Merge branch 'for-3.11/core' of git://git.kernel.dk/linux-block

Pull core block IO updates from Jens Axboe:
 "Here are the core IO block bits for 3.11. It contains:

   - A tweak to the reserved tag logic from Jan, for weirdo devices with
     just 3 free tags.  But for those it improves things substantially
     for random writes.

   - Periodic writeback fix from Jan.  Marked for stable as well.

   - Fix for a race condition in IO scheduler switching from Jianpeng.

   - The hierarchical blk-cgroup support from Tejun.  This is the grunt
     of the series.

   - blk-throttle fix from Vivek.

  Just a note that I'm in the middle of a relocation, whole family is
  flying out tomorrow.  Hence I will be awal the remainder of this week,
  but back at work again on Monday the 15th.  CC'ing Tejun, since any
  potential "surprises" will most likely be from the blk-cgroup work.
  But it's been brewing for a while and sitting in my tree and
  linux-next for a long time, so should be solid."

* 'for-3.11/core' of git://git.kernel.dk/linux-block: (36 commits)
  elevator: Fix a race in elevator switching
  block: Reserve only one queue tag for sync IO if only 3 tags are available
  writeback: Fix periodic writeback after fs mount
  blk-throttle: implement proper hierarchy support
  blk-throttle: implement throtl_grp->has_rules[]
  blk-throttle: Account for child group's start time in parent while bio climbs up
  blk-throttle: add throtl_qnode for dispatch fairness
  blk-throttle: make throtl_pending_timer_fn() ready for hierarchy
  blk-throttle: make tg_dispatch_one_bio() ready for hierarchy
  blk-throttle: make blk_throtl_bio() ready for hierarchy
  blk-throttle: make blk_throtl_drain() ready for hierarchy
  blk-throttle: dispatch from throtl_pending_timer_fn()
  blk-throttle: implement dispatch looping
  blk-throttle: separate out throtl_service_queue->pending_timer from throtl_data->dispatch_work
  blk-throttle: set REQ_THROTTLED from throtl_charge_bio() and gate stats update with it
  blk-throttle: implement sq_to_tg(), sq_to_td() and throtl_log()
  blk-throttle: add throtl_service_queue->parent_sq
  blk-throttle: generalize update_disptime optimization in blk_throtl_bio()
  blk-throttle: dispatch to throtl_data->service_queue.bio_lists[]
  blk-throttle: move bio_lists[] and friends to throtl_service_queue
  ...

12 files changed, 905 insertions, 434 deletions

diff --git a/Documentation/cgroups/blkio-controller.txt b/Documentation/cgroups/blkio-controller.txt
index da272c8f44e..cd556b91478 100644
--- a/Documentation/cgroups/blkio-controller.txt
+++ b/Documentation/cgroups/blkio-controller.txt
@@ -94,11 +94,13 @@ Throttling/Upper Limit policy
 
 Hierarchical Cgroups
 ====================
-- Currently only CFQ supports hierarchical groups. For throttling,
-  cgroup interface does allow creation of hierarchical cgroups and
-  internally it treats them as flat hierarchy.
 
-  If somebody created a hierarchy like as follows.
+Both CFQ and throttling implement hierarchy support; however,
+throttling's hierarchy support is enabled iff "sane_behavior" is
+enabled from cgroup side, which currently is a development option and
+not publicly available.
+
+If somebody created a hierarchy like as follows.
 
 			root
 			/  \
@@ -106,21 +108,20 @@ Hierarchical Cgroups
 			|
 		     test3
 
-  CFQ will handle the hierarchy correctly but and throttling will
-  practically treat all groups at same level. For details on CFQ
-  hierarchy support, refer to Documentation/block/cfq-iosched.txt.
-  Throttling will treat the hierarchy as if it looks like the
-  following.
+CFQ by default and throttling with "sane_behavior" will handle the
+hierarchy correctly.  For details on CFQ hierarchy support, refer to
+Documentation/block/cfq-iosched.txt.  For throttling, all limits apply
+to the whole subtree while all statistics are local to the IOs
+directly generated by tasks in that cgroup.
+
+Throttling without "sane_behavior" enabled from cgroup side will
+practically treat all groups at same level as if it looks like the
+following.
 
 				pivot
 			     /  /   \  \
 			root  test1 test2  test3
 
-  Nesting cgroups, while allowed, isn't officially supported and blkio
-  genereates warning when cgroups nest. Once throttling implements
-  hierarchy support, hierarchy will be supported and the warning will
-  be removed.
-
 Various user visible config options
 ===================================
 CONFIG_BLK_CGROUP
diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
index e8918ffaf96..290792a13e3 100644
--- a/block/blk-cgroup.c
+++ b/block/blk-cgroup.c
@@ -32,26 +32,6 @@ EXPORT_SYMBOL_GPL(blkcg_root);
 
 static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
 
-static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
-				      struct request_queue *q, bool update_hint);
-
-/**
- * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
- * @d_blkg: loop cursor pointing to the current descendant
- * @pos_cgrp: used for iteration
- * @p_blkg: target blkg to walk descendants of
- *
- * Walk @c_blkg through the descendants of @p_blkg.  Must be used with RCU
- * read locked.  If called under either blkcg or queue lock, the iteration
- * is guaranteed to include all and only online blkgs.  The caller may
- * update @pos_cgrp by calling cgroup_rightmost_descendant() to skip
- * subtree.
- */
-#define blkg_for_each_descendant_pre(d_blkg, pos_cgrp, p_blkg)		\
-	cgroup_for_each_descendant_pre((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
-		if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
-					      (p_blkg)->q, false)))
-
 static bool blkcg_policy_enabled(struct request_queue *q,
 				 const struct blkcg_policy *pol)
 {
@@ -71,18 +51,8 @@ static void blkg_free(struct blkcg_gq *blkg)
 	if (!blkg)
 		return;
 
-	for (i = 0; i < BLKCG_MAX_POLS; i++) {
-		struct blkcg_policy *pol = blkcg_policy[i];
-		struct blkg_policy_data *pd = blkg->pd[i];
-
-		if (!pd)
-			continue;
-
-		if (pol && pol->pd_exit_fn)
-			pol->pd_exit_fn(blkg);
-
-		kfree(pd);
-	}
+	for (i = 0; i < BLKCG_MAX_POLS; i++)
+		kfree(blkg->pd[i]);
 
 	blk_exit_rl(&blkg->rl);
 	kfree(blkg);
@@ -134,10 +104,6 @@ static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
 		blkg->pd[i] = pd;
 		pd->blkg = blkg;
 		pd->plid = i;
-
-		/* invoke per-policy init */
-		if (pol->pd_init_fn)
-			pol->pd_init_fn(blkg);
 	}
 
 	return blkg;
@@ -158,8 +124,8 @@ err_free:
  * @q's bypass state.  If @update_hint is %true, the caller should be
  * holding @q->queue_lock and lookup hint is updated on success.
  */
-static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
-				      struct request_queue *q, bool update_hint)
+struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
+			       bool update_hint)
 {
 	struct blkcg_gq *blkg;
 
@@ -234,16 +200,25 @@ static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
 	}
 	blkg = new_blkg;
 
-	/* link parent and insert */
+	/* link parent */
 	if (blkcg_parent(blkcg)) {
 		blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
 		if (WARN_ON_ONCE(!blkg->parent)) {
-			blkg = ERR_PTR(-EINVAL);
+			ret = -EINVAL;
 			goto err_put_css;
 		}
 		blkg_get(blkg->parent);
 	}
 
+	/* invoke per-policy init */
+	for (i = 0; i < BLKCG_MAX_POLS; i++) {
+		struct blkcg_policy *pol = blkcg_policy[i];
+
+		if (blkg->pd[i] && pol->pd_init_fn)
+			pol->pd_init_fn(blkg);
+	}
+
+	/* insert */
 	spin_lock(&blkcg->lock);
 	ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
 	if (likely(!ret)) {
@@ -394,30 +369,38 @@ static void blkg_destroy_all(struct request_queue *q)
 	q->root_rl.blkg = NULL;
 }
 
-static void blkg_rcu_free(struct rcu_head *rcu_head)
+/*
+ * A group is RCU protected, 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 accesses to only values
+ * local to groups like group stats and group rate limits.
+ */
+void __blkg_release_rcu(struct rcu_head *rcu_head)
 {
-	blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head));
-}
+	struct blkcg_gq *blkg = container_of(rcu_head, struct blkcg_gq, rcu_head);
+	int i;
+
+	/* tell policies that this one is being freed */
+	for (i = 0; i < BLKCG_MAX_POLS; i++) {
+		struct blkcg_policy *pol = blkcg_policy[i];
+
+		if (blkg->pd[i] && pol->pd_exit_fn)
+			pol->pd_exit_fn(blkg);
+	}
 
-void __blkg_release(struct blkcg_gq *blkg)
-{
 	/* release the blkcg and parent blkg refs this blkg has been holding */
 	css_put(&blkg->blkcg->css);
-	if (blkg->parent)
+	if (blkg->parent) {
+		spin_lock_irq(blkg->q->queue_lock);
 		blkg_put(blkg->parent);
+		spin_unlock_irq(blkg->q->queue_lock);
+	}
 
-	/*
-	 * 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(&blkg->rcu_head, blkg_rcu_free);
+	blkg_free(blkg);
 }
-EXPORT_SYMBOL_GPL(__blkg_release);
+EXPORT_SYMBOL_GPL(__blkg_release_rcu);
 
 /*
  * The next function used by blk_queue_for_each_rl().  It's a bit tricky
@@ -928,14 +911,6 @@ struct cgroup_subsys blkio_subsys = {
 	.subsys_id = blkio_subsys_id,
 	.base_cftypes = blkcg_files,
 	.module = THIS_MODULE,
-
-	/*
-	 * blkio subsystem is utterly broken in terms of hierarchy support.
-	 * It treats all cgroups equally regardless of where they're
-	 * located in the hierarchy - all cgroups are treated as if they're
-	 * right below the root.  Fix it and remove the following.
-	 */
-	.broken_hierarchy = true,
 };
 EXPORT_SYMBOL_GPL(blkio_subsys);
 
diff --git a/block/blk-cgroup.h b/block/blk-cgroup.h
index 4e595ee8c91..8056c03a338 100644
--- a/block/blk-cgroup.h
+++ b/block/blk-cgroup.h
@@ -266,7 +266,7 @@ static inline void blkg_get(struct blkcg_gq *blkg)
 	blkg->refcnt++;
 }
 
-void __blkg_release(struct blkcg_gq *blkg);
+void __blkg_release_rcu(struct rcu_head *rcu);
 
 /**
  * blkg_put - put a blkg reference
@@ -279,9 +279,43 @@ static inline void blkg_put(struct blkcg_gq *blkg)
 	lockdep_assert_held(blkg->q->queue_lock);
 	WARN_ON_ONCE(blkg->refcnt <= 0);
 	if (!--blkg->refcnt)
-		__blkg_release(blkg);
+		call_rcu(&blkg->rcu_head, __blkg_release_rcu);
 }
 
+struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg, struct request_queue *q,
+			       bool update_hint);
+
+/**
+ * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
+ * @d_blkg: loop cursor pointing to the current descendant
+ * @pos_cgrp: used for iteration
+ * @p_blkg: target blkg to walk descendants of
+ *
+ * Walk @c_blkg through the descendants of @p_blkg.  Must be used with RCU
+ * read locked.  If called under either blkcg or queue lock, the iteration
+ * is guaranteed to include all and only online blkgs.  The caller may
+ * update @pos_cgrp by calling cgroup_rightmost_descendant() to skip
+ * subtree.
+ */
+#define blkg_for_each_descendant_pre(d_blkg, pos_cgrp, p_blkg)		\
+	cgroup_for_each_descendant_pre((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
+		if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
+					      (p_blkg)->q, false)))
+
+/**
+ * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
+ * @d_blkg: loop cursor pointing to the current descendant
+ * @pos_cgrp: used for iteration
+ * @p_blkg: target blkg to walk descendants of
+ *
+ * Similar to blkg_for_each_descendant_pre() but performs post-order
+ * traversal instead.  Synchronization rules are the same.
+ */
+#define blkg_for_each_descendant_post(d_blkg, pos_cgrp, p_blkg)		\
+	cgroup_for_each_descendant_post((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
+		if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
+					      (p_blkg)->q, false)))
+
 /**
  * blk_get_rl - get request_list to use
  * @q: request_queue of interest
diff --git a/block/blk-tag.c b/block/blk-tag.c
index cc345e1d8d4..3f33d867226 100644
--- a/block/blk-tag.c
+++ b/block/blk-tag.c
@@ -348,9 +348,16 @@ int blk_queue_start_tag(struct request_queue *q, struct request *rq)
 	 */
 	max_depth = bqt->max_depth;
 	if (!rq_is_sync(rq) && max_depth > 1) {
-		max_depth -= 2;
-		if (!max_depth)
+		switch (max_depth) {
+		case 2:
 			max_depth = 1;
+			break;
+		case 3:
+			max_depth = 2;
+			break;
+		default:
+			max_depth -= 2;
+		}
 		if (q->in_flight[BLK_RW_ASYNC] > max_depth)
 			return 1;
 	}
diff --git a/block/blk-throttle.c b/block/blk-throttle.c
index 31146225f3d..08a32dfd384 100644
--- a/block/blk-throttle.c
+++ b/block/blk-throttle.c
@@ -25,18 +25,61 @@ static struct blkcg_policy blkcg_policy_throtl;
 
 /* A workqueue to queue throttle related work */
 static struct workqueue_struct *kthrotld_workqueue;
-static void throtl_schedule_delayed_work(struct throtl_data *td,
-				unsigned long delay);
-
-struct throtl_rb_root {
-	struct rb_root rb;
-	struct rb_node *left;
-	unsigned int count;
-	unsigned long min_disptime;
+
+/*
+ * To implement hierarchical throttling, throtl_grps form a tree and bios
+ * are dispatched upwards level by level until they reach the top and get
+ * issued.  When dispatching bios from the children and local group at each
+ * level, if the bios are dispatched into a single bio_list, there's a risk
+ * of a local or child group which can queue many bios at once filling up
+ * the list starving others.
+ *
+ * To avoid such starvation, dispatched bios are queued separately
+ * according to where they came from.  When they are again dispatched to
+ * the parent, they're popped in round-robin order so that no single source
+ * hogs the dispatch window.
+ *
+ * throtl_qnode is used to keep the queued bios separated by their sources.
+ * Bios are queued to throtl_qnode which in turn is queued to
+ * throtl_service_queue and then dispatched in round-robin order.
+ *
+ * It's also used to track the reference counts on blkg's.  A qnode always
+ * belongs to a throtl_grp and gets queued on itself or the parent, so
+ * incrementing the reference of the associated throtl_grp when a qnode is
+ * queued and decrementing when dequeued is enough to keep the whole blkg
+ * tree pinned while bios are in flight.
+ */
+struct throtl_qnode {
+	struct list_head	node;		/* service_queue->queued[] */
+	struct bio_list		bios;		/* queued bios */
+	struct throtl_grp	*tg;		/* tg this qnode belongs to */
 };
 
-#define THROTL_RB_ROOT	(struct throtl_rb_root) { .rb = RB_ROOT, .left = NULL, \
-			.count = 0, .min_disptime = 0}
+struct throtl_service_queue {
+	struct throtl_service_queue *parent_sq;	/* the parent service_queue */
+
+	/*
+	 * Bios queued directly to this service_queue or dispatched from
+	 * children throtl_grp's.
+	 */
+	struct list_head	queued[2];	/* throtl_qnode [READ/WRITE] */
+	unsigned int		nr_queued[2];	/* number of queued bios */
+
+	/*
+	 * RB tree of active children throtl_grp's, which are sorted by
+	 * their ->disptime.
+	 */
+	struct rb_root		pending_tree;	/* RB tree of active tgs */
+	struct rb_node		*first_pending;	/* first node in the tree */
+	unsigned int		nr_pending;	/* # queued in the tree */
+	unsigned long		first_pending_disptime;	/* disptime of the first tg */
+	struct timer_list	pending_timer;	/* fires on first_pending_disptime */
+};
+
+enum tg_state_flags {
+	THROTL_TG_PENDING	= 1 << 0,	/* on parent's pending tree */
+	THROTL_TG_WAS_EMPTY	= 1 << 1,	/* bio_lists[] became non-empty */
+};
 
 #define rb_entry_tg(node)	rb_entry((node), struct throtl_grp, rb_node)
 
@@ -52,9 +95,26 @@ struct throtl_grp {
 	/* must be the first member */
 	struct blkg_policy_data pd;
 
-	/* active throtl group service_tree member */
+	/* active throtl group service_queue member */
 	struct rb_node rb_node;
 
+	/* throtl_data this group belongs to */
+	struct throtl_data *td;
+
+	/* this group's service queue */
+	struct throtl_service_queue service_queue;
+
+	/*
+	 * qnode_on_self is used when bios are directly queued to this
+	 * throtl_grp so that local bios compete fairly with bios
+	 * dispatched from children.  qnode_on_parent is used when bios are
+	 * dispatched from this throtl_grp into its parent and will compete
+	 * with the sibling qnode_on_parents and the parent's
+	 * qnode_on_self.
+	 */
+	struct throtl_qnode qnode_on_self[2];
+	struct throtl_qnode qnode_on_parent[2];
+
 	/*
 	 * Dispatch time in jiffies. This is the estimated time when group
 	 * will unthrottle and is ready to dispatch more bio. It is used as
@@ -64,11 +124,8 @@ struct throtl_grp {
 
 	unsigned int flags;
 
-	/* Two lists for READ and WRITE */
-	struct bio_list bio_lists[2];
-
-	/* Number of queued bios on READ and WRITE lists */
-	unsigned int nr_queued[2];
+	/* are there any throtl rules between this group and td? */
+	bool has_rules[2];
 
 	/* bytes per second rate limits */
 	uint64_t bps[2];
@@ -85,9 +142,6 @@ struct throtl_grp {
 	unsigned long slice_start[2];
 	unsigned long slice_end[2];
 
-	/* Some throttle limits got updated for the group */
-	int limits_changed;
-
 	/* Per cpu stats pointer */
 	struct tg_stats_cpu __percpu *stats_cpu;
 
@@ -98,7 +152,7 @@ struct throtl_grp {
 struct throtl_data
 {
 	/* service tree for active throtl groups */
-	struct throtl_rb_root tg_service_tree;
+	struct throtl_service_queue service_queue;
 
 	struct request_queue *queue;
 
@@ -111,9 +165,7 @@ struct throtl_data
 	unsigned int nr_undestroyed_grps;
 
 	/* Work for dispatching throttled bios */
-	struct delayed_work throtl_work;
-
-	int limits_changed;
+	struct work_struct dispatch_work;
 };
 
 /* list and work item to allocate percpu group stats */
@@ -123,6 +175,8 @@ static LIST_HEAD(tg_stats_alloc_list);
 static void tg_stats_alloc_fn(struct work_struct *);
 static DECLARE_DELAYED_WORK(tg_stats_alloc_work, tg_stats_alloc_fn);
 
+static void throtl_pending_timer_fn(unsigned long arg);
+
 static inline struct throtl_grp *pd_to_tg(struct blkg_policy_data *pd)
 {
 	return pd ? container_of(pd, struct throtl_grp, pd) : NULL;
@@ -143,41 +197,65 @@ static inline struct throtl_grp *td_root_tg(struct throtl_data *td)
 	return blkg_to_tg(td->queue->root_blkg);
 }
 
-enum tg_state_flags {
-	THROTL_TG_FLAG_on_rr = 0,	/* on round-robin busy list */
-};
-
-#define THROTL_TG_FNS(name)						\
-static inline void throtl_mark_tg_##name(struct throtl_grp *tg)		\
-{									\
-	(tg)->flags |= (1 << THROTL_TG_FLAG_##name);			\
-}									\
-static inline void throtl_clear_tg_##name(struct throtl_grp *tg)	\
-{									\
-	(tg)->flags &= ~(1 << THROTL_TG_FLAG_##name);			\
-}									\
-static inline int throtl_tg_##name(const struct throtl_grp *tg)		\
-{									\
-	return ((tg)->flags & (1 << THROTL_TG_FLAG_##name)) != 0;	\
+/**
+ * sq_to_tg - return the throl_grp the specified service queue belongs to
+ * @sq: the throtl_service_queue of interest
+ *
+ * Return the throtl_grp @sq belongs to.  If @sq is the top-level one
+ * embedded in throtl_data, %NULL is returned.
+ */
+static struct throtl_grp *sq_to_tg(struct throtl_service_queue *sq)
+{
+	if (sq && sq->parent_sq)
+		return container_of(sq, struct throtl_grp, service_queue);
+	else
+		return NULL;
 }
 
-THROTL_TG_FNS(on_rr);
+/**
+ * sq_to_td - return throtl_data the specified service queue belongs to
+ * @sq: the throtl_service_queue of interest
+ *
+ * A service_queue can be embeded in either a throtl_grp or throtl_data.
+ * Determine the associated throtl_data accordingly and return it.
+ */
+static struct throtl_data *sq_to_td(struct throtl_service_queue *sq)
+{
+	struct throtl_grp *tg = sq_to_tg(sq);
 
-#define throtl_log_tg(td, tg, fmt, args...)	do {			\
-	char __pbuf[128];						\
+	if (tg)
+		return tg->td;
+	else
+		return container_of(sq, struct throtl_data, service_queue);
+}
+
+/**
+ * throtl_log - log debug message via blktrace
+ * @sq: the service_queue being reported
+ * @fmt: printf format string
+ * @args: printf args
+ *
+ * The messages are prefixed with "throtl BLKG_NAME" if @sq belongs to a
+ * throtl_grp; otherwise, just "throtl".
+ *
+ * TODO: this should be made a function and name formatting should happen
+ * after testing whether blktrace is enabled.
+ */
+#define throtl_log(sq, fmt, args...)	do {				\
+	struct throtl_grp *__tg = sq_to_tg((sq));			\
+	struct throtl_data *__td = sq_to_td((sq));			\
+									\
+	(void)__td;							\
+	if ((__tg)) {							\
+		char __pbuf[128];					\
 									\
-	blkg_path(tg_to_blkg(tg), __pbuf, sizeof(__pbuf));		\
-	blk_add_trace_msg((td)->queue, "throtl %s " fmt, __pbuf, ##args); \
+		blkg_path(tg_to_blkg(__tg), __pbuf, sizeof(__pbuf));	\
+		blk_add_trace_msg(__td->queue, "throtl %s " fmt, __pbuf, ##args); \
+	} else {							\
+		blk_add_trace_msg(__td->queue, "throtl " fmt, ##args);	\
+	}								\
 } while (0)
 
-#define throtl_log(td, fmt, args...)	\
-	blk_add_trace_msg((td)->queue, "throtl " fmt, ##args)
-
-static inline unsigned int total_nr_queued(struct throtl_data *td)
-{
-	return td->nr_queued[0] + td->nr_queued[1];
-}
-
 /*
  * Worker for allocating per cpu stat for tgs. This is scheduled on the
  * system_wq once there are some groups on the alloc_list waiting for
@@ -215,15 +293,141 @@ alloc_stats:
 		goto alloc_stats;
 }
 
+static void throtl_qnode_init(struct throtl_qnode *qn, struct throtl_grp *tg)
+{
+	INIT_LIST_HEAD(&qn->node);
+	bio_list_init(&qn->bios);
+	qn->tg = tg;
+}
+
+/**
+ * throtl_qnode_add_bio - add a bio to a throtl_qnode and activate it
+ * @bio: bio being added
+ * @qn: qnode to add bio to
+ * @queued: the service_queue->queued[] list @qn belongs to
+ *
+ * Add @bio to @qn and put @qn on @queued if it's not already on.
+ * @qn->tg's reference count is bumped when @qn is activated.  See the
+ * comment on top of throtl_qnode definition for details.
+ */
+static void throtl_qnode_add_bio(struct bio *bio, struct throtl_qnode *qn,
+				 struct list_head *queued)
+{
+	bio_list_add(&qn->bios, bio);
+	if (list_empty(&qn->node)) {
+		list_add_tail(&qn->node, queued);
+		blkg_get(tg_to_blkg(qn->tg));
+	}
+}
+
+/**
+ * throtl_peek_queued - peek the first bio on a qnode list
+ * @queued: the qnode list to peek
+ */
+static struct bio *throtl_peek_queued(struct list_head *queued)
+{
+	struct throtl_qnode *qn = list_first_entry(queued, struct throtl_qnode, node);
+	struct bio *bio;
+
+	if (list_empty(queued))
+		return NULL;
+
+	bio = bio_list_peek(&qn->bios);
+	WARN_ON_ONCE(!bio);
+	return bio;
+}
+
+/**
+ * throtl_pop_queued - pop the first bio form a qnode list
+ * @queued: the qnode list to pop a bio from
+ * @tg_to_put: optional out argument for throtl_grp to put
+ *
+ * Pop the first bio from the qnode list @queued.  After popping, the first
+ * qnode is removed from @queued if empty or moved to the end of @queued so
+ * that the popping order is round-robin.
+ *
+ * When the first qnode is removed, its associated throtl_grp should be put
+ * too.  If @tg_to_put is NULL, this function automatically puts it;
+ * otherwise, *@tg_to_put is set to the throtl_grp to put and the caller is
+ * responsible for putting it.
+ */
+static struct bio *throtl_pop_queued(struct list_head *queued,
+				     struct throtl_grp **tg_to_put)
+{
+	struct throtl_qnode *qn = list_first_entry(queued, struct throtl_qnode, node);
+	struct bio *bio;
+
+	if (list_empty(queued))
+		return NULL;
+
+	bio = bio_list_pop(&qn->bios);
+	WARN_ON_ONCE(!bio);
+
+	if (bio_list_empty(&qn->bios)) {
+		list_del_init(&qn->node);
+		if (tg_to_put)
+			*tg_to_put = qn->tg;
+		else
+			blkg_put(tg_to_blkg(qn->tg));
+	} else {
+		list_move_tail(&qn->node, queued);
+	}
+
+	return bio;
+}
+
+/* init a service_queue, assumes the caller zeroed it */
+static void throtl_service_queue_init(struct throtl_service_queue *sq,
+				      struct throtl_service_queue *parent_sq)
+{
+	INIT_LIST_HEAD(&sq->queued[0]);
+	INIT_LIST_HEAD(&sq->queued[1]);
+	sq->pending_tree = RB_ROOT;
+	sq->parent_sq = parent_sq;
+	setup_timer(&sq->pending_timer, throtl_pending_timer_fn,
+		    (unsigned long)sq);
+}
+
+static void throtl_service_queue_exit(struct throtl_service_queue *sq)
+{
+	del_timer_sync(&sq->pending_timer);
+}
+
 static void throtl_pd_init(struct blkcg_gq *blkg)
 {
 	struct throtl_grp *tg = blkg_to_tg(blkg);
+	struct throtl_data *td = blkg->q->td;
+	struct throtl_service_queue *parent_sq;
 	unsigned long flags;
+	int rw;
+
+	/*
+	 * If sane_hierarchy is enabled, we switch to properly hierarchical
+	 * behavior where limits on a given throtl_grp are applied to the
+	 * whole subtree rather than just the group itself.  e.g. If 16M
+	 * read_bps limit is set on the root group, the whole system can't
+	 * exceed 16M for the device.
+	 *
+	 * If sane_hierarchy is not enabled, the broken flat hierarchy
+	 * behavior is retained where all throtl_grps are treated as if
+	 * they're all separate root groups right below throtl_data.
+	 * Limits of a group don't interact with limits of other groups
+	 * regardless of the position of the group in the hierarchy.
+	 */
+	parent_sq = &td->service_queue;
+
+	if (cgroup_sane_behavior(blkg->blkcg->css.cgroup) && blkg->parent)
+		parent_sq = &blkg_to_tg(blkg->parent)->service_queue;
+
+	throtl_service_queue_init(&tg->service_queue, parent_sq);
+
+	for (rw = READ; rw <= WRITE; rw++) {
+		throtl_qnode_init(&tg->qnode_on_self[rw], tg);
+		throtl_qnode_init(&tg->qnode_on_parent[rw], tg);
+	}
 
 	RB_CLEAR_NODE(&tg->rb_node);
-	bio_list_init(&tg->bio_lists[0]);
-	bio_list_init(&tg->bio_lists[1]);
-	tg->limits_changed = false;
+	tg->td = td;
 
 	tg->bps[READ] = -1;
 	tg->bps[WRITE] = -1;
@@ -241,6 +445,30 @@ static void throtl_pd_init(struct blkcg_gq *blkg)
 	spin_unlock_irqrestore(&tg_stats_alloc_lock, flags);
 }
 
+/*
+ * Set has_rules[] if @tg or any of its parents have limits configured.
+ * This doesn't require walking up to the top of the hierarchy as the
+ * parent's has_rules[] is guaranteed to be correct.
+ */
+static void tg_update_has_rules(struct throtl_grp *tg)
+{
+	struct throtl_grp *parent_tg = sq_to_tg(tg->service_queue.parent_sq);
+	int rw;
+
+	for (rw = READ; rw <= WRITE; rw++)
+		tg->has_rules[rw] = (parent_tg && parent_tg->has_rules[rw]) ||
+				    (tg->bps[rw] != -1 || tg->iops[rw] != -1);
+}
+
+static void throtl_pd_online(struct blkcg_gq *blkg)
+{
+	/*
+	 * We don't want new groups to escape the limits of its ancestors.
+	 * Update has_rules[] after a new group is brought online.
+	 */
+	tg_update_has_rules(blkg_to_tg(blkg));
+}
+
 static void throtl_pd_exit(struct blkcg_gq *blkg)
 {
 	struct throtl_grp *tg = blkg_to_tg(blkg);
@@ -251,6 +479,8 @@ static void throtl_pd_exit(struct blkcg_gq *blkg)
 	spin_unlock_irqrestore(&tg_stats_alloc_lock, flags);
 
 	free_percpu(tg->stats_cpu);
+
+	throtl_service_queue_exit(&tg->service_queue);
 }
 
 static void throtl_pd_reset_stats(struct blkcg_gq *blkg)
@@ -309,17 +539,18 @@ static struct throtl_grp *throtl_lookup_create_tg(struct throtl_data *td,
 	return tg;
 }
 
-static struct throtl_grp *throtl_rb_first(struct throtl_rb_root *root)
+static struct throtl_grp *
+throtl_rb_first(struct throtl_service_queue *parent_sq)
 {
 	/* Service tree is empty */
-	if (!root->count)
+	if (!parent_sq->nr_pending)
 		return NULL;
 
-	if (!root->left)
-		root->left = rb_first(&root->rb);
+	if (!parent_sq->first_pending)
+		parent_sq->first_pending = rb_first(&parent_sq->pending_tree);
 
-	if (root->left)
-		return rb_entry_tg(root->left);
+	if (parent_sq->first_pending)
+		return rb_entry_tg(parent_sq->first_pending);
 
 	return NULL;
 }
@@ -330,29 +561,30 @@ static void rb_erase_init(struct rb_node *n, struct rb_root *root)
 	RB_CLEAR_NODE(n);
 }
 
-static void throtl_rb_erase(struct rb_node *n, struct throtl_rb_root *root)
+static void throtl_rb_erase(struct rb_node *n,
+			    struct throtl_service_queue *parent_sq)
 {
-	if (root->left == n)
-		root->left = NULL;
-	rb_erase_init(n, &root->rb);
-	--root->count;
+	if (parent_sq->first_pending == n)
+		parent_sq->first_pending = NULL;
+	rb_erase_init(n, &parent_sq->pending_tree);
+	--parent_sq->nr_pending;
 }
 
-static void update_min_dispatch_time(struct throtl_rb_root *st)
+static void update_min_dispatch_time(struct throtl_service_queue *parent_sq)
 {
 	struct throtl_grp *tg;
 
-	tg = throtl_rb_first(st);
+	tg = throtl_rb_first(parent_sq);
 	if (!tg)
 		return;
 
-	st->min_disptime = tg->disptime;
+	parent_sq->first_pending_disptime = tg->disptime;
 }
 
-static void
-tg_service_tree_add(struct throtl_rb_root *st, struct throtl_grp *tg)
+static void tg_service_queue_add(struct throtl_grp *tg)
 {
-	struct rb_node **node = &st->rb.rb_node;
+	struct throtl_service_queue *parent_sq = tg->service_queue.parent_sq;
+	struct rb_node **node = &parent_sq->pending_tree.rb_node;
 	struct rb_node *parent = NULL;
 	struct throtl_grp *__tg;
 	unsigned long key = tg->disptime;
@@ -371,89 +603,135 @@ tg_service_tree_add(struct throtl_rb_root *st, struct throtl_grp *tg)
 	}
 
 	if (left)
-		st->left = &tg->rb_node;
+		parent_sq->first_pending = &tg->rb_node;
 
 	rb_link_node(&tg->rb_node, parent, node);
-	rb_insert_color(&tg->rb_node, &st->rb);
+	rb_insert_color(&tg->rb_node, &parent_sq->pending_tree);
 }
 
-static void __throtl_enqueue_tg(struct throtl_data *td, struct throtl_grp *tg)
+static void __throtl_enqueue_tg(struct throtl_grp *tg)
 {
-	struct throtl_rb_root *st = &td->tg_service_tree;
+	tg_service_queue_add(tg);
+	tg->flags |= THROTL_TG_PENDING;
+	tg->service_queue.parent_sq->nr_pending++;
+}
 
-	tg_service_tree_add(st, tg);
-	throtl_mark_tg_on_rr(tg);
-	st->count++;
+static void throtl_enqueue_tg(struct throtl_grp *tg)
+{
+	if (!(tg->flags & THROTL_TG_PENDING))
+		__throtl_enqueue_tg(tg);
 }
 
-static void throtl_enqueue_tg(struct throtl_data *td, struct throtl_grp *tg)
+static void __throtl_dequeue_tg(struct throtl_grp *tg)
 {
-	if (!throtl_tg_on_rr(tg))
-		__throtl_enqueue_tg(td, tg);
+	throtl_rb_erase(&tg->rb_node, tg->service_queue.parent_sq);
+	tg->flags &= ~THROTL_TG_PENDING;
 }
 
-static void __throtl_dequeue_tg(struct throtl_data *td, struct throtl_grp *tg)
+static void throtl_dequeue_tg(struct throtl_grp *tg)
 {
-	throtl_rb_erase(&tg->rb_node, &td->tg_service_tree);
-	throtl_clear_tg_on_rr(tg);
+	if (tg->flags & THROTL_TG_PENDING)
+		__throtl_dequeue_tg(tg);
 }
 
-static void throtl_dequeue_tg(struct throtl_data *td, struct throtl_grp *tg)
+/* Call with queue lock held */
+static void throtl_schedule_pending_timer(struct throtl_service_queue *sq,
+					  unsigned long expires)
 {
-	if (throtl_tg_on_rr(tg))
-		__throtl_dequeue_tg(td, tg);
+	mod_timer(&sq->pending_timer, expires);
+	throtl_log(sq, "schedule timer. delay=%lu jiffies=%lu",
+		   expires - jiffies, jiffies);
 }
 
-static void throtl_schedule_next_dispatch(struct throtl_data *td)
+/**
+ * throtl_schedule_next_dispatch - schedule the next dispatch cycle
+ * @sq: the service_queue to schedule dispatch for
+ * @force: force scheduling
+ *
+ * Arm @sq->pending_timer so that the next dispatch cycle starts on the
+ * dispatch time of the first pending child.  Returns %true if either timer
+ * is armed or there's no pending child left.  %false if the current
+ * dispatch window is still open and the caller should continue
+ * dispatching.
+ *
+ * If @force is %true, the dispatch timer is always scheduled and this
+ * function is guaranteed to return %true.  This is to be used when the
+ * caller can't dispatch itself and needs to invoke pending_timer
+ * unconditionally.  Note that forced scheduling is likely to induce short
+ * delay before dispatch starts even if @sq->first_pending_disptime is not
+ * in the future and thus shouldn't be used in hot paths.
+ */
+static bool throtl_schedule_next_dispatch(struct throtl_service_queue *sq,
+					  bool force)
 {
-	struct throtl_rb_root *st = &td->tg_service_tree;
+	/* any pending children left? */
+	if (!sq->nr_pending)
+		return true;
 
-	/*
-	 * If there are more bios pending, schedule more work.
-	 */
-	if (!total_nr_queued(td))
-		return;
+	update_min_dispatch_time(sq);
 
-	BUG_ON(!st->count);
+	/* is the next dispatch time in the future? */
+	if (force || time_after(sq->first_pending_disptime, jiffies)) {
+		throtl_schedule_pending_timer(sq, sq->first_pending_disptime);
+		return true;
+	}
 
-	update_min_dispatch_time(st);
+	/* tell the caller to continue dispatching */
+	return false;
+}
 
-	if (time_before_eq(st->min_disptime, jiffies))
-		throtl_schedule_delayed_work(td, 0);
-	else
-		throtl_schedule_delayed_work(td, (st->min_disptime - jiffies));
+static inline void throtl_start_new_slice_with_credit(struct throtl_grp *tg,
+		bool rw, unsigned long start)
+{
+	tg->bytes_disp[rw] = 0;
+	tg->io_disp[rw] = 0;
+
+	/*
+	 * Previous slice has expired. We must have trimmed it after last
+	 * bio dispatch. Tha