crashme produces hang of linux

[Stefan Nunninger <stefan.nunninger@enst.fr>]

Hello,

I got a Montavista kernel 2.2.14 running on a custom
board with a MPC860 powerpc. The root file system is
mounted over nfs from a PC running an nfs server.

To check the stability of the board I was running several
tests. First I used all kind of applications I could
imagine and tried whether they are working fine. I found
no problems using anyting like that. This includes
basic programs like ls, cd, vi, tar, gzip, top, ftp, ftpd
telnet, telnetd, httpd etc.
Also the board runs for several days when used as websever
even though not under heavy load. So I felt quite confident
everything works fine.

Now I tried to verify the board's stability using crashme.
Crashme is a program that tests the stability of a operating system.
It generates random code and executes it. Obviously this will
generate all kind of errors as segmentation faults, illegal
instructions etc. That is fine and is a wanted property.
However it is expected that crashme may not crash or hang
the operating system.
Unfortunately I found that my system hangs shortly after starting
crashme. The kernel seems to work still fine as it reacts to ping
requests. However it is not possible to connect to the system
using telnet or ftp. Also the console, which is connected via the
serial port (minicom), does not react. The only solution I found
was restarting the system.

Thus this seems to indicates some stability problems on my embedded
device.
It might be that this is nothing serious as such a situation should
not occur during normal operation. Still it would be better if the
kernel would stay useable even in an extrem situation as when using
crashme.

Shurely it would be interesting to know what kind of instruction
produces the hang. There is a possibility to let the program write
a logfile in which the code that is execute is stored. After a crash
the last line in the logfile should give the instruction producing
the crash. However to use this the sync mechanism of linux has to
be switched off. Because syncing would prevent the data be written
immediately to disk. The sync buffer however will be lost after
the crash. For the case of the embedded device there is a further
problem. When the kernel crashes it crashes probably also the
network connection which is necessary for the nfs connection.
Thus quite likely the last instruction will not be transfered by
nfs. Thus I do not know which instruction produces the hang.

After all I would be interested in hearing what you think about all
that.
Do you think crashme is a useful test at all. Should I simple ignore
the result and be happy that so far no other problems occured.
Or is it probable that the board will get unstable in some rare
cases.
What might be the reason for the hang. Is there anything obvious I
should
check. As I've read several times that memory is a difficult task
with
linux I veryfied the UMPA values I'm using. As I have no logic
analyzer
at hand this was only a check for plausability of the values.

And finally has anybody done similar tests. Which further tests
should I do for stability?
Also I'd like to figure out the performance of the board.
I'm especially interested in benchmarks which give an idea of basic
values like raw processing speed, file system performance, memory
and
network performance. And finally I'd like to compare my device to
other
embedded devices and to known PC systems.

Any ideas are welcome - many thanks
	Stefan

--
Stefan Nunninger
Ecole nationale superieure des telecommunications
46, Rue Barrault
75634 Paris Cedex 13
Tel: 01 45 81 7507 (bureau)
     01 45 81 7600 (laboratoire)

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/