The attached patch fixes problems with the "complex" macros in the
scheduler, as discussed about a week ago with Ingo on this mailing list.
On some platforms (at least IA64, sparc64) the RQ lock must be released
before the context switch in order to avoid a potential deadlock. The
macros used in 2.5.26 are buggy and allow the load_balancer to steal the
prev task right before the context switch occurs. If the prev task
exits in the mean time, the kernel crashes.

These fixes are already in Ingo's patches for the SCHED_BATCH extension
but didn't make it into 2.5.26. They are a MUST for the platforms using
complex macros, please consider applying...

Regards,
Erich