OSDN Git Service

IB/rdmavt: Add qp init function
authorMike Marciniszyn <mike.marciniszyn@intel.com>
Tue, 6 Sep 2016 11:37:26 +0000 (04:37 -0700)
committerDoug Ledford <dledford@redhat.com>
Sun, 2 Oct 2016 12:42:10 +0000 (08:42 -0400)
Add an rvt_qp_init() to initialize specific
common fields as the qp is created or reset.

The routine is shared by the rvt_reset_qp() and
the rvt_create_qp().

The intent is that lock dep assertions will only
appear in the rvt_reset_qp().

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
drivers/infiniband/sw/rdmavt/qp.c

index 0a29815..673e2f6 100644 (file)
@@ -493,50 +493,18 @@ static void rvt_remove_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp)
 }
 
 /**
- * reset_qp - initialize the QP state to the reset state
- * @qp: the QP to reset
+ * rvt_init_qp - initialize the QP state to the reset state
+ * @qp: the QP to init or reinit
  * @type: the QP type
- * r and s lock are required to be held by the caller
+ *
+ * This function is called from both rvt_create_qp() and
+ * rvt_reset_qp().   The difference is that the reset
+ * patch the necessary locks to protect against concurent
+ * access.
  */
-static void rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
-                 enum ib_qp_type type)
-       __must_hold(&qp->r_lock)
-       __must_hold(&qp->s_hlock)
-       __must_hold(&qp->s_lock)
+static void rvt_init_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
+                       enum ib_qp_type type)
 {
-       if (qp->state != IB_QPS_RESET) {
-               qp->state = IB_QPS_RESET;
-
-               /* Let drivers flush their waitlist */
-               rdi->driver_f.flush_qp_waiters(qp);
-               qp->s_flags &= ~(RVT_S_TIMER | RVT_S_ANY_WAIT);
-               spin_unlock(&qp->s_lock);
-               spin_unlock(&qp->s_hlock);
-               spin_unlock_irq(&qp->r_lock);
-
-               /* Stop the send queue and the retry timer */
-               rdi->driver_f.stop_send_queue(qp);
-
-               /* Wait for things to stop */
-               rdi->driver_f.quiesce_qp(qp);
-
-               /* take qp out the hash and wait for it to be unused */
-               rvt_remove_qp(rdi, qp);
-               wait_event(qp->wait, !atomic_read(&qp->refcount));
-
-               /* grab the lock b/c it was locked at call time */
-               spin_lock_irq(&qp->r_lock);
-               spin_lock(&qp->s_hlock);
-               spin_lock(&qp->s_lock);
-
-               rvt_clear_mr_refs(qp, 1);
-               /*
-                * Let the driver do any tear down it needs to for a qp
-                * that has been reset
-                */
-               rdi->driver_f.notify_qp_reset(qp);
-       }
-
        qp->remote_qpn = 0;
        qp->qkey = 0;
        qp->qp_access_flags = 0;
@@ -582,6 +550,54 @@ static void rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
 }
 
 /**
+ * rvt_reset_qp - initialize the QP state to the reset state
+ * @qp: the QP to reset
+ * @type: the QP type
+ *
+ * r_lock, s_hlock, and s_lock are required to be held by the caller
+ */
+static void rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
+                        enum ib_qp_type type)
+       __must_hold(&qp->s_lock)
+       __must_hold(&qp->s_hlock)
+       __must_hold(&qp->r_lock)
+{
+       if (qp->state != IB_QPS_RESET) {
+               qp->state = IB_QPS_RESET;
+
+               /* Let drivers flush their waitlist */
+               rdi->driver_f.flush_qp_waiters(qp);
+               qp->s_flags &= ~(RVT_S_TIMER | RVT_S_ANY_WAIT);
+               spin_unlock(&qp->s_lock);
+               spin_unlock(&qp->s_hlock);
+               spin_unlock_irq(&qp->r_lock);
+
+               /* Stop the send queue and the retry timer */
+               rdi->driver_f.stop_send_queue(qp);
+
+               /* Wait for things to stop */
+               rdi->driver_f.quiesce_qp(qp);
+
+               /* take qp out the hash and wait for it to be unused */
+               rvt_remove_qp(rdi, qp);
+               wait_event(qp->wait, !atomic_read(&qp->refcount));
+
+               /* grab the lock b/c it was locked at call time */
+               spin_lock_irq(&qp->r_lock);
+               spin_lock(&qp->s_hlock);
+               spin_lock(&qp->s_lock);
+
+               rvt_clear_mr_refs(qp, 1);
+               /*
+                * Let the driver do any tear down or re-init it needs to for
+                * a qp that has been reset
+                */
+               rdi->driver_f.notify_qp_reset(qp);
+       }
+       rvt_init_qp(rdi, qp, type);
+}
+
+/**
  * rvt_create_qp - create a queue pair for a device
  * @ibpd: the protection domain who's device we create the queue pair for
  * @init_attr: the attributes of the queue pair
@@ -761,7 +777,7 @@ struct ib_qp *rvt_create_qp(struct ib_pd *ibpd,
                }
                qp->ibqp.qp_num = err;
                qp->port_num = init_attr->port_num;
-               rvt_reset_qp(rdi, qp, init_attr->qp_type);
+               rvt_init_qp(rdi, qp, init_attr->qp_type);
                break;
 
        default: