THUMB code detection for ADS 1.2 ELF emages

THUMB code detection for ADS 1.2 ELF emages

[Dmitry Smirnov]

Hi,

I'm trying to use GDB to work with ARM ELF images produced by ADS 1.2 (ARM toolchain, a liitle bit old).
I can do almost all I want with ARM code. But GDB works incorrectly with THUMB code: I cannot disassemle the code, GDB places breakpoints at incorrect address, stepping with 'ni'.

In GDB source code I've found that it completely ignores some special symbols ($t, $a, $d ...) that ADS places to an ELF file.
I'm going to add some processing of this symbols. At first, I would like to handle $t symbols: it looks like they denote THUMB functions (ADS does not use STT_LOPROC) . So, I'm going to cycle though symbols and find a corresponding function name for each $t symbol (they should have the same value, i.e. address).

I'm wondering about the best place for this (I'm just started to learn GDB source).
On my mind it would be better to handle these symbols in elf_slurp_symbol_table() (file elfcode.h). Right now there is no symbol table post processing function (ebd->elf_backend_symbol_table_processing is NULL). So, I'm going to implement such a function. It will cycle over all the symbols and find a symbols that correspond to each $t symbol. If found, this symbol will get STT_LOPROC attribute in isym->st_info. It will replace the STT_FUNC, but I do not know the other way for this. By the time of this assignment, all these symbols will be already processed and sym->symbol.flags will be already updated with BSF_FUNCTION flag.

If you know the better approach, please let me know.

Dmitry

Re: THUMB code detection for ADS 1.2 ELF emages

[Daniel Jacobowitz]

On Fri, Jun 20, 2008 at 12:09:08PM +0400, Dmitry Smirnov wrote:
> In GDB source code I've found that it completely ignores some special symbols ($t, $a, $d ...) that ADS places to an ELF file.

Try a newer version of GDB; this is already supported.  You'll need a
CVS snapshot.

-- 
Daniel Jacobowitz
CodeSourcery

Re: THUMB code detection for ADS 1.2 ELF emages

[Stan Shebs]

Dmitry Smirnov wrote:
> Hi,
>
> I'm trying to use GDB to work with ARM ELF images produced by ADS 1.2 (ARM toolchain, a liitle bit old).
> I can do almost all I want with ARM code. But GDB works incorrectly with THUMB code: I cannot disassemle the code, GDB places breakpoints at incorrect address, stepping with 'ni'.
>
> In GDB source code I've found that it completely ignores some special symbols ($t, $a, $d ...) that ADS places to an ELF file.
>   
Dan Jacobowitz added this bit of machinery recently; see 
http://sourceware.org/ml/gdb-patches/2008-05/msg00113.html . He points 
out that this should be a rare case, only needed if Arm and Thumb code 
are mixed within a single file; so if your GDB is having trouble 
detecting Thumb code, maybe there is some other problem, perhaps the 
toolchain is not marking the executable correctly?

Stan

Re: THUMB code detection for ADS 1.2 ELF emages

[Dmitry Smirnov]

Hi Stan, Dan,

Thanks for pointing this new feature. I hope it will helps me.

P.S. What do you mean by "toolchain is not marking the executable correctly"? I suppose that ADS produce the ELF that is not compartible with latest ARM ELF specs. At least, it specifies ELF as an 0x2000000 version, whereas GDB recognizes 0x4000000+.
I believe executable is correct. It mixes ARM and THUMB code. It runs on a real hardware, I can debug it with TRACE32 using this ELF. I'm trying to run this ELF on skyeye (and it runs!), but cannot debug it "in full-power" with GDB. 


Stan Shebs wrote:
>Dan Jacobowitz added this bit of machinery recently; see 
>http://sourceware.org/ml/gdb-patches/2008-05/msg00113.html . He points 
>out that this should be a rare case, only needed if Arm and Thumb code 
>are mixed within a single file; so if your GDB is having trouble 
>detecting Thumb code, maybe there is some other problem, perhaps the 
>toolchain is not marking the executable correctly?
>
>Stan