nvme-pci: Remove two-pass completions

Completion handling had been done in two steps: find all new completions
under a lock, then handle those completions outside the lock. This was
done to make the locked section as short as possible so that other
threads using the same lock wait less time.

The driver no longer shares locks during completion, and is in fact
lockless for interrupt driven queues, so the optimization no longer
serves its original purpose. Replace the two-pass completion queue
handler with a single pass that completes entries immediately.
Reviewed-by: NSagi Grimberg <sagi@grimberg.me>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NKeith Busch <kbusch@kernel.org>

drivers/nvme/host/pci.c

@@ -971,15 +971,6 @@ static inline void nvme_handle_cqe(struct nvme_queue *nvmeq, u16 idx)
 	nvme_end_request(req, cqe->status, cqe->result);
 }
 
-static void nvme_complete_cqes(struct nvme_queue *nvmeq, u16 start, u16 end)
-{
-	while (start != end) {
-		nvme_handle_cqe(nvmeq, start);
-		if (++start == nvmeq->q_depth)
-			start = 0;
-	}
-}
-
 static inline void nvme_update_cq_head(struct nvme_queue *nvmeq)
 {
 	if (++nvmeq->cq_head == nvmeq->q_depth) {
@@ -988,19 +979,17 @@ static inline void nvme_update_cq_head(struct nvme_queue *nvmeq)
 	}
 }
 
-static inline int nvme_process_cq(struct nvme_queue *nvmeq, u16 *start,
-				  u16 *end)
+static inline int nvme_process_cq(struct nvme_queue *nvmeq)
 {
 	int found = 0;
 
-	*start = nvmeq->cq_head;
 	while (nvme_cqe_pending(nvmeq)) {
 		found++;
+		nvme_handle_cqe(nvmeq, nvmeq->cq_head);
 		nvme_update_cq_head(nvmeq);
 	}
-	*end = nvmeq->cq_head;
 
-	if (*start != *end)
+	if (found)
 		nvme_ring_cq_doorbell(nvmeq);
 	return found;
 }
@@ -1009,21 +998,16 @@ static irqreturn_t nvme_irq(int irq, void *data)
 {
 	struct nvme_queue *nvmeq = data;
 	irqreturn_t ret = IRQ_NONE;
-	u16 start, end;
 
 	/*
 	 * The rmb/wmb pair ensures we see all updates from a previous run of
 	 * the irq handler, even if that was on another CPU.
 	 */
 	rmb();
-	nvme_process_cq(nvmeq, &start, &end);
+	if (nvme_process_cq(nvmeq))
+		ret = IRQ_HANDLED;
 	wmb();
 
-	if (start != end) {
-		nvme_complete_cqes(nvmeq, start, end);
-		return IRQ_HANDLED;
-	}
-
 	return ret;
 }
 
@@ -1042,7 +1026,6 @@ static irqreturn_t nvme_irq_check(int irq, void *data)
 static int nvme_poll_irqdisable(struct nvme_queue *nvmeq)
 {
 	struct pci_dev *pdev = to_pci_dev(nvmeq->dev->dev);
-	u16 start, end;
 	int found;
 
 	/*
@@ -1052,29 +1035,27 @@ static int nvme_poll_irqdisable(struct nvme_queue *nvmeq)
 	 */
 	if (test_bit(NVMEQ_POLLED, &nvmeq->flags)) {
 		spin_lock(&nvmeq->cq_poll_lock);
-		found = nvme_process_cq(nvmeq, &start, &end);
+		found = nvme_process_cq(nvmeq);
 		spin_unlock(&nvmeq->cq_poll_lock);
 	} else {
 		disable_irq(pci_irq_vector(pdev, nvmeq->cq_vector));
-		found = nvme_process_cq(nvmeq, &start, &end);
+		found = nvme_process_cq(nvmeq);
 		enable_irq(pci_irq_vector(pdev, nvmeq->cq_vector));
 	}
 
-	nvme_complete_cqes(nvmeq, start, end);
 	return found;
 }
 
 static int nvme_poll(struct blk_mq_hw_ctx *hctx)
 {
 	struct nvme_queue *nvmeq = hctx->driver_data;
-	u16 start, end;
 	bool found;
 
 	if (!nvme_cqe_pending(nvmeq))
 		return 0;
 
 	spin_lock(&nvmeq->cq_poll_lock);
-	found = nvme_process_cq(nvmeq, &start, &end);
+	found = nvme_process_cq(nvmeq);
 	spin_unlock(&nvmeq->cq_poll_lock);
 
 	return found;
@@ -1406,13 +1387,10 @@ static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown)
  */
 static void nvme_reap_pending_cqes(struct nvme_dev *dev)
 {
-	u16 start, end;
 	int i;
 
-	for (i = dev->ctrl.queue_count - 1; i > 0; i--) {
-		nvme_process_cq(&dev->queues[i], &start, &end);
-		nvme_complete_cqes(&dev->queues[i], start, end);
-	}
+	for (i = dev->ctrl.queue_count - 1; i > 0; i--)
+		nvme_process_cq(&dev->queues[i]);
 }
 
 static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues,