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

Pull block follow-up bits from Jens Axboe:
 "A few minor (but important) fixes for blk-mq for the -rc1 window.

   - Hot removal potential oops fix for single queue devices.  From me.

   - Two merged patches in late May meant that we accidentally lost a
     fix for freeing an active queue.  Fix that up.  From me.

   - A change of the blk_mq_tag_to_rq() API, passing in blk_mq_tags, to
     make life considerably easier for scsi-mq.  From me.

   - A schedule-while-atomic fix from Ming Lei, which would hit if the
     tag space was exhausted.

   - Missing __percpu annotation in one place in blk-mq.  Found by the
     magic Wu compile bot due to code being moved around by the previous
     patch, but it's actually an older issue.  From Ming Lei.

   - Clearing of tag of a flush request at end_io time.  From Ming Lei"

* 'for-linus' of git://git.kernel.dk/linux-block:
  block: mq flush: clear flush_rq's tag in flush_end_io()
  blk-mq: let blk_mq_tag_to_rq() take blk_mq_tags as the main parameter
  blk-mq: fix regression from commit 624dbe475416
  blk-mq: handle NULL req return from blk_map_request in single queue mode
  blk-mq: fix sparse warning on missed __percpu annotation
  blk-mq: fix schedule from atomic context
  blk-mq: move blk_mq_get_ctx/blk_mq_put_ctx to mq private header

8 files changed, 123 insertions, 65 deletions

diff --git a/block/blk-flush.c b/block/blk-flush.c
index ff87c664b7df..8ffee4b5f93d 100644
--- a/block/blk-flush.c
+++ b/block/blk-flush.c
@@ -225,7 +225,7 @@ static void flush_end_io(struct request *flush_rq, int error)
 
 	if (q->mq_ops) {
 		spin_lock_irqsave(&q->mq_flush_lock, flags);
-		q->flush_rq->cmd_flags = 0;
+		q->flush_rq->tag = -1;
 	}
 
 	running = &q->flush_queue[q->flush_running_idx];
diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c
index d90c4aeb7dd3..1aab39f71d95 100644
--- a/block/blk-mq-tag.c
+++ b/block/blk-mq-tag.c
@@ -221,8 +221,10 @@ static struct bt_wait_state *bt_wait_ptr(struct blk_mq_bitmap_tags *bt,
 	return bs;
 }
 
-static int bt_get(struct blk_mq_bitmap_tags *bt, struct blk_mq_hw_ctx *hctx,
-		  unsigned int *last_tag, gfp_t gfp)
+static int bt_get(struct blk_mq_alloc_data *data,
+		struct blk_mq_bitmap_tags *bt,
+		struct blk_mq_hw_ctx *hctx,
+		unsigned int *last_tag)
 {
 	struct bt_wait_state *bs;
 	DEFINE_WAIT(wait);
@@ -232,7 +234,7 @@ static int bt_get(struct blk_mq_bitmap_tags *bt, struct blk_mq_hw_ctx *hctx,
 	if (tag != -1)
 		return tag;
 
-	if (!(gfp & __GFP_WAIT))
+	if (!(data->gfp & __GFP_WAIT))
 		return -1;
 
 	bs = bt_wait_ptr(bt, hctx);
@@ -249,50 +251,62 @@ static int bt_get(struct blk_mq_bitmap_tags *bt, struct blk_mq_hw_ctx *hctx,
 		if (was_empty)
 			atomic_set(&bs->wait_cnt, bt->wake_cnt);
 
+		blk_mq_put_ctx(data->ctx);
+
 		io_schedule();
+
+		data->ctx = blk_mq_get_ctx(data->q);
+		data->hctx = data->q->mq_ops->map_queue(data->q,
+				data->ctx->cpu);
+		if (data->reserved) {
+			bt = &data->hctx->tags->breserved_tags;
+		} else {
+			last_tag = &data->ctx->last_tag;
+			hctx = data->hctx;
+			bt = &hctx->tags->bitmap_tags;
+		}
+		finish_wait(&bs->wait, &wait);
+		bs = bt_wait_ptr(bt, hctx);
 	} while (1);
 
 	finish_wait(&bs->wait, &wait);
 	return tag;
 }
 
-static unsigned int __blk_mq_get_tag(struct blk_mq_tags *tags,
-				     struct blk_mq_hw_ctx *hctx,
-				     unsigned int *last_tag, gfp_t gfp)
+static unsigned int __blk_mq_get_tag(struct blk_mq_alloc_data *data)
 {
 	int tag;
 
-	tag = bt_get(&tags->bitmap_tags, hctx, last_tag, gfp);
+	tag = bt_get(data, &data->hctx->tags->bitmap_tags, data->hctx,
+			&data->ctx->last_tag);
 	if (tag >= 0)
-		return tag + tags->nr_reserved_tags;
+		return tag + data->hctx->tags->nr_reserved_tags;
 
 	return BLK_MQ_TAG_FAIL;
 }
 
-static unsigned int __blk_mq_get_reserved_tag(struct blk_mq_tags *tags,
-					      gfp_t gfp)
+static unsigned int __blk_mq_get_reserved_tag(struct blk_mq_alloc_data *data)
 {
 	int tag, zero = 0;
 
-	if (unlikely(!tags->nr_reserved_tags)) {
+	if (unlikely(!data->hctx->tags->nr_reserved_tags)) {
 		WARN_ON_ONCE(1);
 		return BLK_MQ_TAG_FAIL;
 	}
 
-	tag = bt_get(&tags->breserved_tags, NULL, &zero, gfp);
+	tag = bt_get(data, &data->hctx->tags->breserved_tags, NULL, &zero);
 	if (tag < 0)
 		return BLK_MQ_TAG_FAIL;
 
 	return tag;
 }
 
-unsigned int blk_mq_get_tag(struct blk_mq_hw_ctx *hctx, unsigned int *last_tag,
-			    gfp_t gfp, bool reserved)
+unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data)
 {
-	if (!reserved)
-		return __blk_mq_get_tag(hctx->tags, hctx, last_tag, gfp);
+	if (!data->reserved)
+		return __blk_mq_get_tag(data);
 
-	return __blk_mq_get_reserved_tag(hctx->tags, gfp);
+	return __blk_mq_get_reserved_tag(data);
 }
 
 static struct bt_wait_state *bt_wake_ptr(struct blk_mq_bitmap_tags *bt)
diff --git a/block/blk-mq-tag.h b/block/blk-mq-tag.h
index c959de58d2a5..98696a65d4d4 100644
--- a/block/blk-mq-tag.h
+++ b/block/blk-mq-tag.h
@@ -48,7 +48,7 @@ struct blk_mq_tags {
 extern struct blk_mq_tags *blk_mq_init_tags(unsigned int nr_tags, unsigned int reserved_tags, int node);
 extern void blk_mq_free_tags(struct blk_mq_tags *tags);
 
-extern unsigned int blk_mq_get_tag(struct blk_mq_hw_ctx *hctx, unsigned int *last_tag, gfp_t gfp, bool reserved);
+extern unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data);
 extern void blk_mq_put_tag(struct blk_mq_hw_ctx *hctx, unsigned int tag, unsigned int *last_tag);
 extern bool blk_mq_has_free_tags(struct blk_mq_tags *tags);
 extern ssize_t blk_mq_tag_sysfs_show(struct blk_mq_tags *tags, char *page);
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 0f5879c42dcd..4e4cd6208052 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -33,28 +33,6 @@ static LIST_HEAD(all_q_list);
 
 static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx);
 
-static struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q,
-					   unsigned int cpu)
-{
-	return per_cpu_ptr(q->queue_ctx, cpu);
-}
-
-/*
- * This assumes per-cpu software queueing queues. They could be per-node
- * as well, for instance. For now this is hardcoded as-is. Note that we don't
- * care about preemption, since we know the ctx's are persistent. This does
- * mean that we can't rely on ctx always matching the currently running CPU.
- */
-static struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q)
-{
-	return __blk_mq_get_ctx(q, get_cpu());
-}
-
-static void blk_mq_put_ctx(struct blk_mq_ctx *ctx)
-{
-	put_cpu();
-}
-
 /*
  * Check if any of the ctx's have pending work in this hardware queue
  */
@@ -232,24 +210,23 @@ static void blk_mq_rq_ctx_init(struct request_queue *q, struct blk_mq_ctx *ctx,
 }
 
 static struct request *
-__blk_mq_alloc_request(struct request_queue *q, struct blk_mq_hw_ctx *hctx,
-		struct blk_mq_ctx *ctx, int rw, gfp_t gfp, bool reserved)
+__blk_mq_alloc_request(struct blk_mq_alloc_data *data, int rw)
 {
 	struct request *rq;
 	unsigned int tag;
 
-	tag = blk_mq_get_tag(hctx, &ctx->last_tag, gfp, reserved);
+	tag = blk_mq_get_tag(data);
 	if (tag != BLK_MQ_TAG_FAIL) {
-		rq = hctx->tags->rqs[tag];
+		rq = data->hctx->tags->rqs[tag];
 
 		rq->cmd_flags = 0;
-		if (blk_mq_tag_busy(hctx)) {
+		if (blk_mq_tag_busy(data->hctx)) {
 			rq->cmd_flags = REQ_MQ_INFLIGHT;
-			atomic_inc(&hctx->nr_active);
+			atomic_inc(&data->hctx->nr_active);
 		}
 
 		rq->tag = tag;
-		blk_mq_rq_ctx_init(q, ctx, rq, rw);
+		blk_mq_rq_ctx_init(data->q, data->ctx, rq, rw);
 		return rq;
 	}
 
@@ -262,22 +239,27 @@ struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp,
 	struct blk_mq_ctx *ctx;
 	struct blk_mq_hw_ctx *hctx;
 	struct request *rq;
+	struct blk_mq_alloc_data alloc_data;
 
 	if (blk_mq_queue_enter(q))
 		return NULL;
 
 	ctx = blk_mq_get_ctx(q);
 	hctx = q->mq_ops->map_queue(q, ctx->cpu);
+	blk_mq_set_alloc_data(&alloc_data, q, gfp & ~__GFP_WAIT,
+			reserved, ctx, hctx);
 
-	rq = __blk_mq_alloc_request(q, hctx, ctx, rw, gfp & ~__GFP_WAIT,
-				    reserved);
+	rq = __blk_mq_alloc_request(&alloc_data, rw);
 	if (!rq && (gfp & __GFP_WAIT)) {
 		__blk_mq_run_hw_queue(hctx);
 		blk_mq_put_ctx(ctx);
 
 		ctx = blk_mq_get_ctx(q);
 		hctx = q->mq_ops->map_queue(q, ctx->cpu);
-		rq =  __blk_mq_alloc_request(q, hctx, ctx, rw, gfp, reserved);
+		blk_mq_set_alloc_data(&alloc_data, q, gfp, reserved, ctx,
+				hctx);
+		rq =  __blk_mq_alloc_request(&alloc_data, rw);
+		ctx = alloc_data.ctx;
 	}
 	blk_mq_put_ctx(ctx);
 	return rq;
@@ -547,15 +529,20 @@ void blk_mq_kick_requeue_list(struct request_queue *q)
 }
 EXPORT_SYMBOL(blk_mq_kick_requeue_list);
 
-struct request *blk_mq_tag_to_rq(struct blk_mq_hw_ctx *hctx, unsigned int tag)
+static inline bool is_flush_request(struct request *rq, unsigned int tag)
 {
-	struct request_queue *q = hctx->queue;
+	return ((rq->cmd_flags & REQ_FLUSH_SEQ) &&
+			rq->q->flush_rq->tag == tag);
+}
+
+struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag)
+{
+	struct request *rq = tags->rqs[tag];
 
-	if ((q->flush_rq->cmd_flags & REQ_FLUSH_SEQ) &&
-	    q->flush_rq->tag == tag)
-		return q->flush_rq;
+	if (!is_flush_request(rq, tag))
+		return rq;
 
-	return hctx->tags->rqs[tag];
+	return rq->q->flush_rq;
 }
 EXPORT_SYMBOL(blk_mq_tag_to_rq);
 
@@ -584,7 +571,7 @@ static void blk_mq_timeout_check(void *__data, unsigned long *free_tags)
 		if (tag >= hctx->tags->nr_tags)
 			break;
 
-		rq = blk_mq_tag_to_rq(hctx, tag++);
+		rq = blk_mq_tag_to_rq(hctx->tags, tag++);
 		if (rq->q != hctx->queue)
 			continue;
 		if (!test_bit(REQ_ATOM_STARTED, &rq->atomic_flags))
@@ -1158,6 +1145,7 @@ static struct request *blk_mq_map_request(struct request_queue *q,
 	struct blk_mq_ctx *ctx;
 	struct request *rq;
 	int rw = bio_data_dir(bio);
+	struct blk_mq_alloc_data alloc_data;
 
 	if (unlikely(blk_mq_queue_enter(q))) {
 		bio_endio(bio, -EIO);
@@ -1171,7 +1159,9 @@ static struct request *blk_mq_map_request(struct request_queue *q,
 		rw |= REQ_SYNC;
 
 	trace_block_getrq(q, bio, rw);
-	rq = __blk_mq_alloc_request(q, hctx, ctx, rw, GFP_ATOMIC, false);
+	blk_mq_set_alloc_data(&alloc_data, q, GFP_ATOMIC, false, ctx,
+			hctx);
+	rq = __blk_mq_alloc_request(&alloc_data, rw);
 	if (unlikely(!rq)) {
 		__blk_mq_run_hw_queue(hctx);
 		blk_mq_put_ctx(ctx);
@@ -1179,8 +1169,11 @@ static struct request *blk_mq_map_request(struct request_queue *q,
 
 		ctx = blk_mq_get_ctx(q);
 		hctx = q->mq_ops->map_queue(q, ctx->cpu);
-		rq = __blk_mq_alloc_request(q, hctx, ctx, rw,
-					    __GFP_WAIT|GFP_ATOMIC, false);
+		blk_mq_set_alloc_data(&alloc_data, q,
+				__GFP_WAIT|GFP_ATOMIC, false, ctx, hctx);
+		rq = __blk_mq_alloc_request(&alloc_data, rw);
+		ctx = alloc_data.ctx;
+		hctx = alloc_data.hctx;
 	}
 
 	hctx->queued++;
@@ -1288,6 +1281,8 @@ static void blk_sq_make_request(struct request_queue *q, struct bio *bio)
 		return;
 
 	rq = blk_mq_map_request(q, bio, &data);
+	if (unlikely(!rq))
+		return;
 
 	if (unlikely(is_flush_fua)) {
 		blk_mq_bio_to_request(rq, bio);
@@ -1562,6 +1557,8 @@ static void blk_mq_exit_hw_queues(struct request_queue *q,
 		if (i == nr_queue)
 			break;
 
+		blk_mq_tag_idle(hctx);
+
 		if (set->ops->exit_hctx)
 			set->ops->exit_hctx(hctx, i);
 
@@ -1779,7 +1776,7 @@ static void blk_mq_add_queue_tag_set(struct blk_mq_tag_set *set,
 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 {
 	struct blk_mq_hw_ctx **hctxs;
-	struct blk_mq_ctx *ctx;
+	struct blk_mq_ctx __percpu *ctx;
 	struct request_queue *q;
 	unsigned int *map;
 	int i;
diff --git a/block/blk-mq.h b/block/blk-mq.h
index de7b3bbd5bd6..26460884c6cd 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -69,4 +69,49 @@ struct blk_align_bitmap {
 	unsigned long depth;
 } ____cacheline_aligned_in_smp;
 
+static inline struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q,
+					   unsigned int cpu)
+{
+	return per_cpu_ptr(q->queue_ctx, cpu);
+}
+
+/*
+ * This assumes per-cpu software queueing queues. They could be per-node
+ * as well, for instance. For now this is hardcoded as-is. Note that we don't
+ * care about preemption, since we know the ctx's are persistent. This does
+ * mean that we can't rely on ctx always matching the currently running CPU.
+ */
+static inline struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q)
+{
+	return __blk_mq_get_ctx(q, get_cpu());
+}
+
+static inline void blk_mq_put_ctx(struct blk_mq_ctx *ctx)
+{
+	put_cpu();
+}
+
+struct blk_mq_alloc_data {
+	/* input parameter */
+	struct request_queue *q;
+	gfp_t gfp;
+	bool reserved;
+
+	/* input & output parameter */
+	struct blk_mq_ctx *ctx;
+	struct blk_mq_hw_ctx *hctx;
+};
+
+static inline void blk_mq_set_alloc_data(struct blk_mq_alloc_data *data,
+		struct request_queue *q, gfp_t gfp, bool reserved,
+		struct blk_mq_ctx *ctx,
+		struct blk_mq_hw_ctx *hctx)
+{
+	data->q = q;
+	data->gfp = gfp;
+	data->reserved = reserved;
+	data->ctx = ctx;
+	data->hctx = hctx;
+}
+
 #endif
diff --git a/drivers/block/mtip32xx/mtip32xx.c b/drivers/block/mtip32xx/mtip32xx.c
index abc858b3528b..74abd49fabdc 100644
--- a/drivers/block/mtip32xx/mtip32xx.c
+++ b/drivers/block/mtip32xx/mtip32xx.c
@@ -193,7 +193,9 @@ static void mtip_put_int_command(struct driver_data *dd, struct mtip_cmd *cmd)
 static struct request *mtip_rq_from_tag(struct driver_data *dd,
 					unsigned int tag)
 {
-	return blk_mq_tag_to_rq(dd->queue->queue_hw_ctx[0], tag);
+	struct blk_mq_hw_ctx *hctx = dd->queue->queue_hw_ctx[0];
+
+	return blk_mq_tag_to_rq(hctx->tags, tag);
 }
 
 static struct mtip_cmd *mtip_cmd_from_tag(struct driver_data *dd,
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index c15128833100..0feedebfde48 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -155,7 +155,7 @@ void blk_mq_free_request(struct request *rq);
 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 		gfp_t gfp, bool reserved);
-struct request *blk_mq_tag_to_rq(struct blk_mq_hw_ctx *hctx, unsigned int tag);
+struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 
 struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
 struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int, int);
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 8aba35f46f87..5c6f836afa1b 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -335,7 +335,7 @@ struct request_queue {
 	unsigned int		*mq_map;
 
 	/* sw queues */
-	struct blk_mq_ctx	*queue_ctx;
+	struct blk_mq_ctx __percpu	*queue_ctx;
 	unsigned int		nr_queues;
 
 	/* hw dispatch queues */