emacs crash on SMP notebook only

emacs crash on SMP notebook only

[Joost Helberg]

hi,

I'm running GNU Emacs 21.4 on a Notebook (Sager 9860) with Intel P4 64
bits Hyperthreading.
When using linux-2.6.12 kernel with HT and SMP disabled, Emacs runs
fine. After enabling HT and SMP, Emacs sometimes hangs and often
crashes with a segmentation violation.

Even during init-phase, Emacs sometimes aborts. Today I restarted GNU
Emacs 12 times :-((

Is there anyone out there with the same experience? I've used Emacs
for over 20 years now and only experienced out-of-memory related
crashed, nothing Emacs-self-caused.

Switching kernels makes this problem disappear, but I need to run SMP
of course :-))

Regards,

Joost

-- 
Joost Helberg
Snow B.V.        http://snow.nl Tel 0418-653333 Fax 0418-653666
Voorzitter VOSN  http://www.vosn.nl Tel 0418-653336 Fax 0418-653666

Re: emacs crash on SMP notebook only

[Richard M. Stallman]

    I've investigated further and came to the conclusion that this is
    related to using gcc-4.0.2.
    Using a gcc-3.4 compiled GNU Emacs works fine now.

That is disturbing.  It could be a GCC bug.

    I'm using GNU/Linux of the Debian sort on Intel 4 3.6GHz with
    hyperthreading enabled. Using NO hyperthreading (a kernel option) with
    gcc-4.0.2 compiled emacs shows no crashes.

I am not sure what hyperthreading is.  Does it alter the execution of
machine code?

Please confirm the meaning of these statements.  Are you saying that
the same Emacs executable (made with GCC 4) crashes with
hyperthreading but does not crash when run without hyperthreading?

Does the Emacs executable made with GCC 3.4 run correctly both
with and without hyperthreading?

I think that is what you are saying, but let's make sure there is no
misunderstanding.

    Can I help narrowing this thing down to a gcc or emacs issue?

The way to investigate further is to compile some of the files with
each compiler, and see if it crashes.  That way you can determine
which file has the problem.

Would you like to do that?

Re: emacs crash on SMP notebook only

[Joost Helberg]

Richard,

>>>>> "Richard" == Richard M Stallman <rms@gnu.org> writes:
 > Date: Sat, 26 Nov 2005 19:31:52 -0500

 >     I've investigated further and came to the conclusion that this is
 >     related to using gcc-4.0.2.
 >     Using a gcc-3.4 compiled GNU Emacs works fine now.

 > That is disturbing.  It could be a GCC bug.

 >     I'm using GNU/Linux of the Debian sort on Intel 4 3.6GHz with
 >     hyperthreading enabled. Using NO hyperthreading (a kernel option) with
 >     gcc-4.0.2 compiled emacs shows no crashes.

 > I am not sure what hyperthreading is.  Does it alter the execution of
 > machine code?

It presents the kernel with 2 CPU's. The CPU can then use wait-states
etc. for time-sharing on a CPU-level.
Intel claims a 30% performance gain and better `interactive'
performance. 

 > Please confirm the meaning of these statements.  Are you saying that
 > the same Emacs executable (made with GCC 4) crashes with
 > hyperthreading but does not crash when run without hyperthreading?

I was saying exactly that.

 > Does the Emacs executable made with GCC 3.4 run correctly both
 > with and without hyperthreading?
Now I know that the GCC 3.4 compiled Emacs crashes too, but not as
often. But there is less alarming news:

  Emacs-22 (from CVS) works OK (i.e. with hyperthreading, GCC 4).

  Emacs-21.4 as bundled with my OS but recompiled, crashes often with
                hyperthreading and GCC 4, crashed twice with GCC 3.4.

  Emacs-21.4 as deliverd by the FSF works fine using GCC 4 or
                GCC 3.4.

It looks as if the problem is either gcc or the patches from Debian or
Ubuntu. We can safely assume that gcc deserves more credits than
Debian and Ubuntu together.

I'm currently checking out the patches from Debian and Ubuntu, as I'm
pretty sure these make the crashes appear.

I'm sorry for any hassle I caused, as it looks like the problem sits
in people modifying code for no reason whatsoever.

I'll look further into the problem in order to complete feedback to
the Debian people, I feel a but reluctant though as no good can come
from the way their packaging works.

 > I think that is what you are saying, but let's make sure there is no
 > misunderstanding.

There is a clearer picture now which is a lot less disturbing.

 >     Can I help narrowing this thing down to a gcc or emacs issue?

 > The way to investigate further is to compile some of the files with
 > each compiler, and see if it crashes.  That way you can determine
 > which file has the problem.

 > Would you like to do that?

I will try to pin down the problem by looking further and suspecting
both GCC and the Debian/Ubuntu patches.

Regards,

Joost Helberg

-- 
Joost Helberg
Snow B.V.        http://snow.nl Tel 0418-653333 Fax 0418-653666
Voorzitter VOSN  http://www.vosn.nl Tel 0418-653336 Fax 0418-653666

Re: emacs crash on SMP notebook only

[Richard M. Stallman]

      Emacs-22 (from CVS) works OK (i.e. with hyperthreading, GCC 4).

      Emacs-21.4 as bundled with my OS but recompiled, crashes often with
		    hyperthreading and GCC 4, crashed twice with GCC 3.4.

      Emacs-21.4 as deliverd by the FSF works fine using GCC 4 or
		    GCC 3.4.

    It looks as if the problem is either gcc or the patches from Debian or
    Ubuntu. We can safely assume that gcc deserves more credits than
    Debian and Ubuntu together.

That is a very useful clue.  If you can determine which patch it is,
we would be on the home stretch.

Re: emacs crash on SMP notebook only

[Lőrentey Károly]

Joost Helberg <joost@snow.nl> writes:
>   Emacs-22 (from CVS) works OK (i.e. with hyperthreading, GCC 4).
>
>   Emacs-21.4 as bundled with my OS but recompiled, crashes often with
>                 hyperthreading and GCC 4, crashed twice with GCC 3.4.
>
>   Emacs-21.4 as deliverd by the FSF works fine using GCC 4 or
>                 GCC 3.4.

> I'm currently checking out the patches from Debian and Ubuntu, as I'm
> pretty sure these make the crashes appear.

Did you check what GCC options are used to compile Emacs in
Debian/Ubuntu?  Perhaps the target processor (-mcpu) or the
optimization level is different than what you use when you compile the
package yourself.

The last I checked, there were only harmless-looking bugfixes in the
Debian patches.

-- 
Károly

Re: emacs crash on SMP notebook only

[Richard M. Stallman]

    The last I checked, there were only harmless-looking bugfixes in the
    Debian patches.

Harmless-looking changes can cause real trouble when there is a
compiler bug.