IB/ipath: Fix potential deadlock with multicast spinlocks
authorRoland Dreier <rolandd@cisco.com>
Sat, 19 May 2007 15:51:53 +0000 (08:51 -0700)
committerRoland Dreier <rolandd@cisco.com>
Sat, 19 May 2007 15:51:53 +0000 (08:51 -0700)
commit8b8c8bca3a63073bac20f0fca178e00fdf7f5a09
tree59c7f4b6a502e99e620d20d9c251fdebb79cc34a
parent7b82cd8ee7374f803a3daf9a6cbc6eb4bbb10a63
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@cisco.com>
drivers/infiniband/hw/ipath/ipath_verbs_mcast.c