Re: Possible serious bug in glibc, assistance needed
W liście z pon, 15-12-2003, godz. 00:49, Daniel Jacobowitz pisze:
> On Mon, Dec 15, 2003 at 12:07:39AM +0100, Mathrick wrote:
> > W li?cie z nie, 14-12-2003, godz. 20:21, Daniel Jacobowitz pisze:
> >
> > > > So to sum up, it looks like libc bug because:
> > >
> > > So take a simple testcase that shows the problem, attack it with strace
> > > and gdb, and figure out what's going on.
> >
> > It's hard because it shows mostly in complex control flow scenarios,
> > like python interpreter running bittorrent or Galeon. I wasn't able to
> > come up with simple thing showing this behavior. If you have any
> > suggestions how to increase stress put on system w/o increasing code's
> > complexity too much, I'd be grateful. Also, any other suggestions how to
> > isolate the problem will be appreciated.
>
> Strace on an application which shows the problem may be enlightening.
Any ideas what I should be looking for? Also, it tends to happen mostly
on occasions when it's hard to hook strace, like swapping a lot on
session start, or randomly during usage. Some practical tips would help
a lot, I'm not too fond of possibly running strace for several hours.
> > > I don't think this could be a libc bug. The memory allocation code is
> > > some of the best, most-frequently tested code in the library and no one
> > > else has seen anything similar.
> >
> > I know of at least one other person experiencing virtually identical
> > problems on totally independent setup, he's being cc'ed. If you're
> > absolutely positive there is no way for this to be libc issue, I'd like
> > to hear your bets on what else could have similar impact on system and
> > show on wide range of apps.
>
> Could be anything. Any LD_PRELOAD'd libraries, interesting kernel
> patches?
Nope, pretty boring stuff, except maybe for devfs enabled. Also, I've
tried it on several kernels: custom 2.4.18, distro 2.4.18, distro
2.4.20, distro 2.4.21, custom 2.4.21, 2.4.21-ck, still the same. 2.4.18
is the same stuff I had going flawlessly on Woody - unlikely to be
direct kernel bug, some twisted interaction occuring on that number of
different kernels would also be pretty perverse.
> I am quite convinced that a bug of this magnitude would show up for
> more than two people. GNOME runs just fine here.
What's unusual here is that symptoms are identical for both of us. It
may be something rare triggered by some specific combination of whatever
things. I'll get some more info on the second machine, do cross
comparision, maybe that will show something.
Also a question: are you running gDesklets w/o trouble? They expose the
problem a lot, maybe it would be possible to use them as testcase.
--
"Tautologizm to coś tautologicznego"
Maciej Katafiasz <mnews2@wp.pl>
http://mathrick.blog.pl
Reply to: