Merge branch 'for-linus' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "This contains a set of fixes for xen-blkback by way of Konrad, and a
  performance regression fix for blk-mq for shared tags.

  The latter could account for as much as a 50x reduction in
  performance, with the test case from the user with 500 name spaces. A
  more realistic setup on my end with 32 drives showed a 3.5x drop. The
  fix has been thoroughly tested before being committed"

* 'for-linus' of git://git.kernel.dk/linux-block:
  blk-mq: fix performance regression with shared tags
  xen-blkback: don't leak stack data via response ring
  xen/blkback: don't use xen_blkif_get() in xen-blkback kthread
  xen/blkback: don't free be structure too early
  xen/blkback: fix disconnect while I/Os in flight

Author: torvalds

block/blk-mq-sched.c

@@ -68,6 +68,45 @@ static void blk_mq_sched_assign_ioc(struct request_queue *q,
 		__blk_mq_sched_assign_ioc(q, rq, bio, ioc);
 }
 
+/*
+ * Mark a hardware queue as needing a restart. For shared queues, maintain
+ * a count of how many hardware queues are marked for restart.
+ */
+static void blk_mq_sched_mark_restart_hctx(struct blk_mq_hw_ctx *hctx)
+{
+	if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
+		return;
+
+	if (hctx->flags & BLK_MQ_F_TAG_SHARED) {
+		struct request_queue *q = hctx->queue;
+
+		if (!test_and_set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
+			atomic_inc(&q->shared_hctx_restart);
+	} else
+		set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);
+}
+
+static bool blk_mq_sched_restart_hctx(struct blk_mq_hw_ctx *hctx)
+{
+	if (!test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
+		return false;
+
+	if (hctx->flags & BLK_MQ_F_TAG_SHARED) {
+		struct request_queue *q = hctx->queue;
+
+		if (test_and_clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
+			atomic_dec(&q->shared_hctx_restart);
+	} else
+		clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);
+
+	if (blk_mq_hctx_has_pending(hctx)) {
+		blk_mq_run_hw_queue(hctx, true);
+		return true;
+	}
+
+	return false;
+}
+
 struct request *blk_mq_sched_get_request(struct request_queue *q,
 					 struct bio *bio,
 					 unsigned int op,
@@ -266,18 +305,6 @@ static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx,
 	return true;
 }
 
-static bool blk_mq_sched_restart_hctx(struct blk_mq_hw_ctx *hctx)
-{
-	if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) {
-		clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);
-		if (blk_mq_hctx_has_pending(hctx)) {
-			blk_mq_run_hw_queue(hctx, true);
-			return true;
-		}
-	}
-	return false;
-}
-
 /**
  * list_for_each_entry_rcu_rr - iterate in a round-robin fashion over rcu list
  * @pos:    loop cursor.
@@ -309,6 +336,13 @@ void blk_mq_sched_restart(struct blk_mq_hw_ctx *const hctx)
 	unsigned int i, j;
 
 	if (set->flags & BLK_MQ_F_TAG_SHARED) {
+		/*
+		 * If this is 0, then we know that no hardware queues
+		 * have RESTART marked. We're done.
+		 */
+		if (!atomic_read(&queue->shared_hctx_restart))
+			return;
+
 		rcu_read_lock();
 		list_for_each_entry_rcu_rr(q, queue, &set->tag_list,
 					   tag_set_list) {

block/blk-mq-sched.h

@@ -115,15 +115,6 @@ static inline bool blk_mq_sched_has_work(struct blk_mq_hw_ctx *hctx)
 	return false;
 }
 
-/*
- * Mark a hardware queue as needing a restart.
- */
-static inline void blk_mq_sched_mark_restart_hctx(struct blk_mq_hw_ctx *hctx)
-{
-	if (!test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
-		set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);
-}
-
 static inline bool blk_mq_sched_needs_restart(struct blk_mq_hw_ctx *hctx)
 {
 	return test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);

block/blk-mq.c

@@ -2103,20 +2103,30 @@ static void blk_mq_map_swqueue(struct request_queue *q,
 	}
 }
 
+/*
+ * Caller needs to ensure that we're either frozen/quiesced, or that
+ * the queue isn't live yet.
+ */
 static void queue_set_hctx_shared(struct request_queue *q, bool shared)
 {
 	struct blk_mq_hw_ctx *hctx;
 	int i;
 
 	queue_for_each_hw_ctx(q, hctx, i) {
-		if (shared)
+		if (shared) {
+			if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
+				atomic_inc(&q->shared_hctx_restart);
 			hctx->flags |= BLK_MQ_F_TAG_SHARED;
-		else
+		} else {
+			if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
+				atomic_dec(&q->shared_hctx_restart);
 			hctx->flags &= ~BLK_MQ_F_TAG_SHARED;
+		}
 	}
 }
 
-static void blk_mq_update_tag_set_depth(struct blk_mq_tag_set *set, bool shared)
+static void blk_mq_update_tag_set_depth(struct blk_mq_tag_set *set,
+					bool shared)
 {
 	struct request_queue *q;
 

drivers/block/xen-blkback/blkback.c

@@ -609,8 +609,6 @@ int xen_blkif_schedule(void *arg)
 	unsigned long timeout;
 	int ret;
 
-	xen_blkif_get(blkif);
-
 	set_freezable();
 	while (!kthread_should_stop()) {
 		if (try_to_freeze())
@@ -665,7 +663,6 @@ int xen_blkif_schedule(void *arg)
 		print_stats(ring);
 
 	ring->xenblkd = NULL;
-	xen_blkif_put(blkif);
 
 	return 0;
 }
@@ -1436,34 +1433,35 @@ static int dispatch_rw_block_io(struct xen_blkif_ring *ring,
 static void make_response(struct xen_blkif_ring *ring, u64 id,
 			  unsigned short op, int st)
 {
-	struct blkif_response  resp;
+	struct blkif_response *resp;
 	unsigned long     flags;
 	union blkif_back_rings *blk_rings;
 	int notify;
 
-	resp.id        = id;
-	resp.operation = op;
-	resp.status    = st;
-
 	spin_lock_irqsave(&ring->blk_ring_lock, flags);
 	blk_rings = &ring->blk_rings;
 	/* Place on the response ring for the relevant domain. */
 	switch (ring->blkif->blk_protocol) {
 	case BLKIF_PROTOCOL_NATIVE:
-		memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt),
-		       &resp, sizeof(resp));
+		resp = RING_GET_RESPONSE(&blk_rings->native,
+					 blk_rings->native.rsp_prod_pvt);
 		break;
 	case BLKIF_PROTOCOL_X86_32:
-		memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt),
-		       &resp, sizeof(resp));
+		resp = RING_GET_RESPONSE(&blk_rings->x86_32,
+					 blk_rings->x86_32.rsp_prod_pvt);
 		break;
 	case BLKIF_PROTOCOL_X86_64:
-		memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt),
-		       &resp, sizeof(resp));
+		resp = RING_GET_RESPONSE(&blk_rings->x86_64,
+					 blk_rings->x86_64.rsp_prod_pvt);
 		break;
 	default:
 		BUG();
 	}
+
+	resp->id        = id;
+	resp->operation = op;
+	resp->status    = st;
+
 	blk_rings->common.rsp_prod_pvt++;
 	RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify);
 	spin_unlock_irqrestore(&ring->blk_ring_lock, flags);

drivers/block/xen-blkback/common.h

@@ -75,9 +75,8 @@ extern unsigned int xenblk_max_queues;
 struct blkif_common_request {
 	char dummy;
 };
-struct blkif_common_response {
-	char dummy;
-};
+
+/* i386 protocol version */
 
 struct blkif_x86_32_request_rw {
 	uint8_t        nr_segments;  /* number of segments                   */
@@ -129,14 +128,6 @@ struct blkif_x86_32_request {
 	} u;
 } __attribute__((__packed__));
 
-/* i386 protocol version */
-#pragma pack(push, 4)
-struct blkif_x86_32_response {
-	uint64_t        id;              /* copied from request */
-	uint8_t         operation;       /* copied from request */
-	int16_t         status;          /* BLKIF_RSP_???       */
-};
-#pragma pack(pop)
 /* x86_64 protocol version */
 
 struct blkif_x86_64_request_rw {
@@ -193,18 +184,12 @@ struct blkif_x86_64_request {
 	} u;
 } __attribute__((__packed__));
 
-struct blkif_x86_64_response {
-	uint64_t       __attribute__((__aligned__(8))) id;
-	uint8_t         operation;       /* copied from request */
-	int16_t         status;          /* BLKIF_RSP_???       */
-};
-
 DEFINE_RING_TYPES(blkif_common, struct blkif_common_request,
-		  struct blkif_common_response);
+		  struct blkif_response);
 DEFINE_RING_TYPES(blkif_x86_32, struct blkif_x86_32_request,
-		  struct blkif_x86_32_response);
+		  struct blkif_response __packed);
 DEFINE_RING_TYPES(blkif_x86_64, struct blkif_x86_64_request,
-		  struct blkif_x86_64_response);
+		  struct blkif_response);
 
 union blkif_back_rings {
 	struct blkif_back_ring        native;
@@ -281,6 +266,7 @@ struct xen_blkif_ring {
 
 	wait_queue_head_t	wq;
 	atomic_t		inflight;
+	bool			active;
 	/* One thread per blkif ring. */
 	struct task_struct	*xenblkd;
 	unsigned int		waiting_reqs;

drivers/block/xen-blkback/xenbus.c

@@ -159,7 +159,7 @@ static int xen_blkif_alloc_rings(struct xen_blkif *blkif)
 		init_waitqueue_head(&ring->shutdown_wq);
 		ring->blkif = blkif;
 		ring->st_print = jiffies;
-		xen_blkif_get(blkif);
+		ring->active = true;
 	}
 
 	return 0;
@@ -249,10 +249,12 @@ static int xen_blkif_disconnect(struct xen_blkif *blkif)
 		struct xen_blkif_ring *ring = &blkif->rings[r];
 		unsigned int i = 0;
 
+		if (!ring->active)
+			continue;
+
 		if (ring->xenblkd) {
 			kthread_stop(ring->xenblkd);
 			wake_up(&ring->shutdown_wq);
-			ring->xenblkd = NULL;
 		}
 
 		/* The above kthread_stop() guarantees that at this point we
@@ -296,7 +298,7 @@ static int xen_blkif_disconnect(struct xen_blkif *blkif)
 		BUG_ON(ring->free_pages_num != 0);
 		BUG_ON(ring->persistent_gnt_c != 0);
 		WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
-		xen_blkif_put(blkif);
+		ring->active = false;
 	}
 	blkif->nr_ring_pages = 0;
 	/*
@@ -312,9 +314,10 @@ static int xen_blkif_disconnect(struct xen_blkif *blkif)
 
 static void xen_blkif_free(struct xen_blkif *blkif)
 {
-
-	xen_blkif_disconnect(blkif);
+	WARN_ON(xen_blkif_disconnect(blkif));
 	xen_vbd_free(&blkif->vbd);
+	kfree(blkif->be->mode);
+	kfree(blkif->be);
 
 	/* Make sure everything is drained before shutting down */
 	kmem_cache_free(xen_blkif_cachep, blkif);
@@ -511,8 +514,6 @@ static int xen_blkbk_remove(struct xenbus_device *dev)
 		xen_blkif_put(be->blkif);
 	}
 
-	kfree(be->mode);
-	kfree(be);
 	return 0;
 }
 

include/linux/blkdev.h

@@ -391,6 +391,8 @@ struct request_queue {
 	int			nr_rqs[2];	/* # allocated [a]sync rqs */
 	int			nr_rqs_elvpriv;	/* # allocated rqs w/ elvpriv */
 
+	atomic_t		shared_hctx_restart;
+
 	struct blk_queue_stats	*stats;
 	struct rq_wb		*rq_wb;