o	Parse a "--skipcpus n" argument that allows tests to be run in
	parallel on a single machine.  Also a "--limcpus" to limit
	the number that may be used.

o	Parse a "--threads n" argument that captures # threads to be
	created in multithreaded tests.

o	Parse a "--processes n" argument that captures # processes per
	process group for process-group tests.

o	Make mapmem() correctly handle pathname arguments.
	Maybe also provide a way to create the file.

o	Additional tests:

	o	sigpg_exit, sigpg_exec, sigpg_leaderexit

	o	sigpt_exit, sigpt_exec, sigpt_fork, sigpt_create,
		sigpt_reparent

	o	sigcatch variants of above.

	o	sigign variants of above.

	x	sig_reparent, sigpg_forkstorm, sig_exec

o	Run some tests on PREEMPT_RT.  Upon first failure, publish to LKML.


x	Make sig_exit have a separate process do the kill in order to
	test race with cleanup as well as exit() itself.

x	Consider creating an array indexed by (pid-childpid) so as
	to explicitly track all forkstormed processes without needing
	atomic instructions.  This would allow faster checking for
	processes surviving a fatal group signal.

	Rejected -- adds obnoxious dependency on consecutive PID
	numbering.  The current non-atomic manipulations are not
	a big deal -- it is all approximate anyway.  That said, it
	would be good to have user-level atomic instructions in a
	CPU-independent form!!!

x	Add "_" in name for existing tests.

x	Parse a "--verbose" argument that prints progress.

x	Have some way to get additional CPUs for tests needing them
	(e.g., fork-storm tests).  Keep track of available CPUs.

x	pass the childspin down as an argument?

x	separate common functions into common.h or wherever.

x	reduce overkill 33791 after: 52639

x	Create a combined do_steamroller() that finds and rolls the race.

x	Create a makefile.

