Here are the newest results of my tests with "crashme". I was
quite busy the last two days running the tests on different
Linux systems, but I think the results are quite interesting:
Results of crashme tests wit Linux-2.0.34prexy:
==============================================
Side note: all systems are carefully set up: CPU cooler+fan,
thermal grease, correct CPU clocking, correctly terminated
SCSI bus. All systems under test were used for heavy stress
testing Linux-2.0.34prexy before, without a single problem.
The command used to run the test was:
% ./crashme +2000 666 100 15:00:00 3
It was executed for an unprivileged user (UID 100), using
the "bash" command-interpreter.
Systems under test and results:
==============================
1.) MB ASUS P/I-P55T2P4, 256MB RAM, AMD K6-233, 2xAHA2940UW,
2GB SCSI + 16GB UW-SCSI RAID5 (5xIBM DCAS-34330), 120MB Swap,
Linux-2.0.34pre16, egcs-1.0.2
-> Crash after about 15 minutes
2.) Notebook Panasonic CF62, 48MB RAM, Intel i586-133, 1.2GB IDE HDD
32MB Swap
-> NO crash in 15 hours (about 114000 Processes)
3.) MB ASUS P/I-P55T2P4, 64MB RAM, AMD K6-200, 2GB IDE HDD,
64MB Swap, Linux-2.0.34pre15, egcs-1.0.2
-> Crash after about 1 hour
4.) MB ASUS P/I-P55T2P4, 32MB RAM, AMD K6-200, 2GB IDE HDD,
64MB Swap, Linux-2.0.34pre15, egcs-1.0.2
(same system as 3., but booted with "mem=32M")
-> Crash after about 17000 Processes
5.) MB ASUS P/I-P55T2P4, 32MB RAM, Intel i586-100, 2.5GB IDE HDD,
120MB Swap, Linux-2.0.34pre16+ISDN patches (maximum configuration),
egcs-1.0.2
-> NO Crash in 15 hours (about 137000 Processes)
6.) MB ASUS P/I-P55T2P4, 256MB RAM, AMD K6-233, 2xAHA2940UW,
2GB SCSI + 16GB UW-SCSI RAID5 (5xIBM DCAS-34330), 120MB Swap,
Linux-2.0.34pre16, gcc-2.7.3.2
(same system as 1., but kernel compiled with GNU GCC-2.7.3.2)
-> Crash after about 15 minutes
7.) MB ASUS P/I-P55T2P4, 64MB RAM, AMD K6-200, 1.2GB IDE HDD,
120MB Swap, Linux-2.0.34pre16, egcs-1.0.2
-> Crash after about 15 minutes
8.) MB ASUS P/I-P55T2P4, 32MB RAM, Intel i586-133, 1.2GB IDE HDD,
64MB Swap, Linux-2.0.34pre15, egcs-1.0.2
-> NO Crash after 3 hours (about 28000 Processes)
9.) MB ASUS P/I-P55T2P4, 256MB RAM, Intel i586-200 MMX, 2xAHA2940UW,
2GB SCSI + 16GB UW-SCSI RAID5 (5xIBM DCAS-34330), 120MB Swap,
Linux-2.0.34pre16, gcc-2.7.3.2
(same system as 6., but with Intel CPU)
-> NO Crash after 3 hours (about 30000 Processes), and still
running...
Interesting facts:
=================
1.) All systems with AMD K6 CPU crashed
2.) No system with Intel Pentium CPU crashed
3.) Systems crashed regardless of compiler used to compile kernel
(egcs-1.0.2 and gcc-2.7.2.3)
4.) Systems crashed regardless of kernel configuration
5.) Systems crashed regardless of RAM size (from 32MB to 256MB)
6.) Systems crashed regardless of RAM:Swap ratio (from 2:1 to 1:2)
7.) Systems crashed regardless of disk subsystem (IDE and SCSI)
Intrepretation:
==============
It seems the problem is the AMD CPU, not Linux!
I think most impressive is test #6 compared to test #9:
both tests use the same, quite advanced, setup, except the CPU:
With AMD K6 it crashed within 15 minutes, with Intel Pentium it
didn't crash in more than 3 hours!
It seems, "crashme" triggers some problem within the AMD CPU, maybe
some illegal code halts or crashes the CPU or something like that.
(Doesn't this sound familiar? We all remember the problems AMD CPU's
had about a year ago!)
The AMD CPU's in test are quite recent one's, though.
I don't have the production dates at hand right now, but I'm
quite sure they aren't older than 6 months.
Further actions:
===============
I would like to ask everyone to prove this interpretation! Please,
run crashme on a Linux box where it doesn't matter if it locks up,
and mail me your results. Include the type of CPU this system has.
Especially I would be interested in the following cases:
a) AMD K6 CPU's not crashing
b) Intel Pentium CPU's crashing
In these cases please report any detail you think might be interesting
(CPU production dates, type of motherboard, RAM, Cache, software versions)
If you can't find crashme on a ftp-server near you, you can get it from
ftp://ftp.xss.co.at/pub/security/crashme-2.4.tar.gz
I'll also try to add a logging facility to "crashme". Maybe I can find
the illegal instructions, which crash the AMD CPU.
What do you all think about this? Am I hunting ghosts or could this
be a real, new K6 problem?
- andreas
-- Andreas Haumer | email: andreas@xss.co.at | PGP key available *x Software + Systeme | phone: +43.1.6001508 | on request. Buchengasse 67/8 | +43.664.3004449 | A-1100 Vienna, Austria | fax: +43.1.6001507 |- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu