I've compiled and ran the test (below) on a virgin 2.4.27 kernel patched with uml-patch-2.4.27-1.bz2.

Result: the bug is still readily reproducible and is not fixed in this patch.

P.S. Jeff Dike: if CONFIG_NETFILTER=y is set in .config, the patched kernel fails to compile.

Blaisorblade wrote:

On Sunday 14 November 2004 11:47, Struan Bartlett wrote:

I've now tested for this bug using a simplified methodology on the
latest patched-versions of the host (2.4.27 and 2.6.8) and guest (2.4.27
and 2.6.9).

_New methodology_

My Debian guest system raises two xterms at startup. On each I log in as
root. On one, I enter:

while true; do echo >/dev/null; done

On the second, I enter:

cat /dev/urandom

The devastating symptom: within 5 seconds the guest kernel has paniced
and exited with the message (below) "kernel BUG at sched.c:564!".

_Findings_

guest 2.4.27: patched with uml-2.4.27-bs1.patch and
uml-patch-2.4.24-1base.patch - no change; this bug is readily
reproducible and so IS NOT FIXED, regardless of whether the guest is run
on host 2.4.27 or 2.6.8

_Conclusion_

If such a simple pair of commands as this causes such devastating
consequences, it seems the 2.4.27 kernel sched.c line 564 needs to be
taken a look at - unless, that is, Jeff Dike's recent post about
uml-patch-2.4.27-1 "a nasty scheduler race fixed" addresses this bug.

I may be wrong, but I think I should have picked that patch into 2.4.27-bs1.

I'll take a look at your report - there is one patch applied on 2.6 that 
seemed unneeded on 2.4 until now - this may be reconsidered...