Procstate patches for RedHat's 2.4.20 (+patches) kernel
--------------------------------------------------------

This is not a port of pset to newer kernels.  It is a new system call pair
for Linux which makes some of the pset functionality possible.

	int sched_getprocstate(int cpu, int *state);
	int sched_setprocstate(int cpu, int state);

These calls allow you to handle the state of a processor.  States are
currently defined as:

	PROC_ENABLED		can run any task
	PROC_RESTRICTED		can only run tasks that have their affinity
				(see sched_setaffinity) set to that CPU

In this directory you will find the kernel patch against Red Hat kernel
2.4.20-8.  I think it all works.  It is significantly reduced in size and
complexity from the Red Hat 8.0 release due to a major semantic change.

In the RH8 version (and earlier RH9 versions), a task which was not
eligible on ANY processor would be blocked until a CPU became available.
This was VERY fiddly, and had a major problem - a task might be holding a
semaphore when it became unrunnable.  Livelocks ensued.  So after
discussion with several Linux kernel luminaries, and folks at Veraxx (who
funded this), we changed the unrunnable semantics.

A task which becomes unrunnable will have it's affinity set to -1 (any CPU
may run the task) and will be sent a SIGPWR.  The default behavior for a
process receiving SIGPWR is to die.