.hash section oddity

.hash section oddity

[Smith, Stephen (SwCOE)]

I have a situation when a component is compiled on several different host platforms the binary is different when it is not expected.  

In all cases the host is a Cygwin box (with the same Cygwin versions and the cross compiler is for powerpc-motorol-elf or i686-pc-elf plaforms). The binutils package is version 2.17 (we upgraded from 2.10 to 2.17 so that we could create debug link files).  In this case the executable (libansi.so) points at libansi.so.dbg which is what has the symbolic debugging information.

With that as the background I took one of the builds, disassembled (using the -D switch) one of the binaries in question, and diff'ed it and got:

$ diff -u ppc-libansi-300.txt ppc-libansi-3873.txt
--- ppc-libansi-300.txt 2007-09-05 10:45:48.871022500 -0700
+++ ppc-libansi-3873.txt        2007-09-05 10:44:53.152272500 -0700
@@ -1,5 +1,5 @@
   
-../../ansi-3_0_0/ppc/appbin/dbg/libansi.so:  file format elf32-powerpc
+ppc/appbin/dbg/libansi.so:     file format elf32-powerpc

Disassembly of section .hash:
  
@@ -13354,4 +13354,4 @@
    4:  6e 73 69 2e     xoris   r19,r19,26926
    8:  73 6f 2e 64     andi.   r15,r27,11876
    c:  62 67 00 6e     ori     r7,r19,110
-  10:  2a af b8 77     cmpldi  cr5,r15,47223
+  10:  13 10 72 26     vmsumuhm v24,v16,v14,v8

This told me that a DWORD is different and that the variable is aligned on a DWORD boundary.

Next I used the "od -a" command on both binaries, compared the output and found:

--- ppc-libansi-300-od.txt      2007-09-05 10:56:13.699147500 -0700
+++ ppc-libansi-3873-od.txt     2007-09-05 10:56:55.949147500 -0700
@@ -3230,11 +3230,11 @@
 0145200   l   ) nul nul   G   C   C   :  sp   (   G   N   U   )  sp   2
 0145220   .   9   5   .   2   -   6  sp   1   9   9   9   1   0   2   4
 0145240  sp   (   c   y   g   w   i   n  sp   e   x   p   e   r   i   m
 0145260   e   n   t   a   l   ) nul nul   G   N   U  sp   C  sp   d   c
 0145300   r   t   n   .   S nul   l   i   b   a   n   s   i   .   s   o
-0145320   .   d   b   g nul   n   *   /   8   w nul   .   s   y   m   t
+0145320   .   d   b   g nul   n dc3 dle   r   & nul   .   s   y   m   t
 0145340   a   b nul   .   s   t   r   t   a   b nul   .   s   h   s   t
 0145360   r   t   a   b nul   .   h   a   s   h nul   .   d   y   n   s
 0145400   y   m nul   .   d   y   n   s   t   r nul   .   r   e   l   a
 0145420   .   d   y   n nul   .   r   e   l   a   .   p   l   t nul   .
 0145440   i   n   i   t nul   .   t   e   x   t nul   .   f   i   n   i

What is interesting is that the difference is following the debug link name libansi.so.dbg.   

What is at the point in the file where the two binaries differ?  Why don't I see similar results in other binaries we build?

Re: .hash section oddity

[Alan Modra]

On Wed, Sep 05, 2007 at 04:23:48PM -0700, Smith, Stephen (SwCOE) wrote:
> $ diff -u ppc-libansi-300.txt ppc-libansi-3873.txt
> --- ppc-libansi-300.txt 2007-09-05 10:45:48.871022500 -0700
> +++ ppc-libansi-3873.txt        2007-09-05 10:44:53.152272500 -0700
> @@ -1,5 +1,5 @@
>    
> -../../ansi-3_0_0/ppc/appbin/dbg/libansi.so:  file format elf32-powerpc
> +ppc/appbin/dbg/libansi.so:     file format elf32-powerpc
> 
> Disassembly of section .hash:
>   
> @@ -13354,4 +13354,4 @@
>     4:  6e 73 69 2e     xoris   r19,r19,26926
>     8:  73 6f 2e 64     andi.   r15,r27,11876
>     c:  62 67 00 6e     ori     r7,r19,110
> -  10:  2a af b8 77     cmpldi  cr5,r15,47223
> +  10:  13 10 72 26     vmsumuhm v24,v16,v14,v8
> 
> This told me that a DWORD is different and that the variable is aligned on a DWORD boundary.

This difference isn't in the .hash section.

> What is at the point in the file where the two binaries differ?

You tell us.  :-)  Look in ppc-libansi-300.txt at around line 13350
for the section name.

-- 
Alan Modra
Australia Development Lab, IBM

RE: .hash section oddity

[Smith, Stephen (SWCOE)]

Thanks - It is in the debuglink section.

-----Original Message-----
From: Alan Modra [mailto:amodra@bigpond.net.au] 
Sent: Wednesday, September 05, 2007 5:44 PM
To: Smith, Stephen (SWCOE)
Cc: binutils@sources.redhat.com
Subject: Re: .hash section oddity

On Wed, Sep 05, 2007 at 04:23:48PM -0700, Smith, Stephen (SwCOE) wrote:
> $ diff -u ppc-libansi-300.txt ppc-libansi-3873.txt
> --- ppc-libansi-300.txt 2007-09-05 10:45:48.871022500 -0700
> +++ ppc-libansi-3873.txt        2007-09-05 10:44:53.152272500 -0700
> @@ -1,5 +1,5 @@
>    
> -../../ansi-3_0_0/ppc/appbin/dbg/libansi.so:  file format
elf32-powerpc
> +ppc/appbin/dbg/libansi.so:     file format elf32-powerpc
> 
> Disassembly of section .hash:
>   
> @@ -13354,4 +13354,4 @@
>     4:  6e 73 69 2e     xoris   r19,r19,26926
>     8:  73 6f 2e 64     andi.   r15,r27,11876
>     c:  62 67 00 6e     ori     r7,r19,110
> -  10:  2a af b8 77     cmpldi  cr5,r15,47223
> +  10:  13 10 72 26     vmsumuhm v24,v16,v14,v8
> 
> This told me that a DWORD is different and that the variable is
aligned on a DWORD boundary.

This difference isn't in the .hash section.

> What is at the point in the file where the two binaries differ?

You tell us.  :-)  Look in ppc-libansi-300.txt at around line 13350
for the section name.

-- 
Alan Modra
Australia Development Lab, IBM

Re: .hash section oddity

[Alan Modra]

On Thu, Sep 06, 2007 at 09:10:35AM -0700, Smith, Stephen (SWCOE) wrote:
> Thanks - It is in the debuglink section.

That's what I thought you would find.  The difference is the CRC for
the debug info file, so somehow your debug info is different.

-- 
Alan Modra
Australia Development Lab, IBM