Re: [PATCH]: Use core regset when possible in linux-nat.c

[Daniel Jacobowitz <drow@false.org>]

On Thu, Apr 06, 2006 at 03:03:30PM -0700, David S. Miller wrote:
> *1: The test in corefile.exp to look at the mmap() area pointed
>     to by buf2 fails due to a Linux kernel bug, it doesn't dump
>     mmap() areas which have not been written to into the core
>     file so gdb can't look at it.  I've posted a patch to linux-kernel
>     already suggesting to take away that check in the ELF core dumper.

Or, we could just change the test.  I don't know...

> 2006-04-06  David S. Miller  <davem@sunset.davemloft.net>
> 
> 	* linux-nat.c (linux_nat_do_thread_registers): Use the
> 	regset_from_core_section infrastructure if the target
> 	supports it.
> 	* Makefile.in: Update dependencies.

Almost.  I know I suggested changing the fallback case from your
original suggestion, but this isn't quite what I meant:

> -  fill_fpregset (&fpregs, -1);
> +  if (core_regset_p)
> +    {
> +      regset = gdbarch_regset_from_core_section (gdbarch, ".reg2",
> +						 sizeof (fpregs));
> +      regset->collect_regset (regset, current_regcache, -1,
> +			      &fpregs, sizeof (fpregs));
> +    }
> +  else
> +    fill_fpregset (&fpregs, -1);
>    note_data = (char *) elfcore_write_prfpreg (obfd,
>  					      note_data,
>  					      note_size,
>  					      &fpregs, sizeof (fpregs));

A target which only defines .reg and not .reg2 will crash here.
We'd still have to check for non-NULL; I just think that if the
target supports regset_from_core_section, and still returns NULL,
that that's a pretty good hint we have nothing to write.

The other thing that bugs me about this interface may be harder to fix
- we're still assuming the two are the same, more or less, because the
regset type is required here and we use its sizeof.  A problem for
another day?


-- 
Daniel Jacobowitz
CodeSourcery