2003-11-25 19:45:54

by Ali Akcaagac

[permalink] [raw]
Subject: 2.6.0-preX causes memory corruption

Hello,

Sorry for the subject but I have been noticing this problem for quite
some time now under 2 totally different machines and I now belive that
this may be a Kernel issue.

Machine 1 (old):

Elitegroup k7s5a, 256mb, g400, 15gb

After installing 2.6.0-pre9 the System seemed to work normally, all the
stuff I did before worked normally but when doing large fileoperation
including crunching stuff using bzip2 (e.g. checking out modules from
CVS and tar'ing them up) the archives get corrupt. I was first assuming
that this was a onetime mistake and thus I deleted the corrupt file and
re-run my normal operations. But after a while I noticed that this
problem occoured more and more and I was starting to worry. Archives are
showing to be corrupted but after an reset these archives can be
unpacked normally again.

I was really worrying myself if this could be my machine e.g. defective
Ram modules or something thus I ran memtest86 for 3 passes and the
memory was ok. Later on after this problem showed up again I thought
that this may be something else e.g. Motherboard (dying capacitors) my
CPU or whatever.

Anyways I bought totally new hardware (not only because of the problem,
because I wanted to do this anyways and this problem was a good excuse
to go for new stuff 2 days later). So I bought brand new mobo, ram,
harddisk and stuff like that and build the system up:

Machine 2 (new):

Shuttle AK39N, Radeon 9200, 512mb, 40gb Harddisk

Updated my System to 2.6.0-pre10 but the problem still exists. I'm now
really worried what the cause of this problem could be. I know there
could be dozen of reasons such as compiler used, stuff compiled and
things like this but the really strange thing is that I was doing CVS
(and tar'ing up stuff) on a daily basis even with my old machine and
earlier 2.6.0-preX kernels without any problems, the only stuff that I
consistently update is mainly GNOME or KDE, basically NO system updates
happened during 2.6.0-pre5/6 and all in all my system is quite clean.

I know all this is quite vague but now that I totally changed 2
different hardware and that this problem showed up first time with
2.6.0-pre9 (and exists in 10) I may have the tendency to shift this
problem to the Kernel. Also normal operations such as compiling stuff
sometimes end in e.g. telling me that libraries are NOT ELF or that
libraries show wrong stuff etc. and a normal reset usually solved this
which tells me that this is not filesystem related and that the files
itself are in best shape.

Anyways I hope I didn't caused any worries or something but I thought to
let you know about this issue, chances may be that this may indeed be a
kernel issue. May or may not...

greets.

PS: CC me, I'm not subscribed.


2003-11-25 20:09:32

by Måns Rullgård

[permalink] [raw]
Subject: Re: 2.6.0-preX causes memory corruption

Ali Akcaagac <[email protected]> writes:

> After installing 2.6.0-pre9 the System seemed to work normally, all the

You mean 2.6.0-test9, don't you?

> stuff I did before worked normally but when doing large fileoperation
> including crunching stuff using bzip2 (e.g. checking out modules from
> CVS and tar'ing them up) the archives get corrupt. I was first assuming
> that this was a onetime mistake and thus I deleted the corrupt file and
> re-run my normal operations. But after a while I noticed that this
> problem occoured more and more and I was starting to worry. Archives are
> showing to be corrupted but after an reset these archives can be
> unpacked normally again.

Do you have preemptive kernel enabled (CONFIG_PREEMPT=y)? There's
been some discussion about it possibly causing strange things in some
configurations. If it helps to disable it, please post your .config,
so we can compare with others.

--
M?ns Rullg?rd
[email protected]

2003-11-25 20:26:06

by Ali Akcaagac

[permalink] [raw]
Subject: Re: 2.6.0-preX causes memory corruption

> > After installing 2.6.0-pre9 the System seemed to work normally,
> > all the

> You mean 2.6.0-test9, don't you?

Yes

> > stuff I did before worked normally but when doing large
> > fileoperation including crunching stuff using bzip2 (e.g.
> > checking out modules from CVS and tar'ing them up) the
> > archives get corrupt. I was first assuming that this was
> > a onetime mistake and thus I deleted the corrupt file and
> > re-run my normal operations. But after a while I noticed
> > that this problem occoured more and more and I was starting
> > to worry. Archives are showing to be corrupted but after an
> > reset these archives can be unpacked normally again.

> Do you have preemptive kernel enabled (CONFIG_PREEMPT=y)?

Yes

> There's been some discussion about it possibly causing strange
> things in some configurations. If it helps to disable it, please
> post your .config, so we can compare with others.

Yes, saw it after I've posted my mail. But here is my config.

http://www.akcaagac.com/.config

Hope it helps.

2003-11-26 05:52:02

by Ali Akcaagac

[permalink] [raw]
Subject: Re: 2.6.0-preX causes memory corruption

> > Sorry for the subject but I have been noticing this problem for
> > quite some time now under 2 totally different machines and I now
> > belive that this may be a Kernel issue.

> Do you have pre-emption enabled, by any chance? If so, can you
> reproduce these problems with it disabled?

Yes it's enabled, will disable it now and report back.

2003-11-26 06:00:09

by Ali Akcaagac

[permalink] [raw]
Subject: Re: 2.6.0-preX causes memory corruption

> > to go for new stuff 2 days later). So I bought brand new mobo,
> > ram, harddisk and stuff like that and build the system up:

> new power supply?

Yes that was the first thing I've replaced. My old 250W for a new 433W
although I must admit that the old PSU still works fine as it did for
the past 3.5 years for my old Gigabyte BX2000 and old Elitegroup K7S5A
(including all the stuff like HD, CD, GRAKA etc.) it consumed less than
250W. I'm educated electronics guy and tested the old PSU under stress
conditions e.g. with consumer connected only to check whether the
voltage or current break down or something. The new PSU should satisfy
my system more than one time.

> > really worried what the cause of this problem could be. I know
> > there could be dozen of reasons such as compiler used, stuff
> > compiled ...

> and which compiler did you use?

gcc (GCC) 3.3.2

But I doubt that this is Compiler related since I use this one even
before pre9 (where the system worked stable). This problem showed up the
first time within pre9 and still exists in pre10 and now that other
folks confirmed that there is some 'strange magic' inside the Kernel I
now feel more that this may be the case. It's no permanent thing or
something it happens every now and then.

I now gonna disable preempt and report back.

2003-11-26 09:50:57

by Michael Buesch

[permalink] [raw]
Subject: Re: 2.6.0-preX causes memory corruption

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Tuesday 25 November 2003 21:26, Ali Akcaagac wrote:
> > There's been some discussion about it possibly causing strange
> > things in some configurations. If it helps to disable it, please
> > post your .config, so we can compare with others.
>
> Yes, saw it after I've posted my mail. But here is my config.
>
> http://www.akcaagac.com/.config
>
> Hope it helps.

I ran many tests on almost any testX kernel, but it's
rock-stable for me. CONFIG_PREEMPT is enabled, so I thought
I'll send my .config, so you can compare.

My build environment is:

Linux lfs 2.6.0-test10-bttvpatch #2 Mon Nov 24 18:15:25 CET 2003 i686 unknown

Gnu C 3.3.2
Gnu make 3.80
util-linux 2.12
mount 2.12
module-init-tools 0.9.15-pre4
e2fsprogs 1.27
reiserfsprogs 3.6.8
isdn4k-utils 3.2p1
nfs-utils 1.0.5
Linux C Library 2.3.1
Dynamic linker (ldd) 2.3.1
Linux C++ Library 5.0.5
Procps 3.1.5
Net-tools 1.60
Kbd 1.06
Sh-utils 5.0
Modules Loaded nvidia ne2k_pci 8390

(it's stable with and without nvidia loaded)

- --
Regards Michael Buesch [ http://www.tuxsoft.de.vu ]
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)

iD8DBQE/xHNXoxoigfggmSgRAlS7AJ91TWk7u5k8t8fzZRrPHCVbibI9xwCfdSYn
p7Nc/5I/qaP79kAzYYWUmJM=
=m6iH
-----END PGP SIGNATURE-----


Attachments:
(No filename) (1.44 kB)
.config (21.62 kB)
Download all attachments