gas: undefined symbol in immediate operand

gas: undefined symbol in immediate operand

[Jeremy Dexter]

Hi

I'm finding that if I accidentally enter an undefined symbol name as an
*immediate* operand, then as will accept it without any warning. I've tried
this on binutils 2.14 and 2.17 for ARM and PowerPC.

For example, the ARM assembler produces the following:

                                .equ    Moo,   0x42 
 297 0148 000080E3 	                orr    r0,r0,#moo

Note the typo in the symbol name. The assembler accepts it and adds an
undefined symbol "moo" to the object file, but the linker will not flag it
as undefined, which I presume is because there is no valid reference to it.
So you end up with code that ORs the value 0x00 rather than 0x42.

I would have thought that the assembler is at fault here. An immediate
operand should resolve to a *constant* - an external symbol reference should
not be accepted here. Or am I missing something?

Thanks in advance!

Jeremy

Re: gas: undefined symbol in immediate operand

[Richard Earnshaw]

On Thu, 2006-08-24 at 04:51, Jeremy Dexter wrote:
> Hi
> 
> I'm finding that if I accidentally enter an undefined symbol name as an
> *immediate* operand, then as will accept it without any warning. I've tried
> this on binutils 2.14 and 2.17 for ARM and PowerPC.
> 
> For example, the ARM assembler produces the following:
> 
>                                 .equ    Moo,   0x42 
>  297 0148 000080E3 	                orr    r0,r0,#moo
> 
> Note the typo in the symbol name. The assembler accepts it and adds an
> undefined symbol "moo" to the object file, but the linker will not flag it
> as undefined, which I presume is because there is no valid reference to it.
> So you end up with code that ORs the value 0x00 rather than 0x42.
> 
> I would have thought that the assembler is at fault here. An immediate
> operand should resolve to a *constant* - an external symbol reference should
> not be accepted here. Or am I missing something?

Not quite.  On some processors an undefined symbol can be resolved by
the assembler to use a relocation (so that some external file can define
the value).  So the rule in the assembler is

If there is no acceptable relocation for the missing symbol generate an
error.  Otherwise, generate that relocation and the linker must raise
the error if the symbol is undefined (or can't fit the value into the
space available).

R.

Re: gas: undefined symbol in immediate operand

[Nick Clifton]

Hi Jeremy,

> I'm finding that if I accidentally enter an undefined symbol name as an
> *immediate* operand, then as will accept it without any warning. I've tried
> this on binutils 2.14 and 2.17 for ARM and PowerPC.

Are you sure about this ?  I just tried an arm-eabi toolchain built from 
the 2.17 sources and using your test case:

>   .equ    Moo,   0x42 
>    orr    r0,r0,#moo

I received this error message:

   jim.s: Assembler messages:
   jim.s:5: Error: undefined symbol moo used as an immediate value

> Note the typo in the symbol name. The assembler accepts it and adds an
> undefined symbol "moo" to the object file, but the linker will not flag it
> as undefined, which I presume is because there is no valid reference to it.

No if the linker does behave this way then it is a bug.  Assuming that 
the assembler does pass the reference to the undefined "moo" symbol on 
to the linker, then the linker should complain about it.  The only 
reason it would not is if the reference to the symbol is in a section of 
code that is being discarded and not included in the final executable.

Can you put together a full test case to demonstrate the problem, 
including specifying which toolchain(s) you use ?

Cheers
   Nick

RE: gas: undefined symbol in immediate operand

[Jeremy Dexter]

Thanks for the replies Nick and Richard

I'm using a pre-built ARM toolchain (supplied with an embedded development
system) running on Windows under cygwin. Using a very simple test case I
get:


C:\jd\svn\can_trunk>arm-uclibc-as -a test.s -o test.o
ARM GAS  test.s                         page 1


   1                        .equ     Moo,   0x42
   2 0000 000080E3          orr      r0,r0,#moo
   3 0004 420080E3          orr      r0,r0,#Moo
   4 0008 00009FE5          ldr      r0, =bruce
   5 000c FEFFFFEA          b        freddy
   6 0010 00000000          .end
?ARM GAS  test.s                        page 2


DEFINED SYMBOLS
              test.s:1      *ABS*:00000042 Moo

UNDEFINED SYMBOLS
moo
bruce
freddy

C:\jd\svn\can_trunk>arm-uclibc-objdump -xtr test.o

test.o:     file format elf32-littlearm
test.o
architecture: arm, flags 0x00000011:
HAS_RELOC, HAS_SYMS
start address 0x00000000
private flags = 0: [APCS-32] [FPA float format]

Sections:
Idx Name          Size      VMA       LMA       File off  Algn
  0 .text         00000014  00000000  00000000  00000034  2**2
                  CONTENTS, ALLOC, LOAD, RELOC, READONLY, CODE
  1 .data         00000000  00000000  00000000  00000048  2**0
                  CONTENTS, ALLOC, LOAD, DATA
  2 .bss          00000000  00000000  00000000  00000048  2**0
                  ALLOC
SYMBOL TABLE:
00000000 l    d  .text  00000000
00000000 l    d  .data  00000000
00000000 l    d  .bss   00000000
00000042 l       *ABS*  00000000 Moo
00000000         *UND*  00000000 moo
00000000         *UND*  00000000 bruce
00000000         *UND*  00000000 freddy


RELOCATION RECORDS FOR [.text]:
OFFSET   TYPE              VALUE
0000000c R_ARM_PC24        freddy
00000010 R_ARM_ABS32       bruce

------------------

We've also tried this on a more recent PowerPC toolchain, this time running
on Linux and get a similar result:

jeff@trantor:~/prj/dt8x/bantha/Src/Bootstrap$ powerpc-eabi-as -a -mregnames
tmp.s
GAS LISTING tmp.s                       page 1


   1                    .equ MOO, 0xff000000
   2
   3                    #    .org    0x0100          # reset/NMI vector
   4                        .global _start
   5                    _start:
   6
   7 0000 3FE00000          lis     r31,moo@h
   8 0004 3FE0FF00          lis     r31,MOO@h
   9 0008 4E800020          blr
  10
  11

GAS LISTING tmp.s                       page 2


DEFINED SYMBOLS
               tmp.s:1      *ABS*:ff000000 MOO
               tmp.s:5      .text:00000000 _start

UNDEFINED SYMBOLS
moo

jeff@trantor:~/prj/dt8x/bantha/Src/Bootstrap$ powerpc-eabi-as --version
GNU assembler 060810 20060810
Copyright 2005 Free Software Foundation, Inc.
This program is free software; you may redistribute it under the terms of
the GNU General Public License.  This program has absolutely no warranty.
This assembler was configured for a target of `powerpc-eabi'.

----------------

The objdump output shows that the assembler hasn't created a relocation
entry for the undefined "moo" symbol, unlike the undefined "bruce" and
"freddy" symbols, but it hasn't generated an error either. This does not
seem right.

It's interesting that you got an error message using the current ARM
version, so maybe this has been fixed between 2.14 & 2.17, although I didn't
see anything in the release notes. However the PowerPC version we have very
recent and also gives no error message.

cheers

Jeremy.



-----Original Message-----
From: Nick Clifton
Sent: Thursday, 24 August 2006 8:05 pm
To: Jeremy Dexter
Cc: binutils@sourceware.org
Subject: Re: gas: undefined symbol in immediate operand


Hi Jeremy,

> I'm finding that if I accidentally enter an undefined symbol name as an
> *immediate* operand, then as will accept it without any warning. I've
tried
> this on binutils 2.14 and 2.17 for ARM and PowerPC.

Are you sure about this ?  I just tried an arm-eabi toolchain built from 
the 2.17 sources and using your test case:

>   .equ    Moo,   0x42 
>    orr    r0,r0,#moo

I received this error message:

   jim.s: Assembler messages:
   jim.s:5: Error: undefined symbol moo used as an immediate value

> Note the typo in the symbol name. The assembler accepts it and adds an
> undefined symbol "moo" to the object file, but the linker will not flag it
> as undefined, which I presume is because there is no valid reference to
it.

No if the linker does behave this way then it is a bug.  Assuming that 
the assembler does pass the reference to the undefined "moo" symbol on 
to the linker, then the linker should complain about it.  The only 
reason it would not is if the reference to the symbol is in a section of 
code that is being discarded and not included in the final executable.

Can you put together a full test case to demonstrate the problem, 
including specifying which toolchain(s) you use ?

Cheers
   Nick

Re: gas: undefined symbol in immediate operand

[Nick Clifton]

Hi Jeremy,

> I'm using a pre-built ARM toolchain (supplied with an embedded development
> system) running on Windows under cygwin. 

I think that you must have an old version of the ARM assembler.  The 
current versions definitely do report an error for the use of an 
undefined symbol as an immediate operand.

> We've also tried this on a more recent PowerPC toolchain, this time running
> on Linux and get a similar result:

> The objdump output shows that the assembler hasn't created a relocation
> entry for the undefined "moo" symbol,

What objdump output ?  When I tried running objdump I saw this:

   $ objdump -Dr test.o

   test.o:     file format elf32-powerpc

   Disassembly of section .text:

   00000000 <_start>:
      0:   3f e0 00 00     lis     r31,0
                         2: R_PPC_ADDR16_HI      moo
      4:   3f e0 ff 00     lis     r31,-256
      8:   4e 80 00 20     blr

Which clearly does show the reloc against symbol "moo".  Furthermore, 
when I tried to link this object file, I received an error message:

   $ ld test.o
   test.o: In function `_start':
   (.text+0x2): undefined reference to `moo'

This is which an assembler and linker compiled from today's binutils 
sources.

Cheers
   Nick

RE: gas: undefined symbol in immediate operand

[Jeremy Dexter]

Hi Nick

OK, I tried this on our PowerPC assembler and it works as you describe, ie.
a relocation entry is generated.

So basically it looks like the "Error: undefined symbol moo used as an
immediate value" message was added to the ARM assembler sometime after 2.14.
It's not really practical for me to upgrade to the latest ARM tools so I'll
just live with it for now.

Thanks again for your assistance!

cheers

Jeremy

-----Original Message-----
From: Nick Clifton
Sent: Friday, 25 August 2006 5:44 pm
To: Jeremy Dexter
Cc: binutils@sourceware.org
Subject: Re: gas: undefined symbol in immediate operand


Hi Jeremy,

> I'm using a pre-built ARM toolchain (supplied with an embedded development
> system) running on Windows under cygwin. 

I think that you must have an old version of the ARM assembler.  The 
current versions definitely do report an error for the use of an 
undefined symbol as an immediate operand.

> We've also tried this on a more recent PowerPC toolchain, this time
running
> on Linux and get a similar result:

> The objdump output shows that the assembler hasn't created a relocation
> entry for the undefined "moo" symbol,

What objdump output ?  When I tried running objdump I saw this:

   $ objdump -Dr test.o

   test.o:     file format elf32-powerpc

   Disassembly of section .text:

   00000000 <_start>:
      0:   3f e0 00 00     lis     r31,0
                         2: R_PPC_ADDR16_HI      moo
      4:   3f e0 ff 00     lis     r31,-256
      8:   4e 80 00 20     blr

Which clearly does show the reloc against symbol "moo".  Furthermore, 
when I tried to link this object file, I received an error message:

   $ ld test.o
   test.o: In function `_start':
   (.text+0x2): undefined reference to `moo'

This is which an assembler and linker compiled from today's binutils 
sources.

Cheers
   Nick