Merge branch 'for-2.6.40/core' of git://git.kernel.dk/linux-2.6-block

* 'for-2.6.40/core' of git://git.kernel.dk/linux-2.6-block: (40 commits)
cfq-iosched: free cic_index if cfqd allocation fails
cfq-iosched: remove unused 'group_changed' in cfq_service_tree_add()
cfq-iosched: reduce bit operations in cfq_choose_req()
cfq-iosched: algebraic simplification in cfq_prio_to_maxrq()
blk-cgroup: Initialize ioc->cgroup_changed at ioc creation time
block: move bd_set_size() above rescan_partitions() in __blkdev_get()
block: call elv_bio_merged() when merged
cfq-iosched: Make IO merge related stats per cpu
cfq-iosched: Fix a memory leak of per cpu stats for root group
backing-dev: Kill set but not used var in bdi_debug_stats_show()
block: get rid of on-stack plugging debug checks
blk-throttle: Make no throttling rule group processing lockless
blk-cgroup: Make cgroup stat reset path blkg->lock free for dispatch stats
blk-cgroup: Make 64bit per cpu stats safe on 32bit arch
blk-throttle: Make dispatch stats per cpu
blk-throttle: Free up a group only after one rcu grace period
blk-throttle: Use helper function to add root throtl group to lists
blk-throttle: Introduce a helper function to fill in device details
blk-throttle: Dynamically allocate root group
blk-cgroup: Allow sleeping while dynamically allocating a group
...

Documentation/ABI/testing/sysfs-block

···
 		with the previous I/O request are enabled. When set to 2,
 		all merge tries are disabled. The default value is 0 -
 		which enables all types of merge tries.
+
+What:		/sys/block/<disk>/discard_alignment
+Date:		May 2011
+Contact:	Martin K. Petersen <[e-mail]>
+Description:
+		Devices that support discard functionality may
+		internally allocate space in units that are bigger than
+		the exported logical block size. The discard_alignment
+		parameter indicates how many bytes the beginning of the
+		device is offset from the internal allocation unit's
+		natural alignment.
+
+What:		/sys/block/<disk>/<partition>/discard_alignment
+Date:		May 2011
+Contact:	Martin K. Petersen <[e-mail]>
+Description:
+		Devices that support discard functionality may
+		internally allocate space in units that are bigger than
+		the exported logical block size. The discard_alignment
+		parameter indicates how many bytes the beginning of the
+		partition is offset from the internal allocation unit's
+		natural alignment.
+
+What:		/sys/block/<disk>/queue/discard_granularity
+Date:		May 2011
+Contact:	Martin K. Petersen <[e-mail]>
+Description:
+		Devices that support discard functionality may
+		internally allocate space using units that are bigger
+		than the logical block size. The discard_granularity
+		parameter indicates the size of the internal allocation
+		unit in bytes if reported by the device. Otherwise the
+		discard_granularity will be set to match the device's
+		physical block size. A discard_granularity of 0 means
+		that the device does not support discard functionality.
+
+What:		/sys/block/<disk>/queue/discard_max_bytes
+Date:		May 2011
+Contact:	Martin K. Petersen <[e-mail]>
+Description:
+		Devices that support discard functionality may have
+		internal limits on the number of bytes that can be
+		trimmed or unmapped in a single operation. Some storage
+		protocols also have inherent limits on the number of
+		blocks that can be described in a single command. The
+		discard_max_bytes parameter is set by the device driver
+		to the maximum number of bytes that can be discarded in
+		a single operation. Discard requests issued to the
+		device must not exceed this limit. A discard_max_bytes
+		value of 0 means that the device does not support
+		discard functionality.
+
+What:		/sys/block/<disk>/queue/discard_zeroes_data
+Date:		May 2011
+Contact:	Martin K. Petersen <[e-mail]>
+Description:
+		Devices that support discard functionality may return
+		stale or random data when a previously discarded block
+		is read back. This can cause problems if the filesystem
+		expects discarded blocks to be explicitly cleared. If a
+		device reports that it deterministically returns zeroes
+		when a discarded area is read the discard_zeroes_data
+		parameter will be set to one. Otherwise it will be 0 and
+		the result of reading a discarded area is undefined.

block/blk-cgroup.c

···
 
 	spin_lock_irqsave(&blkg->stats_lock, flags);
 	blkg->stats.time += time;
+#ifdef CONFIG_DEBUG_BLK_CGROUP
 	blkg->stats.unaccounted_time += unaccounted_time;
+#endif
 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
 }
 EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
 
+/*
+ * should be called under rcu read lock or queue lock to make sure blkg pointer
+ * is valid.
+ */
 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
 				uint64_t bytes, bool direction, bool sync)
 {
-	struct blkio_group_stats *stats;
+	struct blkio_group_stats_cpu *stats_cpu;
 	unsigned long flags;
 
-	spin_lock_irqsave(&blkg->stats_lock, flags);
-	stats = &blkg->stats;
-	stats->sectors += bytes >> 9;
-	blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICED], 1, direction,
-			sync);
-	blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_BYTES], bytes,
-			direction, sync);
-	spin_unlock_irqrestore(&blkg->stats_lock, flags);
+	/*
+	 * Disabling interrupts to provide mutual exclusion between two
+	 * writes on same cpu. It probably is not needed for 64bit. Not
+	 * optimizing that case yet.
+	 */
+	local_irq_save(flags);
+
+	stats_cpu = this_cpu_ptr(blkg->stats_cpu);
+
+	u64_stats_update_begin(&stats_cpu->syncp);
+	stats_cpu->sectors += bytes >> 9;
+	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
+			1, direction, sync);
+	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
+			bytes, direction, sync);
+	u64_stats_update_end(&stats_cpu->syncp);
+	local_irq_restore(flags);
 }
 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
 
···
 }
 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
 
+/*  Merged stats are per cpu.  */
 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
 					bool sync)
 {
+	struct blkio_group_stats_cpu *stats_cpu;
 	unsigned long flags;
 
-	spin_lock_irqsave(&blkg->stats_lock, flags);
-	blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_MERGED], 1, direction,
-			sync);
-	spin_unlock_irqrestore(&blkg->stats_lock, flags);
+	/*
+	 * Disabling interrupts to provide mutual exclusion between two
+	 * writes on same cpu. It probably is not needed for 64bit. Not
+	 * optimizing that case yet.
+	 */
+	local_irq_save(flags);
+
+	stats_cpu = this_cpu_ptr(blkg->stats_cpu);
+
+	u64_stats_update_begin(&stats_cpu->syncp);
+	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1,
+				direction, sync);
+	u64_stats_update_end(&stats_cpu->syncp);
+	local_irq_restore(flags);
 }
 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
+
+/*
+ * This function allocates the per cpu stats for blkio_group. Should be called
+ * from sleepable context as alloc_per_cpu() requires that.
+ */
+int blkio_alloc_blkg_stats(struct blkio_group *blkg)
+{
+	/* Allocate memory for per cpu stats */
+	blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu);
+	if (!blkg->stats_cpu)
+		return -ENOMEM;
+	return 0;
+}
+EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats);
 
 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
 		struct blkio_group *blkg, void *key, dev_t dev,
···
 }
 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
 
+static void blkio_reset_stats_cpu(struct blkio_group *blkg)
+{
+	struct blkio_group_stats_cpu *stats_cpu;
+	int i, j, k;
+	/*
+	 * Note: On 64 bit arch this should not be an issue. This has the
+	 * possibility of returning some inconsistent value on 32bit arch
+	 * as 64bit update on 32bit is non atomic. Taking care of this
+	 * corner case makes code very complicated, like sending IPIs to
+	 * cpus, taking care of stats of offline cpus etc.
+	 *
+	 * reset stats is anyway more of a debug feature and this sounds a
+	 * corner case. So I am not complicating the code yet until and
+	 * unless this becomes a real issue.
+	 */
+	for_each_possible_cpu(i) {
+		stats_cpu = per_cpu_ptr(blkg->stats_cpu, i);
+		stats_cpu->sectors = 0;
+		for(j = 0; j < BLKIO_STAT_CPU_NR; j++)
+			for (k = 0; k < BLKIO_STAT_TOTAL; k++)
+				stats_cpu->stat_arr_cpu[j][k] = 0;
+	}
+}
+
 static int
 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
 {
···
 		}
 #endif
 		spin_unlock(&blkg->stats_lock);
+
+		/* Reset Per cpu stats which don't take blkg->stats_lock */
+		blkio_reset_stats_cpu(blkg);
 	}
+
 	spin_unlock_irq(&blkcg->lock);
 	return 0;
 }
···
 	return val;
 }
 
+
+static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg,
+			enum stat_type_cpu type, enum stat_sub_type sub_type)
+{
+	int cpu;
+	struct blkio_group_stats_cpu *stats_cpu;
+	u64 val = 0, tval;
+
+	for_each_possible_cpu(cpu) {
+		unsigned int start;
+		stats_cpu  = per_cpu_ptr(blkg->stats_cpu, cpu);
+
+		do {
+			start = u64_stats_fetch_begin(&stats_cpu->syncp);
+			if (type == BLKIO_STAT_CPU_SECTORS)
+				tval = stats_cpu->sectors;
+			else
+				tval = stats_cpu->stat_arr_cpu[type][sub_type];
+		} while(u64_stats_fetch_retry(&stats_cpu->syncp, start));
+
+		val += tval;
+	}
+
+	return val;
+}
+
+static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg,
+		struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type)
+{
+	uint64_t disk_total, val;
+	char key_str[MAX_KEY_LEN];
+	enum stat_sub_type sub_type;
+
+	if (type == BLKIO_STAT_CPU_SECTORS) {
+		val = blkio_read_stat_cpu(blkg, type, 0);
+		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev);
+	}
+
+	for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
+			sub_type++) {
+		blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
+		val = blkio_read_stat_cpu(blkg, type, sub_type);
+		cb->fill(cb, key_str, val);
+	}
+
+	disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) +
+			blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE);
+
+	blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
+	cb->fill(cb, key_str, disk_total);
+	return disk_total;
+}
+
 /* This should be called with blkg->stats_lock held */
 static uint64_t blkio_get_stat(struct blkio_group *blkg,
 		struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
···
 	if (type == BLKIO_STAT_TIME)
 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
 					blkg->stats.time, cb, dev);
-	if (type == BLKIO_STAT_SECTORS)
-		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
-					blkg->stats.sectors, cb, dev);
 #ifdef CONFIG_DEBUG_BLK_CGROUP
 	if (type == BLKIO_STAT_UNACCOUNTED_TIME)
 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
···
 }
 
 static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
-		struct cftype *cft, struct cgroup_map_cb *cb, enum stat_type type,
-		bool show_total)
+		struct cftype *cft, struct cgroup_map_cb *cb,
+		enum stat_type type, bool show_total, bool pcpu)
 {
 	struct blkio_group *blkg;
 	struct hlist_node *n;
···
 		if (blkg->dev) {
 			if (!cftype_blkg_same_policy(cft, blkg))
 				continue;
-			spin_lock_irq(&blkg->stats_lock);
-			cgroup_total += blkio_get_stat(blkg, cb, blkg->dev,
-						type);
-			spin_unlock_irq(&blkg->stats_lock);
+			if (pcpu)
+				cgroup_total += blkio_get_stat_cpu(blkg, cb,
+						blkg->dev, type);
+			else {
+				spin_lock_irq(&blkg->stats_lock);
+				cgroup_total += blkio_get_stat(blkg, cb,
+						blkg->dev, type);
+				spin_unlock_irq(&blkg->stats_lock);
+			}
 		}
 	}
 	if (show_total)
···
 		switch(name) {
 		case BLKIO_PROP_time:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_TIME, 0);
+						BLKIO_STAT_TIME, 0, 0);
 		case BLKIO_PROP_sectors:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_SECTORS, 0);
+						BLKIO_STAT_CPU_SECTORS, 0, 1);
 		case BLKIO_PROP_io_service_bytes:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_SERVICE_BYTES, 1);
+					BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
 		case BLKIO_PROP_io_serviced:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_SERVICED, 1);
+						BLKIO_STAT_CPU_SERVICED, 1, 1);
 		case BLKIO_PROP_io_service_time:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_SERVICE_TIME, 1);
+						BLKIO_STAT_SERVICE_TIME, 1, 0);
 		case BLKIO_PROP_io_wait_time:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_WAIT_TIME, 1);
+						BLKIO_STAT_WAIT_TIME, 1, 0);
 		case BLKIO_PROP_io_merged:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_MERGED, 1);
+						BLKIO_STAT_CPU_MERGED, 1, 1);
 		case BLKIO_PROP_io_queued:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_QUEUED, 1);
+						BLKIO_STAT_QUEUED, 1, 0);
 #ifdef CONFIG_DEBUG_BLK_CGROUP
 		case BLKIO_PROP_unaccounted_time:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_UNACCOUNTED_TIME, 0);
+					BLKIO_STAT_UNACCOUNTED_TIME, 0, 0);
 		case BLKIO_PROP_dequeue:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_DEQUEUE, 0);
+						BLKIO_STAT_DEQUEUE, 0, 0);
 		case BLKIO_PROP_avg_queue_size:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_AVG_QUEUE_SIZE, 0);
+					BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0);
 		case BLKIO_PROP_group_wait_time:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_GROUP_WAIT_TIME, 0);
+					BLKIO_STAT_GROUP_WAIT_TIME, 0, 0);
 		case BLKIO_PROP_idle_time:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_IDLE_TIME, 0);
+						BLKIO_STAT_IDLE_TIME, 0, 0);
 		case BLKIO_PROP_empty_time:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_EMPTY_TIME, 0);
+						BLKIO_STAT_EMPTY_TIME, 0, 0);
 #endif
 		default:
 			BUG();
···
 		switch(name){
 		case BLKIO_THROTL_io_service_bytes:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_SERVICE_BYTES, 1);
+						BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
 		case BLKIO_THROTL_io_serviced:
 			return blkio_read_blkg_stats(blkcg, cft, cb,
-						BLKIO_STAT_SERVICED, 1);
+						BLKIO_STAT_CPU_SERVICED, 1, 1);
 		default:
 			BUG();
 		}

block/blk-cgroup.h

···
  */
 
 #include <linux/cgroup.h>
+#include <linux/u64_stats_sync.h>
 
 enum blkio_policy_id {
 	BLKIO_POLICY_PROP = 0,		/* Proportional Bandwidth division */
···
 	 * request completion for IOs doen by this cgroup. This may not be
 	 * accurate when NCQ is turned on. */
 	BLKIO_STAT_SERVICE_TIME = 0,
-	/* Total bytes transferred */
-	BLKIO_STAT_SERVICE_BYTES,
-	/* Total IOs serviced, post merge */
-	BLKIO_STAT_SERVICED,
 	/* Total time spent waiting in scheduler queue in ns */
 	BLKIO_STAT_WAIT_TIME,
-	/* Number of IOs merged */
-	BLKIO_STAT_MERGED,
 	/* Number of IOs queued up */
 	BLKIO_STAT_QUEUED,
 	/* All the single valued stats go below this */
 	BLKIO_STAT_TIME,
-	BLKIO_STAT_SECTORS,
+#ifdef CONFIG_DEBUG_BLK_CGROUP
 	/* Time not charged to this cgroup */
 	BLKIO_STAT_UNACCOUNTED_TIME,
-#ifdef CONFIG_DEBUG_BLK_CGROUP
 	BLKIO_STAT_AVG_QUEUE_SIZE,
 	BLKIO_STAT_IDLE_TIME,
 	BLKIO_STAT_EMPTY_TIME,
 	BLKIO_STAT_GROUP_WAIT_TIME,
 	BLKIO_STAT_DEQUEUE
 #endif
+};
+
+/* Per cpu stats */
+enum stat_type_cpu {
+	BLKIO_STAT_CPU_SECTORS,
+	/* Total bytes transferred */
+	BLKIO_STAT_CPU_SERVICE_BYTES,
+	/* Total IOs serviced, post merge */
+	BLKIO_STAT_CPU_SERVICED,
+	/* Number of IOs merged */
+	BLKIO_STAT_CPU_MERGED,
+	BLKIO_STAT_CPU_NR
 };
 
 enum stat_sub_type {
···
 struct blkio_group_stats {
 	/* total disk time and nr sectors dispatched by this group */
 	uint64_t time;
-	uint64_t sectors;
-	/* Time not charged to this cgroup */
-	uint64_t unaccounted_time;
 	uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL];
 #ifdef CONFIG_DEBUG_BLK_CGROUP
+	/* Time not charged to this cgroup */
+	uint64_t unaccounted_time;
+
 	/* Sum of number of IOs queued across all samples */
 	uint64_t avg_queue_size_sum;
 	/* Count of samples taken for average */
···
 #endif
 };
 
+/* Per cpu blkio group stats */
+struct blkio_group_stats_cpu {
+	uint64_t sectors;
+	uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL];
+	struct u64_stats_sync syncp;
+};
+
 struct blkio_group {
 	/* An rcu protected unique identifier for the group */
 	void *key;
···
 	/* Need to serialize the stats in the case of reset/update */
 	spinlock_t stats_lock;
 	struct blkio_group_stats stats;
+	/* Per cpu stats pointer */
+	struct blkio_group_stats_cpu __percpu *stats_cpu;
 };
 
 struct blkio_policy_node {
···
 extern void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
 	struct blkio_group *blkg, void *key, dev_t dev,
 	enum blkio_policy_id plid);
+extern int blkio_alloc_blkg_stats(struct blkio_group *blkg);
 extern int blkiocg_del_blkio_group(struct blkio_group *blkg);
 extern struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg,
 						void *key);
···
 static inline void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
 		struct blkio_group *blkg, void *key, dev_t dev,
 		enum blkio_policy_id plid) {}
+
+static inline int blkio_alloc_blkg_stats(struct blkio_group *blkg) { return 0; }
 
 static inline int
 blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; }

block/blk-core.c

···
 
 static inline void blk_free_request(struct request_queue *q, struct request *rq)
 {
-	BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
-
 	if (rq->cmd_flags & REQ_ELVPRIV)
 		elv_put_request(q, rq);
 	mempool_free(rq, q->rq.rq_pool);
···
 {
 	const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
 
-	/*
-	 * Debug stuff, kill later
-	 */
-	if (!rq_mergeable(req)) {
-		blk_dump_rq_flags(req, "back");
-		return false;
-	}
-
 	if (!ll_back_merge_fn(q, req, bio))
 		return false;
 
···
 	req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
 
 	drive_stat_acct(req, 0);
+	elv_bio_merged(q, req, bio);
 	return true;
 }
 
···
 {
 	const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
 	sector_t sector;
-
-	/*
-	 * Debug stuff, kill later
-	 */
-	if (!rq_mergeable(req)) {
-		blk_dump_rq_flags(req, "front");
-		return false;
-	}
 
 	if (!ll_front_merge_fn(q, req, bio))
 		return false;
···
 	req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
 
 	drive_stat_acct(req, 0);
+	elv_bio_merged(q, req, bio);
 	return true;
 }
 
···
 
 	el_ret = elv_merge(q, &req, bio);
 	if (el_ret == ELEVATOR_BACK_MERGE) {
-		BUG_ON(req->cmd_flags & REQ_ON_PLUG);
 		if (bio_attempt_back_merge(q, req, bio)) {
 			if (!attempt_back_merge(q, req))
 				elv_merged_request(q, req, el_ret);
 			goto out_unlock;
 		}
 	} else if (el_ret == ELEVATOR_FRONT_MERGE) {
-		BUG_ON(req->cmd_flags & REQ_ON_PLUG);
 		if (bio_attempt_front_merge(q, req, bio)) {
 			if (!attempt_front_merge(q, req))
 				elv_merged_request(q, req, el_ret);
···
 			if (__rq->q != q)
 				plug->should_sort = 1;
 		}
-		/*
-		 * Debug flag, kill later
-		 */
-		req->cmd_flags |= REQ_ON_PLUG;
 		list_add_tail(&req->queuelist, &plug->list);
 		drive_stat_acct(req, 1);
 	} else {
···
 			goto end_io;
 		}
 
-		blk_throtl_bio(q, &bio);
+		if (blk_throtl_bio(q, &bio))
+			goto end_io;
 
 		/*
 		 * If bio = NULL, bio has been throttled and will be submitted
···
 	while (!list_empty(&list)) {
 		rq = list_entry_rq(list.next);
 		list_del_init(&rq->queuelist);
-		BUG_ON(!(rq->cmd_flags & REQ_ON_PLUG));
 		BUG_ON(!rq->q);
 		if (rq->q != q) {
 			/*
···
 			depth = 0;
 			spin_lock(q->queue_lock);
 		}
-		rq->cmd_flags &= ~REQ_ON_PLUG;
-
 		/*
 		 * rq is already accounted, so use raw insert
 		 */

block/blk-exec.c

···
 	spin_lock_irq(q->queue_lock);
 	__elv_add_request(q, rq, where);
 	__blk_run_queue(q);
-	/* the queue is stopped so it won't be plugged+unplugged */
+	/* the queue is stopped so it won't be run */
 	if (rq->cmd_type == REQ_TYPE_PM_RESUME)
 		q->request_fn(q);
 	spin_unlock_irq(q->queue_lock);

block/blk-flush.c

···
 	}
 
 	/*
-	 * Moving a request silently to empty queue_head may stall the
-	 * queue.  Kick the queue in those cases.  This function is called
-	 * from request completion path and calling directly into
-	 * request_fn may confuse the driver.  Always use kblockd.
+	 * Kick the queue to avoid stall for two cases:
+	 * 1. Moving a request silently to empty queue_head may stall the
+	 * queue.
+	 * 2. When flush request is running in non-queueable queue, the
+	 * queue is hold. Restart the queue after flush request is finished
+	 * to avoid stall.
+	 * This function is called from request completion path and calling
+	 * directly into request_fn may confuse the driver.  Always use
+	 * kblockd.
 	 */
-	if (queued)
+	if (queued || q->flush_queue_delayed)
 		blk_run_queue_async(q);
+	q->flush_queue_delayed = 0;
 }
 
 /**

block/blk-ioc.c

···
 		INIT_RADIX_TREE(&ret->radix_root, GFP_ATOMIC | __GFP_HIGH);
 		INIT_HLIST_HEAD(&ret->cic_list);
 		ret->ioc_data = NULL;
+#if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
+		ret->cgroup_changed = 0;
+#endif
 	}
 
 	return ret;

block/blk-lib.c

···
 
 #include "blk.h"
 
-static void blkdev_discard_end_io(struct bio *bio, int err)
+struct bio_batch {
+	atomic_t		done;
+	unsigned long		flags;
+	struct completion	*wait;
+};
+
+static void bio_batch_end_io(struct bio *bio, int err)
 {
-	if (err) {
-		if (err == -EOPNOTSUPP)
-			set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
-		clear_bit(BIO_UPTODATE, &bio->bi_flags);
-	}
+	struct bio_batch *bb = bio->bi_private;
 
-	if (bio->bi_private)
-		complete(bio->bi_private);
-
+	if (err && (err != -EOPNOTSUPP))
+		clear_bit(BIO_UPTODATE, &bb->flags);
+	if (atomic_dec_and_test(&bb->done))
+		complete(bb->wait);
 	bio_put(bio);
 }
 
···
 	struct request_queue *q = bdev_get_queue(bdev);
 	int type = REQ_WRITE | REQ_DISCARD;
 	unsigned int max_discard_sectors;
+	struct bio_batch bb;
 	struct bio *bio;
 	int ret = 0;
 
···
 		type |= REQ_SECURE;
 	}
 
-	while (nr_sects && !ret) {
+	atomic_set(&bb.done, 1);
+	bb.flags = 1 << BIO_UPTODATE;
+	bb.wait = &wait;
+
+	while (nr_sects) {
 		bio = bio_alloc(gfp_mask, 1);
 		if (!bio) {
 			ret = -ENOMEM;
···
 		}
 
 		bio->bi_sector = sector;
-		bio->bi_end_io = blkdev_discard_end_io;
+		bio->bi_end_io = bio_batch_end_io;
 		bio->bi_bdev = bdev;
-		bio->bi_private = &wait;
+		bio->bi_private = &bb;
 
 		if (nr_sects > max_discard_sectors) {
 			bio->bi_size = max_discard_sectors << 9;
···
 			nr_sects = 0;
 		}
 
-		bio_get(bio);
+		atomic_inc(&bb.done);
 		submit_bio(type, bio);
+	}
 
+	/* Wait for bios in-flight */
+	if (!atomic_dec_and_test(&bb.done))
 		wait_for_completion(&wait);
 
-		if (bio_flagged(bio, BIO_EOPNOTSUPP))
-			ret = -EOPNOTSUPP;
-		else if (!bio_flagged(bio, BIO_UPTODATE))
-			ret = -EIO;
-		bio_put(bio);
-	}
+	if (!test_bit(BIO_UPTODATE, &bb.flags))
+		ret = -EIO;
 
 	return ret;
 }
 EXPORT_SYMBOL(blkdev_issue_discard);
-
-struct bio_batch
-{
-	atomic_t 		done;
-	unsigned long 		flags;
-	struct completion 	*wait;
-};
-
-static void bio_batch_end_io(struct bio *bio, int err)
-{
-	struct bio_batch *bb = bio->bi_private;
-
-	if (err) {
-		if (err == -EOPNOTSUPP)
-			set_bit(BIO_EOPNOTSUPP, &bb->flags);
-		else
-			clear_bit(BIO_UPTODATE, &bb->flags);
-	}
-	if (bb)
-		if (atomic_dec_and_test(&bb->done))
-			complete(bb->wait);
-	bio_put(bio);
-}
 
 /**
  * blkdev_issue_zeroout - generate number of zero filed write bios
···
 	bb.flags = 1 << BIO_UPTODATE;
 	bb.wait = &wait;
 
-submit:
 	ret = 0;
 	while (nr_sects != 0) {
 		bio = bio_alloc(gfp_mask,
···
 
 		while (nr_sects != 0) {
 			sz = min((sector_t) PAGE_SIZE >> 9 , nr_sects);
-			if (sz == 0)
-				/* bio has maximum size possible */
-				break;
 			ret = bio_add_page(bio, ZERO_PAGE(0), sz << 9, 0);
 			nr_sects -= ret >> 9;
 			sector += ret >> 9;
···
 		/* One of bios in the batch was completed with error.*/
 		ret = -EIO;
 
-	if (ret)
-		goto out;
-
-	if (test_bit(BIO_EOPNOTSUPP, &bb.flags)) {
-		ret = -EOPNOTSUPP;
-		goto out;
-	}
-	if (nr_sects != 0)
-		goto submit;
-out:
 	return ret;
 }
 EXPORT_SYMBOL(blkdev_issue_zeroout);

block/blk-settings.c

···
 	lim->discard_granularity = 0;
 	lim->discard_alignment = 0;
 	lim->discard_misaligned = 0;
-	lim->discard_zeroes_data = -1;
+	lim->discard_zeroes_data = 1;
 	lim->logical_block_size = lim->physical_block_size = lim->io_min = 512;
 	lim->bounce_pfn = (unsigned long)(BLK_BOUNCE_ANY >> PAGE_SHIFT);
 	lim->alignment_offset = 0;
···
 
 	blk_set_default_limits(&q->limits);
 	blk_queue_max_hw_sectors(q, BLK_SAFE_MAX_SECTORS);
+	q->limits.discard_zeroes_data = 0;
 
 	/*
 	 * by default assume old behaviour and bounce for any highmem page
···
 	q->flush_flags = flush & (REQ_FLUSH | REQ_FUA);
 }
 EXPORT_SYMBOL_GPL(blk_queue_flush);
+
+void blk_queue_flush_queueable(struct request_queue *q, bool queueable)
+{
+	q->flush_not_queueable = !queueable;
+}
+EXPORT_SYMBOL_GPL(blk_queue_flush_queueable);
 
 static int __init blk_settings_init(void)
 {

block/blk-sysfs.c

···
 
 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
 {
-	return queue_var_show(q->limits.max_discard_sectors << 9, page);
+	return sprintf(page, "%llu\n",
+		       (unsigned long long)q->limits.max_discard_sectors << 9);
 }
 
 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)

block/blk-throttle.c

···
 
 	/* Some throttle limits got updated for the group */
 	int limits_changed;
+
+	struct rcu_head rcu_head;
 };
 
 struct throtl_data
···
 	/* service tree for active throtl groups */
 	struct throtl_rb_root tg_service_tree;
 
-	struct throtl_grp root_tg;
+	struct throtl_grp *root_tg;
 	struct request_queue *queue;
 
 	/* Total Number of queued bios on READ and WRITE lists */
···
 	return tg;
 }
 
+static void throtl_free_tg(struct rcu_head *head)
+{
+	struct throtl_grp *tg;
+
+	tg = container_of(head, struct throtl_grp, rcu_head);
+	free_percpu(tg->blkg.stats_cpu);
+	kfree(tg);
+}
+
 static void throtl_put_tg(struct throtl_grp *tg)
 {
 	BUG_ON(atomic_read(&tg->ref) <= 0);
 	if (!atomic_dec_and_test(&tg->ref))
 		return;
-	kfree(tg);
+
+	/*
+	 * A group is freed in rcu manner. But having an rcu lock does not
+	 * mean that one can access all the fields of blkg and assume these
+	 * are valid. For example, don't try to follow throtl_data and
+	 * request queue links.
+	 *
+	 * Having a reference to blkg under an rcu allows acess to only
+	 * values local to groups like group stats and group rate limits
+	 */
+	call_rcu(&tg->rcu_head, throtl_free_tg);
 }
 
-static struct throtl_grp * throtl_find_alloc_tg(struct throtl_data *td,
-			struct blkio_cgroup *blkcg)
+static void throtl_init_group(struct throtl_grp *tg)
 {
-	struct throtl_grp *tg = NULL;
-	void *key = td;
-	struct backing_dev_info *bdi = &td->queue->backing_dev_info;
-	unsigned int major, minor;
-
-	/*
-	 * TODO: Speed up blkiocg_lookup_group() by maintaining a radix
-	 * tree of blkg (instead of traversing through hash list all
-	 * the time.
-	 */
-
-	/*
-	 * This is the common case when there are no blkio cgroups.
- 	 * Avoid lookup in this case
- 	 */
-	if (blkcg == &blkio_root_cgroup)
-		tg = &td->root_tg;
-	else
-		tg = tg_of_blkg(blkiocg_lookup_group(blkcg, key));
-
-	/* Fill in device details for root group */
-	if (tg && !tg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
-		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
-		tg->blkg.dev = MKDEV(major, minor);
-		goto done;
-	}
-
-	if (tg)
-		goto done;
-
-	tg = kzalloc_node(sizeof(*tg), GFP_ATOMIC, td->queue->node);
-	if (!tg)
-		goto done;
-
 	INIT_HLIST_NODE(&tg->tg_node);
 	RB_CLEAR_NODE(&tg->rb_node);
 	bio_list_init(&tg->bio_lists[0]);
 	bio_list_init(&tg->bio_lists[1]);
-	td->limits_changed = false;
+	tg->limits_changed = false;
+
+	/* Practically unlimited BW */
+	tg->bps[0] = tg->bps[1] = -1;
+	tg->iops[0] = tg->iops[1] = -1;
 
 	/*
 	 * Take the initial reference that will be released on destroy
···
 	 * exit or cgroup deletion path depending on who is exiting first.
 	 */
 	atomic_set(&tg->ref, 1);
+}
+
+/* Should be called with rcu read lock held (needed for blkcg) */
+static void
+throtl_add_group_to_td_list(struct throtl_data *td, struct throtl_grp *tg)
+{
+	hlist_add_head(&tg->tg_node, &td->tg_list);
+	td->nr_undestroyed_grps++;
+}
+
+static void
+__throtl_tg_fill_dev_details(struct throtl_data *td, struct throtl_grp *tg)
+{
+	struct backing_dev_info *bdi = &td->queue->backing_dev_info;
+	unsigned int major, minor;
+
+	if (!tg || tg->blkg.dev)
+		return;
+
+	/*
+	 * Fill in device details for a group which might not have been
+	 * filled at group creation time as queue was being instantiated
+	 * and driver had not attached a device yet
+	 */
+	if (bdi->dev && dev_name(bdi->dev)) {
+		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
+		tg->blkg.dev = MKDEV(major, minor);
+	}
+}
+
+/*
+ * Should be called with without queue lock held. Here queue lock will be
+ * taken rarely. It will be taken only once during life time of a group
+ * if need be
+ */
+static void
+throtl_tg_fill_dev_details(struct throtl_data *td, struct throtl_grp *tg)
+{
+	if (!tg || tg->blkg.dev)
+		return;
+
+	spin_lock_irq(td->queue->queue_lock);
+	__throtl_tg_fill_dev_details(td, tg);
+	spin_unlock_irq(td->queue->queue_lock);
+}
+
+static void throtl_init_add_tg_lists(struct throtl_data *td,
+			struct throtl_grp *tg, struct blkio_cgroup *blkcg)
+{
+	__throtl_tg_fill_dev_details(td, tg);
 
 	/* Add group onto cgroup list */
-	sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
 	blkiocg_add_blkio_group(blkcg, &tg->blkg, (void *)td,
-				MKDEV(major, minor), BLKIO_POLICY_THROTL);
+				tg->blkg.dev, BLKIO_POLICY_THROTL);
 
 	tg->bps[READ] = blkcg_get_read_bps(blkcg, tg->blkg.dev);
 	tg->bps[WRITE] = blkcg_get_write_bps(blkcg, tg->blkg.dev);
 	tg->iops[READ] = blkcg_get_read_iops(blkcg, tg->blkg.dev);
 	tg->iops[WRITE] = blkcg_get_write_iops(blkcg, tg->blkg.dev);
 
-	hlist_add_head(&tg->tg_node, &td->tg_list);
-	td->nr_undestroyed_grps++;
-done:
+	throtl_add_group_to_td_list(td, tg);
+}
+
+/* Should be called without queue lock and outside of rcu period */
+static struct throtl_grp *throtl_alloc_tg(struct throtl_data *td)
+{
+	struct throtl_grp *tg = NULL;
+	int ret;
+
+	tg = kzalloc_node(sizeof(*tg), GFP_ATOMIC, td->queue->node);
+	if (!tg)
+		return NULL;
+
+	ret = blkio_alloc_blkg_stats(&tg->blkg);
+
+	if (ret) {
+		kfree(tg);
+		return NULL;
+	}
+
+	throtl_init_group(tg);
 	return tg;
 }
 
-static struct throtl_grp * throtl_get_tg(struct throtl_data *td)
+static struct
+throtl_grp *throtl_find_tg(struct throtl_data *td, struct blkio_cgroup *blkcg)
 {
 	struct throtl_grp *tg = NULL;
+	void *key = td;
+
+	/*
+	 * This is the common case when there are no blkio cgroups.
+ 	 * Avoid lookup in this case
+ 	 */
+	if (blkcg == &blkio_root_cgroup)
+		tg = td->root_tg;
+	else
+		tg = tg_of_blkg(blkiocg_lookup_group(blkcg, key));
+
+	__throtl_tg_fill_dev_details(td, tg);
+	return tg;
+}
+
+/*
+ * This function returns with queue lock unlocked in case of error, like
+ * request queue is no more
+ */
+static struct throtl_grp * throtl_get_tg(struct throtl_data *td)
+{
+	struct throtl_grp *tg = NULL, *__tg = NULL;
 	struct blkio_cgroup *blkcg;
+	struct request_queue *q = td->queue;
 
 	rcu_read_lock();
 	blkcg = task_blkio_cgroup(current);
-	tg = throtl_find_alloc_tg(td, blkcg);
-	if (!tg)
-		tg = &td->root_tg;
+	tg = throtl_find_tg(td, blkcg);
+	if (tg) {
+		rcu_read_unlock();
+		return tg;
+	}
+
+	/*
+	 * Need to allocate a group. Allocation of group also needs allocation
+	 * of per cpu stats which in-turn takes a mutex() and can block. Hence
+	 * we need to drop rcu lock and queue_lock before we call alloc
+	 *
+	 * Take the request queue reference to make sure queue does not
+	 * go away once we return from allocation.
+	 */
+	blk_get_queue(q);
+	rcu_read_unlock();
+	spin_unlock_irq(q->queue_lock);
+
+	tg = throtl_alloc_tg(td);
+	/*
+	 * We might have slept in group allocation. Make sure queue is not
+	 * dead
+	 */
+	if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
+		blk_put_queue(q);
+		if (tg)
+			kfree(tg);
+
+		return ERR_PTR(-ENODEV);
+	}
+	blk_put_queue(q);
+
+	/* Group allocated and queue is still alive. take the lock */
+	spin_lock_irq(q->queue_lock);
+
+	/*
+	 * Initialize the new group. After sleeping, read the blkcg again.
+	 */
+	rcu_read_lock();
+	blkcg = task_blkio_cgroup(current);
+
+	/*
+	 * If some other thread already allocated the group while we were
+	 * not holding queue lock, free up the group
+	 */
+	__tg = throtl_find_tg(td, blkcg);
+
+	if (__tg) {
+		kfree(tg);
+		rcu_read_unlock();
+		return __tg;
+	}
+
+	/* Group allocation failed. Account the IO to root group */
+	if (!tg) {
+		tg = td->root_tg;
+		return tg;
+	}
+
+	throtl_init_add_tg_lists(td, tg, blkcg);
 	rcu_read_unlock();
 	return tg;
 }
···
 	return 0;
 }
 
+static bool tg_no_rule_group(struct throtl_grp *tg, bool rw) {
+	if (tg->bps[rw] == -1 && tg->iops[rw] == -1)
+		return 1;
+	return 0;
+}
+
 /*
  * Returns whether one can dispatch a bio or not. Also returns approx number
  * of jiffies to wait before this bio is with-in IO rate and can be dispatched
···
 	tg->bytes_disp[rw] += bio->bi_size;
 	tg->io_disp[rw]++;
 
-	/*
-	 * TODO: This will take blkg->stats_lock. Figure out a way
-	 * to avoid this cost.
-	 */
 	blkiocg_update_dispatch_stats(&tg->blkg, bio->bi_size, rw, sync);
 }
 
···
 	struct throtl_grp *tg;
 	struct bio *bio = *biop;
 	bool rw = bio_data_dir(bio), update_disptime = true;
+	struct blkio_cgroup *blkcg;
 
 	if (bio->bi_rw & REQ_THROTTLED) {
 		bio->bi_rw &= ~REQ_THROTTLED;
 		return 0;
 	}
 
+	/*
+	 * A throtl_grp pointer retrieved under rcu can be used to access
+	 * basic fields like stats and io rates. If a group has no rules,
+	 * just update the dispatch stats in lockless manner and return.
+	 */
+
+	rcu_read_lock();
+	blkcg = task_blkio_cgroup(current);
+	tg = throtl_find_tg(td, blkcg);
+	if (tg) {
+		throtl_tg_fill_dev_details(td, tg);
+
+		if (tg_no_rule_group(tg, rw)) {
+			blkiocg_update_dispatch_stats(&tg->blkg, bio->bi_size,
+					rw, bio->bi_rw & REQ_SYNC);
+			rcu_read_unlock();
+			return 0;
+		}
+	}
+	rcu_read_unlock();
+
+	/*
+	 * Either group has not been allocated yet or it is not an unlimited
+	 * IO group
+	 */
+
 	spin_lock_irq(q->queue_lock);
 	tg = throtl_get_tg(td);
+
+	if (IS_ERR(tg)) {
+		if (PTR_ERR(tg)	== -ENODEV) {
+			/*
+			 * Queue is gone. No queue lock held here.
+			 */
+			return -ENODEV;
+		}
+	}
 
 	if (tg->nr_queued[rw]) {
 		/*
···
 	INIT_HLIST_HEAD(&td->tg_list);
 	td->tg_service_tree = THROTL_RB_ROOT;
 	td->limits_changed = false;
-
-	/* Init root group */
-	tg = &td->root_tg;
-	INIT_HLIST_NODE(&tg->tg_node);
-	RB_CLEAR_NODE(&tg->rb_node);
-	bio_list_init(&tg->bio_lists[0]);
-	bio_list_init(&tg->bio_lists[1]);
-
-	/* Practically unlimited BW */
-	tg->bps[0] = tg->bps[1] = -1;
-	tg->iops[0] = tg->iops[1] = -1;
-	td->limits_changed = false;
-
-	/*
-	 * Set root group reference to 2. One reference will be dropped when
-	 * all groups on tg_list are being deleted during queue exit. Other
-	 * reference will remain there as we don't want to delete this group
-	 * as it is statically allocated and gets destroyed when throtl_data
-	 * goes away.
-	 */
-	atomic_set(&tg->ref, 2);
-	hlist_add_head(&tg->tg_node, &td->tg_list);
-	td->nr_undestroyed_grps++;
-
 	INIT_DELAYED_WORK(&td->throtl_work, blk_throtl_work);
 
+	/* alloc and Init root group. */
+	td->queue = q;
+	tg = throtl_alloc_tg(td);
+
+	if (!tg) {
+		kfree(td);
+		return -ENOMEM;
+	}
+
+	td->root_tg = tg;
+
 	rcu_read_lock();
-	blkiocg_add_blkio_group(&blkio_root_cgroup, &tg->blkg, (void *)td,
-					0, BLKIO_POLICY_THROTL);
+	throtl_init_add_tg_lists(td, tg, &blkio_root_cgroup);
 	rcu_read_unlock();
 
 	/* Attach throtl data to request queue */
-	td->queue = q;
 	q->td = td;
 	return 0;
 }

block/blk.h

···
 			return rq;
 		}
 
-		if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
+		/*
+		 * Flush request is running and flush request isn't queueable
+		 * in the drive, we can hold the queue till flush request is
+		 * finished. Even we don't do this, driver can't dispatch next
+		 * requests and will requeue them. And this can improve
+		 * throughput too. For example, we have request flush1, write1,
+		 * flush 2. flush1 is dispatched, then queue is hold, write1
+		 * isn't inserted to queue. After flush1 is finished, flush2
+		 * will be dispatched. Since disk cache is already clean,
+		 * flush2 will be finished very soon, so looks like flush2 is
+		 * folded to flush1.
+		 * Since the queue is hold, a flag is set to indicate the queue
+		 * should be restarted later. Please see flush_end_io() for
+		 * details.
+		 */
+		if (q->flush_pending_idx != q->flush_running_idx &&
+				!queue_flush_queueable(q)) {
+			q->flush_queue_delayed = 1;
+			return NULL;
+		}
+		if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags) ||
+		    !q->elevator->ops->elevator_dispatch_fn(q, 0))
 			return NULL;
 	}
 }

block/cfq-iosched.c

···
 
 	/* List of cfq groups being managed on this device*/
 	struct hlist_head cfqg_list;
-	struct rcu_head rcu;
+
+	/* Number of groups which are on blkcg->blkg_list */
+	unsigned int nr_blkcg_linked_grps;
 };
 
 static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd);
···
 	if (rq2 == NULL)
 		return rq1;
 
-	if (rq_is_sync(rq1) && !rq_is_sync(rq2))
-		return rq1;
-	else if (rq_is_sync(rq2) && !rq_is_sync(rq1))
-		return rq2;
-	if ((rq1->cmd_flags & REQ_META) && !(rq2->cmd_flags & REQ_META))
-		return rq1;
-	else if ((rq2->cmd_flags & REQ_META) &&
-		 !(rq1->cmd_flags & REQ_META))
-		return rq2;
+	if (rq_is_sync(rq1) != rq_is_sync(rq2))
+		return rq_is_sync(rq1) ? rq1 : rq2;
+
+	if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_META)
+		return rq1->cmd_flags & REQ_META ? rq1 : rq2;
 
 	s1 = blk_rq_pos(rq1);
 	s2 = blk_rq_pos(rq2);
···
 	cfqg->needs_update = true;
 }
 
-static struct cfq_group * cfq_find_alloc_cfqg(struct cfq_data *cfqd,
-		struct blkio_cgroup *blkcg, int create)
+static void cfq_init_add_cfqg_lists(struct cfq_data *cfqd,
+			struct cfq_group *cfqg, struct blkio_cgroup *blkcg)
 {
-	struct cfq_group *cfqg = NULL;
-	void *key = cfqd;
-	int i, j;
-	struct cfq_rb_root *st;
 	struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
 	unsigned int major, minor;
 
-	cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key));
-	if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
+	/*
+	 * Add group onto cgroup list. It might happen that bdi->dev is
+	 * not initialized yet. Initialize this new group without major
+	 * and minor info and this info will be filled in once a new thread
+	 * comes for IO.
+	 */
+	if (bdi->dev) {
 		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
-		cfqg->blkg.dev = MKDEV(major, minor);
-		goto done;
-	}
-	if (cfqg || !create)
-		goto done;
+		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg,
+					(void *)cfqd, MKDEV(major, minor));
+	} else
+		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg,
+					(void *)cfqd, 0);
+
+	cfqd->nr_blkcg_linked_grps++;
+	cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev);
+
+	/* Add group on cfqd list */
+	hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list);
+}
+
+/*
+ * Should be called from sleepable context. No request queue lock as per
+ * cpu stats are allocated dynamically and alloc_percpu needs to be called
+ * from sleepable context.
+ */
+static struct cfq_group * cfq_alloc_cfqg(struct cfq_data *cfqd)
+{
+	struct cfq_group *cfqg = NULL;
+	int i, j, ret;
+	struct cfq_rb_root *st;
 
 	cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node);
 	if (!cfqg)
-		goto done;
+		return NULL;
 
 	for_each_cfqg_st(cfqg, i, j, st)
 		*st = CFQ_RB_ROOT;
···
 	 */
 	cfqg->ref = 1;
 
+	ret = blkio_alloc_blkg_stats(&cfqg->blkg);
+	if (ret) {
+		kfree(cfqg);
+		return NULL;
+	}
+
+	return cfqg;
+}
+
+static struct cfq_group *
+cfq_find_cfqg(struct cfq_data *cfqd, struct blkio_cgroup *blkcg)
+{
+	struct cfq_group *cfqg = NULL;
+	void *key = cfqd;
+	struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info;
+	unsigned int major, minor;
+
 	/*
-	 * Add group onto cgroup list. It might happen that bdi->dev is
-	 * not initialized yet. Initialize this new group without major
-	 * and minor info and this info will be filled in once a new thread
-	 * comes for IO. See code above.
+	 * This is the common case when there are no blkio cgroups.
+	 * Avoid lookup in this case
 	 */
-	if (bdi->dev) {
+	if (blkcg == &blkio_root_cgroup)
+		cfqg = &cfqd->root_group;
+	else
+		cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key));
+
+	if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) {
 		sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor);
-		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd,
-					MKDEV(major, minor));
-	} else
-		cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, (void *)cfqd,
-					0);
+		cfqg->blkg.dev = MKDEV(major, minor);
+	}
 
-	cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev);
-
-	/* Add group on cfqd list */
-	hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list);
-
-done:
 	return cfqg;
 }
 
 /*
- * Search for the cfq group current task belongs to. If create = 1, then also
- * create the cfq group if it does not exist. request_queue lock must be held.
+ * Search for the cfq group current task belongs to. request_queue lock must
+ * be held.
  */
-static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create)
+static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd)
 {
 	struct blkio_cgroup *blkcg;
-	struct cfq_group *cfqg = NULL;
+	struct cfq_group *cfqg = NULL, *__cfqg = NULL;
+	struct request_queue *q = cfqd->queue;
 
 	rcu_read_lock();
 	blkcg = task_blkio_cgroup(current);
-	cfqg = cfq_find_alloc_cfqg(cfqd, blkcg, create);
-	if (!cfqg && create)
+	cfqg = cfq_find_cfqg(cfqd, blkcg);
+	if (cfqg) {
+		rcu_read_unlock();
+		return cfqg;
+	}
+
+	/*
+	 * Need to allocate a group. Allocation of group also needs allocation
+	 * of per cpu stats which in-turn takes a mutex() and can block. Hence
+	 * we need to drop rcu lock and queue_lock before we call alloc.
+	 *
+	 * Not taking any queue reference here and assuming that queue is
+	 * around by the time we return. CFQ queue allocation code does
+	 * the same. It might be racy though.
+	 */
+
+	rcu_read_unlock();
+	spin_unlock_irq(q->queue_lock);
+
+	cfqg = cfq_alloc_cfqg(cfqd);
+
+	spin_lock_irq(q->queue_lock);
+
+	rcu_read_lock();
+	blkcg = task_blkio_cgroup(current);
+
+	/*
+	 * If some other thread already allocated the group while we were
+	 * not holding queue lock, free up the group
+	 */
+	__cfqg = cfq_find_cfqg(cfqd, blkcg);
+
+	if (__cfqg) {
+		kfree(cfqg);
+		rcu_read_unlock();
+		return __cfqg;
+	}
+
+	if (!cfqg)
 		cfqg = &cfqd->root_group;
+
+	cfq_init_add_cfqg_lists(cfqd, cfqg, blkcg);
 	rcu_read_unlock();
 	return cfqg;
 }
···
 		return;
 	for_each_cfqg_st(cfqg, i, j, st)
 		BUG_ON(!RB_EMPTY_ROOT(&st->rb));
+	free_percpu(cfqg->blkg.stats_cpu);
 	kfree(cfqg);
 }
 
···
 }
 
 #else /* GROUP_IOSCHED */
-static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd, int create)
+static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd)
 {
 	return &cfqd->root_group;
 }
···
 	struct cfq_rb_root *service_tree;
 	int left;
 	int new_cfqq = 1;
-	int group_changed = 0;
 
 	service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq),
 						cfqq_type(cfqq));
···
 	rb_link_node(&cfqq->rb_node, parent, p);
 	rb_insert_color(&cfqq->rb_node, &service_tree->rb);
 	service_tree->count++;
-	if ((add_front || !new_cfqq) && !group_changed)
+	if (add_front || !new_cfqq)
 		return;
 	cfq_group_notify_queue_add(cfqd, cfqq->cfqg);
 }
···
 
 	WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR);
 
-	return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio));
+	return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio);
 }
 
 /*
···
 	struct cfq_group *cfqg;
 
 retry:
-	cfqg = cfq_get_cfqg(cfqd, 1);
+	cfqg = cfq_get_cfqg(cfqd);
 	cic = cfq_cic_lookup(cfqd, ioc);
 	/* cic always exists here */
 	cfqq = cic_to_cfqq(cic, is_sync);
···
 		cfq_put_queue(cfqd->async_idle_cfqq);
 }
 
-static void cfq_cfqd_free(struct rcu_head *head)
-{
-	kfree(container_of(head, struct cfq_data, rcu));
-}
-
 static void cfq_exit_queue(struct elevator_queue *e)
 {
 	struct cfq_data *cfqd = e->elevator_data;
 	struct request_queue *q = cfqd->queue;
+	bool wait = false;
 
 	cfq_shutdown_timer_wq(cfqd);
 
···
 
 	cfq_put_async_queues(cfqd);
 	cfq_release_cfq_groups(cfqd);
-	cfq_blkiocg_del_blkio_group(&cfqd->root_group.blkg);
+
+	/*
+	 * If there are groups which we could not unlink from blkcg list,
+	 * wait for a rcu period for them to be freed.
+	 */
+	if (cfqd->nr_blkcg_linked_grps)
+		wait = true;
 
 	spin_unlock_irq(q->queue_lock);
 
···
 	ida_remove(&cic_index_ida, cfqd->cic_index);
 	spin_unlock(&cic_index_lock);
 
-	/* Wait for cfqg->blkg->key accessors to exit their grace periods. */
-	call_rcu(&cfqd->rcu, cfq_cfqd_free);
+	/*
+	 * Wait for cfqg->blkg->key accessors to exit their grace periods.
+	 * Do this wait only if there are other unlinked groups out
+	 * there. This can happen if cgroup deletion path claimed the
+	 * responsibility of cleaning up a group before queue cleanup code
+	 * get to the group.
+	 *
+	 * Do not call synchronize_rcu() unconditionally as there are drivers
+	 * which create/delete request queue hundreds of times during scan/boot
+	 * and synchronize_rcu() can take significant time and slow down boot.
+	 */
+	if (wait)
+		synchronize_rcu();
+
+#ifdef CONFIG_CFQ_GROUP_IOSCHED
+	/* Free up per cpu stats for root group */
+	free_percpu(cfqd->root_group.blkg.stats_cpu);
+#endif
+	kfree(cfqd);
 }
 
 static int cfq_alloc_cic_index(void)
···
 		return NULL;
 
 	cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node);
-	if (!cfqd)
+	if (!cfqd) {
+		spin_lock(&cic_index_lock);
+		ida_remove(&cic_index_ida, i);
+		spin_unlock(&cic_index_lock);
 		return NULL;
+	}
 
 	/*
 	 * Don't need take queue_lock in the routine, since we are
···
 
 #ifdef CONFIG_CFQ_GROUP_IOSCHED
 	/*
-	 * Take a reference to root group which we never drop. This is just
-	 * to make sure that cfq_put_cfqg() does not try to kfree root group
+	 * Set root group reference to 2. One reference will be dropped when
+	 * all groups on cfqd->cfqg_list are being deleted during queue exit.
+	 * Other reference will remain there as we don't want to delete this
+	 * group as it is statically allocated and gets destroyed when
+	 * throtl_data goes away.
 	 */
-	cfqg->ref = 1;
+	cfqg->ref = 2;
+
+	if (blkio_alloc_blkg_stats(&cfqg->blkg)) {
+		kfree(cfqg);
+		kfree(cfqd);
+		return NULL;
+	}
+
 	rcu_read_lock();
+
 	cfq_blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg,
 					(void *)cfqd, 0);
 	rcu_read_unlock();
+	cfqd->nr_blkcg_linked_grps++;
+
+	/* Add group on cfqd->cfqg_list */
+	hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list);
 #endif
 	/*
 	 * Not strictly needed (since RB_ROOT just clears the node and we

block/elevator.c

···
 
 	e = elevator_find(name);
 	if (!e) {
-		char elv[ELV_NAME_MAX + strlen("-iosched")];
-
 		spin_unlock(&elv_list_lock);
-
-		snprintf(elv, sizeof(elv), "%s-iosched", name);
-
-		request_module("%s", elv);
+		request_module("%s-iosched", name);
 		spin_lock(&elv_list_lock);
 		e = elevator_find(name);
 	}
···
 	struct list_head *entry;
 	int stop_flags;
 
-	BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
-
 	if (q->last_merge == rq)
 		q->last_merge = NULL;
 
···
 	trace_block_rq_insert(q, rq);
 
 	rq->q = q;
-
-	BUG_ON(rq->cmd_flags & REQ_ON_PLUG);
 
 	if (rq->cmd_flags & REQ_SOFTBARRIER) {
 		/* barriers are scheduling boundary, update end_sector */

drivers/ata/libata-scsi.c

···
 static int ata_scsi_dev_config(struct scsi_device *sdev,
 			       struct ata_device *dev)
 {
+	struct request_queue *q = sdev->request_queue;
+
 	if (!ata_id_has_unload(dev->id))
 		dev->flags |= ATA_DFLAG_NO_UNLOAD;
 
 	/* configure max sectors */
-	blk_queue_max_hw_sectors(sdev->request_queue, dev->max_sectors);
+	blk_queue_max_hw_sectors(q, dev->max_sectors);
 
 	if (dev->class == ATA_DEV_ATAPI) {
-		struct request_queue *q = sdev->request_queue;
 		void *buf;
 
 		sdev->sector_size = ATA_SECT_SIZE;
 
 		/* set DMA padding */
-		blk_queue_update_dma_pad(sdev->request_queue,
-					 ATA_DMA_PAD_SZ - 1);
+		blk_queue_update_dma_pad(q, ATA_DMA_PAD_SZ - 1);
 
 		/* configure draining */
 		buf = kmalloc(ATAPI_MAX_DRAIN, q->bounce_gfp | GFP_KERNEL);
···
 			"sector_size=%u > PAGE_SIZE, PIO may malfunction\n",
 			sdev->sector_size);
 
-	blk_queue_update_dma_alignment(sdev->request_queue,
-				       sdev->sector_size - 1);
+	blk_queue_update_dma_alignment(q, sdev->sector_size - 1);
 
 	if (dev->flags & ATA_DFLAG_AN)
 		set_bit(SDEV_EVT_MEDIA_CHANGE, sdev->supported_events);
···
 		depth = min(ATA_MAX_QUEUE - 1, depth);
 		scsi_adjust_queue_depth(sdev, MSG_SIMPLE_TAG, depth);
 	}
+
+	blk_queue_flush_queueable(q, false);
 
 	dev->sdev = sdev;
 	return 0;

drivers/block/paride/pcd.c

···
 		disk->first_minor = unit;
 		strcpy(disk->disk_name, cd->name);	/* umm... */
 		disk->fops = &pcd_bdops;
+		disk->flags = GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE;
+		disk->events = DISK_EVENT_MEDIA_CHANGE;
 	}
 }
 

drivers/cdrom/viocd.c

···
 	blk_queue_max_hw_sectors(q, 4096 / 512);
 	gendisk->queue = q;
 	gendisk->fops = &viocd_fops;
-	gendisk->flags = GENHD_FL_CD|GENHD_FL_REMOVABLE;
+	gendisk->flags = GENHD_FL_CD | GENHD_FL_REMOVABLE |
+			 GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE;
+	gendisk->events = DISK_EVENT_MEDIA_CHANGE;
 	set_capacity(gendisk, 0);
 	gendisk->private_data = d;
 	d->viocd_disk = gendisk;

drivers/ide/ide-cd.c

···
 
 	ide_cd_read_toc(drive, &sense);
 	g->fops = &idecd_ops;
-	g->flags |= GENHD_FL_REMOVABLE;
+	g->flags |= GENHD_FL_REMOVABLE | GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE;
+	g->events = DISK_EVENT_MEDIA_CHANGE;
 	add_disk(g);
 	return 0;
 

drivers/scsi/sr.c

···
 	disk->first_minor = minor;
 	sprintf(disk->disk_name, "sr%d", minor);
 	disk->fops = &sr_bdops;
-	disk->flags = GENHD_FL_CD;
+	disk->flags = GENHD_FL_CD | GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE;
 	disk->events = DISK_EVENT_MEDIA_CHANGE | DISK_EVENT_EJECT_REQUEST;
 
 	blk_queue_rq_timeout(sdev->request_queue, SR_TIMEOUT);

fs/block_dev.c

···
 	res = __blkdev_get(bdev, mode, 0);
 
 	if (whole) {
+		struct gendisk *disk = whole->bd_disk;
+
 		/* finish claiming */
 		mutex_lock(&bdev->bd_mutex);
 		spin_lock(&bdev_lock);
···
 		spin_unlock(&bdev_lock);
 
 		/*
-		 * Block event polling for write claims.  Any write
-		 * holder makes the write_holder state stick until all
-		 * are released.  This is good enough and tracking
-		 * individual writeable reference is too fragile given
-		 * the way @mode is used in blkdev_get/put().
+		 * Block event polling for write claims if requested.  Any
+		 * write holder makes the write_holder state stick until
+		 * all are released.  This is good enough and tracking
+		 * individual writeable reference is too fragile given the
+		 * way @mode is used in blkdev_get/put().
 		 */
-		if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder) {
+		if ((disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE) &&
+		    !res && (mode & FMODE_WRITE) && !bdev->bd_write_holder) {
 			bdev->bd_write_holder = true;
-			disk_block_events(bdev->bd_disk);
+			disk_block_events(disk);
 		}
 
 		mutex_unlock(&bdev->bd_mutex);

fs/partitions/check.c

···
 				   struct device_attribute *attr, char *buf)
 {
 	struct hd_struct *p = dev_to_part(dev);
-	return sprintf(buf, "%u\n", p->discard_alignment);
+	struct gendisk *disk = dev_to_disk(dev);
+
+	return sprintf(buf, "%u\n",
+			queue_limit_discard_alignment(&disk->queue->limits,
+							p->start_sect));
 }
 
 ssize_t part_stat_show(struct device *dev,
···
 	p->start_sect = start;
 	p->alignment_offset =
 		queue_limit_alignment_offset(&disk->queue->limits, start);
-	p->discard_alignment =
-		queue_limit_discard_alignment(&disk->queue->limits, start);
 	p->nr_sects = len;
 	p->partno = partno;
 	p->policy = get_disk_ro(disk);

include/linux/blk_types.h

···
 	__REQ_IO_STAT,		/* account I/O stat */
 	__REQ_MIXED_MERGE,	/* merge of different types, fail separately */
 	__REQ_SECURE,		/* secure discard (used with __REQ_DISCARD) */
-	__REQ_ON_PLUG,		/* on plug list */
 	__REQ_NR_BITS,		/* stops here */
 };
 
···
 #define REQ_IO_STAT		(1 << __REQ_IO_STAT)
 #define REQ_MIXED_MERGE		(1 << __REQ_MIXED_MERGE)
 #define REQ_SECURE		(1 << __REQ_SECURE)
-#define REQ_ON_PLUG		(1 << __REQ_ON_PLUG)
 
 #endif /* __LINUX_BLK_TYPES_H */

include/linux/blkdev.h

···
 	unsigned char		misaligned;
 	unsigned char		discard_misaligned;
 	unsigned char		cluster;
-	signed char		discard_zeroes_data;
+	unsigned char		discard_zeroes_data;
 };
 
 struct request_queue
···
 	 * for flush operations
 	 */
 	unsigned int		flush_flags;
+	unsigned int		flush_not_queueable:1;
+	unsigned int		flush_queue_delayed:1;
 	unsigned int		flush_pending_idx:1;
 	unsigned int		flush_running_idx:1;
 	unsigned long		flush_pending_since;
···
 extern void blk_queue_rq_timed_out(struct request_queue *, rq_timed_out_fn *);
 extern void blk_queue_rq_timeout(struct request_queue *, unsigned int);
 extern void blk_queue_flush(struct request_queue *q, unsigned int flush);
+extern void blk_queue_flush_queueable(struct request_queue *q, bool queueable);
 extern struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev);
 
 extern int blk_rq_map_sg(struct request_queue *, struct request *, struct scatterlist *);
···
 {
 	unsigned int alignment = (sector << 9) & (lim->discard_granularity - 1);
 
+	if (!lim->max_discard_sectors)
+		return 0;
+
 	return (lim->discard_granularity + lim->discard_alignment - alignment)
 		& (lim->discard_granularity - 1);
 }
 
 static inline unsigned int queue_discard_zeroes_data(struct request_queue *q)
 {
-	if (q->limits.discard_zeroes_data == 1)
+	if (q->limits.max_discard_sectors && q->limits.discard_zeroes_data == 1)
 		return 1;
 
 	return 0;
···
 static inline unsigned int block_size(struct block_device *bdev)
 {
 	return bdev->bd_block_size;
+}
+
+static inline bool queue_flush_queueable(struct request_queue *q)
+{
+	return !q->flush_not_queueable;
 }
 
 typedef struct {struct page *v;} Sector;

include/linux/genhd.h

···
 	sector_t start_sect;
 	sector_t nr_sects;
 	sector_t alignment_offset;
-	unsigned int discard_alignment;
 	struct device __dev;
 	struct kobject *holder_dir;
 	int policy, partno;
···
 #define GENHD_FL_SUPPRESS_PARTITION_INFO	32
 #define GENHD_FL_EXT_DEVT			64 /* allow extended devt */
 #define GENHD_FL_NATIVE_CAPACITY		128
+#define GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE	256
 
 enum {
 	DISK_EVENT_MEDIA_CHANGE			= 1 << 0, /* media changed */

mm/backing-dev.c

···
 	unsigned long background_thresh;
 	unsigned long dirty_thresh;
 	unsigned long bdi_thresh;
-	unsigned long nr_dirty, nr_io, nr_more_io, nr_wb;
+	unsigned long nr_dirty, nr_io, nr_more_io;
 	struct inode *inode;
 
-	nr_wb = nr_dirty = nr_io = nr_more_io = 0;
+	nr_dirty = nr_io = nr_more_io = 0;
 	spin_lock(&inode_wb_list_lock);
 	list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
 		nr_dirty++;