[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#507845: Dedibox server with VIA C7 cpu freeze with the last libc6



Aurelien Jarno a écrit :
On Tue, Jan 06, 2009 at 10:53:34PM +0100, Maelvon HAWK wrote:
Aurelien Jarno a écrit :
On Fri, Dec 05, 2008 at 09:26:06PM +0100, Maelvon HAWK wrote:
Aurelien Jarno a écrit :
Maelvon HAWK a écrit :
Aurelien Jarno a écrit :
tag 507845 + unreproducible
tag 507845 + moreinfo
thanks

Maelvon HAWK a écrit :
Package: libc6
Version: 2.3.6.ds1-3
This is a very old version. Try using at least the latest version from
stable, that is 2.3.6.ds1-13etch7.
If I put the stable one, it freeze, with no clues…

Severity: important

I've a Dedibox server, with a VIA C7 cpu, running a debian ETCH, and it seems I've done a dist-upgrade without rebooting that's put the mess in the server. After that it freeze when it want, for no special apparent reason. I said it
seems because I remember to have done that, and found some links about that.

The actual solution I found is to downgrade my libc6, specifing some preferences in /etc/apt/preferences and the server is working well now, but I never update the libc6 or the bug comes back.

But I've tested to install mod_python today and the libc6 downgrade seems to impact it. I've done a remove of the
package "php5-recode" to make the mod_python running.

In the other hand, I've found a similar bug report in Gentoo [1], if it can help to resolve this.
As clearly explained in the gentoo bug report, this is not a glibc
problem, but rather a kernel problem. The glibc may trigger a kernel
bug, but the bug has to be fixed in the kernel.

First of all, please try to use a standard Debian kernel instead of the
kernel 2.6.24.2dedibox-r8-1-c7 one, which is known to be problematic. If
you are still be able to reproduce the bug with a Debian kernel, I'll
reassign the bug to the kernel package.

Aurelien

Which kernel should I install. An "apt-cache search linux-image-2.6" give me :

linux-image-2.6-openvz-686 - Linux 2.6 image on PPro/Celeron/PII/PIII/P4
linux-image-2.6-vserver-686-bigmem - Linux 2.6 image on PPro/Celeron/PII/PIII/P4
linux-image-2.6.26-bpo.1-486 - Linux 2.6.26 image on x86
linux-image-2.6.26-bpo.1-686 - Linux 2.6.26 image on PPro/Celeron/PII/PIII/P4 linux-image-2.6.26-bpo.1-686-bigmem - Linux 2.6.26 image on PPro/Celeron/PII/PIII/P4
linux-image-2.6.26-bpo.1-amd64 - Linux 2.6.26 image on AMD64
linux-image-2.6.26-bpo.1-openvz-686 - Linux 2.6.26 image on PPro/Celeron/PII/PIII/P4, OpenVZ support linux-image-2.6.26-bpo.1-vserver-686 - Linux 2.6.26 image on PPro/Celeron/PII/PIII/P4, Linux-VServer support linux-image-2.6.26-bpo.1-vserver-686-bigmem - Linux 2.6.26 image on PPro/Celeron/PII/PIII/P4, Linux-VServer support
linux-image-2.6.26-bpo.1-xen-686 - Linux 2.6.26 image on i686
linux-image-2.6.23.13dedibox-r8 - Linux kernel binary image for version 2.6.23.13dedibox-r8
You should use the stable one, either:
- linux-image-2.6.18-6-686
- linux-image-2.6.24-etchnhalf.1-686

I had a dedibox with a VIA C7 CPU running the first one and glibc
2.3.6.ds1-13etchX for more than a year without any problem.

Ok, first freeze!
With the linux-image-2.6.18-6-686, what's the next step?
Nothing at all in the /var/log/messages.

I'll test now the "2.6.24.2-by-myckeul"[1] kernel.

It's a production server so I should found a solution for this bug or I return to the last solution, the downgrade of libc6.

Now that we now it freeze with a normal kernel, the best is probably to
determine which version introduced the change. We really have to know
the exact version that has introduced the change, because comparing the
sources haven't shown anything, so we will have to compare the generated
assembly code.

First of all, do you have libc6-i686 installed? If yes, it may be worth
removing it and see if the problem still occurs.

If I understand correctly, the version 2.3.6.ds1-3 is the latest known version that works. The stable version is 2.3.6.ds1-13etch6. That means
there is 16 different versions in between (ds1-4 to ds1-13 and
ds1-13etch1 to ds1-13etch6). Using a dichotomy process, you can find the
first bad version in 4 tries.

You can access the previous versions of the glibc by adding this line to
/etc/apt/sources.list:

deb http://snapshot.debian.net/archive pool glibc

then run:
# apt-get update
# apt-get install libc6=version libc6-dev=version locales=version

When I know the first bad version, I'll try to see what has changed at
the binary level.

Aurelien,

Thanks for your time, but I've passed to an unstable version of debian, for libc6, and especially for the python2.5.2 library.

And how does it work? Do you still see the freezes?


I keep you in touch if it evolves, but for now I'm at 8 days uptime and no freeze, crossing fingers.

Linux 2.6.24-etchnhalf.1-686
libc6:		Installé : 2.7-16
libc6-dev: 	Installé : 2.7-16
libc6-i686:	Installé : 2.7-16
locales:	Installé : 2.7-16




Reply to: