llvm dev - Sep 2016 - [RFC] Support disassembly of ARM and thumb mixed in single ELF file

Hi,

 

The llvm-objdump tool at the moment disassembles ARM ELF binary but with lot
of extra user supplied data

*         The triple supplied on the command line for tools has to be be
correct (ARM, thumb etc)

*         The ELF file cannot have mix of ARM and thumb

*         There is no direct way of using it such as llvm-objdump -d
elf_file. This works for architectures such as Hexagon, X86 etc.

 

Is there a way to enhance the tool to make the use more friendly by removing
the need to specify these extra option on command line since it is possible
to find the information in the file.

The ELF file for arm does not have any field that can be easily inspected to
see if the file of type ET_ARM is either a thumb only binary, ARM only
binary or a mix. There is however an optional ARM specific section
.ARM.attributes that can contain CPU and architecture information such as
ISA in use. BFD and gold linker use this section while linking to find any
incompatibilities and then emit out a merged (as per ARM ABI extensions)
section in final binary. 

 

Any ideas to add this functionality to the without completely redesigning
the llvm-objdump will be appreciated. There is already a patch that uses
data from ELF file to try doing this. 

It is at https://reviews.llvm.org/D24636

 

Thanks,

--

Hemant Kulkarni

khemant at codeaurora.org

Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by
the Linux Foundation

 

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160916/f101641c/attachment.html>

On Fri, Sep 16, 2016 at 04:51:51PM -0500, Hemant Kulkarni via llvm-dev
wrote:
> The llvm-objdump tool at the moment disassembles ARM ELF binary but with
lot
> of extra user supplied data
> 
> *         The ELF file cannot have mix of ARM and thumb
This can be done by looking for the marker symbols, at least if the file
was not stripped.

Joerg

On Fri, Sep 16, 2016 at 2:51 PM, Hemant Kulkarni <khemant at
codeaurora.org>
wrote:
> Hi,
>
>
>
> The llvm-objdump tool at the moment disassembles ARM ELF binary but with
> lot of extra user supplied data
>
> ·         The triple supplied on the command line for tools has to be be
> correct (ARM, thumb etc)
>
> ·         The ELF file cannot have mix of ARM and thumb
>
> ·         There is no direct way of using it such as llvm-objdump –d
> elf_file. This works for architectures such as Hexagon, X86 etc.
>
>
>
> Is there a way to enhance the tool to make the use more friendly by
> removing the need to specify these extra option on command line since it is
> possible to find the information in the file.
>
> The ELF file for arm does not have any field that can be easily inspected
> to see if the file of type ET_ARM is either a thumb only binary, ARM only
> binary or a mix. There is however an optional ARM specific section
> .ARM.attributes that can contain CPU and architecture information such as
> ISA in use. BFD and gold linker use this section while linking to find any
> incompatibilities and then emit out a merged (as per ARM ABI extensions)
> section in final binary.
>
>
>
> Any ideas to add this functionality to the without completely redesigning
> the llvm-objdump will be appreciated. There is already a patch that uses
> data from ELF file to try doing this.
>
> It is at https://reviews.llvm.org/D24636
>
>
>
> Thanks,
>
>
>
objdump is currently in a pretty bad state. The disassembler was basically
ripped from llvm-mc and has been hacked on since without needed
refactoring. I feel that at this point we need to stop and restructure it
before throwing more complexity onto the pile.

- Michael Spencer
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160917/57ee10fb/attachment.html>

One option for detecting arm/thumb is to look at the LSB of the symbol
addresses (SymbolRef::SF_Thumb flag) what specifies the instruction set for
the (beginning of the) function. It will work more or less for stripped
object files as well based on the symbols in the dynsym section. I think
the disassembler in the Mach-O part of objdump (same binary different code
path) already solves the problem this way.

Tamas

On Sun, Sep 18, 2016 at 12:20 AM Michael Spencer via llvm-dev <
llvm-dev at lists.llvm.org> wrote:

On Fri, Sep 16, 2016 at 2:51 PM, Hemant Kulkarni <khemant at
codeaurora.org>
wrote:

Hi,



The llvm-objdump tool at the moment disassembles ARM ELF binary but with
lot of extra user supplied data

·         The triple supplied on the command line for tools has to be be
correct (ARM, thumb etc)

·         The ELF file cannot have mix of ARM and thumb

·         There is no direct way of using it such as llvm-objdump –d
elf_file. This works for architectures such as Hexagon, X86 etc.



Is there a way to enhance the tool to make the use more friendly by
removing the need to specify these extra option on command line since it is
possible to find the information in the file.

The ELF file for arm does not have any field that can be easily inspected
to see if the file of type ET_ARM is either a thumb only binary, ARM only
binary or a mix. There is however an optional ARM specific section
.ARM.attributes that can contain CPU and architecture information such as
ISA in use. BFD and gold linker use this section while linking to find any
incompatibilities and then emit out a merged (as per ARM ABI extensions)
section in final binary.



Any ideas to add this functionality to the without completely redesigning
the llvm-objdump will be appreciated. There is already a patch that uses
data from ELF file to try doing this.

It is at https://reviews.llvm.org/D24636



Thanks,



objdump is currently in a pretty bad state. The disassembler was basically
ripped from llvm-mc and has been hacked on since without needed
refactoring. I feel that at this point we need to stop and restructure it
before throwing more complexity onto the pile.

- Michael Spencer

_______________________________________________
LLVM Developers mailing list
llvm-dev at lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160921/5d11fb75/attachment.html>