
space on disk.

Ivan...