Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 15 Sep 2009 19:14:25 +0000 (UTC)
From:      Attilio Rao <attilio@FreeBSD.org>
To:        src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-8@freebsd.org
Subject:   svn commit: r197226 - stable/8/sys/kern
Message-ID:  <200909151914.n8FJEPGw000116@svn.freebsd.org>

next in thread | raw e-mail | index | archive | help
Author: attilio
Date: Tue Sep 15 19:14:25 2009
New Revision: 197226
URL: http://svn.freebsd.org/changeset/base/197226

Log:
  MFC r197223:
  Fix sched_switch_migrate() by assuming locks cannot be shared and a
  deadlock between 3 different threads by acquiring both runqueue locks
  when doing the migration.
  
  Please note that this is a special condition as we want this fix in
  before RC1 as we assume it is critical and so it has been handled
  as an instant-merge.  For the STABLE_7 branch, 1 week before the MFC
  is assumed.
  
  Approved by:	re (kib)

Modified:
  stable/8/sys/kern/sched_ule.c

Modified: stable/8/sys/kern/sched_ule.c
==============================================================================
--- stable/8/sys/kern/sched_ule.c	Tue Sep 15 18:39:27 2009	(r197225)
+++ stable/8/sys/kern/sched_ule.c	Tue Sep 15 19:14:25 2009	(r197226)
@@ -1749,19 +1749,19 @@ sched_switch_migrate(struct tdq *tdq, st
 	 */
 	spinlock_enter();
 	thread_block_switch(td);	/* This releases the lock on tdq. */
-	TDQ_LOCK(tdn);
-	tdq_add(tdn, td, flags);
-	tdq_notify(tdn, td);
+
 	/*
-	 * After we unlock tdn the new cpu still can't switch into this
-	 * thread until we've unblocked it in cpu_switch().  The lock
-	 * pointers may match in the case of HTT cores.  Don't unlock here
-	 * or we can deadlock when the other CPU runs the IPI handler.
+	 * Acquire both run-queue locks before placing the thread on the new
+	 * run-queue to avoid deadlocks created by placing a thread with a
+	 * blocked lock on the run-queue of a remote processor.  The deadlock
+	 * occurs when a third processor attempts to lock the two queues in
+	 * question while the target processor is spinning with its own
+	 * run-queue lock held while waiting for the blocked lock to clear.
 	 */
-	if (TDQ_LOCKPTR(tdn) != TDQ_LOCKPTR(tdq)) {
-		TDQ_UNLOCK(tdn);
-		TDQ_LOCK(tdq);
-	}
+	tdq_lock_pair(tdn, tdq);
+	tdq_add(tdn, td, flags);
+	tdq_notify(tdn, td);
+	TDQ_UNLOCK(tdn);
 	spinlock_exit();
 #endif
 	return (TDQ_LOCKPTR(tdn));



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200909151914.n8FJEPGw000116>