Re: [RFC] RT-patch update to remove the global pi_lock

[Steven Rostedt <rostedt@goodmis.org>]

On Tue, 2005-08-30 at 11:00 -0400, Steven Rostedt wrote:
> OK, I'm looking for a second set of eyes (or third or more :-) to see if
> there's a danger of a deadlock here.

Unless someone sees a problem with the patch, here it is.  I noticed
that I unlocked the lock->wait_lock when I should not have.  (scares me
since I'm currently running the kernel with that bug!)

Ingo,  in the pi_setprio, I can stop the loop if the owner of the BKL
has a task lock_depth >= 0 correct?  This means that the owner is either
running, or is going to go to sleep.  If it was blocked on another task,
via a mutex, then this count would be -1.  Correct?  I may need to move
the setting of lock_depth = -1 to inside the protection of the task
pi_lock.

-- Steve

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

Index: linux_realtime_goliath/kernel/rt.c
===================================================================
--- linux_realtime_goliath/kernel/rt.c	(revision 310)
+++ linux_realtime_goliath/kernel/rt.c	(working copy)
@@ -1089,11 +1089,21 @@
 		__raw_spin_unlock(&new_owner->task->pi_lock);
 		goto try_again;
 	}
+	/*
+	 * Once again the BKL comes to play.  Since the BKL can be grabbed and released
+	 * out of the normal P1->L1->P2 order, there's a chance that someone has the
+	 * BKL owner's lock and is waiting on the new owner lock.
+	 */
+	if (unlikely(lock == &kernel_sem.lock)) {
+		if (!__raw_spin_trylock(&old_owner->task->pi_lock)) {
+			__raw_spin_unlock(&new_owner->task->pi_lock);
+			goto try_again;
+		}
+	} else
 #endif
+		__raw_spin_lock(&old_owner->task->pi_lock);
+
 	plist_del_init(&waiter->list, &lock->wait_list);
-
-	__raw_spin_lock(&old_owner->task->pi_lock);
-
 	plist_del(&waiter->pi_list, &old_owner->task->pi_waiters);
 	plist_init(&waiter->pi_list, waiter->ti->task->prio);