IB/hfi1: Optimize kthread pointer locking when queuing CQ entries

All threads queuing CQ entries on different CQs are unnecessarily
synchronized by a spin lock to check if the CQ kthread worker hasn't
been destroyed before queuing an CQ entry.

The lock used in 6efaf10f ("IB/rdmavt: Avoid queuing work into a
destroyed cq kthread worker") is a device global lock and will have
poor performance at scale as completions are entered from a large
number of CPUs.

Convert to use RCU where the read side of RCU is rvt_cq_enter() to
determine that the worker is alive prior to triggering the
completion event.
Apply write side RCU semantics in rvt_driver_cq_init() and
rvt_cq_exit().

Fixes: 6efaf10f ("IB/rdmavt: Avoid queuing work into a destroyed cq kthread worker")
Cc: <stable@vger.kernel.org> # 4.14.x
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

drivers/infiniband/sw/rdmavt/cq.c

@@ -120,17 +120,20 @@ void rvt_cq_enter(struct rvt_cq *cq, struct ib_wc *entry, bool solicited)
 	if (cq->notify == IB_CQ_NEXT_COMP ||
 	    (cq->notify == IB_CQ_SOLICITED &&
 	     (solicited || entry->status != IB_WC_SUCCESS))) {
+		struct kthread_worker *worker;
+
 		/*
 		 * This will cause send_complete() to be called in
 		 * another thread.
 		 */
-		spin_lock(&cq->rdi->n_cqs_lock);
-		if (likely(cq->rdi->worker)) {
+		rcu_read_lock();
+		worker = rcu_dereference(cq->rdi->worker);
+		if (likely(worker)) {
 			cq->notify = RVT_CQ_NONE;
 			cq->triggered++;
-			kthread_queue_work(cq->rdi->worker, &cq->comptask);
+			kthread_queue_work(worker, &cq->comptask);
 		}
-		spin_unlock(&cq->rdi->n_cqs_lock);
+		rcu_read_unlock();
 	}
 
 	spin_unlock_irqrestore(&cq->lock, flags);
@@ -512,7 +515,7 @@ int rvt_driver_cq_init(struct rvt_dev_info *rdi)
 	int cpu;
 	struct kthread_worker *worker;
 
-	if (rdi->worker)
+	if (rcu_access_pointer(rdi->worker))
 		return 0;
 
 	spin_lock_init(&rdi->n_cqs_lock);
@@ -524,7 +527,7 @@ int rvt_driver_cq_init(struct rvt_dev_info *rdi)
 		return PTR_ERR(worker);
 
 	set_user_nice(worker->task, MIN_NICE);
-	rdi->worker = worker;
+	RCU_INIT_POINTER(rdi->worker, worker);
 	return 0;
 }
 
@@ -536,15 +539,19 @@ void rvt_cq_exit(struct rvt_dev_info *rdi)
 {
 	struct kthread_worker *worker;
 
-	/* block future queuing from send_complete() */
-	spin_lock_irq(&rdi->n_cqs_lock);
-	worker = rdi->worker;
+	if (!rcu_access_pointer(rdi->worker))
+		return;
+
+	spin_lock(&rdi->n_cqs_lock);
+	worker = rcu_dereference_protected(rdi->worker,
+					   lockdep_is_held(&rdi->n_cqs_lock));
 	if (!worker) {
-		spin_unlock_irq(&rdi->n_cqs_lock);
+		spin_unlock(&rdi->n_cqs_lock);
 		return;
 	}
-	rdi->worker = NULL;
-	spin_unlock_irq(&rdi->n_cqs_lock);
+	RCU_INIT_POINTER(rdi->worker, NULL);
+	spin_unlock(&rdi->n_cqs_lock);
+	synchronize_rcu();
 
 	kthread_destroy_worker(worker);
 }

include/rdma/rdma_vt.h

@@ -402,7 +402,7 @@ struct rvt_dev_info {
 	spinlock_t pending_lock; /* protect pending mmap list */
 
 	/* CQ */
-	struct kthread_worker *worker; /* per device cq worker */
+	struct kthread_worker __rcu *worker; /* per device cq worker */
 	u32 n_cqs_allocated;    /* number of CQs allocated for device */
 	spinlock_t n_cqs_lock; /* protect count of in use cqs */