The following patch, against 2.4.19-pre8-ac4, implements the task CPU
affinity syscalls found in 2.5.  This is not suggested for inclusion in
2.4 until the 2.5 interface is assuredly what we want.  I post it now
for discussion and those who are interested.

The syscalls are:

        sched_getaffinity(pid_t pid, unsigned long len, unsigned long *mask)
        sched_setaffinity(pid_t pid, unsigned long len, unsigned long *mask)

which get and set the affinity of any given task (provided proper
permission) to the given CPU bitmask.

I believe we solved the locking issue I presented before: handle it as
in 2.5.  Acquire tasklist_lock, translate the pid into a valid
task_struct, call get_task_struct to bump the task_struct's page
reference count, drop tasklist_lock, call set_cpus_allowed, then call
free_task_struct which does free_pages.  Since the pages usage count was
bumped, it will not be freed unless the task prematurely deallocated in
the middle.

Comments?  Hopefully this works and - assuming the 2.5 interface stays
as is - I can put this aside until it is time to merge it.

	Robert Love