IB/ehca: Fix sync between completion handler and destroy cq

This patch fixes two issues reported by Roland Dreier and Christoph Hellwig:

- Mismatched sync/locking between completion handler and destroy cq We
  introduced a counter nr_events per cq to track number of irq events
  seen. This counter is incremented when an event queue entry is seen
  and decremented after completion handler has been called regardless
  if scaling code is active or not. Note that nr_callbacks tracks
  number of events assigned to a cpu and both counters can potentially
  diverge.

  The sync between running completion handler and destroy cq is done
  by using the global spin lock ehca_cq_idr_lock.

- Replace yield by wait_event on the counter above to become zero.

Signed-off-by: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

1 files changed, 39 insertions, 20 deletions

diff --git a/drivers/infiniband/hw/ehca/ehca_irq.c b/drivers/infiniband/hw/ehca/ehca_irq.c
index 3ec53c6..20f36bf 100644
--- a/drivers/infiniband/hw/ehca/ehca_irq.c
+++ b/drivers/infiniband/hw/ehca/ehca_irq.c
@@ -404,10 +404,11 @@ static inline void process_eqe(struct ehca_shca *shca, struct ehca_eqe *eqe)
 	u32 token;
 	unsigned long flags;
 	struct ehca_cq *cq;
+
 	eqe_value = eqe->entry;
 	ehca_dbg(&shca->ib_device, "eqe_value=%lx", eqe_value);
 	if (EHCA_BMASK_GET(EQE_COMPLETION_EVENT, eqe_value)) {
-		ehca_dbg(&shca->ib_device, "... completion event");
+		ehca_dbg(&shca->ib_device, "Got completion event");
 		token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe_value);
 		spin_lock_irqsave(&ehca_cq_idr_lock, flags);
 		cq = idr_find(&ehca_cq_idr, token);
@@ -419,16 +420,20 @@ static inline void process_eqe(struct ehca_shca *shca, struct ehca_eqe *eqe)
 			return;
 		}
 		reset_eq_pending(cq);
-		if (ehca_scaling_code) {
+		cq->nr_events++;
+		spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+		if (ehca_scaling_code)
 			queue_comp_task(cq);
-			spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
-		} else {
-			spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+		else {
 			comp_event_callback(cq);
+			spin_lock_irqsave(&ehca_cq_idr_lock, flags);
+			cq->nr_events--;
+			if (!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");
+		ehca_dbg(&shca->ib_device, "Got non completion event");
 		parse_identifier(shca, eqe_value);
 	}
 }
@@ -478,6 +483,7 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 					 "token=%x", token);
 				continue;
 			}
+			eqe_cache[eqe_cnt].cq->nr_events++;
 			spin_unlock(&ehca_cq_idr_lock);
 		} else
 			eqe_cache[eqe_cnt].cq = NULL;
@@ -504,12 +510,18 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 	/* call completion handler for cached eqes */
 	for (i = 0; i < eqe_cnt; i++)
 		if (eq->eqe_cache[i].cq) {
-			if (ehca_scaling_code) {
-				spin_lock(&ehca_cq_idr_lock);
+			if (ehca_scaling_code)
 				queue_comp_task(eq->eqe_cache[i].cq);
-				spin_unlock(&ehca_cq_idr_lock);
-			} else
-				comp_event_callback(eq->eqe_cache[i].cq);
+			else {
+				struct ehca_cq *cq = eq->eqe_cache[i].cq;
+				comp_event_callback(cq);
+				spin_lock_irqsave(&ehca_cq_idr_lock, flags);
+				cq->nr_events--;
+				if (!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");
 			parse_identifier(shca, eq->eqe_cache[i].eqe->entry);
@@ -523,7 +535,6 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 		if (!eqe)
 			break;
 		process_eqe(shca, eqe);
-		eqe_cnt++;
 	} while (1);
 
 unlock_irq_spinlock:
@@ -567,8 +578,7 @@ static void __queue_comp_task(struct ehca_cq *__cq,
 		list_add_tail(&__cq->entry, &cct->cq_list);
 		cct->cq_jobs++;
 		wake_up(&cct->wait_queue);
-	}
-	else
+	} else
 		__cq->nr_callbacks++;
 
 	spin_unlock(&__cq->task_lock);
@@ -577,18 +587,21 @@ static void __queue_comp_task(struct ehca_cq *__cq,
 
 static void queue_comp_task(struct ehca_cq *__cq)
 {
-	int cpu;
 	int cpu_id;
 	struct ehca_cpu_comp_task *cct;
+	int cq_jobs;
+	unsigned long flags;
 
-	cpu = get_cpu();
 	cpu_id = find_next_online_cpu(pool);
 	BUG_ON(!cpu_online(cpu_id));
 
 	cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu_id);
 	BUG_ON(!cct);
 
-	if (cct->cq_jobs > 0) {
+	spin_lock_irqsave(&cct->task_lock, flags);
+	cq_jobs = cct->cq_jobs;
+	spin_unlock_irqrestore(&cct->task_lock, flags);
+	if (cq_jobs > 0) {
 		cpu_id = find_next_online_cpu(pool);
 		cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu_id);
 		BUG_ON(!cct);
@@ -608,11 +621,17 @@ static void run_comp_task(struct ehca_cpu_comp_task* cct)
 		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(&cct->task_lock, flags);
 
+		spin_lock_irqsave(&ehca_cq_idr_lock, flags);
+		cq->nr_events--;
+		if (!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);
 		cq->nr_callbacks--;
-		if (cq->nr_callbacks == 0) {
+		if (!cq->nr_callbacks) {
 			list_del_init(cct->cq_list.next);
 			cct->cq_jobs--;
 		}