X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=block%2Fcfq-iosched.c;h=6a062eebbd15301320e7491b5dd45d17f2204a3c;hb=99cd3386f290eaf61f2b7596d5a4cc2007771174;hp=ca198e61fa65c555bf936d2d628876313b62dbb1;hpb=fe094d98e79351344c9e0e2c1446794240d247a4;p=safe%2Fjmp%2Flinux-2.6

diff --git a/block/cfq-iosched.c b/block/cfq-iosched.c
index ca198e6..6a062ee 100644
--- a/block/cfq-iosched.c
+++ b/block/cfq-iosched.c
@@ -11,6 +11,7 @@
 #include <linux/elevator.h>
 #include <linux/rbtree.h>
 #include <linux/ioprio.h>
+#include <linux/blktrace_api.h>
 
 /*
  * tunables
@@ -38,16 +39,18 @@ static int cfq_slice_idle = HZ / 125;
 #define CFQ_MIN_TT		(2)
 
 #define CFQ_SLICE_SCALE		(5)
+#define CFQ_HW_QUEUE_MIN	(5)
 
 #define RQ_CIC(rq)		\
 	((struct cfq_io_context *) (rq)->elevator_private)
-#define RQ_CFQQ(rq)		((rq)->elevator_private2)
+#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elevator_private2)
 
 static struct kmem_cache *cfq_pool;
 static struct kmem_cache *cfq_ioc_pool;
 
 static DEFINE_PER_CPU(unsigned long, ioc_count);
 static struct completion *ioc_gone;
+static DEFINE_SPINLOCK(ioc_gone_lock);
 
 #define CFQ_PRIO_LISTS		IOPRIO_BE_NR
 #define cfq_class_idle(cfqq)	((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
@@ -84,7 +87,14 @@ struct cfq_data {
 
 	int rq_in_driver;
 	int sync_flight;
+
+	/*
+	 * queue-depth detection
+	 */
+	int rq_queued;
 	int hw_tag;
+	int hw_tag_samples;
+	int rq_in_driver_peak;
 
 	/*
 	 * idle window management
@@ -124,6 +134,8 @@ struct cfq_data {
 struct cfq_queue {
 	/* reference count */
 	atomic_t ref;
+	/* various state flags, see below */
+	unsigned int flags;
 	/* parent cfq_data */
 	struct cfq_data *cfqd;
 	/* service_tree member */
@@ -138,14 +150,14 @@ struct cfq_queue {
 	int queued[2];
 	/* currently allocated requests */
 	int allocated[2];
-	/* pending metadata requests */
-	int meta_pending;
 	/* fifo list of requests in sort_list */
 	struct list_head fifo;
 
 	unsigned long slice_end;
 	long slice_resid;
 
+	/* pending metadata requests */
+	int meta_pending;
 	/* number of requests that are on the dispatch list or inside driver */
 	int dispatched;
 
@@ -153,8 +165,7 @@ struct cfq_queue {
 	unsigned short ioprio, org_ioprio;
 	unsigned short ioprio_class, org_ioprio_class;
 
-	/* various state flags, see below */
-	unsigned int flags;
+	pid_t pid;
 };
 
 enum cfqq_state_flags {
@@ -198,6 +209,11 @@ CFQ_CFQQ_FNS(slice_new);
 CFQ_CFQQ_FNS(sync);
 #undef CFQ_CFQQ_FNS
 
+#define cfq_log_cfqq(cfqd, cfqq, fmt, args...)	\
+	blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args)
+#define cfq_log(cfqd, fmt, args...)	\
+	blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)
+
 static void cfq_dispatch_insert(struct request_queue *, struct request *);
 static struct cfq_queue *cfq_get_queue(struct cfq_data *, int,
 				       struct io_context *, gfp_t);
@@ -234,8 +250,10 @@ static inline int cfq_bio_sync(struct bio *bio)
  */
 static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
 {
-	if (cfqd->busy_queues)
-		kblockd_schedule_work(&cfqd->unplug_work);
+	if (cfqd->busy_queues) {
+		cfq_log(cfqd, "schedule dispatch");
+		kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work);
+	}
 }
 
 static int cfq_queue_empty(struct request_queue *q)
@@ -270,6 +288,7 @@ static inline void
 cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 {
 	cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
+	cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
 }
 
 /*
@@ -539,6 +558,7 @@ static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
  */
 static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 {
+	cfq_log_cfqq(cfqd, cfqq, "add_to_rr");
 	BUG_ON(cfq_cfqq_on_rr(cfqq));
 	cfq_mark_cfqq_on_rr(cfqq);
 	cfqd->busy_queues++;
@@ -552,6 +572,7 @@ static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
  */
 static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 {
+	cfq_log_cfqq(cfqd, cfqq, "del_from_rr");
 	BUG_ON(!cfq_cfqq_on_rr(cfqq));
 	cfq_clear_cfqq_on_rr(cfqq);
 
@@ -638,15 +659,8 @@ static void cfq_activate_request(struct request_queue *q, struct request *rq)
 	struct cfq_data *cfqd = q->elevator->elevator_data;
 
 	cfqd->rq_in_driver++;
-
-	/*
-	 * If the depth is larger 1, it really could be queueing. But lets
-	 * make the mark a little higher - idling could still be good for
-	 * low queueing, and a low queueing number could also just indicate
-	 * a SCSI mid layer like behaviour where limit+1 is often seen.
-	 */
-	if (!cfqd->hw_tag && cfqd->rq_in_driver > 4)
-		cfqd->hw_tag = 1;
+	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d",
+						cfqd->rq_in_driver);
 
 	cfqd->last_position = rq->hard_sector + rq->hard_nr_sectors;
 }
@@ -657,6 +671,8 @@ static void cfq_deactivate_request(struct request_queue *q, struct request *rq)
 
 	WARN_ON(!cfqd->rq_in_driver);
 	cfqd->rq_in_driver--;
+	cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d",
+						cfqd->rq_in_driver);
 }
 
 static void cfq_remove_request(struct request *rq)
@@ -669,6 +685,7 @@ static void cfq_remove_request(struct request *rq)
 	list_del_init(&rq->queuelist);
 	cfq_del_rq_rb(rq);
 
+	cfqq->cfqd->rq_queued--;
 	if (rq_is_meta(rq)) {
 		WARN_ON(!cfqq->meta_pending);
 		cfqq->meta_pending--;
@@ -746,6 +763,7 @@ static void __cfq_set_active_queue(struct cfq_data *cfqd,
 				   struct cfq_queue *cfqq)
 {
 	if (cfqq) {
+		cfq_log_cfqq(cfqd, cfqq, "set_active");
 		cfqq->slice_end = 0;
 		cfq_clear_cfqq_must_alloc_slice(cfqq);
 		cfq_clear_cfqq_fifo_expire(cfqq);
@@ -763,6 +781,8 @@ static void
 __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 		    int timed_out)
 {
+	cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out);
+
 	if (cfq_cfqq_wait_request(cfqq))
 		del_timer(&cfqd->idle_slice_timer);
 
@@ -772,8 +792,10 @@ __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 	/*
 	 * store what was left of this slice, if the queue idled/timed out
 	 */
-	if (timed_out && !cfq_cfqq_slice_new(cfqq))
+	if (timed_out && !cfq_cfqq_slice_new(cfqq)) {
 		cfqq->slice_resid = cfqq->slice_end - jiffies;
+		cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid);
+	}
 
 	cfq_resort_rr_list(cfqd, cfqq);
 
@@ -856,6 +878,14 @@ static void cfq_arm_slice_timer(struct cfq_data *cfqd)
 	struct cfq_io_context *cic;
 	unsigned long sl;
 
+	/*
+	 * SSD device without seek penalty, disable idling. But only do so
+	 * for devices that support queuing, otherwise we still have a problem
+	 * with sync vs async workloads.
+	 */
+	if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)
+		return;
+
 	WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
 	WARN_ON(cfq_cfqq_slice_new(cfqq));
 
@@ -866,6 +896,12 @@ static void cfq_arm_slice_timer(struct cfq_data *cfqd)
 		return;
 
 	/*
+	 * still requests with the driver, don't idle
+	 */
+	if (cfqd->rq_in_driver)
+		return;
+
+	/*
 	 * task has exited, don't wait
 	 */
 	cic = cfqd->active_cic;
@@ -892,6 +928,7 @@ static void cfq_arm_slice_timer(struct cfq_data *cfqd)
 		sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT));
 
 	mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
+	cfq_log(cfqd, "arm_idle: %lu", sl);
 }
 
 /*
@@ -902,6 +939,8 @@ static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
 	struct cfq_data *cfqd = q->elevator->elevator_data;
 	struct cfq_queue *cfqq = RQ_CFQQ(rq);
 
+	cfq_log_cfqq(cfqd, cfqq, "dispatch_insert");
+
 	cfq_remove_request(rq);
 	cfqq->dispatched++;
 	elv_dispatch_sort(q, rq);
@@ -931,8 +970,9 @@ static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
 	rq = rq_entry_fifo(cfqq->fifo.next);
 
 	if (time_before(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo]))
-		return NULL;
+		rq = NULL;
 
+	cfq_log_cfqq(cfqd, cfqq, "fifo=%p", rq);
 	return rq;
 }
 
@@ -1072,6 +1112,7 @@ static int cfq_forced_dispatch(struct cfq_data *cfqd)
 
 	BUG_ON(cfqd->busy_queues);
 
+	cfq_log(cfqd, "forced_dispatch=%d\n", dispatched);
 	return dispatched;
 }
 
@@ -1112,6 +1153,7 @@ static int cfq_dispatch_requests(struct request_queue *q, int force)
 		dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch);
 	}
 
+	cfq_log(cfqd, "dispatched=%d", dispatched);
 	return dispatched;
 }
 
@@ -1130,6 +1172,7 @@ static void cfq_put_queue(struct cfq_queue *cfqq)
 	if (!atomic_dec_and_test(&cfqq->ref))
 		return;
 
+	cfq_log_cfqq(cfqd, cfqq, "put_queue");
 	BUG_ON(rb_first(&cfqq->sort_list));
 	BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
 	BUG_ON(cfq_cfqq_on_rr(cfqq));
@@ -1143,43 +1186,58 @@ static void cfq_put_queue(struct cfq_queue *cfqq)
 }
 
 /*
- * Call func for each cic attached to this ioc. Returns number of cic's seen.
+ * Must always be called with the rcu_read_lock() held
+ */
+static void
+__call_for_each_cic(struct io_context *ioc,
+		    void (*func)(struct io_context *, struct cfq_io_context *))
+{
+	struct cfq_io_context *cic;
+	struct hlist_node *n;
+
+	hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list)
+		func(ioc, cic);
+}
+
+/*
+ * Call func for each cic attached to this ioc.
  */
-#define CIC_GANG_NR	16
-static unsigned int
+static void
 call_for_each_cic(struct io_context *ioc,
 		  void (*func)(struct io_context *, struct cfq_io_context *))
 {
-	struct cfq_io_context *cics[CIC_GANG_NR];
-	unsigned long index = 0;
-	unsigned int called = 0;
-	int nr;
-
 	rcu_read_lock();
+	__call_for_each_cic(ioc, func);
+	rcu_read_unlock();
+}
 
-	do {
-		int i;
-
-		/*
-		 * Perhaps there's a better way - this just gang lookups from
-		 * 0 to the end, restarting after each CIC_GANG_NR from the
-		 * last key + 1.
-		 */
-		nr = radix_tree_gang_lookup(&ioc->radix_root, (void **) cics,
-						index, CIC_GANG_NR);
-		if (!nr)
-			break;
+static void cfq_cic_free_rcu(struct rcu_head *head)
+{
+	struct cfq_io_context *cic;
 
-		called += nr;
-		index = 1 + (unsigned long) cics[nr - 1]->key;
+	cic = container_of(head, struct cfq_io_context, rcu_head);
 
-		for (i = 0; i < nr; i++)
-			func(ioc, cics[i]);
-	} while (nr == CIC_GANG_NR);
+	kmem_cache_free(cfq_ioc_pool, cic);
+	elv_ioc_count_dec(ioc_count);
 
-	rcu_read_unlock();
+	if (ioc_gone) {
+		/*
+		 * CFQ scheduler is exiting, grab exit lock and check
+		 * the pending io context count. If it hits zero,
+		 * complete ioc_gone and set it back to NULL
+		 */
+		spin_lock(&ioc_gone_lock);
+		if (ioc_gone && !elv_ioc_count_read(ioc_count)) {
+			complete(ioc_gone);
+			ioc_gone = NULL;
+		}
+		spin_unlock(&ioc_gone_lock);
+	}
+}
 
-	return called;
+static void cfq_cic_free(struct cfq_io_context *cic)
+{
+	call_rcu(&cic->rcu_head, cfq_cic_free_rcu);
 }
 
 static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic)
@@ -1190,26 +1248,26 @@ static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic)
 
 	spin_lock_irqsave(&ioc->lock, flags);
 	radix_tree_delete(&ioc->radix_root, cic->dead_key);
+	hlist_del_rcu(&cic->cic_list);
 	spin_unlock_irqrestore(&ioc->lock, flags);
 
-	kmem_cache_free(cfq_ioc_pool, cic);
+	cfq_cic_free(cic);
 }
 
+/*
+ * Must be called with rcu_read_lock() held or preemption otherwise disabled.
+ * Only two callers of this - ->dtor() which is called with the rcu_read_lock(),
+ * and ->trim() which is called with the task lock held
+ */
 static void cfq_free_io_context(struct io_context *ioc)
 {
-	int freed;
-
 	/*
-	 * ioc->refcount is zero here, so no more cic's are allowed to be
-	 * linked into this ioc. So it should be ok to iterate over the known
-	 * list, we will see all cic's since no new ones are added.
+	 * ioc->refcount is zero here, or we are called from elv_unregister(),
+	 * so no more cic's are allowed to be linked into this ioc.  So it
+	 * should be ok to iterate over the known list, we will see all cic's
+	 * since no new ones are added.
 	 */
-	freed = call_for_each_cic(ioc, cic_free_func);
-
-	elv_ioc_count_mod(ioc_count, -freed);
-
-	if (ioc_gone && !elv_ioc_count_read(ioc_count))
-		complete(ioc_gone);
+	__call_for_each_cic(ioc, cic_free_func);
 }
 
 static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
@@ -1225,6 +1283,8 @@ static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 static void __cfq_exit_single_io_context(struct cfq_data *cfqd,
 					 struct cfq_io_context *cic)
 {
+	struct io_context *ioc = cic->ioc;
+
 	list_del_init(&cic->queue_list);
 
 	/*
@@ -1234,6 +1294,9 @@ static void __cfq_exit_single_io_context(struct cfq_data *cfqd,
 	cic->dead_key = (unsigned long) cic->key;
 	cic->key = NULL;
 
+	if (ioc->ioc_data == cic)
+		rcu_assign_pointer(ioc->ioc_data, NULL);
+
 	if (cic->cfqq[ASYNC]) {
 		cfq_exit_cfqq(cfqd, cic->cfqq[ASYNC]);
 		cic->cfqq[ASYNC] = NULL;
@@ -1266,7 +1329,6 @@ static void cfq_exit_single_io_context(struct io_context *ioc,
  */
 static void cfq_exit_io_context(struct io_context *ioc)
 {
-	rcu_assign_pointer(ioc->ioc_data, NULL);
 	call_for_each_cic(ioc, cfq_exit_single_io_context);
 }
 
@@ -1280,6 +1342,7 @@ cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
 	if (cic) {
 		cic->last_end_request = jiffies;
 		INIT_LIST_HEAD(&cic->queue_list);
+		INIT_HLIST_NODE(&cic->cic_list);
 		cic->dtor = cfq_free_io_context;
 		cic->exit = cfq_exit_io_context;
 		elv_ioc_count_inc(ioc_count);
@@ -1302,10 +1365,10 @@ static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc)
 		printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
 	case IOPRIO_CLASS_NONE:
 		/*
-		 * no prio set, place us in the middle of the BE classes
+		 * no prio set, inherit CPU scheduling settings
 		 */
 		cfqq->ioprio = task_nice_ioprio(tsk);
-		cfqq->ioprio_class = IOPRIO_CLASS_BE;
+		cfqq->ioprio_class = task_nice_ioclass(tsk);
 		break;
 	case IOPRIO_CLASS_RT:
 		cfqq->ioprio = task_ioprio(ioc);
@@ -1418,6 +1481,8 @@ retry:
 				cfq_mark_cfqq_idle_window(cfqq);
 			cfq_mark_cfqq_sync(cfqq);
 		}
+		cfqq->pid = current->pid;
+		cfq_log_cfqq(cfqd, cfqq, "alloced");
 	}
 
 	if (new_cfqq)
@@ -1475,15 +1540,6 @@ cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc,
 	return cfqq;
 }
 
-static void cfq_cic_free(struct cfq_io_context *cic)
-{
-	kmem_cache_free(cfq_ioc_pool, cic);
-	elv_ioc_count_dec(ioc_count);
-
-	if (ioc_gone && !elv_ioc_count_read(ioc_count))
-		complete(ioc_gone);
-}
-
 /*
  * We drop cfq io contexts lazily, so we may find a dead one.
  */
@@ -1497,10 +1553,10 @@ cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc,
 
 	spin_lock_irqsave(&ioc->lock, flags);
 
-	if (ioc->ioc_data == cic)
-		rcu_assign_pointer(ioc->ioc_data, NULL);
+	BUG_ON(ioc->ioc_data == cic);
 
 	radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd);
+	hlist_del_rcu(&cic->cic_list);
 	spin_unlock_irqrestore(&ioc->lock, flags);
 
 	cfq_cic_free(cic);
@@ -1510,20 +1566,24 @@ static struct cfq_io_context *
 cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc)
 {
 	struct cfq_io_context *cic;
+	unsigned long flags;
 	void *k;
 
 	if (unlikely(!ioc))
 		return NULL;
 
+	rcu_read_lock();
+
 	/*
 	 * we maintain a last-hit cache, to avoid browsing over the tree
 	 */
 	cic = rcu_dereference(ioc->ioc_data);
-	if (cic && cic->key == cfqd)
+	if (cic && cic->key == cfqd) {
+		rcu_read_unlock();
 		return cic;
+	}
 
 	do {
-		rcu_read_lock();
 		cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd);
 		rcu_read_unlock();
 		if (!cic)
@@ -1532,10 +1592,13 @@ cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc)
 		k = cic->key;
 		if (unlikely(!k)) {
 			cfq_drop_dead_cic(cfqd, ioc, cic);
+			rcu_read_lock();
 			continue;
 		}
 
+		spin_lock_irqsave(&ioc->lock, flags);
 		rcu_assign_pointer(ioc->ioc_data, cic);
+		spin_unlock_irqrestore(&ioc->lock, flags);
 		break;
 	} while (1);
 
@@ -1561,6 +1624,8 @@ static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
 		spin_lock_irqsave(&ioc->lock, flags);
 		ret = radix_tree_insert(&ioc->radix_root,
 						(unsigned long) cfqd, cic);
+		if (!ret)
+			hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list);
 		spin_unlock_irqrestore(&ioc->lock, flags);
 
 		radix_tree_preload_end();
@@ -1666,7 +1731,7 @@ static void
 cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 		       struct cfq_io_context *cic)
 {
-	int enable_idle;
+	int old_idle, enable_idle;
 
 	/*
 	 * Don't idle for async or idle io prio class
@@ -1674,7 +1739,7 @@ cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 	if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
 		return;
 
-	enable_idle = cfq_cfqq_idle_window(cfqq);
+	enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
 
 	if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle ||
 	    (cfqd->hw_tag && CIC_SEEKY(cic)))
@@ -1686,10 +1751,13 @@ cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 			enable_idle = 1;
 	}
 
-	if (enable_idle)
-		cfq_mark_cfqq_idle_window(cfqq);
-	else
-		cfq_clear_cfqq_idle_window(cfqq);
+	if (old_idle != enable_idle) {
+		cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle);
+		if (enable_idle)
+			cfq_mark_cfqq_idle_window(cfqq);
+		else
+			cfq_clear_cfqq_idle_window(cfqq);
+	}
 }
 
 /*
@@ -1748,6 +1816,7 @@ cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
  */
 static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 {
+	cfq_log_cfqq(cfqd, cfqq, "preempt");
 	cfq_slice_expired(cfqd, 1);
 
 	/*
@@ -1772,6 +1841,7 @@ cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 {
 	struct cfq_io_context *cic = RQ_CIC(rq);
 
+	cfqd->rq_queued++;
 	if (rq_is_meta(rq))
 		cfqq->meta_pending++;
 
@@ -1809,6 +1879,7 @@ static void cfq_insert_request(struct request_queue *q, struct request *rq)
 	struct cfq_data *cfqd = q->elevator->elevator_data;
 	struct cfq_queue *cfqq = RQ_CFQQ(rq);
 
+	cfq_log_cfqq(cfqd, cfqq, "insert_request");
 	cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc);
 
 	cfq_add_rq_rb(rq);
@@ -1818,6 +1889,31 @@ static void cfq_insert_request(struct request_queue *q, struct request *rq)
 	cfq_rq_enqueued(cfqd, cfqq, rq);
 }
 
+/*
+ * Update hw_tag based on peak queue depth over 50 samples under
+ * sufficient load.
+ */
+static void cfq_update_hw_tag(struct cfq_data *cfqd)
+{
+	if (cfqd->rq_in_driver > cfqd->rq_in_driver_peak)
+		cfqd->rq_in_driver_peak = cfqd->rq_in_driver;
+
+	if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN &&
+	    cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN)
+		return;
+
+	if (cfqd->hw_tag_samples++ < 50)
+		return;
+
+	if (cfqd->rq_in_driver_peak >= CFQ_HW_QUEUE_MIN)
+		cfqd->hw_tag = 1;
+	else
+		cfqd->hw_tag = 0;
+
+	cfqd->hw_tag_samples = 0;
+	cfqd->rq_in_driver_peak = 0;
+}
+
 static void cfq_completed_request(struct request_queue *q, struct request *rq)
 {
 	struct cfq_queue *cfqq = RQ_CFQQ(rq);
@@ -1826,6 +1922,9 @@ static void cfq_completed_request(struct request_queue *q, struct request *rq)
 	unsigned long now;
 
 	now = jiffies;
+	cfq_log_cfqq(cfqd, cfqq, "complete");
+
+	cfq_update_hw_tag(cfqd);
 
 	WARN_ON(!cfqd->rq_in_driver);
 	WARN_ON(!cfqq->dispatched);
@@ -1995,6 +2094,7 @@ queue_fail:
 
 	cfq_schedule_dispatch(cfqd);
 	spin_unlock_irqrestore(q->queue_lock, flags);
+	cfq_log(cfqd, "set_request fail");
 	return 1;
 }
 
@@ -2020,6 +2120,8 @@ static void cfq_idle_slice_timer(unsigned long data)
 	unsigned long flags;
 	int timed_out = 1;
 
+	cfq_log(cfqd, "idle timer fired");
+
 	spin_lock_irqsave(cfqd->queue->queue_lock, flags);
 
 	cfqq = cfqd->active_queue;
@@ -2134,12 +2236,17 @@ static void *cfq_init_queue(struct request_queue *q)
 	cfqd->cfq_slice[1] = cfq_slice_sync;
 	cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
 	cfqd->cfq_slice_idle = cfq_slice_idle;
+	cfqd->hw_tag = 1;
 
 	return cfqd;
 }
 
 static void cfq_slab_kill(void)
 {
+	/*
+	 * Caller already ensured that pending RCU callbacks are completed,
+	 * so we should have no busy allocations at this point.
+	 */
 	if (cfq_pool)
 		kmem_cache_destroy(cfq_pool);
 	if (cfq_ioc_pool)
@@ -2152,7 +2259,7 @@ static int __init cfq_slab_setup(void)
 	if (!cfq_pool)
 		goto fail;
 
-	cfq_ioc_pool = KMEM_CACHE(cfq_io_context, SLAB_DESTROY_BY_RCU);
+	cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0);
 	if (!cfq_ioc_pool)
 		goto fail;
 
@@ -2298,9 +2405,13 @@ static void __exit cfq_exit(void)
 	ioc_gone = &all_gone;
 	/* ioc_gone's update must be visible before reading ioc_count */
 	smp_wmb();
+
+	/*
+	 * this also protects us from entering cfq_slab_kill() with
+	 * pending RCU callbacks
+	 */
 	if (elv_ioc_count_read(ioc_count))
-		wait_for_completion(ioc_gone);
-	synchronize_rcu();
+		wait_for_completion(&all_gone);
 	cfq_slab_kill();
 }