[PATCH] posix timers: RCU optimization for clock_gettime()

Use RCU to avoid the need to acquire tasklist_lock in the single-threaded
case of clock_gettime().  It still acquires tasklist_lock when for a
(potentially multithreaded) process.  This change allows realtime
applications to frequently monitor CPU consumption of individual tasks, as
requested (and now deployed) by some off-list users.

This has been in Ingo Molnar's -rt patchset since late 2005 with no
problems reported, and tests successfully on 2.6.20-rc6, so I believe that
it is long-since ready for mainline adoption.

[paulmck@linux.vnet.ibm.com: fix exit()/posix_cpu_clock_get() race spotted by Oleg]
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Roman Zippel <zippel@linux-m68k.org>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel/posix-cpu-timers.c

@@ -304,20 +304,25 @@ int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
 		 * should be able to see it.
 		 */
 		struct task_struct *p;
-		read_lock(&tasklist_lock);
+		rcu_read_lock();
 		p = find_task_by_pid(pid);
 		if (p) {
 			if (CPUCLOCK_PERTHREAD(which_clock)) {
 				if (p->tgid == current->tgid) {
 					error = cpu_clock_sample(which_clock,
 								 p, &rtn);
 				}
-			} else if (p->tgid == pid && p->signal) {
-				error = cpu_clock_sample_group(which_clock,
-							       p, &rtn);
+			} else {
+				read_lock(&tasklist_lock);
+				if (p->tgid == pid && p->signal) {
+					error =
+					    cpu_clock_sample_group(which_clock,
+							           p, &rtn);
+				}
+				read_unlock(&tasklist_lock);
 			}
 		}
-		read_unlock(&tasklist_lock);
+		rcu_read_unlock();
 	}
 
 	if (error)