Reproducable user mode system hang

Reproducable user mode system hang

[Charlie Brej]

I have seem to have found a method to hang the kernel from user mode. The system 
hangs and does not print an Oops. It still responds to network pings but nothing 
else.

I have successfully crashed the 2.6 kernels on three different machines (all 
athlon 2.6 kernels):
Linux rain.cs.man.ac.uk 2.6.5-1.358 #1 Sat May 8 09:04:50 EDT 2004 i686 athlon 
i386 GNU/Linux
Linux solem.cs.man.ac.uk 2.6.6-1.374 #1 Wed May 19 12:44:14 EDT 2004 i686 athlon 
i386 GNU/Linux
Linux hogshead 2.4.20-8 #1 Thu Mar 13 17:18:24 EST 2003 i686 athlon i386 GNU/Linux

This problem does not occur on any 2.4 kernel machines I have tried.

Reproducing the problem:

Unfortunately the problem occurs in the middle of a program execution and I have 
been unable to track it down.

Download kmd 0.9.19pre1 from (If anyone wants I could distribute my binarys):
http://www.cs.man.ac.uk/~brejc8/kmd/dist/KMD-0.9.19.pre1.tar.gz
In user mode configure, compile and execute from the source directory:
"./kmd -e ./jimulator"
In the memory windows in the address box type in "E1000000" and press return.
This should now crash the system.

Kmd sporns an emulator (jimulator) with which it communicates using stdin/out 
pipes. I suspect it is a problem in the pipe communication. It occurs even when 
run under valgrind. I don't know of many methods of narrowing down the search.

-- 
         Charlie Brej
APT Group, Dept. Computer Science, University of Manchester
Web: http://www.cs.man.ac.uk/~brejc8/ Tel: +44 161 275 6844
Mail: IT302, Manchester University, Manchester, M13 9PL, UK

Re: Reproducable user mode system hang

[Marcelo Tosatti]

On Thu, Aug 12, 2004 at 08:21:29PM +0100, Charlie Brej wrote:
> I have seem to have found a method to hang the kernel from user mode. The 
> system hangs and does not print an Oops. It still responds to network pings 
> but nothing else.
> 
> I have successfully crashed the 2.6 kernels on three different machines 
> (all athlon 2.6 kernels):
> Linux rain.cs.man.ac.uk 2.6.5-1.358 #1 Sat May 8 09:04:50 EDT 2004 i686 
> athlon i386 GNU/Linux
> Linux solem.cs.man.ac.uk 2.6.6-1.374 #1 Wed May 19 12:44:14 EDT 2004 i686 
> athlon i386 GNU/Linux
> Linux hogshead 2.4.20-8 #1 Thu Mar 13 17:18:24 EST 2003 i686 athlon i386 
> GNU/Linux
> 
> This problem does not occur on any 2.4 kernel machines I have tried.
> 
> Reproducing the problem:
> 
> Unfortunately the problem occurs in the middle of a program execution and I 
> have been unable to track it down.
> 
> Download kmd 0.9.19pre1 from (If anyone wants I could distribute my 
> binarys):
> http://www.cs.man.ac.uk/~brejc8/kmd/dist/KMD-0.9.19.pre1.tar.gz
> In user mode configure, compile and execute from the source directory:
> "./kmd -e ./jimulator"
> In the memory windows in the address box type in "E1000000" and press 
> return.
> This should now crash the system.
> 
> Kmd sporns an emulator (jimulator) with which it communicates using 
> stdin/out pipes. I suspect it is a problem in the pipe communication. It 
> occurs even when run under valgrind. I don't know of many methods of 
> narrowing down the search.

Can you get any kind of trace (ctrl+sysrq+p or NMI oopser) ? 

And also make sure to rerun the tests with newer v2.6's. 

Re: Reproducable user mode system hang

[Charlie Brej]

Marcelo Tosatti wrote:
> Can you get any kind of trace (ctrl+sysrq+p or NMI oopser) ? 

I will be able to tell you tomorrow when I get back to work

 > And also make sure to rerun the tests with newer v2.6's.

Well you are right. It seems to be either fixed in the newer kernel or not 
effecting the newer kernel. These two both worked fine with the redhat compiled 
2.6.7:

Linux solem.cs.man.ac.uk 2.6.7-1.517 #1 Wed Aug 11 16:28:33 EDT 2004 i686 athlon 
i386 GNU/Linux
Linux hilly.house 2.6.7-1.517 #1 Wed Aug 11 16:28:33 EDT 2004 i686 i686 i386 
GNU/Linux

Did you have an idea of what was causing this?
-- 
         Charlie Brej
APT Group, Dept. Computer Science, University of Manchester
Web: http://www.cs.man.ac.uk/~brejc8/ Tel: +44 161 275 6844
Mail: IT302, Manchester University, Manchester, M13 9PL, UK

Re: Reproducable user mode system hang

[Marcelo Tosatti]

On Thu, Aug 12, 2004 at 10:21:55PM +0100, Charlie Brej wrote:
> Marcelo Tosatti wrote:
> >Can you get any kind of trace (ctrl+sysrq+p or NMI oopser) ? 
> 
> I will be able to tell you tomorrow when I get back to work
> 
> > And also make sure to rerun the tests with newer v2.6's.
> 
> Well you are right. It seems to be either fixed in the newer kernel or not 
> effecting the newer kernel. These two both worked fine with the redhat 
> compiled 2.6.7:
> 
> Linux solem.cs.man.ac.uk 2.6.7-1.517 #1 Wed Aug 11 16:28:33 EDT 2004 i686 
> athlon i386 GNU/Linux
> Linux hilly.house 2.6.7-1.517 #1 Wed Aug 11 16:28:33 EDT 2004 i686 i686 
> i386 GNU/Linux
> 
> Did you have an idea of what was causing this?

Not a clue - but it seems to be fixed in recent kernels :)