Port maintainers, please check: makecontext with uc_link == NULL

[Thomas Schwinge <thomas at codesourcery dot com>]

Hi!

Port maintainers, please check your ports:

On Tue, 17 Jul 2012 09:12:52 +0200, Andreas Jaeger <aj@suse.com> wrote:
> On 07/16/2012 04:28 PM, Thomas Schwinge wrote:
> > On Mon, 16 Jul 2012 13:25:07 +0200, Andreas Krebbel <krebbel@linux.vnet.ibm.com> wrote:
> >> On 07/16/2012 10:19 AM, Thomas Schwinge wrote:
> >>> On Thu, 12 Jul 2012 09:38:35 -0400, Carlos O'Donell <carlos_odonell@mentor.com> wrote:
> >>>> On 7/12/2012 8:19 AM, Andreas Krebbel wrote:
> >>>>> stdlib/tst-makecontext already calls makecontext with uc_link == NULL
> >>>>> but the function invoked in the context does explicitly call exit (0).
> >>>>> Removing this enables the testcase to cover that problem as well.
> >>>
> >>>>> 	* stdlib/tst-makecontext.c: Remove explicit exit call.
> >>>>
> >>>> Excellent. The testcase changes look good to me and match what
> >>>> SuSv2 says about a returning from a context where uc_link is
> >>>> zero e.g. "the thread will exit when this context returns."
> >>>>
> >>>> I'm happy with this, thanks for enhancing the testcase to cover
> >>>> the failure scenario.
> >>>
> >>> I already raised this topic in
> >>> <http://news.gmane.org/find-root.php?message_id=%3C87lijq5h72.fsf@schwinge.name%3E>:
> >>>
> >>> | [a bug is only seen] when returning from a context with Âuc_link ==
> >>> | NULLÂ, which is not exercised in the testsuite.
> >>> |
> >>> | I first though about simply removing the Âexit (0)Â from
> >>> | stdlib/tst-makecontext.c:cf (which would then test exactly this case),
> >>> | but apparently it is not specified which status value to use for exit in
> >>> | this case -- libc.info: ÂIf `uc_link' was a null pointer the application
> >>> | terminates in this case. -- so it is not trivial to test for.  (Maybe
> >>> | worth specifying?  EXIT_SUCCESS (0)?)
> >> Ok. Thanks for the pointer. I think everything other than 0 doesn't make much sense since
> >> exiting that way is not caused by any kind of error.
> >
> > Ack.
> >
> >> I could check in the testcase change removing the exit and we could make the back-end
> >> maintainers aware of the problem with their target. What do you think?
> >
> > How about this one?
> >
> > manual: setcontext: Clarify termination when uc_link is the null pointer.
> >
> > 2012-07-16  Thomas Schwinge  <thomas@codesourcery.com>
> > 	    Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>
> >
> > 	* manual/setjmp.texi (setcontext): Clarify normal process
> > 	termination when uc_link is the null pointer.
> > 	* stdlib/tst-makecontext.c (cf): Exercise this: remove explicit
> > 	exit call.
> >
> > diff --git a/manual/setjmp.texi b/manual/setjmp.texi
> > index a5a7ce6..f13ac7b 100644
> > --- a/manual/setjmp.texi
> > +++ b/manual/setjmp.texi
> > @@ -354,7 +354,8 @@ specified parameters passed.  If this function returns execution is
> >   resumed in the context which was referenced by the @code{uc_link}
> >   element of the context structure passed to @code{makecontext} at the
> >   time of the call.  If @code{uc_link} was a null pointer the application
> > -terminates in this case.
> > +terminates normally with an exit status value of @code{EXIT_SUCCESS}
> > +(@pxref{Program Termination}).
> >
> >   Since the context contains information about the stack no two threads
> >   should use the same context at the same time.  The result in most cases
> > diff --git a/stdlib/tst-makecontext.c b/stdlib/tst-makecontext.c
> > index 3185900..eb6e89b 100644
> > --- a/stdlib/tst-makecontext.c
> > +++ b/stdlib/tst-makecontext.c
> > @@ -35,7 +35,9 @@ cf (int i)
> >         printf ("i %d thr %d\n", i, thr);
> >         exit (1);
> >       }
> > -  exit (0);
> > +
> > +  /* Since uc_link below has been set to NULL, setcontext is supposed to
> > +     terminate the process normally after this function returns.  */
> >   }

Pushed as dc97c227c95dd521594f1eaa472399b9fba03b2a.

> > x86_64: makecontext: exit (0) if uc_link is the null pointer.
> >
> > 2012-07-16  Thomas Schwinge  <thomas@codesourcery.com>
> >
> > 	* sysdeps/unix/sysv/linux/x86_64/__start_context.S
> > 	(__start_context): Preserve zero value for regular exit case.
> 
> Yes, this is fine.

Pushed as f7db31703ab2e11b162d4e0e3b4af0c1c971b6cd.

> > SH: makecontext: exit (0) if uc_link is the null pointer.
> >
> > 2012-07-16  Thomas Schwinge  <thomas@codesourcery.com>
> >
> > 	* sysdeps/unix/sysv/linux/sh/makecontext.S (.Lexitcode): Preserve
> > 	zero value for regular exit case.

> I suggest the sh maintainer to review this, it looks fine to me,

That'd be me ;-) (and Kaz, of course).  Pushed as
07cbfc23683827c1b92d0bc62b15a77a48b09a17.

> > Further (incomplete) notes for port maintainers have been gives as
> > follows:
> >
> >      i386: sysdeps/unix/sysv/linux/i386/makecontext.S:L(exitcode): seems OK
> >      x86_64: sysdeps/unix/sysv/linux/x86_64/__start_context.S: 2: is misplaced, should be after the movq %rax, %rdi
> >      SPARC32: sysdeps/unix/sysv/linux/sparc/sparc32/setcontext.S:__start_context can't tell
> >      SPARC64: sysdeps/unix/sysv/linux/sparc/sparc64/makecontext.c:__makecontext_ret missing handling completely
> >      S390-32: sysdeps/unix/sysv/linux/s390/s390-32/makecontext.c:__makecontext_ret missing handling completely
> >      S390-64: sysdeps/unix/sysv/linux/s390/s390-64/makecontext.c:__makecontext_ret missing handling completely


GrÃÃe,
 Thomas

Attachment: pgp00000.pgp
Description: PGP signature