Is That what Alignment is?

Index > Main > Is That what Alignment is?

Author

Thread

Nameless

Joined: 30 Apr 2010
Posts: 95

Nameless

i was readying about Alignment (http://www.asmcommunity.net/book/Alignment)
thats how i understand ALIGN command
is it correct?
it starts the next declared data from the next address of memory thats divisible by 4
is that correct?

Image Attached

Description:
Filesize:	6.87 KB
Viewed:	4294 Time(s)

02 Oct 2010, 08:27

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 17935
Location: In your JS exploiting you and your system

revolution

Nameless wrote:

it starts the next declared data from the next address of memory thats divisible by 4

Yes

02 Oct 2010, 09:20

Nameless

Joined: 30 Apr 2010
Posts: 95

Nameless

thanks

my learning is being accelerated, as i go further things gets kinda easier Very Happy

02 Oct 2010, 12:28

Tomasz Grysztar

Joined: 16 Jun 2003
Posts: 7861
Location: Kraków, Poland

Tomasz Grysztar

One detail: after "g" there should be a "0", because you have byte of such value defined at the end of your string. So it really should be "0" and two "x" instead of three "x".

02 Oct 2010, 13:13

edfed

Joined: 20 Feb 2006
Posts: 4250
Location: Now

edfed

align fills the empty bytes with NOP (90h), not X, but really nops.
because align can be used inside code to align huge instructions, it can speed up the code to do so, because nops are frequentlly ignored in recent X86.

a formula for align can be that:

Code:

Align:
;input:
;eax = boundary parameter = 2^N
;es:edi = current offset before align
;return:
;eax = count of bytes 'NOPed'
;es:edi = current offset after align
push ecx edx edi ;save trashed registers
xor ecx,ecx ;ecx=0
mov cx,es ;ecx=es
shl ecx,4 ;ecx*16
add edi,ecx ;edi = linear adress
xchg eax,edi ;eax=linear adress, edi=bound param
xor edx,edx ;expand eax to unsigned qword for div
div edi ;offset/bound param
mov ecx,edx ;ecx=modulus = count of bytes to nop
mov al,90h ;al=opcode for nop
pop edi ;restore offset
rep stosb ;nop bytes
mov eax,edx ;eax=count of bytes noped
pop edx ecx ;restore trashed registers
ret ;quit

02 Oct 2010, 14:06

bitRAKE

Joined: 21 Jul 2003
Posts: 3076
Location: vpcmipstrm

bitRAKE

We also see variations of ALIGN. Sometimes it is desired to have other data fill the gap. Data might compress better, or be more secure after encryption, etc...

Code:

db $ and 1 dup 0 ; same as align 2

FASM's default behavior is to put NOPs in the code sections, and ? (0) in the data sections.

Code:

dw (512-$+$$)/2 dup $AA55 ; pattern fill

Parallelism within the CPU makes alignment important. There is even an alignment exception which can be turned on for Ring-3 applications. Not to mention instructions which only work on aligned data!

Code:

times (4-$) and 3 db $FF

At the source code level it is important to remember that alignment can only be guaranteed at a granularity of the present section. This is mostly important with object files. If the section is only aligned on a 16 byte boundary then it is impossible to get cacheline alignment at assemble-time (of course, it could be done programmatically at run-time).

There is much to be said about run-time alignment, but I'll leave that for another post... Wink

03 Oct 2010, 01:32

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 17935
Location: In your JS exploiting you and your system

revolution

bitRAKE wrote:

FASM's default behavior is to put NOPs in the code sections, and ? (0) in the data sections.

Nope, NOPs everywhere.

03 Oct 2010, 01:53

mindcooler

Joined: 01 Dec 2009
Posts: 423
Location: Västerås, Sweden

mindcooler

90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 ...

03 Oct 2010, 02:24

bitRAKE

Joined: 21 Jul 2003
Posts: 3076
Location: vpcmipstrm

bitRAKE

revolution wrote:

bitRAKE wrote:
FASM's default behavior is to put NOPs in the code sections, and ? (0) in the data sections.
Nope, NOPs everywhere.

Did that change, or am I imagining things again? Embarassed

(Maybe, there was a discussion regarding this?)

Code:

macro ALIGN0 n {
  ; assuming (n) is a power of two
  rb (n-$) and (n-1)
}

_________________
¯\(°_o)/¯ unlicense.org

03 Oct 2010, 02:42

mindcooler

Joined: 01 Dec 2009
Posts: 423
Location: Västerås, Sweden

mindcooler

All I ever saw was 90 90 90 90 everywhere.

But I do not use sections.

03 Oct 2010, 03:04

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 17935
Location: In your JS exploiting you and your system

revolution

bitRAKE wrote:

Did that change, ...

No change that I am aware of.

The alignment code does not know what section attributes the align is assembled within. Besides, not all file formats have separate data and code sections. Not all file formats have sections.

Anyhow, it is easy to test what really happens.

03 Oct 2010, 03:24

bitRAKE

Joined: 21 Jul 2003
Posts: 3076
Location: vpcmipstrm

bitRAKE

I just did,

Code:

rb 3
align 16
rb 123
align 16
rb 256

...and was surprised to find data produced anywhere I put it. Luckily, this is easy to resolve. I must have stumbled on this before.

03 Oct 2010, 04:10

Tomasz Grysztar

Joined: 16 Jun 2003
Posts: 7861
Location: Kraków, Poland

Tomasz Grysztar

fasm's ALIGN generates bytes that have value 90h and are marked as uninitialized at the same time.

Thus if they occur in the data that is not included in the output file (like large uninitialized block at the end of PE section, or any uninitialized data at the end of binary format file) they are not included in output either; however when they happen to be in the output, they always have value 90h each.

03 Oct 2010, 10:05

< Last Thread | Next Thread >

Forum Rules:

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You cannot attach files in this forum
You can download files in this forum