>Date: Tue, 25 Jan 2000 13:15:52 +0100
>From: Jaromir Dolecek <dolecek@ics.muni.cz>
>Reply-To: dolecek@ics.muni.cz
>X-Accept-Language: cs,en
>CC: mason@acheron.middleboro.ma.us, current-users@netbsd.org
>Content-Type: text/plain; charset=iso-8859-2
>
>maximum entropy wrote:
>> >Date: Mon, 24 Jan 2000 12:23:59 -0500
>> >From: Mason Loring Bliss <mason@acheron.middleboro.ma.us>
>> >
>> >In the past couple days I've also seen some stuff like, for instance, gcc
>> >dying with a "signal 10." I haven't seen this very recently, though - I
>> >believe I'm running a new kernel since the last time I saw this.
>
>> Is it possible that this heavier use of the math coprocessor is
>> tickling a latent kernel bug?
>> 
>> I nuked /etc/ld.so.conf on my system, and it seems healthy.  But with
>> this sort of nondeterministic lossage it's really hard to say, as the
>> system could go berzerk 2.3 seconds after I send this message...
>
>Note that e.g. egcs/gcc doesn't use libm at all.

I'm aware of that.  But it seems to have a lot to do with
concurrency.  So I was thinking more along the lines of a different
process triggering the bug, and with cc1 being the victim.  Or some
other process -- I've seen awk, install, make and a few other programs
get killed in the same way.

I'm running kde, so there are always several active programs linked
with libm:  XF86_SVGA, kdm, kcontrol, kattraction.kss, kcmikbd...

>Perhaps it's bug in ld_elf.so only .... 

Could be...

>I use libm compiled with USE_I387 on and I haven't
>noticed any problems so far, FWIW.

I can't make enough sense of the Makefiles in /usr/src/lib to
determine whether USE_I387 would cause precisely the same routines to
be used (as a matter of fact, I can't even find any references to
USE_I387 at all).

In any case, my system just finished a "make -j 12 build" without
incident...I'm really becoming convinced that I've stumbled onto
something.

--
entropy