ppc/powerpc, remote debugging (ptrace), floating point variables

[Kumar Gala]

On Aug 28, 2007, at 4:40 AM, Visser, Udo RD-IS-E23 wrote:

>
> Hello world, ;-)
>
> when stepping through my own software, which is using floating  
> point variables, gdb showed weird results (mostly NAN´s) for these  
> variables. Running the software without gdb did not show any  
> problems. The software is compiled with floating point instructions  
> and the results (without gdb) are OK.
>
> My setup:
>
> MPC8349 target, "home grown", much like the MPC8349EMDS, running  
> 2.6.16 (ppc)
> GDB, gdbserver 6.3 connected via ethernet with my Linux Host.
>
>
> What I did to get rid of the described problem:
>
> In module:     arch/powerpc/kernel/process.c
> In function:   void flush_fp_to_thread(struct tast_struct *tsk)
>
> I changed the line:  giveup_fpu(current);
> to:                  giveup_fpu(tsk);
>
>
> Now my questions: Has anybody else had such problems?
>                   Have I found a bug, or will I encounter any yet  
> unknown problems in the near future?
>
> I would appreciate any comment on my changes, especially from the  
> maintainers.

I have a patch to fix just this issue.

take a look at:

http://patchwork.ozlabs.org/linuxppc/patch?id=13219

- k