[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [patch] thread-aware coredumps, 2.5.43-C3



I suppose it'll get into some version.  

This version is kind-of new to me.  I would like to see this implementation 
get some more mileage on the 2.5 tree before talking too much about back 
porting it to 2.4x or the Carrier Grade Linux thingy.

Also, there is was some talk about the SIGKILL's getting sent to the thread 
group as being a possible issue.  I'm not expert enough in POSIX to say one 
way or the other.  Whether the pthreads helper thread or the kernel sends the 
SIGKILL's shouldn't make any difference.  I was hoping to see that question 
closed out as well.

--mgross

On Friday 18 October 2002 10:11 am, Howell, David P wrote:
> Mark,
> We'll likely need this for CGLE 1.0 as well, right?
>
> Dave Howell
>
> -----Original Message-----
> From: Mark Gross [mailto:mark thegnar org]
> Sent: Friday, October 18, 2002 9:58 AM
> To: phil-list redhat com; Daniel Jacobowitz; Mark Kettenis; mgross
> Cc: linux-kernel vger kernel org; NPT library mailing list
> Subject: Re: [patch] thread-aware coredumps, 2.5.43-C3
>
> I think I fixed it to set namesz to 5, with the +1 it was making it 6.
> My
> patch is supposed to remove the +1.
>
> The value for men-name for the extended registers case is "LINUX".
>
> --mgross
>
> On Thursday 17 October 2002 05:48 pm, Daniel Jacobowitz wrote:
> > You'd have to ask Mark Kettenis about that.  Mark, it looks like you
> > updated the kernel to write namesz == 6, but BFD still expects 5 (and
> > elfcore_write_note writes 6)?  Shouldn't we accept both, anyway?
> >
> > On Thu, Oct 17, 2002 at 05:07:41PM -0400, mgross wrote:
> > > This patch is working pretty well for me with only one problem,
>
> which
>
> > > seems to have happened indepently of Ingo's patch.
> > >
> > > Extended floating point note sections are no longer getting
>
> recognized by
>
> > > GDB 5.x.
> > >
> > > After a bit of poking around in the GDB 5.2.1 code (line 6399
>
> bfd/efl.c)
>
> > > I noticed that there is a n_namesz test for the reg-xfp section
>
> parsing.
>
> > > The following minor tweak on top of Ingo's patch  to binfmt_elf.c
>
> fixes
>
> > > the problem.
> > >
> > > diff -urN -X dontdiff linux-2.5.43/fs/binfmt_elf.c
> > > linux-2.5.43.xfp/fs/binfmt_elf.c
> > > --- linux-2.5.43/fs/binfmt_elf.c      Thu Oct 17 16:54:13 2002
> > > +++ linux-2.5.43.xfp/fs/binfmt_elf.c  Thu Oct 17 16:53:00 2002
> > > @@ -964,7 +964,9 @@
> > >  {
> > >       struct elf_note en;
> > >
> > > -     en.n_namesz = strlen(men->name) + 1;
> > > +     en.n_namesz = strlen(men->name);
> > > +     /* en.n_namesz = strlen(men->name) + 1;  gdb checks namez and
>
> the +
>
> > > 1 */ +                                     /*breaks xfp core file
>
> note
>
> > > sections. */ en.n_descsz = men->datasz;
> > >       en.n_type = men->type;
> > >
> > >
> > >
> > > I don't know when this +1 was added to writenote binfmt_elf.c or
>
> why.  I
>
> > > do know that newer than July and it isn't in 2.4
> > >
> > > If there the + 1 is needed of others then we may need to special
>
> case the
>
> > > NT_PRXFPREG note sections.
> > >
> > > --mgross





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]