On Wed, Oct 22, 2003 at 02:00:50PM -0200, Felipe Almeida Lessa wrote: > Segundo os pergaminhos sagrados, algu�m chamado Daniel Jacobowitz <[EMAIL > PROTECTED]> escreveu: > > > Now, we need to find the problems I _don't_ know about. I'd like for those > > who feel comfortable testing a new glibc - usual caveats about the quality > > of experimental software! - to give it a try. Use it, let us know what > > happens. Don't upload Debian packages to the archive built against it, > > obviously - they won't be installable because of the >= 2.3.2.ds1 dependency > > in shlibs. If you have a kernel above 2.4.something, below 2.6, and an > > i686, you'll get i686 optimized libraries - er, I don't know if they'll work > > on a Via C3, I didn't check :( Probably not. If you have a kernel at least > > 2.6.0-test on an i486 or above, you'll get NPTL. > > [Please CC me, I'm subscribed to debian-devel only] > > I'm using 2.6.0-test8 (compiled myself, just don't like distro > kernels) on a K6-II (i586) and NTPL just doesn't work. After dpkg > unpackaged the new libc6, it stopped working. I tried to open > aptitude again, and it said "Illegal instruction". I tried to reboot > but init failed just like the others -- had to SysRq-S SysRq-U > SysRq-B. After the (forced) reboot, init doesn't even print any > message. > > After all these things, I tried my ol' 2.4.22. Yeah, it works! But > not optimized (i586 < i686)... So I downgraded libc6* and locales to > unstable ones.
OK, sounds like something's wrong with NPTL on a 586... could you test further? What I'd really like would be a core dump; unless core dumps are disabled (ulimit -c) you ought to have gotten one when aptitude crashed. FYI, with the new libc6 installed and 2.6.0-test8 you should be able to say "LD_ASSUME_KERNEL=2.4.22 aptitude" and that should work (disabling NPTL). So if you can get up a core file, load it up using LD_ASSUME_KERNEL=2.4.22 gdb aptitude core and give us the output of info registers disassemble -- Daniel Jacobowitz MontaVista Software Debian GNU/Linux Developer -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
