2005-02-04 19:17:15

by Gary Smith

[permalink] [raw]
Subject: Post install 2.4.29 causes many apps to seg fault.

Hello,

I have been running RHEL3 update 3 for some time and need to patch netfilter
for PPTP. After doing so and installing the kernel I found that certain
applications (such as MySQL, nslook, etc) began to segfault. Rolling the
kernel back fixed the problem.

I have since then gone back and recompiled the vanilla 2.4.29 kernel (without
additing any patches this time - clean from tarball) and installed it and all
of the the applications that failed on the custom kernel (with the PPTP
patches) continue to fail (clean box as well).

Is there something more that I need to compile besides the kernel for
compatability or is this a sign of some type of bug. I do realize that RHEL3
itself has some proprietary items added to their kernel but replacing it
shouldn't make other applications fails.

Any assistance would be greatly appreciated.

Gary Smith



2005-02-04 19:45:12

by jurriaan

[permalink] [raw]
Subject: Re: Post install 2.4.29 causes many apps to seg fault.

From: Gary Smith <[email protected]>
Date: Fri, Feb 04, 2005 at 11:10:19AM -0800
> Hello,
>
> I have been running RHEL3 update 3 for some time and need to patch netfilter
> for PPTP. After doing so and installing the kernel I found that certain
> applications (such as MySQL, nslook, etc) began to segfault. Rolling the
> kernel back fixed the problem.
>
> I have since then gone back and recompiled the vanilla 2.4.29 kernel (without
> additing any patches this time - clean from tarball) and installed it and all
> of the the applications that failed on the custom kernel (with the PPTP
> patches) continue to fail (clean box as well).
>
> Is there something more that I need to compile besides the kernel for
> compatability or is this a sign of some type of bug. I do realize that RHEL3
> itself has some proprietary items added to their kernel but replacing it
> shouldn't make other applications fails.
>
> Any assistance would be greatly appreciated.
>
If lots of things fail in strange places, I would wonder if your glibc
is compatible with your kernel. I suggest you take it up on a redhat
mailinglist.

Good luck,
Jurriaan
--
Maybe I shall start calling my tricorder Sally
DS9 - Miles O'Brien
Debian (Unstable) GNU/Linux 2.6.10-mm1 2x6078 bogomips load 0.45

2005-02-04 20:50:43

by Nick Warne

[permalink] [raw]
Subject: Re: Post install 2.4.29 causes many apps to seg fault.

>> Is there something more that I need to compile besides the kernel for
>> compatability or is this a sign of some type of bug. I do realize that
RHEL3
>> itself has some proprietary items added to their kernel but replacing it
>> shouldn't make other applications fails.
>>
>> Any assistance would be greatly appreciated.
>>
> If lots of things fail in strange places, I would wonder if your glibc
> is compatible with your kernel. I suggest you take it up on a redhat
> mailinglist.

Yes, I had a similar problems at work when I up2date'd the latest GLIBC for
RHEL 3 late last year. A colleague in Montreal (I am in UK) sussed what was
going on. I will _presume_ you are seeing similar problems with a kernel
build.

Here is the link that explains it... what to do with many processes
segfaulting, I don't know. RHEL support is _very_ good - give them a ring.

http://people.redhat.com/drepper/assumekernel.html

Nick
--
"When you're chewing on life's gristle,
Don't grumble, Give a whistle..."

2005-02-05 02:09:50

by Gary Smith

[permalink] [raw]
Subject: Re: Post install 2.4.29 causes many apps to seg fault.

Quoting Nick Warne <[email protected]>:

> Yes, I had a similar problems at work when I up2date'd the latest GLIBC for
>
> RHEL 3 late last year. A colleague in Montreal (I am in UK) sussed what was
>
> going on. I will _presume_ you are seeing similar problems with a kernel
> build.
>
> Here is the link that explains it... what to do with many processes
> segfaulting, I don't know. RHEL support is _very_ good - give them a ring.
>
> http://people.redhat.com/drepper/assumekernel.html
>
> Nick

Nick,

The article seems to make sense about the versioning. I was wondering what
your resolution was.

Gary

2005-02-05 02:24:31

by Barry K. Nathan

[permalink] [raw]
Subject: Re: Post install 2.4.29 causes many apps to seg fault.

On Fri, Feb 04, 2005 at 11:10:19AM -0800, Gary Smith wrote:
> Hello,
>
> I have been running RHEL3 update 3 for some time and need to patch netfilter
> for PPTP. After doing so and installing the kernel I found that certain
> applications (such as MySQL, nslook, etc) began to segfault. Rolling the
> kernel back fixed the problem.
>
> I have since then gone back and recompiled the vanilla 2.4.29 kernel (without
> additing any patches this time - clean from tarball) and installed it and all
> of the the applications that failed on the custom kernel (with the PPTP
> patches) continue to fail (clean box as well).
>
> Is there something more that I need to compile besides the kernel for
> compatability or is this a sign of some type of bug. I do realize that RHEL3
> itself has some proprietary items added to their kernel but replacing it
> shouldn't make other applications fails.

RHEL3 glibc (and some application binaries IIRC) assume the existence of
futex support in the kernel. Upstream, futex support only exists in 2.6,
but Red Hat backported it to their 2.4 kernels for Red Hat 9, RHEL 3,
and Fedora Core 1.

Red Hat's assumption with RHEL is that you will be running *their* kernel
so it's OK for them to make their glibc and application binaries depend
on their kernel.

If you must run a vanilla kernel, it would be best to use RHEL 2.1
instead, or to run a 2.6 kernel. Note that you'll need to install
module-init-tools to get modules working under 2.6. One way to do this
is to install a newer "modutils" package (I *think* the one from Fedora
Core 2 should work, but I don't remember for certain) -- despite the name,
the modutils package from Fedora Core 2 actually has module-init-tools in
it as well.

You may run into other gotchas when trying to run 2.6 on RHEL3; in that
case it may be easiest to wait for RHEL4 (Valentine's Day is rumored),
as that will work with 2.6 out-of-the-box and RHEL4's 2.6 kernel will
likely be far less heavily patched than RHEL3's 2.4 kernel. If you can't
wait for RHEL4, you could try the RHEL4 beta or Fedora Core in the
meantime.

If you *really* need to run a vanilla 2.4 kernel on RHEL3, you could try
installing the glibc packages from Fedora Core 1 or 2 (*maybe* FC3 glibc
packages would work too, but I'm less sure about that). Then recompile
any applications that still segfault or freeze with a vanilla kernel
(note that, for all I know, rpm itself could be one of these!)

I hope this helps...

-Barry K. Nathan <[email protected]>

2005-02-05 12:59:14

by Nick Warne

[permalink] [raw]
Subject: Re: Post install 2.4.29 causes many apps to seg fault.

On Saturday 05 February 2005 02:01, Gary Smith wrote:
> Quoting Nick Warne <[email protected]>:
> > Here is the link that explains it... what to do with many processes
> > segfaulting, I don't know. RHEL support is _very_ good - give them a
> > ring.
> >
> > http://people.redhat.com/drepper/assumekernel.html
> >
> > Nick
>
> Nick,
>
> The article seems to make sense about the versioning. I was wondering what
> your resolution was.

I run two boxes in the UK with RHEL 3 for DNS/DHCP running Lucent's QIP/QMS
services (Montreal admins run that). All I am is local SysAdmin on all the
other stuff.

When I up2dated GLIBC the only thing that went wonky at first was
smartmontools (built from src), everything else appeared OK. About 4 days
later, I was asked to check why one of the QIP monitoring tools wasn't
running (it's a JVM thing, precompiled binaries). I then called the Montreal
Linux guy, and he sussed it and told me it was the assume_kernel problem.

As it is their area, I never asked what he done exactly to fix it, but he did
say that although the GLIBC upgrade broke the tool, once he fixed the
problem, it was now working correctly using pthreads (or something) that it
wouldn't use before the upgrade. I will ask for more info Monday when back
at work.

I think Barry K. Nathan reason/problem/solution looks more likely though, re
futex in this thread.

Nick
--
"When you're chewing on life's gristle,
Don't grumble, Give a whistle..."