IB/qib: Fix race between qib_error_qp() and receive packet processing
When transitioning a QP to the error state, in progress RWQEs need to
be marked complete. This also involves releasing the reference count
to the memory regions referenced in the SGEs. The locking in the
receive packet processing wasn't sufficient to prevent qib_error_qp()
from modifying the r_sge state at the same time, thus leading to
kernel panics.
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
diff --git a/drivers/infiniband/hw/qib/qib_sdma.c b/drivers/infiniband/hw/qib/qib_sdma.c
index b845688..cad4449 100644
--- a/drivers/infiniband/hw/qib/qib_sdma.c
+++ b/drivers/infiniband/hw/qib/qib_sdma.c
@@ -656,6 +656,7 @@
}
qp = tx->qp;
qib_put_txreq(tx);
+ spin_lock(&qp->r_lock);
spin_lock(&qp->s_lock);
if (qp->ibqp.qp_type == IB_QPT_RC) {
/* XXX what about error sending RDMA read responses? */
@@ -664,6 +665,7 @@
} else if (qp->s_wqe)
qib_send_complete(qp, qp->s_wqe, IB_WC_GENERAL_ERR);
spin_unlock(&qp->s_lock);
+ spin_unlock(&qp->r_lock);
/* return zero to process the next send work request */
goto unlock;