Re: PATCH: PR corefiles/11467: amd64 gdb generates corrupted 32bit core file

["H.J. Lu" <hjl.tools@gmail.com>]

On Sun, Apr 11, 2010 at 9:49 AM, Mark Kettenis <mark.kettenis@xs4all.nl> wrote:
>> Date: Sat, 10 Apr 2010 15:27:42 -0700
>> From: "H.J. Lu" <hongjiu.lu@intel.com>
>>
>> On Sat, Apr 10, 2010 at 03:19:43PM -0700, H.J. Lu wrote:
>> > Hi,
>> >
>> > I am checking in this patch to support 32bit core note sections on
>> > Linux/x86-64. I will submit a separate gdb patch.
>> >
>> >
>> > H.J.
>> > ---
>> > 2010-04-10  H.J. Lu  <hongjiu.lu@intel.com>
>> >
>> >     PR corefiles/11467
>> >     * configure.in (CORE_HEADER): New. Set to hosts/x86-64linux.h
>> >     for x86_64-*-linux*.
>> >     * config.in: Regenerated.
>> >     * configure: Likewise.
>> >
>> >     * elf.c: Include CORE_HEADER if it is defined.
>> >
>> > 2010-04-10  H.J. Lu  <hongjiu.lu@intel.com>
>> >         Jan Kratochvil  <jan.kratochvil@redhat.com>
>> >
>> >     * hosts/x86-64linux.h: New.
>>
>> Hi,
>>
>> Hi,
>>
>> Here is the gdb patch to properly generate 32bit coredumps on
>> Linux/x86-64. The key here is to use the right register offset
>> for gcore. OK to install?
>
> No, this is wrong.  This should all be handled by -tdep.c code.  The
> gcore code in linux-nat.c nly falls back on the fetch_gregset() if for
> some reason the gdbarch_regset_from_core_section() function doesn't
> return a valid register set.  I don't immediately see why, but that is
> what you need to fix.
>

I will prepare a patch. I will move codes from i386-linux-nat.c to
i386-linux-tdep.c.  I will add i386_linux_regset_from_core_section
and call

 set_gdbarch_regset_from_core_section (gdbarch,
              i386_linux_regset_from_core_section);

in i386_linux_init_abi. i386_linux_regset_from_core_section will
call fill_gregset originally in i386-linux-nat.c.

Thanks.


-- 
H.J.