IB/ehca: Refactor sync between completions and destroy_cq using atomic_t

- ehca_cq.nr_events is made an atomic_t, eliminating a lot of locking.
- The CQ is removed from the CQ idr first now to make sure no more
  completions are scheduled on that CQ. The "wait for all completions to
  end" code becomes much simpler this way.

Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

1 files changed, 17 insertions, 19 deletions

diff --git a/drivers/infiniband/hw/ehca/ehca_irq.c b/drivers/infiniband/hw/ehca/ehca_irq.c
index 100329b..3e790a3 100644
--- a/drivers/infiniband/hw/ehca/ehca_irq.c
+++ b/drivers/infiniband/hw/ehca/ehca_irq.c
@@ -5,6 +5,8 @@
  *
  *  Authors: Heiko J Schick <schickhj@de.ibm.com>
  *           Khadija Souissi <souissi@de.ibm.com>
+ *           Hoang-Nam Nguyen <hnguyen@de.ibm.com>
+ *           Joachim Fenkes <fenkes@de.ibm.com>
  *
  *  Copyright (c) 2005 IBM Corporation
  *
@@ -212,6 +214,8 @@ static void cq_event_callback(struct ehca_shca *shca,
 
 	spin_lock_irqsave(&ehca_cq_idr_lock, flags);
 	cq = idr_find(&ehca_cq_idr, token);
+	if (cq)
+		atomic_inc(&cq->nr_events);
 	spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 
 	if (!cq)
@@ -219,6 +223,9 @@ static void cq_event_callback(struct ehca_shca *shca,
 
 	ehca_error_data(shca, cq, cq->ipz_cq_handle.handle);
 
+	if (atomic_dec_and_test(&cq->nr_events))
+		wake_up(&cq->wait_completion);
+
 	return;
 }
 
@@ -414,25 +421,22 @@ static inline void process_eqe(struct ehca_shca *shca, struct ehca_eqe *eqe)
 		token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe_value);
 		spin_lock_irqsave(&ehca_cq_idr_lock, flags);
 		cq = idr_find(&ehca_cq_idr, token);
+		if (cq)
+			atomic_inc(&cq->nr_events);
+		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 		if (cq == NULL) {
-			spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 			ehca_err(&shca->ib_device,
 				 "Invalid eqe for non-existing cq token=%x",
 				 token);
 			return;
 		}
 		reset_eq_pending(cq);
-		cq->nr_events++;
-		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 		if (ehca_scaling_code)
 			queue_comp_task(cq);
 		else {
 			comp_event_callback(cq);
-			spin_lock_irqsave(&ehca_cq_idr_lock, flags);
-			cq->nr_events--;
-			if (!cq->nr_events)
+			if (atomic_dec_and_test(&cq->nr_events))
 				wake_up(&cq->wait_completion);
-			spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 		}
 	} else {
 		ehca_dbg(&shca->ib_device, "Got non completion event");
@@ -478,15 +482,15 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 			token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe_value);
 			spin_lock(&ehca_cq_idr_lock);
 			eqe_cache[eqe_cnt].cq = idr_find(&ehca_cq_idr, token);
+			if (eqe_cache[eqe_cnt].cq)
+				atomic_inc(&eqe_cache[eqe_cnt].cq->nr_events);
+			spin_unlock(&ehca_cq_idr_lock);
 			if (!eqe_cache[eqe_cnt].cq) {
-				spin_unlock(&ehca_cq_idr_lock);
 				ehca_err(&shca->ib_device,
 					 "Invalid eqe for non-existing cq "
 					 "token=%x", token);
 				continue;
 			}
-			eqe_cache[eqe_cnt].cq->nr_events++;
-			spin_unlock(&ehca_cq_idr_lock);
 		} else
 			eqe_cache[eqe_cnt].cq = NULL;
 		eqe_cnt++;
@@ -517,11 +521,8 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 			else {
 				struct ehca_cq *cq = eq->eqe_cache[i].cq;
 				comp_event_callback(cq);
-				spin_lock(&ehca_cq_idr_lock);
-				cq->nr_events--;
-				if (!cq->nr_events)
+				if (atomic_dec_and_test(&cq->nr_events))
 					wake_up(&cq->wait_completion);
-				spin_unlock(&ehca_cq_idr_lock);
 			}
 		} else {
 			ehca_dbg(&shca->ib_device, "Got non completion event");
@@ -621,13 +622,10 @@ static void run_comp_task(struct ehca_cpu_comp_task* cct)
 	while (!list_empty(&cct->cq_list)) {
 		cq = list_entry(cct->cq_list.next, struct ehca_cq, entry);
 		spin_unlock_irqrestore(&cct->task_lock, flags);
-		comp_event_callback(cq);
 
-		spin_lock_irqsave(&ehca_cq_idr_lock, flags);
-		cq->nr_events--;
-		if (!cq->nr_events)
+		comp_event_callback(cq);
+		if (atomic_dec_and_test(&cq->nr_events))
 			wake_up(&cq->wait_completion);
-		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
 
 		spin_lock_irqsave(&cct->task_lock, flags);
 		spin_lock(&cq->task_lock);