[ofa-general] possible irq lock inversion dependency detected

Roland Dreier rdreier at cisco.com
Tue May 15 11:58:54 PDT 2007


Thanks for the report... looks like a real bug.

Can you check whether this patch makes the lockdep warnings go away?

commit 4b7eed244c032ce963be543a63e3100b96bc2d87
Author: Roland Dreier <rolandd at cisco.com>
Date:   Tue May 15 11:56:05 2007 -0700

    IB/ipath: Fix potential deadlock with multicast spinlocks
    
    Lockdep found the following potential deadlock between mcast_lock and
    n_mcast_grps_lock: mcast_lock is taken from both interrupt context and
    process context, so spin_lock_irqsave() must be used to take it.
    n_mcast_grps_lock is only taken from process context, so at first it
    seems safe to take it with plain spin_lock(); however, it also nests
    inside mcast_lock, and hence we could deadlock:
    
      cpu A                                   cpu B
        ipath_mcast_add():
          spin_lock_irq(&mcast_lock);
    
                                                ipath_mcast_detach():
                                                  spin_lock(&n_mcast_grps_lock);
    
                                                <enter interrupt>
    
                                                ipath_mcast_find():
                                                  spin_lock_irqsave(&mcast_lock);
    
          spin_lock(&n_mcast_grps_lock);
    
    Fix this by using spin_lock_irq() to take n_mcast_grps_lock.
    
    Signed-off-by: Roland Dreier <rolandd at cisco.com>

diff --git a/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c b/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c
index 085e28b..dd691cf 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c
+++ b/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c
@@ -165,10 +165,9 @@ static int ipath_mcast_add(struct ipath_ibdev *dev,
 {
 	struct rb_node **n = &mcast_tree.rb_node;
 	struct rb_node *pn = NULL;
-	unsigned long flags;
 	int ret;
 
-	spin_lock_irqsave(&mcast_lock, flags);
+	spin_lock_irq(&mcast_lock);
 
 	while (*n) {
 		struct ipath_mcast *tmcast;
@@ -228,7 +227,7 @@ static int ipath_mcast_add(struct ipath_ibdev *dev,
 	ret = 0;
 
 bail:
-	spin_unlock_irqrestore(&mcast_lock, flags);
+	spin_unlock_irq(&mcast_lock);
 
 	return ret;
 }
@@ -289,17 +288,16 @@ int ipath_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 	struct ipath_mcast *mcast = NULL;
 	struct ipath_mcast_qp *p, *tmp;
 	struct rb_node *n;
-	unsigned long flags;
 	int last = 0;
 	int ret;
 
-	spin_lock_irqsave(&mcast_lock, flags);
+	spin_lock_irq(&mcast_lock);
 
 	/* Find the GID in the mcast table. */
 	n = mcast_tree.rb_node;
 	while (1) {
 		if (n == NULL) {
-			spin_unlock_irqrestore(&mcast_lock, flags);
+			spin_unlock_irq(&mcast_lock);
 			ret = -EINVAL;
 			goto bail;
 		}
@@ -334,7 +332,7 @@ int ipath_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 		break;
 	}
 
-	spin_unlock_irqrestore(&mcast_lock, flags);
+	spin_unlock_irq(&mcast_lock);
 
 	if (p) {
 		/*
@@ -348,9 +346,9 @@ int ipath_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 		atomic_dec(&mcast->refcount);
 		wait_event(mcast->wait, !atomic_read(&mcast->refcount));
 		ipath_mcast_free(mcast);
-		spin_lock(&dev->n_mcast_grps_lock);
+		spin_lock_irq(&dev->n_mcast_grps_lock);
 		dev->n_mcast_grps_allocated--;
-		spin_unlock(&dev->n_mcast_grps_lock);
+		spin_unlock_irq(&dev->n_mcast_grps_lock);
 	}
 
 	ret = 0;



More information about the general mailing list