gas' nop(s)

gas' nop(s)

[m_hrebien]

Hi there,

did You noticed that GNU as adds some padding nop(s) like:

lea (%esi),%esi
mov %esi,%esi
nop

at the end of .text section? Is this really necessary? The .text section
mustn't be padded to 4 like it seems to be or am i wrong? How can i
avoid this? Is there any gas option to switch this off?

Sorry if this topic is doubled by me but i couldn't find something
similar in the archive :(

-- 
Maciej Hrebien

Re: gas' nop(s)

[m_hrebien]

Jani Monoses wrote:
> 
> Hi
> look at the .align directive in info gas

.align, oh, of course!! :) Thank You!!

> there is no requirement for text to be packed at 4

I thought so. Thanks for Your confirmation!

> but it is the default (or maybe even 16)

16? Possibly it depends on system/compilation, don't You think so? On my
gas it seems to be 4 as i wrote.

-- 
Maciej Hrebien

Re: gas' nop(s)

[h-peter recktenwald]

On Sun, 04 Aug 2002 22:28:47 +0200
m_hrebien@wp.pl wrote:

> Hi there,
> 
> did You noticed that GNU as adds some padding nop(s) like:
> 
> lea (%esi),%esi
> mov %esi,%esi
> nop
> 
> at the end of .text section? Is this really necessary? The .text section
> mustn't be padded to 4 like it seems to be or am i wrong? How can i
> avoid this? Is there any gas option to switch this off?

some time back, i tried AS as an intermediate program, w. piped i/o, and 
encountered the same problem, which i didn't manage to solve, too. there 
is no option + no other idea. those gods of gnu binutils devel might know 
but, i myself never received an answer to any AS related question...

anyway, AS wsn't meant to do plain, stand alone assembly but is designed 
to produceing a set of data which can be combined to something, probably 
executeable, with the "ld" linker. so the implicite alignment might even 
be very helpful for faster processing.

i'd (highly) recommend the "fasm" assembler for the more 'direct' work:)
	http://fasm.metro-nt.pl/

> Sorry if this topic is doubled by me but i couldn't find something
> similar in the archive :(
> 
> -- 
> Maciej Hrebien
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-assembly" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html


-- 
Linux,Assembly,Forth: http://www.lxhp.in-berlin.de/index-lx.shtml en/de
NO abusive software patents http://petition.eurolinux.org/pr/pr17.html
-
To unsubscribe from this list: send the line "unsubscribe linux-assembly" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: gas' nop(s)

[m_hrebien]

h-peter recktenwald wrote:
> 
> > Hi there,
> >
> > did You noticed that GNU as adds some padding nop(s) like:
> >
> > lea (%esi),%esi
> > mov %esi,%esi
> > nop
> >
> > at the end of .text section? Is this really necessary? The .text section
> > mustn't be padded to 4 like it seems to be or am i wrong? How can i
> > avoid this? Is there any gas option to switch this off?
> 
> some time back, i tried AS as an intermediate program, w. piped i/o, and
> encountered the same problem, which i didn't manage to solve, too. there
> is no option + no other idea. those gods of gnu binutils devel might know
> but, i myself never received an answer to any AS related question...
>
> anyway, AS wsn't meant to do plain, stand alone assembly but is designed
> to produceing a set of data which can be combined to something, probably
> executeable, with the "ld" linker.

Yes, it's gcc's back-end rather then stand alone assembler.

> so the implicite alignment might even
> be very helpful for faster processing.
>
> i'd (highly) recommend the "fasm" assembler for the more 'direct' work:)
>         http://fasm.metro-nt.pl/

:) Thanks, i'll check it...
 
> > Sorry if this topic is doubled by me but i couldn't find something
> > similar in the archive :(

Guess what? I've tried this .align directive as Jani Monoses suggested
like this:

.text

.align 1

.globl _start
_start:
	mov	$1,%eax
	xor	%ebx,%ebx
	int	$0x80

and it's not working as i wish - gas adds lea (%esi),%esi after int
$0x80 to the output :( I've tried with 2, 4, 8 & 16 and it seems that
.align x works good if x is greater equal than 4 :( But why??

-- 
Maciej Hrebien

Re: gas' nop(s)

[h-peter recktenwald]

On Mon, 05 Aug 2002 19:53:41 +0200
m_hrebien@wp.pl wrote:
 
> Guess what? I've tried this .align directive as Jani Monoses suggested
> like this:
> 
> .text
> 
> .align 1
> 
> .globl _start
> _start:
> 	mov	$1,%eax
> 	xor	%ebx,%ebx
> 	int	$0x80
> 
> and it's not working as i wish - gas adds lea (%esi),%esi after int
> $0x80 to the output :( I've tried with 2, 4, 8 & 16 and it seems that
> .align x works good if x is greater equal than 4 :( But why??

.text section for ELF format defined to be page aligned, so it's always a 
multiple of 4, and the end-alignment added for what "bu" gnu-ers seem to 
think is #the# appropriate thing or, just more convenient to further asm 
output processing - just guessing, has anyone else a more thorough idea?

a work-around for piped output may be defining a label at code end and 
taking the actually assmbled code length from the listing file. tested:

echo ".att_syntax noprefix;.global XLAB;jnc XLAB;movl eax,ebx;\
      movl ebx,edx;XLAB:;.end"|as -achnlsd m=l --gstabs |tee t

the listing:
----------------------------------------------------
   1 0000 0F83FCFF      .att_syntax noprefix
   1      FFFF89C3
   1      89DA89F6
DEFINED SYMBOLS
    {standard input}:1      .text:0000000a XLAB

NO UNDEFINED SYMBOLS
--------------------------------------------------
asm 4 bytes shorter w/o the ".global" statement, <jnc> then w. byte disp.

regards,
	hp
 
-- 
Linux,Assembly,Forth: http://www.lxhp.in-berlin.de/index-lx.shtml en/de
NO abusive software patents http://petition.eurolinux.org/pr/pr17.html
-
To unsubscribe from this list: send the line "unsubscribe linux-assembly" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: gas' nop(s)

[m_hrebien]

h-peter recktenwald wrote:
> 
> > .text
> >
> > .align 1
> >
> > .globl _start
> > _start:
> >       mov     $1,%eax
> >       xor     %ebx,%ebx
> >       int     $0x80
> >
> 
> echo ".att_syntax noprefix;.global XLAB;jnc XLAB;movl eax,ebx;\
>       movl ebx,edx;XLAB:;.end"|as -achnlsd m=l --gstabs |tee t
> 
> the listing:
> ----------------------------------------------------
>    1 0000 0F83FCFF      .att_syntax noprefix
>    1      FFFF89C3
>    1      89DA89F6
> DEFINED SYMBOLS
>     {standard input}:1      .text:0000000a XLAB
> 
> NO UNDEFINED SYMBOLS
> --------------------------------------------------
> asm 4 bytes shorter w/o the ".global" statement, <jnc> then w. byte disp.

Yes, right! But it still adds "89F6" padding which stands for "mov
%esi,%esi" & my question was how to avoid this.

-- 
Maciej Hrebien

Re: gas' nop(s)

[h-peter recktenwald]

On Tue, 06 Aug 2002 11:30:56 +0200
m_hrebien@wp.pl wrote:

> > asm 4 bytes shorter w/o the ".global" statement, <jnc> then w. byte disp.
> 
> Yes, right! But it still adds "89F6" padding which stands for "mov
> %esi,%esi" & my question was how to avoid this.

ok. you can not.

hp

Re: gas' nop(s)

[h-peter recktenwald]

On Wed, 7 Aug 2002 07:41:36 +0000
h-peter recktenwald <lx@lxhp.in-berlin.de> wrote:

> On Tue, 06 Aug 2002 11:30:56 +0200
> m_hrebien@wp.pl wrote:
> 
> > > asm 4 bytes shorter w/o the ".global" statement, <jnc> then w. byte disp.
> > 
> > Yes, right! But it still adds "89F6" padding which stands for "mov
> > %esi,%esi" & my question was how to avoid this.
> 
> ok. you can not.

just tried: apparently, the ".end" directive inhibits those alignemt bytes, 
(if not a label at end of code).


hp

-- 
Linux,Assembly,Forth: http://www.lxhp.in-berlin.de/index-lx.shtml en/de
NO abusive software patents http://petition.eurolinux.org/pr/pr17.html
-
To unsubscribe from this list: send the line "unsubscribe linux-assembly" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: gas' nop(s)

[h-peter recktenwald]

On Thu, 8 Aug 2002 15:50:47 +0000
h-peter recktenwald <*@lxhp.in-berlin.de> wrote:

> > > 
> > > Yes, right! But it still adds "89F6" padding which stands for "mov
> > > %esi,%esi" & my question was how to avoid this.
> 
> just tried: apparently, the ".end" directive inhibits those alignemt bytes, 
> (if not a label at end of code).

that was an error. though i wonder how some program of mine ends up 
with an odd address. probably due to its .data end. so i tried asm 
into a .data section, which works and, does #not# align. so, since you 
seem not to asm to any fixed 'format', that might be the alternative.

hp

-- 
Linux,Assembly,Forth: http://www.lxhp.in-berlin.de/index-lx.shtml en/de
NO abusive software patents http://petition.eurolinux.org/pr/pr17.html
-
To unsubscribe from this list: send the line "unsubscribe linux-assembly" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: gas' nop(s)

[Maciej Hrebien]

h-peter recktenwald wrote:
> 
> > > > Yes, right! But it still adds "89F6" padding which stands for "mov
> > > > %esi,%esi" & my question was how to avoid this.
> >
> > just tried: apparently, the ".end" directive inhibits those alignemt bytes,
> > (if not a label at end of code).
> 
> that was an error. though i wonder how some program of mine ends up
> with an odd address. probably due to its .data end. so i tried asm
> into a .data section, which works and, does #not# align. so, since you
> seem not to asm to any fixed 'format', that might be the alternative.

Good idea!! I've tried to move our example (this with simple exit call)
to the .data section but it still adds padding - it's: add %al,(%eax)
(coded as "00 00") on my gas even with .align directive set to 1. But!
I've followed Your way of thinking and guess what?! :) "why not to move
this code to the .rodata section?" and... it works just fine!! No
paddings!! :)

Thank You very much for Your suggestion Peter (is that Your name?)!!!

-- 
Maciej Hrebien