[PATCH 10/10] xen/blkfront: use work queue to fast blkif interrupt return

[Bob Liu]

Move the request complete logic out of blkif_interrupt() to a work queue,
after that we can replace 'spin_lock_irq' with 'spin_lock' so that irq won't
be disabled too long in blk_mq_queue_rq().

No more warning like this:
INFO: rcu_sched detected stalls on CPUs/tasks: { 7} (detected by 0,
t=15002 jiffies, g=1018, c=1017, q=0)
Task dump for CPU 7:
swapper/7       R  running task        0     0      1 0x00080000
ffff88028f4edf50 0000000000000086 ffff88028f4ee330 ffff880283df3e18
ffffffff8108836a 0000000183f75438 0000000000000040 000000000000df50
0000008bde2dd600 ffff88028f4ee330 0000000000000086 ffff880283f75038
Call Trace:
[<ffffffff8108836a>] ? __hrtimer_start_range_ns+0x269/0x27b
[<ffffffff8108838f>] ? hrtimer_start+0x13/0x15
[<ffffffff81085298>] ? rcu_eqs_enter+0x66/0x79
[<ffffffff81013847>] ? default_idle+0x9/0xd
[<ffffffff81013f2d>] ? arch_cpu_idle+0xa/0xc
[<ffffffff810746ad>] ? cpu_startup_entry+0x118/0x253
[<ffffffff81030f57>] ? start_secondary+0x12e/0x132

Signed-off-by: Bob Liu <bob.liu@xxxxxxxxxx>
---
 drivers/block/xen-blkfront.c | 47 ++++++++++++++++++++++++++------------------
 1 file changed, 28 insertions(+), 19 deletions(-)

diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index 32caf85..bdd9a15 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -116,6 +116,7 @@ struct blkfront_ring_info {
 	struct blkif_front_ring ring;
 	unsigned int evtchn, irq;
 	struct work_struct work;
+	struct work_struct done_work;
 	struct gnttab_free_callback callback;
 	struct blk_shadow shadow[BLK_RING_SIZE];
 	struct list_head grants;
@@ -630,29 +631,29 @@ static int blk_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
 	int ret = BLK_MQ_RQ_QUEUE_OK;
 
 	blk_mq_start_request(qd->rq);
-	spin_lock_irq(&rinfo->io_lock);
+	spin_lock(&rinfo->io_lock);
 	if (RING_FULL(&rinfo->ring)) {
-		spin_unlock_irq(&rinfo->io_lock);
+		spin_unlock(&rinfo->io_lock);
 		blk_mq_stop_hw_queue(hctx);
 		ret = BLK_MQ_RQ_QUEUE_BUSY;
 		goto out;
 	}
 
 	if (blkif_request_flush_invalid(qd->rq, rinfo->info)) {
-		spin_unlock_irq(&rinfo->io_lock);
+		spin_unlock(&rinfo->io_lock);
 		ret = BLK_MQ_RQ_QUEUE_ERROR;
 		goto out;
 	}
 
 	if (blkif_queue_request(qd->rq, rinfo)) {
-		spin_unlock_irq(&rinfo->io_lock);
+		spin_unlock(&rinfo->io_lock);
 		blk_mq_stop_hw_queue(hctx);
 		ret = BLK_MQ_RQ_QUEUE_BUSY;
 		goto out;
 	}
 
 	flush_requests(rinfo);
-	spin_unlock_irq(&rinfo->io_lock);
+	spin_unlock(&rinfo->io_lock);
 out:
 	return ret;
 }
@@ -937,6 +938,7 @@ static void xlvbd_release_gendisk(struct blkfront_info *info)
 
 		/* Flush gnttab callback work. Must be done with no locks held. */
 		flush_work(&rinfo->work);
+		flush_work(&rinfo->done_work);
 	}
 
 	del_gendisk(info->gd);
@@ -955,15 +957,13 @@ static void xlvbd_release_gendisk(struct blkfront_info *info)
 
 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
 {
-	unsigned long flags;
-
-	spin_lock_irqsave(&rinfo->io_lock, flags);
+	spin_lock(&rinfo->io_lock);
 	if (!RING_FULL(&rinfo->ring)) {
-		spin_unlock_irqrestore(&rinfo->io_lock, flags);
+		spin_unlock(&rinfo->io_lock);
 		blk_mq_start_stopped_hw_queues(rinfo->info->rq, true);
 		return;
 	}
-	spin_unlock_irqrestore(&rinfo->io_lock, flags);
+	spin_unlock(&rinfo->io_lock);
 }
 
 static void blkif_restart_queue(struct work_struct *work)
@@ -1070,6 +1070,7 @@ free_shadow:
 
 		/* Flush gnttab callback work. Must be done with no locks held. */
 		flush_work(&rinfo->work);
+		flush_work(&rinfo->done_work);
 
 		/* Free resources associated with old device channel. */
 		if (rinfo->ring_ref != GRANT_INVALID_REF) {
@@ -1168,19 +1169,15 @@ static void blkif_completion(struct blk_shadow *s, struct blkfront_ring_info *ri
 	}
 }
 
-static irqreturn_t blkif_interrupt(int irq, void *dev_id)
+static void blkif_done_req(struct work_struct *work)
 {
+	struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, done_work);
 	struct request *req;
 	struct blkif_response *bret;
 	RING_IDX i, rp;
-	unsigned long flags;
-	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
 	struct blkfront_info *info = rinfo->info;
 
-	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
-		return IRQ_HANDLED;
-
-	spin_lock_irqsave(&rinfo->io_lock, flags);
+	spin_lock(&rinfo->io_lock);
  again:
 	rp = rinfo->ring.sring->rsp_prod;
 	rmb(); /* Ensure we see queued responses up to 'rp'. */
@@ -1271,9 +1268,20 @@ static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 	} else
 		rinfo->ring.sring->rsp_event = i + 1;
 
-	spin_unlock_irqrestore(&rinfo->io_lock, flags);
-	kick_pending_request_queues(rinfo);
+	if (!RING_FULL(&rinfo->ring))
+		blk_mq_start_stopped_hw_queues(rinfo->info->rq, true);
+	spin_unlock(&rinfo->io_lock);
+}
+
+static irqreturn_t blkif_interrupt(int irq, void *dev_id)
+{
+	struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
+	struct blkfront_info *info = rinfo->info;
+
+	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
+		return IRQ_HANDLED;
 
+	schedule_work(&rinfo->done_work);
 	return IRQ_HANDLED;
 }
 
@@ -1535,6 +1543,7 @@ static int blkfront_probe(struct xenbus_device *dev,
 		rinfo->persistent_gnts_c = 0;
 		rinfo->info = info;
 		INIT_WORK(&rinfo->work, blkif_restart_queue);
+		INIT_WORK(&rinfo->done_work, blkif_done_req);
 
 		for (i = 0; i < BLK_RING_SIZE; i++)
 			rinfo->shadow[i].req.u.rw.id = i+1;
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/