Skip to contentSkip to conversation

cleanup io_context interface

PATCH In review Tejun Heo 2 participants 18 messages Oct 29, 2011 – Nov 14, 2011 Block layer

lore
Share & tools
lore

Tejun Heo

author
#0 lore raw

PATCH 1/1 cleanup io_context interface

Hello,

Over time, cfq and io_context has grown into each other and currently it isn'tq clear which part belongs where. Some of cfq internals live in io_context code and vice-versa. This makes the code error-prone, difficult to understand.

With locking simplified, io_context interface can be cleaned up now. This patchset moves bulk of io_cq management from cfq to block core. This provides clean separation between iosched algorithm implementation and ioc / ioc-q association (icq, formerly cic's) management which is a bit complex due to the locking requirements.

Elevators now can simply set .icq_size and align in elevator_type and block layer will automatically manage icq's and guarantee correct behavior across ioc, q exits and elevator switches.

Comments are added to iocontext.h to explain the rules for using icq.

0001-block-cfq-replace-current_io_context-with-create_io_.patch 0002-block-reorder-elevator-switch-sequence.patch 0003-block-remove-elevator_queue-ops.patch 0004-block-cfq-reorganize-cfq_io_context-into-generic-and.patch 0005-block-cfq-move-cfqd-icq_list-to-request_queue-and-ad.patch 0006-block-cfq-move-io_cq-lookup-to-blk-ioc.c.patch 0007-block-cfq-move-icq-cache-management-to-block-core.patch 0008-block-cfq-move-io_cq-exit-release-to-blk-ioc.c.patch 0009-block-cfq-restructure-io_cq-creation-path-for-io_con.patch 0010-block-cfq-move-icq-creation-and-rq-elv.icq-associati.patch

This patchset is on top of

block:for-3.2/core 334c2b0b8b "blk-throttle: use queue_is_locked() instead..."
+ [1] patchset "further updates to blk_cleanup_queue(), take#2"
+ [2] patchset "rescue cfq from RCU death spiral"

and available in the following git branch.

git://git.kernel.org/pub/scm/linux/kernel/git/tj/misc.git block-ioc

Jens, this patchset is the last of io_context related updates; however, as I wrote earlier, I'm thinking about doing similar updates to blkcg code which has basically the same issues. I'll wait for reviews on the pending patchsets and then proceed.

diffstat follows. Added lines are from comments. Without them total LOC is negative.

 block/blk-core.c          |   72 ++++++--
 block/blk-ioc.c           |  282 ++++++++++++++++++++++---------
 block/blk-sysfs.c         |    6 
 block/blk.h               |   49 ++++-
 block/cfq-iosched.c       |  408 +++++++++++-----------------------------------
 block/deadline-iosched.c  |    4 
 block/elevator.c          |  203 ++++++++++++----------
 block/noop-iosched.c      |    4 
 include/linux/blkdev.h    |   10 -
 include/linux/elevator.h  |   23 ++
 include/linux/iocontext.h |  114 +++++++++---
 11 files changed, 630 insertions(+), 545 deletions(-)

Thank you.

--
tejun

[1] http://thread.gmane.org/gmane.linux.kernel/1207608
[2] http://thread.gmane.org/gmane.linux.kernel/1207624

author
#1 lore raw

PATCH 1/10 block, cfq: replace current_io_context() with create_io_context()

When called under queue_lock, current_io_context() triggers lockdep warning if it hits allocation path. This is because io_context installation is protected by task_lock which is not IRQ safe, so it triggers irq-unsafe-lock -> irq -> irq-safe-lock -> irq-unsafe-lock deadlock warning.

Given the restriction, accessor + creator rolled into one doesn't work too well. Drop current_io_context() and let the users access task->io_context directly inside queue_lock combined with explicit creation using create_io_context().

Future ioc updates will further consolidate ioc access and the create interface will be unexported.

While at it, relocate ioc internal interface declarations in blk.h and add section comments before and after.

This patch does not introduce functional change.

---
 block/blk-core.c    |   25 ++++++++++++++++----
 block/blk-ioc.c     |   62 ++++++++++++++-------------------------------------
 block/blk.h         |   36 +++++++++++++++++++++++++++--
 block/cfq-iosched.c |    2 +-
 4 files changed, 71 insertions(+), 54 deletions(-)
block/blk-core.c +20 -5
index 39f1289..74b44ab 100644
@@ -766,9 +766,12 @@ static struct request *get_request(struct request_queue *q, int rw_flags,
 {
 	struct request *rq = NULL;
 	struct request_list *rl = &q->rq;
-	struct io_context *ioc = NULL;
+	struct io_context *ioc;
 	const bool is_sync = rw_is_sync(rw_flags) != 0;
+	bool retried = false;
 	int may_queue;
+retry:
+	ioc = current->io_context;
 
 	if (unlikely(blk_queue_dead(q)))
 		return NULL;
@@ -779,7 +782,20 @@ static struct request *get_request(struct request_queue *q, int rw_flags,
 
 	if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
 		if (rl->count[is_sync]+1 >= q->nr_requests) {
-			ioc = current_io_context(GFP_ATOMIC, q->node);
+			/*
+			 * We want ioc to record batching state.  If it's
+			 * not already there, creating a new one requires
+			 * dropping queue_lock, which in turn requires
+			 * retesting conditions to avoid queue hang.
+			 */
+			if (!ioc && !retried) {
+				spin_unlock_irq(q->queue_lock);
+				create_io_context(current, gfp_mask, q->node);
+				spin_lock_irq(q->queue_lock);
+				retried = true;
+				goto retry;
+			}
+
 			/*
 			 * The queue will fill after this allocation, so set
 			 * it as full, and mark this process as "batching".
@@ -887,7 +903,6 @@ static struct request *get_request_wait(struct request_queue *q, int rw_flags,
 	rq = get_request(q, rw_flags, bio, GFP_NOIO);
 	while (!rq) {
 		DEFINE_WAIT(wait);
-		struct io_context *ioc;
 		struct request_list *rl = &q->rq;
 
 		if (unlikely(blk_queue_dead(q)))
@@ -907,8 +922,8 @@ static struct request *get_request_wait(struct request_queue *q, int rw_flags,
 		 * up to a big batch of them for a small period time.
 		 * See ioc_batching, ioc_set_batching
 		 */
-		ioc = current_io_context(GFP_NOIO, q->node);
-		ioc_set_batching(q, ioc);
+		create_io_context(current, GFP_NOIO, q->node);
+		ioc_set_batching(q, current->io_context);
 
 		spin_lock_irq(q->queue_lock);
 		finish_wait(&rl->wait[is_sync], &wait);
block/blk-ioc.c +17 -45

98 lines not on this page · raw message

block/blk.h +33 -3
index 8d42115..5bca266 100644
@@ -127,9 +127,6 @@ static inline int blk_should_fake_timeout(struct request_queue *q)
 }
 #endif
 
-void get_io_context(struct io_context *ioc);
-struct io_context *current_io_context(gfp_t gfp_flags, int node);
-
 int ll_back_merge_fn(struct request_queue *q, struct request *req,
 		     struct bio *bio);
 int ll_front_merge_fn(struct request_queue *q, struct request *req, 
@@ -198,6 +195,39 @@ static inline int blk_do_io_stat(struct request *rq)
 	        (rq->cmd_flags & REQ_DISCARD));
 }
 
+/*
+ * Internal io_context interface
+ */
+void get_io_context(struct io_context *ioc);
+
+void create_io_context_slowpath(struct task_struct *task, gfp_t gfp_mask,
+				int node);
+
+/**
+ * create_io_context - try to create task->io_context
+ * @task: target task
+ * @gfp_mask: allocation mask
+ * @node: allocation node
+ *
+ * If @task->io_context is %NULL, allocate a new io_context and install it.
+ * Returns the current @task->io_context which may be %NULL if allocation
+ * failed.
+ *
+ * Note that this function can't be called with IRQ disabled because
+ * task_lock which protects @task->io_context is IRQ-unsafe.
+ */
+static inline struct io_context *create_io_context(struct task_struct *task,
+						   gfp_t gfp_mask, int node)
+{
+	WARN_ON_ONCE(irqs_disabled());
+	if (unlikely(!task->io_context))
+		create_io_context_slowpath(task, gfp_mask, node);
+	return task->io_context;
+}
+
+/*
+ * Internal throttling interface
+ */
 #ifdef CONFIG_BLK_DEV_THROTTLING
 extern bool blk_throtl_bio(struct request_queue *q, struct bio *bio);
 extern void blk_throtl_drain(struct request_queue *q);
block/cfq-iosched.c +1 -1
index 3b07ce1..5f7e4d1 100644
@@ -3012,7 +3012,7 @@ static int cfq_create_cic(struct cfq_data *cfqd, gfp_t gfp_mask)
 	might_sleep_if(gfp_mask & __GFP_WAIT);
 
 	/* allocate stuff */
-	ioc = current_io_context(gfp_mask, q->node);
+	ioc = create_io_context(current, gfp_mask, q->node);
 	if (!ioc)
 		goto out;
 
Signature · 1 line
1.7.3.1
author
#2 lore raw

PATCH 2/10 block: reorder elevator switch sequence

In reply to Tejun Heo

Elevator switch sequence first attached the new elevator, then tried registering it (sysfs) and if that failed attached back the old elevator. However, sysfs registration doesn't require the elevator to be attached, so there is no reason to do the "detach, attach new, register, maybe re-attach old" sequence. It can just do "register, detach, attach".

* elevator_init_queue() is updated to set ->elevator_data directly and return 0 / -errno. This allows elevator_exit() on an unattached elevator.

* __elv_unregister_queue() which was necessary to unregister unattached q is removed in favor of __elv_register_queue() which can register unattached q.

* elevator_attach() becomes a single assignment and obscures more then it helps. Dropped.

This will help cleaning up io_context handling across elevator switch.

This patch doesn't introduce visible behavior change.

---
 block/elevator.c |   91 ++++++++++++++++++++++--------------------------------
 1 files changed, 37 insertions(+), 54 deletions(-)
block/elevator.c +37 -54
index 6a343e8..a16c2d1 100644
@@ -168,17 +168,13 @@ static struct elevator_type *elevator_get(const char *name)
 	return e;
 }
 
-static void *elevator_init_queue(struct request_queue *q,
-				 struct elevator_queue *eq)
+static int elevator_init_queue(struct request_queue *q,
+			       struct elevator_queue *eq)
 {
-	return eq->ops->elevator_init_fn(q);
-}
-
-static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
-			   void *data)
-{
-	q->elevator = eq;
-	eq->elevator_data = data;
+	eq->elevator_data = eq->ops->elevator_init_fn(q);
+	if (eq->elevator_data)
+		return 0;
+	return -ENOMEM;
 }
 
 static char chosen_elevator[ELV_NAME_MAX];
@@ -241,7 +237,7 @@ int elevator_init(struct request_queue *q, char *name)
 {
 	struct elevator_type *e = NULL;
 	struct elevator_queue *eq;
-	void *data;
+	int err;
 
 	if (unlikely(q->elevator))
 		return 0;
@@ -278,13 +274,13 @@ int elevator_init(struct request_queue *q, char *name)
 	if (!eq)
 		return -ENOMEM;
 
-	data = elevator_init_queue(q, eq);
-	if (!data) {
+	err = elevator_init_queue(q, eq);
+	if (err) {
 		kobject_put(&eq->kobj);
-		return -ENOMEM;
+		return err;
 	}
 
-	elevator_attach(q, eq, data);
+	q->elevator = eq;
 	return 0;
 }
 EXPORT_SYMBOL(elevator_init);
@@ -856,9 +852,8 @@ static struct kobj_type elv_ktype = {
 	.release	= elevator_release,
 };
 
-int elv_register_queue(struct request_queue *q)
+int __elv_register_queue(struct request_queue *q, struct elevator_queue *e)
 {
-	struct elevator_queue *e = q->elevator;
 	int error;
 
 	error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
@@ -876,19 +871,22 @@ int elv_register_queue(struct request_queue *q)
 	}
 	return error;
 }
-EXPORT_SYMBOL(elv_register_queue);
 
-static void __elv_unregister_queue(struct elevator_queue *e)
+int elv_register_queue(struct request_queue *q)
 {
-	kobject_uevent(&e->kobj, KOBJ_REMOVE);
-	kobject_del(&e->kobj);
-	e->registered = 0;
+	return __elv_register_queue(q, q->elevator);
 }
+EXPORT_SYMBOL(elv_register_queue);
 
 void elv_unregister_queue(struct request_queue *q)
 {
-	if (q)
-		__elv_unregister_queue(q->elevator);
+	if (q) {
+		struct elevator_queue *e = q->elevator;
+
+		kobject_uevent(&e->kobj, KOBJ_REMOVE);
+		kobject_del(&e->kobj);
+		e->registered = 0;
+	}
 }
 EXPORT_SYMBOL(elv_unregister_queue);
 
@@ -928,50 +926,36 @@ EXPORT_SYMBOL_GPL(elv_unregister);
 static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
 {
 	struct elevator_queue *old_elevator, *e;
-	void *data;
 	int err;
 
-	/*
-	 * Allocate new elevator
-	 */
+	/* allocate new elevator */
 	e = elevator_alloc(q, new_e);
 	if (!e)
 		return -ENOMEM;
 
-	data = elevator_init_queue(q, e);
-	if (!data) {
+	err = elevator_init_queue(q, e);
+	if (err) {
 		kobject_put(&e->kobj);
-		return -ENOMEM;
+		return err;
 	}
 
-	/*
-	 * Turn on BYPASS and drain all requests w/ elevator private data
-	 */
+	/* turn on BYPASS and drain all requests w/ elevator private data */
 	elv_quiesce_start(q);
 
-	/*
-	 * Remember old elevator.
-	 */
-	old_elevator = q->elevator;
-
-	/*
-	 * attach and start new elevator
-	 */
-	spin_lock_irq(q->queue_lock);
-	elevator_attach(q, e, data);
-	spin_unlock_irq(q->queue_lock);
-
-	if (old_elevator->registered) {
-		__elv_unregister_queue(old_elevator);
-
-		err = elv_register_queue(q);
+	/* unregister old queue, register new one and kill old elevator */
+	if (q->elevator->registered) {
+		elv_unregister_queue(q);
+		err = __elv_register_queue(q, e);
 		if (err)
 			goto fail_register;
 	}
 
-	/*
-	 * finally exit old elevator and turn off BYPASS.
-	 */
+	/* done, replace the old one with new one and turn off BYPASS */
+	spin_lock_irq(q->queue_lock);
+	old_elevator = q->elevator;
+	q->elevator = e;
+	spin_unlock_irq(q->queue_lock);
+
 	elevator_exit(old_elevator);
 	elv_quiesce_end(q);
 
@@ -985,7 +969,6 @@ fail_register:
 	 * one again (along with re-adding the sysfs dir)
 	 */
 	elevator_exit(e);
-	q->elevator = old_elevator;
 	elv_register_queue(q);
 	elv_quiesce_end(q);
 
Signature · 1 line
1.7.3.1
author
#3 lore raw

PATCH 3/10 block: remove elevator_queue->ops

In reply to Tejun Heo

elevator_queue->ops points to the same ops struct ->elevator_type.ops is pointing to. The only effect of caching it in elevator_queue is shorter notation - it doesn't save any indirect derefence.

Relocate elevator_type->list which used only during module init/exit to the end of the structure, rename elevator_queue->elevator_type to ->type, and replace elevator_queue->ops with elevator_queue->type.ops.

This doesn't introduce any functional difference.

---
 block/blk.h              |   10 +++---
 block/elevator.c         |   74 ++++++++++++++++++++++-----------------------
 include/linux/elevator.h |    5 +--
 3 files changed, 43 insertions(+), 46 deletions(-)
block/blk.h +5 -5
index 5bca266..4943770 100644
@@ -94,7 +94,7 @@ static inline struct request *__elv_next_request(struct request_queue *q)
 			return NULL;
 		}
 		if (unlikely(blk_queue_dead(q)) ||
-		    !q->elevator->ops->elevator_dispatch_fn(q, 0))
+		    !q->elevator->type->ops.elevator_dispatch_fn(q, 0))
 			return NULL;
 	}
 }
@@ -103,16 +103,16 @@ static inline void elv_activate_rq(struct request_queue *q, struct request *rq)
 {
 	struct elevator_queue *e = q->elevator;
 
-	if (e->ops->elevator_activate_req_fn)
-		e->ops->elevator_activate_req_fn(q, rq);
+	if (e->type->ops.elevator_activate_req_fn)
+		e->type->ops.elevator_activate_req_fn(q, rq);
 }
 
 static inline void elv_deactivate_rq(struct request_queue *q, struct request *rq)
 {
 	struct elevator_queue *e = q->elevator;
 
-	if (e->ops->elevator_deactivate_req_fn)
-		e->ops->elevator_deactivate_req_fn(q, rq);
+	if (e->type->ops.elevator_deactivate_req_fn)
+		e->type->ops.elevator_deactivate_req_fn(q, rq);
 }
 
 #ifdef CONFIG_FAIL_IO_TIMEOUT
block/elevator.c +36 -38

216 lines not on this page · raw message

include/linux/elevator.h +2 -3
index 02604c8..04958ef 100644
@@ -78,11 +78,11 @@ struct elv_fs_entry {
  */
 struct elevator_type
 {
-	struct list_head list;
 	struct elevator_ops ops;
 	struct elv_fs_entry *elevator_attrs;
 	char elevator_name[ELV_NAME_MAX];
 	struct module *elevator_owner;
+	struct list_head list;
 };
 
 /*
@@ -90,10 +90,9 @@ struct elevator_type
  */
 struct elevator_queue
 {
-	struct elevator_ops *ops;
+	struct elevator_type *type;
 	void *elevator_data;
 	struct kobject kobj;
-	struct elevator_type *elevator_type;
 	struct mutex sysfs_lock;
 	struct hlist_head *hash;
 	unsigned int registered:1;
Signature · 1 line
1.7.3.1
author
#4 lore raw

PATCH 4/10 block, cfq: reorganize cfq_io_context into generic and cfq specific parts

In reply to Tejun Heo

Currently io_context and cfq logics are mixed without clear boundary. Most of io_context is independent from cfq but cfq_io_context handling logic is dispersed between generic ioc code and cfq.

cfq_io_context represents association between an io_context and a request_queue, which is a concept useful outside of cfq, but it also contains fields which are useful only to cfq.

This patch takes out generic part and put it into io_cq (io context-queue) and the rest into cfq_io_cq (cic moniker remains the same) which contains io_cq. The following changes are made together.

* cfq_ttime and cfq_io_cq now live in cfq-iosched.c.

* All related fields, functions and constants are renamed accordingly.

* ioc->ioc_data is now "struct io_cq *" instead of "void *" and renamed to icq_hint.

This prepares for io_context API cleanup. Documentation is currently sparse. It will be added later.

Changes in this patch are mechanical and don't cause functional change.

---
 block/blk-ioc.c           |   58 +++++------
 block/cfq-iosched.c       |  248 ++++++++++++++++++++++++---------------------
 include/linux/iocontext.h |   43 +++-----
 3 files changed, 175 insertions(+), 174 deletions(-)
block/blk-ioc.c +28 -30
index e23c797..dc5e69d 100644
@@ -46,7 +46,7 @@ EXPORT_SYMBOL(get_io_context);
 
 /*
  * Slow path for ioc release in put_io_context().  Performs double-lock
- * dancing to unlink all cic's and then frees ioc.
+ * dancing to unlink all icq's and then frees ioc.
  */
 static void ioc_release_fn(struct work_struct *work)
 {
@@ -56,11 +56,10 @@ static void ioc_release_fn(struct work_struct *work)
 
 	spin_lock_irq(&ioc->lock);
 
-	while (!hlist_empty(&ioc->cic_list)) {
-		struct cfq_io_context *cic = hlist_entry(ioc->cic_list.first,
-							 struct cfq_io_context,
-							 cic_list);
-		struct request_queue *this_q = cic->q;
+	while (!hlist_empty(&ioc->icq_list)) {
+		struct io_cq *icq = hlist_entry(ioc->icq_list.first,
+						struct io_cq, ioc_node);
+		struct request_queue *this_q = icq->q;
 
 		if (this_q != last_q) {
 			/*
@@ -89,8 +88,8 @@ static void ioc_release_fn(struct work_struct *work)
 			continue;
 		}
 		ioc_release_depth_inc(this_q);
-		cic->exit(cic);
-		cic->release(cic);
+		icq->exit(icq);
+		icq->release(icq);
 		ioc_release_depth_dec(this_q);
 	}
 
@@ -131,10 +130,10 @@ void put_io_context(struct io_context *ioc, struct request_queue *locked_q)
 		return;
 
 	/*
-	 * Destroy @ioc.  This is a bit messy because cic's are chained
+	 * Destroy @ioc.  This is a bit messy because icq's are chained
 	 * from both ioc and queue, and ioc->lock nests inside queue_lock.
-	 * The inner ioc->lock should be held to walk our cic_list and then
-	 * for each cic the outer matching queue_lock should be grabbed.
+	 * The inner ioc->lock should be held to walk our icq_list and then
+	 * for each icq the outer matching queue_lock should be grabbed.
 	 * ie. We need to do reverse-order double lock dancing.
 	 *
 	 * Another twist is that we are often called with one of the
@@ -153,11 +152,10 @@ void put_io_context(struct io_context *ioc, struct request_queue *locked_q)
 	spin_lock_irqsave_nested(&ioc->lock, flags,
 				 ioc_release_depth(locked_q));
 
-	while (!hlist_empty(&ioc->cic_list)) {
-		struct cfq_io_context *cic = hlist_entry(ioc->cic_list.first,
-							 struct cfq_io_context,
-							 cic_list);
-		struct request_queue *this_q = cic->q;
+	while (!hlist_empty(&ioc->icq_list)) {
+		struct io_cq *icq = hlist_entry(ioc->icq_list.first,
+						struct io_cq, ioc_node);
+		struct request_queue *this_q = icq->q;
 
 		if (this_q != last_q) {
 			if (last_q && last_q != locked_q)
@@ -170,8 +168,8 @@ void put_io_context(struct io_context *ioc, struct request_queue *locked_q)
 			continue;
 		}
 		ioc_release_depth_inc(this_q);
-		cic->exit(cic);
-		cic->release(cic);
+		icq->exit(icq);
+		icq->release(icq);
 		ioc_release_depth_dec(this_q);
 	}
 
@@ -180,8 +178,8 @@ void put_io_context(struct io_context *ioc, struct request_queue *locked_q)
 
 	spin_unlock_irqrestore(&ioc->lock, flags);
 
-	/* if no cic's left, we're done; otherwise, kick release_work */
-	if (hlist_empty(&ioc->cic_list))
+	/* if no icq is left, we're done; otherwise, kick release_work */
+	if (hlist_empty(&ioc->icq_list))
 		kmem_cache_free(iocontext_cachep, ioc);
 	else
 		schedule_work(&ioc->release_work);
@@ -219,8 +217,8 @@ void create_io_context_slowpath(struct task_struct *task, gfp_t gfp_flags,
 	atomic_long_set(&ioc->refcount, 1);
 	atomic_set(&ioc->nr_tasks, 1);
 	spin_lock_init(&ioc->lock);
-	INIT_RADIX_TREE(&ioc->radix_root, GFP_ATOMIC | __GFP_HIGH);
-	INIT_HLIST_HEAD(&ioc->cic_list);
+	INIT_RADIX_TREE(&ioc->icq_tree, GFP_ATOMIC | __GFP_HIGH);
+	INIT_HLIST_HEAD(&ioc->icq_list);
 	INIT_WORK(&ioc->release_work, ioc_release_fn);
 
 	/* try to install, somebody might already have beaten us to it */
@@ -270,11 +268,11 @@ EXPORT_SYMBOL(get_task_io_context);
 
 void ioc_set_changed(struct io_context *ioc, int which)
 {
-	struct cfq_io_context *cic;
+	struct io_cq *icq;
 	struct hlist_node *n;
 
-	hlist_for_each_entry(cic, n, &ioc->cic_list, cic_list)
-		set_bit(which, &cic->changed);
+	hlist_for_each_entry(icq, n, &ioc->icq_list, ioc_node)
+		set_bit(which, &icq->changed);
 }
 
 /**
@@ -282,8 +280,8 @@ void ioc_set_changed(struct io_context *ioc, int which)
  * @ioc: io_context of interest
  * @ioprio: new ioprio
  *
- * @ioc's ioprio has changed to @ioprio.  Set %CIC_IOPRIO_CHANGED for all
- * cic's.  iosched is responsible for checking the bit and applying it on
+ * @ioc's ioprio has changed to @ioprio.  Set %ICQ_IOPRIO_CHANGED for all
+ * icq's.  iosched is responsible for checking the bit and applying it on
  * request issue path.
  */
 void ioc_ioprio_changed(struct io_context *ioc, int ioprio)
@@ -292,7 +290,7 @@ void ioc_ioprio_changed(struct io_context *ioc, int ioprio)
 
 	spin_lock_irqsave(&ioc->lock, flags);
 	ioc->ioprio = ioprio;
-	ioc_set_changed(ioc, CIC_IOPRIO_CHANGED);
+	ioc_set_changed(ioc, ICQ_IOPRIO_CHANGED);
 	spin_unlock_irqrestore(&ioc->lock, flags);
 }
 
@@ -300,7 +298,7 @@ void ioc_ioprio_changed(struct io_context *ioc, int ioprio)
  * ioc_cgroup_changed - notify cgroup change
  * @ioc: io_context of interest
  *
- * @ioc's cgroup has changed.  Set %CIC_CGROUP_CHANGED for all cic's.
+ * @ioc's cgroup has changed.  Set %ICQ_CGROUP_CHANGED for all icq's.
  * iosched is responsible for checking the bit and applying it on request
  * issue path.
  */
@@ -309,7 +307,7 @@ void ioc_cgroup_changed(struct io_context *ioc)
 	unsigned long flags;
 
 	spin_lock_irqsave(&ioc->lock, flags);
-	ioc_set_changed(ioc, CIC_CGROUP_CHANGED);
+	ioc_set_changed(ioc, ICQ_CGROUP_CHANGED);
 	spin_unlock_irqrestore(&ioc->lock, flags);
 }
 
block/cfq-iosched.c +133 -115

604 lines not on this page · raw message

include/linux/iocontext.h +14 -29

61 lines not on this page · raw message

Signature · 1 line
1.7.3.1
author
#5 lore raw

PATCH 5/10 block, cfq: move cfqd->icq_list to request_queue and add request->elv.icq

In reply to Tejun Heo

Most of icq management is about to be moved out of cfq into blk-ioc. This patch prepares for it.

* Move cfqd->icq_list to request_queue->icq_list

* Make request explicitly point to icq instead of through elevator private data. ->elevator_private[3] is replaced with sub struct elv which contains icq pointer and priv[2]. cfq is updated accordingly.

* Meaningless clearing of ->elevator_private[0] removed from elv_set_request(). At that point in code, the field was guaranteed to be %NULL anyway.

This patch doesn't introduce any functional change.

---
 block/blk-core.c       |    1 +
 block/cfq-iosched.c    |   28 +++++++++++-----------------
 block/elevator.c       |    2 --
 include/linux/blkdev.h |   10 ++++++++--
 4 files changed, 20 insertions(+), 21 deletions(-)
block/blk-core.c +1 -0
index 74b44ab..5315f09 100644
@@ -492,6 +492,7 @@ struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 		    laptop_mode_timer_fn, (unsigned long) q);
 	setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
 	INIT_LIST_HEAD(&q->timeout_list);
+	INIT_LIST_HEAD(&q->icq_list);
 	INIT_LIST_HEAD(&q->flush_queue[0]);
 	INIT_LIST_HEAD(&q->flush_queue[1]);
 	INIT_LIST_HEAD(&q->flush_data_in_flight);
block/cfq-iosched.c +11 -17
index d2f16fc..9bc5ecc 100644
@@ -54,9 +54,9 @@ static const int cfq_hist_divisor = 4;
 #define CFQQ_SECT_THR_NONROT	(sector_t)(2 * 32)
 #define CFQQ_SEEKY(cfqq)	(hweight32(cfqq->seek_history) > 32/8)
 
-#define RQ_CIC(rq)		icq_to_cic((rq)->elevator_private[0])
-#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elevator_private[1])
-#define RQ_CFQG(rq)		(struct cfq_group *) ((rq)->elevator_private[2])
+#define RQ_CIC(rq)		icq_to_cic((rq)->elv.icq)
+#define RQ_CFQQ(rq)		(struct cfq_queue *) ((rq)->elv.priv[0])
+#define RQ_CFQG(rq)		(struct cfq_group *) ((rq)->elv.priv[1])
 
 static struct kmem_cache *cfq_pool;
 static struct kmem_cache *cfq_icq_pool;
@@ -297,8 +297,6 @@ struct cfq_data {
 	unsigned int cfq_group_idle;
 	unsigned int cfq_latency;
 
-	struct list_head icq_list;
-
 	/*
 	 * Fallback dummy cfqq for extreme OOM conditions
 	 */
@@ -3053,7 +3051,7 @@ static int cfq_create_cic(struct cfq_data *cfqd, gfp_t gfp_mask)
 	ret = radix_tree_insert(&ioc->icq_tree, q->id, icq);
 	if (likely(!ret)) {
 		hlist_add_head(&icq->ioc_node, &ioc->icq_list);
-		list_add(&icq->q_node, &cfqd->icq_list);
+		list_add(&icq->q_node, &q->icq_list);
 		icq = NULL;
 	} else if (ret == -EEXIST) {
 		/* someone else already did it */
@@ -3605,12 +3603,10 @@ static void cfq_put_request(struct request *rq)
 
 		put_io_context(RQ_CIC(rq)->icq.ioc, cfqq->cfqd->queue);
 
-		rq->elevator_private[0] = NULL;
-		rq->elevator_private[1] = NULL;
-
 		/* Put down rq reference on cfqg */
 		cfq_put_cfqg(RQ_CFQG(rq));
-		rq->elevator_private[2] = NULL;
+		rq->elv.priv[0] = NULL;
+		rq->elv.priv[1] = NULL;
 
 		cfq_put_queue(cfqq);
 	}
@@ -3696,9 +3692,9 @@ new_queue:
 	cfqq->allocated[rw]++;
 
 	cfqq->ref++;
-	rq->elevator_private[0] = &cic->icq;
-	rq->elevator_private[1] = cfqq;
-	rq->elevator_private[2] = cfq_ref_get_cfqg(cfqq->cfqg);
+	rq->elv.icq = &cic->icq;
+	rq->elv.priv[0] = cfqq;
+	rq->elv.priv[1] = cfq_ref_get_cfqg(cfqq->cfqg);
 	spin_unlock_irq(q->queue_lock);
 	return 0;
 
@@ -3810,8 +3806,8 @@ static void cfq_exit_queue(struct elevator_queue *e)
 	if (cfqd->active_queue)
 		__cfq_slice_expired(cfqd, cfqd->active_queue, 0);
 
-	while (!list_empty(&cfqd->icq_list)) {
-		struct io_cq *icq = list_entry(cfqd->icq_list.next,
+	while (!list_empty(&q->icq_list)) {
+		struct io_cq *icq = list_entry(q->icq_list.next,
 					       struct io_cq, q_node);
 		struct io_context *ioc = icq->ioc;
 
@@ -3922,8 +3918,6 @@ static void *cfq_init_queue(struct request_queue *q)
 	cfqd->oom_cfqq.ref++;
 	cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group);
 
-	INIT_LIST_HEAD(&cfqd->icq_list);
-
 	cfqd->queue = q;
 
 	init_timer(&cfqd->idle_slice_timer);
block/elevator.c +0 -2
index 31ffe76..c5c6214 100644
@@ -745,8 +745,6 @@ int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
 
 	if (e->type->ops.elevator_set_req_fn)
 		return e->type->ops.elevator_set_req_fn(q, rq, gfp_mask);
-
-	rq->elevator_private[0] = NULL;
 	return 0;
 }
 
include/linux/blkdev.h +8 -2
index e4c10a0..6c7e09c 100644
@@ -111,10 +111,14 @@ struct request {
 	 * Three pointers are available for the IO schedulers, if they need
 	 * more they have to dynamically allocate it.  Flush requests are
 	 * never put on the IO scheduler. So let the flush fields share
-	 * space with the three elevator_private pointers.
+	 * space with the elevator data.
 	 */
 	union {
-		void *elevator_private[3];
+		struct {
+			struct io_cq		*icq;
+			void			*priv[2];
+		} elv;
+
 		struct {
 			unsigned int		seq;
 			struct list_head	list;
@@ -357,6 +361,8 @@ struct request_queue {
 	struct timer_list	timeout;
 	struct list_head	timeout_list;
 
+	struct list_head	icq_list;
+
 	struct queue_limits	limits;
 
 	/*
Signature · 1 line
1.7.3.1
author
#6 lore raw

PATCH 6/10 block, cfq: move io_cq lookup to blk-ioc.c

In reply to Tejun Heo

Now that all io_cq related data structures are in block core layer, io_cq lookup can be moved from cfq-iosched.c to blk-ioc.c.

Lookup logic from cfq_cic_lookup() is moved to ioc_lookup_icq() with parameter return type changes (cfqd -> request_queue, cfq_io_cq -> io_cq) and cfq_cic_lookup() becomes thin wrapper around cfq_cic_lookup().

---
 block/blk-ioc.c     |   36 ++++++++++++++++++++++++++++++++++++
 block/blk.h         |    1 +
 block/cfq-iosched.c |   48 ++++++++----------------------------------------
 3 files changed, 45 insertions(+), 40 deletions(-)
block/blk-ioc.c +36 -0
index dc5e69d..87ecc98 100644
@@ -266,6 +266,42 @@ struct io_context *get_task_io_context(struct task_struct *task,
 }
 EXPORT_SYMBOL(get_task_io_context);
 
+/**
+ * ioc_lookup_icq - lookup io_cq from ioc
+ * @ioc: the associated io_context
+ * @q: the associated request_queue
+ *
+ * Look up io_cq associated with @ioc - @q pair from @ioc.  Must be called
+ * with @q->queue_lock held.
+ */
+struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q)
+{
+	struct io_cq *icq;
+
+	lockdep_assert_held(q->queue_lock);
+
+	/*
+	 * icq's are indexed from @ioc using radix tree and hint pointer,
+	 * both of which are protected with RCU.  All removals are done
+	 * holding both q and ioc locks, and we're holding q lock - if we
+	 * find a icq which points to us, it's guaranteed to be valid.
+	 */
+	rcu_read_lock();
+	icq = rcu_dereference(ioc->icq_hint);
+	if (icq && icq->q == q)
+		goto out;
+
+	icq = radix_tree_lookup(&ioc->icq_tree, q->id);
+	if (icq && icq->q == q)
+		rcu_assign_pointer(ioc->icq_hint, icq);	/* allowed to race */
+	else
+		icq = NULL;
+out:
+	rcu_read_unlock();
+	return icq;
+}
+EXPORT_SYMBOL(ioc_lookup_icq);
+
 void ioc_set_changed(struct io_context *ioc, int which)
 {
 	struct io_cq *icq;
block/blk.h +1 -0
index 4943770..3c510a4 100644
@@ -199,6 +199,7 @@ static inline int blk_do_io_stat(struct request *rq)
  * Internal io_context interface
  */
 void get_io_context(struct io_context *ioc);
+struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
 
 void create_io_context_slowpath(struct task_struct *task, gfp_t gfp_mask,
 				int node);
block/cfq-iosched.c +8 -40

66 lines not on this page · raw message

Signature · 1 line
1.7.3.1
author
#7 lore raw

PATCH 7/10 block, cfq: move icq cache management to block core

In reply to Tejun Heo

Let elevators set ->icq_size and ->icq_align in elevator_type and elv_register() and elv_unregister() respectively create and destroy kmem_cache for icq.

* elv_register() now can return failure. All callers updated.

* icq caches are automatically named "ELVNAME_io_cq".

* cfq_slab_setup/kill() are collapsed into cfq_init/exit().

* While at it, minor indentation change for iosched_cfq.elevator_name for consistency.

This will help moving icq management to block core. This doesn't introduce any functional change.

---
 block/cfq-iosched.c      |   48 ++++++++++++++-------------------------------
 block/deadline-iosched.c |    4 +--
 block/elevator.c         |   37 +++++++++++++++++++++++++++++++++-
 block/noop-iosched.c     |    4 +--
 include/linux/elevator.h |   11 +++++++++-
 5 files changed, 62 insertions(+), 42 deletions(-)
block/cfq-iosched.c +15 -33
index 048fa69..06e59ab 100644
@@ -3914,34 +3914,6 @@ static void *cfq_init_queue(struct request_queue *q)
 	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_icq_pool)
-		kmem_cache_destroy(cfq_icq_pool);
-}
-
-static int __init cfq_slab_setup(void)
-{
-	cfq_pool = KMEM_CACHE(cfq_queue, 0);
-	if (!cfq_pool)
-		goto fail;
-
-	cfq_icq_pool = KMEM_CACHE(cfq_io_cq, 0);
-	if (!cfq_icq_pool)
-		goto fail;
-
-	return 0;
-fail:
-	cfq_slab_kill();
-	return -ENOMEM;
-}
-
 /*
  * sysfs parts below -->
  */
@@ -4053,8 +4025,10 @@ static struct elevator_type iosched_cfq = {
 		.elevator_init_fn =		cfq_init_queue,
 		.elevator_exit_fn =		cfq_exit_queue,
 	},
+	.icq_size	=	sizeof(struct cfq_io_cq),
+	.icq_align	=	__alignof__(struct cfq_io_cq),
 	.elevator_attrs =	cfq_attrs,
-	.elevator_name =	"cfq",
+	.elevator_name	=	"cfq",
 	.elevator_owner =	THIS_MODULE,
 };
 
@@ -4072,6 +4046,8 @@ static struct blkio_policy_type blkio_policy_cfq;
 
 static int __init cfq_init(void)
 {
+	int ret;
+
 	/*
 	 * could be 0 on HZ < 1000 setups
 	 */
@@ -4086,10 +4062,17 @@ static int __init cfq_init(void)
 #else
 		cfq_group_idle = 0;
 #endif
-	if (cfq_slab_setup())
+	cfq_pool = KMEM_CACHE(cfq_queue, 0);
+	if (!cfq_pool)
 		return -ENOMEM;
 
-	elv_register(&iosched_cfq);
+	ret = elv_register(&iosched_cfq);
+	if (ret) {
+		kmem_cache_destroy(cfq_pool);
+		return ret;
+	}
+	cfq_icq_pool = iosched_cfq.icq_cache;
+
 	blkio_policy_register(&blkio_policy_cfq);
 
 	return 0;
@@ -4099,8 +4082,7 @@ static void __exit cfq_exit(void)
 {
 	blkio_policy_unregister(&blkio_policy_cfq);
 	elv_unregister(&iosched_cfq);
-	rcu_barrier();	/* make sure all cic RCU frees are complete */
-	cfq_slab_kill();
+	kmem_cache_destroy(cfq_pool);
 }
 
 module_init(cfq_init);
block/deadline-iosched.c +1 -3
index c644137..7bf12d7 100644
@@ -448,9 +448,7 @@ static struct elevator_type iosched_deadline = {
 
 static int __init deadline_init(void)
 {
-	elv_register(&iosched_deadline);
-
-	return 0;
+	return elv_register(&iosched_deadline);
 }
 
 static void __exit deadline_exit(void)
block/elevator.c +35 -2
index c5c6214..cca049f 100644
@@ -886,15 +886,36 @@ void elv_unregister_queue(struct request_queue *q)
 }
 EXPORT_SYMBOL(elv_unregister_queue);
 
-void elv_register(struct elevator_type *e)
+int elv_register(struct elevator_type *e)
 {
 	char *def = "";
 
+	/* create icq_cache if requested */
+	if (e->icq_size) {
+		if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
+		    WARN_ON(e->icq_align < __alignof__(struct io_cq)))
+			return -EINVAL;
+
+		snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
+			 "%s_io_cq", e->elevator_name);
+		e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
+						 e->icq_align, 0, NULL);
+		if (!e->icq_cache)
+			return -ENOMEM;
+	}
+
+	/* register, don't allow duplicate names */
 	spin_lock(&elv_list_lock);
-	BUG_ON(elevator_find(e->elevator_name));
+	if (elevator_find(e->elevator_name)) {
+		spin_unlock(&elv_list_lock);
+		if (e->icq_cache)
+			kmem_cache_destroy(e->icq_cache);
+		return -EBUSY;
+	}
 	list_add_tail(&e->list, &elv_list);
 	spin_unlock(&elv_list_lock);
 
+	/* print pretty message */
 	if (!strcmp(e->elevator_name, chosen_elevator) ||
 			(!*chosen_elevator &&
 			 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
@@ -902,14 +923,26 @@ void elv_register(struct elevator_type *e)
 
 	printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
 								def);
+	return 0;
 }
 EXPORT_SYMBOL_GPL(elv_register);
 
 void elv_unregister(struct elevator_type *e)
 {
+	/* unregister */
 	spin_lock(&elv_list_lock);
 	list_del_init(&e->list);
 	spin_unlock(&elv_list_lock);
+
+	/*
+	 * Destroy icq_cache if it exists.  icq's are RCU managed.  Make
+	 * sure all RCU operations are complete before proceeding.
+	 */
+	if (e->icq_cache) {
+		rcu_barrier();
+		kmem_cache_destroy(e->icq_cache);
+		e->icq_cache = NULL;
+	}
 }
 EXPORT_SYMBOL_GPL(elv_unregister);
 
block/noop-iosched.c +1 -3
index 06389e9..413a0b1 100644
@@ -94,9 +94,7 @@ static struct elevator_type elevator_noop = {
 
 static int __init noop_init(void)
 {
-	elv_register(&elevator_noop);
-
-	return 0;
+	return elv_register(&elevator_noop);
 }
 
 static void __exit noop_exit(void)
include/linux/elevator.h +10 -1
index 04958ef..d3d3e28 100644
@@ -78,10 +78,19 @@ struct elv_fs_entry {
  */
 struct elevator_type
 {
+	/* managed by elevator core */
+	struct kmem_cache *icq_cache;
+
+	/* fields provided by elevator implementation */
 	struct elevator_ops ops;
+	size_t icq_size;
+	size_t icq_align;
 	struct elv_fs_entry *elevator_attrs;
 	char elevator_name[ELV_NAME_MAX];
 	struct module *elevator_owner;
+
+	/* managed by elevator core */
+	char icq_cache_name[ELV_NAME_MAX + 5];	/* elvname + "_io_cq" */
 	struct list_head list;
 };
 
@@ -127,7 +136,7 @@ extern void elv_drain_elevator(struct request_queue *);
 /*
  * io scheduler registration
  */
-extern void elv_register(struct elevator_type *);
+extern int elv_register(struct elevator_type *);
 extern void elv_unregister(struct elevator_type *);
 
 /*
Signature · 1 line
1.7.3.1
author
#8 lore raw

PATCH 8/10 block, cfq: move io_cq exit/release to blk-ioc.c

In reply to Tejun Heo

With kmem_cache managed by blk-ioc, io_cq exit/release can be moved to blk-ioc too. The odd ->io_cq->exit/release() callbacks are replaced with elevator_ops->elevator_exit_icq_fn() with unlinking from both ioc and q, and freeing automatically handled by blk-ioc. The elevator operation only need to perform exit operation specific to the elevator
- in cfq's case, exiting the cfqq's.

Also, clearing of io_cq's on q detach is moved to block core and automatically performed on elevator switch and q release.

Because the q io_cq points to might be freed before RCU callback for the io_cq runs, blk-ioc code should remember to which cache the io_cq needs to be freed when the io_cq is released. New field io_cq->__rcu_icq_cache is added for this purpose. As both the new field and rcu_head are used only after io_cq is released and the q/ioc_node fields aren't, they are put into unions.

---
 block/blk-ioc.c           |   76 ++++++++++++++++++++++++++++++++++++++++-----
 block/blk-sysfs.c         |    6 +++-
 block/blk.h               |    1 +
 block/cfq-iosched.c       |   47 +--------------------------
 block/elevator.c          |    3 +-
 include/linux/elevator.h  |    5 +++
 include/linux/iocontext.h |   20 ++++++++---
 7 files changed, 97 insertions(+), 61 deletions(-)
block/blk-ioc.c +68 -8
index 87ecc98..0910a55 100644
@@ -44,6 +44,51 @@ EXPORT_SYMBOL(get_io_context);
 #define ioc_release_depth_dec(q)	do { } while (0)
 #endif
 
+static void icq_free_icq_rcu(struct rcu_head *head)
+{
+	struct io_cq *icq = container_of(head, struct io_cq, __rcu_head);
+
+	kmem_cache_free(icq->__rcu_icq_cache, icq);
+}
+
+/*
+ * Exit and free an icq.  Called with both ioc and q locked.
+ */
+static void ioc_exit_icq(struct io_cq *icq)
+{
+	struct io_context *ioc = icq->ioc;
+	struct request_queue *q = icq->q;
+	struct elevator_type *et = q->elevator->type;
+
+	lockdep_assert_held(&ioc->lock);
+	lockdep_assert_held(q->queue_lock);
+
+	radix_tree_delete(&ioc->icq_tree, icq->q->id);
+	hlist_del_init(&icq->ioc_node);
+	list_del_init(&icq->q_node);
+
+	/*
+	 * Both setting lookup hint to and clearing it from @icq are done
+	 * under queue_lock.  If it's not pointing to @icq now, it never
+	 * will.  Hint assignment itself can race safely.
+	 */
+	if (rcu_dereference_raw(ioc->icq_hint) == icq)
+		rcu_assign_pointer(ioc->icq_hint, NULL);
+
+	if (et->ops.elevator_exit_icq_fn) {
+		ioc_release_depth_inc(q);
+		et->ops.elevator_exit_icq_fn(icq);
+		ioc_release_depth_dec(q);
+	}
+
+	/*
+	 * @icq->q might have gone away by the time RCU callback runs
+	 * making it impossible to determine icq_cache.  Record it in @icq.
+	 */
+	icq->__rcu_icq_cache = et->icq_cache;
+	call_rcu(&icq->__rcu_head, icq_free_icq_rcu);
+}
+
 /*
  * Slow path for ioc release in put_io_context().  Performs double-lock
  * dancing to unlink all icq's and then frees ioc.
@@ -87,10 +132,7 @@ static void ioc_release_fn(struct work_struct *work)
 			spin_lock(&ioc->lock);
 			continue;
 		}
-		ioc_release_depth_inc(this_q);
-		icq->exit(icq);
-		icq->release(icq);
-		ioc_release_depth_dec(this_q);
+		ioc_exit_icq(icq);
 	}
 
 	if (last_q) {
@@ -167,10 +209,7 @@ void put_io_context(struct io_context *ioc, struct request_queue *locked_q)
 			last_q = this_q;
 			continue;
 		}
-		ioc_release_depth_inc(this_q);
-		icq->exit(icq);
-		icq->release(icq);
-		ioc_release_depth_dec(this_q);
+		ioc_exit_icq(icq);
 	}
 
 	if (last_q && last_q != locked_q)
@@ -203,6 +242,27 @@ void exit_io_context(struct task_struct *task)
 	put_io_context(ioc, NULL);
 }
 
+/**
+ * ioc_clear_queue - break any ioc association with the specified queue
+ * @q: request_queue being cleared
+ *
+ * Walk @q->icq_list and exit all io_cq's.  Must be called with @q locked.
+ */
+void ioc_clear_queue(struct request_queue *q)
+{
+	lockdep_assert_held(q->queue_lock);
+
+	while (!list_empty(&q->icq_list)) {
+		struct io_cq *icq = list_entry(q->icq_list.next,
+					       struct io_cq, q_node);
+		struct io_context *ioc = icq->ioc;
+
+		spin_lock(&ioc->lock);
+		ioc_exit_icq(icq);
+		spin_unlock(&ioc->lock);
+	}
+}
+
 void create_io_context_slowpath(struct task_struct *task, gfp_t gfp_flags,
 				int node)
 {
block/blk-sysfs.c +5 -1
index 5b4b4ab..cf15001 100644
@@ -479,8 +479,12 @@ static void blk_release_queue(struct kobject *kobj)
 
 	blk_sync_queue(q);
 
-	if (q->elevator)
+	if (q->elevator) {
+		spin_lock_irq(q->queue_lock);
+		ioc_clear_queue(q);
+		spin_unlock_irq(q->queue_lock);
 		elevator_exit(q->elevator);
+	}
 
 	blk_throtl_exit(q);
 
block/blk.h +1 -0
index 3c510a4..ed4d9bf 100644
@@ -200,6 +200,7 @@ static inline int blk_do_io_stat(struct request *rq)
  */
 void get_io_context(struct io_context *ioc);
 struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
+void ioc_clear_queue(struct request_queue *q);
 
 void create_io_context_slowpath(struct task_struct *task, gfp_t gfp_mask,
 				int node);
block/cfq-iosched.c +2 -45

83 lines not on this page · raw message

block/elevator.c +2 -1
index cca049f..91e18f8 100644
@@ -979,8 +979,9 @@ static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
 			goto fail_register;
 	}
 
-	/* done, replace the old one with new one and turn off BYPASS */
+	/* done, clear io_cq's, switch elevators and turn off BYPASS */
 	spin_lock_irq(q->queue_lock);
+	ioc_clear_queue(q);
 	old_elevator = q->elevator;
 	q->elevator = e;
 	spin_unlock_irq(q->queue_lock);
include/linux/elevator.h +5 -0
index d3d3e28..06e4dd56 100644
@@ -5,6 +5,8 @@
 
 #ifdef CONFIG_BLOCK
 
+struct io_cq;
+
 typedef int (elevator_merge_fn) (struct request_queue *, struct request **,
 				 struct bio *);
 
@@ -24,6 +26,7 @@ typedef struct request *(elevator_request_list_fn) (struct request_queue *, stru
 typedef void (elevator_completed_req_fn) (struct request_queue *, struct request *);
 typedef int (elevator_may_queue_fn) (struct request_queue *, int);
 
+typedef void (elevator_exit_icq_fn) (struct io_cq *);
 typedef int (elevator_set_req_fn) (struct request_queue *, struct request *, gfp_t);
 typedef void (elevator_put_req_fn) (struct request *);
 typedef void (elevator_activate_req_fn) (struct request_queue *, struct request *);
@@ -56,6 +59,8 @@ struct elevator_ops
 	elevator_request_list_fn *elevator_former_req_fn;
 	elevator_request_list_fn *elevator_latter_req_fn;
 
+	elevator_exit_icq_fn *elevator_exit_icq_fn;
+
 	elevator_set_req_fn *elevator_set_req_fn;
 	elevator_put_req_fn *elevator_put_req_fn;
 
include/linux/iocontext.h +14 -6
index d15ca65..ac390a3 100644
@@ -14,14 +14,22 @@ struct io_cq {
 	struct request_queue	*q;
 	struct io_context	*ioc;
 
-	struct list_head	q_node;
-	struct hlist_node	ioc_node;
+	/*
+	 * q_node and ioc_node link io_cq through icq_list of q and ioc
+	 * respectively.  Both fields are unused once ioc_exit_icq() is
+	 * called and shared with __rcu_icq_cache and __rcu_head which are
+	 * used for RCU free of io_cq.
+	 */
+	union {
+		struct list_head	q_node;
+		struct kmem_cache	*__rcu_icq_cache;
+	};
+	union {
+		struct hlist_node	ioc_node;
+		struct rcu_head		__rcu_head;
+	};
 
 	unsigned long		changed;
-	struct rcu_head		rcu_head;
-
-	void (*exit)(struct io_cq *);
-	void (*release)(struct io_cq *);
 };
 
 /*
Signature · 1 line
1.7.3.1
author
#9 lore raw

PATCH 9/10 block, cfq: restructure io_cq creation path for io_context interface cleanup

In reply to Tejun Heo

Add elevator_ops->elevator_init_icq_fn() and restructure cfq_create_cic() and rename it to ioc_create_icq().

The new function expects its caller to pass in io_context, uses elevator_type->icq_cache, handles generic init, calls the new elevator operation for elevator specific initialization, and returns pointer to created or looked up icq. This leaves cfq_icq_pool variable without any user. Removed.

This prepares for io_context interface cleanup and doesn't introduce any functional difference.

---
 block/cfq-iosched.c      |   94 ++++++++++++++++++++--------------------------
 include/linux/elevator.h |    2 +
 2 files changed, 43 insertions(+), 53 deletions(-)
block/cfq-iosched.c +41 -53
index f6d3155..11f49d0 100644
@@ -59,7 +59,6 @@ static const int cfq_hist_divisor = 4;
 #define RQ_CFQG(rq)		(struct cfq_group *) ((rq)->elv.priv[1])
 
 static struct kmem_cache *cfq_pool;
-static struct kmem_cache *cfq_icq_pool;
 
 #define CFQ_PRIO_LISTS		IOPRIO_BE_NR
 #define cfq_class_idle(cfqq)	((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE)
@@ -2707,6 +2706,13 @@ static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 	cfq_put_queue(cfqq);
 }
 
+static void cfq_init_icq(struct io_cq *icq)
+{
+	struct cfq_io_cq *cic = icq_to_cic(icq);
+
+	cic->ttime.last_end_request = jiffies;
+}
+
 static void cfq_exit_icq(struct io_cq *icq)
 {
 	struct cfq_io_cq *cic = icq_to_cic(icq);
@@ -2723,21 +2729,6 @@ static void cfq_exit_icq(struct io_cq *icq)
 	}
 }
 
-static struct cfq_io_cq *cfq_alloc_cic(struct cfq_data *cfqd, gfp_t gfp_mask)
-{
-	struct cfq_io_cq *cic;
-
-	cic = kmem_cache_alloc_node(cfq_icq_pool, gfp_mask | __GFP_ZERO,
-							cfqd->queue->node);
-	if (cic) {
-		cic->ttime.last_end_request = jiffies;
-		INIT_LIST_HEAD(&cic->icq.q_node);
-		INIT_HLIST_NODE(&cic->icq.ioc_node);
-	}
-
-	return cic;
-}
-
 static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc)
 {
 	struct task_struct *tsk = current;
@@ -2945,64 +2936,62 @@ cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc,
 }
 
 /**
- * cfq_create_cic - create and link a cfq_io_cq
- * @cfqd: cfqd of interest
+ * ioc_create_icq - create and link io_cq
+ * @q: request_queue of interest
  * @gfp_mask: allocation mask
  *
- * Make sure cfq_io_cq linking %current->io_context and @cfqd exists.  If
- * ioc and/or cic doesn't exist, they will be created using @gfp_mask.
+ * Make sure io_cq linking %current->io_context and @q exists.  If either
+ * io_context and/or icq don't exist, they will be created using @gfp_mask.
+ *
+ * The caller is responsible for ensuring @ioc won't go away and @q is
+ * alive and will stay alive until this function returns.
  */
-static int cfq_create_cic(struct cfq_data *cfqd, gfp_t gfp_mask)
+static struct io_cq *ioc_create_icq(struct request_queue *q, gfp_t gfp_mask)
 {
-	struct request_queue *q = cfqd->queue;
-	struct io_cq *icq = NULL;
-	struct cfq_io_cq *cic;
+	struct elevator_type *et = q->elevator->type;
 	struct io_context *ioc;
-	int ret = -ENOMEM;
-
-	might_sleep_if(gfp_mask & __GFP_WAIT);
+	struct io_cq *icq;
 
 	/* allocate stuff */
 	ioc = create_io_context(current, gfp_mask, q->node);
 	if (!ioc)
-		goto out;
+		return NULL;
 
-	cic = cfq_alloc_cic(cfqd, gfp_mask);
-	if (!cic)
-		goto out;
-	icq = &cic->icq;
+	icq = kmem_cache_alloc_node(et->icq_cache, gfp_mask | __GFP_ZERO,
+				    q->node);
+	if (!icq)
+		return NULL;
 
-	ret = radix_tree_preload(gfp_mask);
-	if (ret)
-		goto out;
+	if (radix_tree_preload(gfp_mask) < 0) {
+		kmem_cache_free(et->icq_cache, icq);
+		return NULL;
+	}
 
 	icq->ioc = ioc;
-	icq->q = cfqd->queue;
+	icq->q = q;
+	INIT_LIST_HEAD(&icq->q_node);
+	INIT_HLIST_NODE(&icq->ioc_node);
 
 	/* lock both q and ioc and try to link @icq */
 	spin_lock_irq(q->queue_lock);
 	spin_lock(&ioc->lock);
 
-	ret = radix_tree_insert(&ioc->icq_tree, q->id, icq);
-	if (likely(!ret)) {
+	if (likely(!radix_tree_insert(&ioc->icq_tree, q->id, icq))) {
 		hlist_add_head(&icq->ioc_node, &ioc->icq_list);
 		list_add(&icq->q_node, &q->icq_list);
-		icq = NULL;
-	} else if (ret == -EEXIST) {
-		/* someone else already did it */
-		ret = 0;
+		if (et->ops.elevator_init_icq_fn)
+			et->ops.elevator_init_icq_fn(icq);
+	} else {
+		kmem_cache_free(et->icq_cache, icq);
+		icq = ioc_lookup_icq(ioc, q);
+		if (!icq)
+			printk(KERN_ERR "cfq: icq link failed!\n");
 	}
 
 	spin_unlock(&ioc->lock);
 	spin_unlock_irq(q->queue_lock);
-
 	radix_tree_preload_end();
-out:
-	if (ret)
-		printk(KERN_ERR "cfq: icq link failed!\n");
-	if (icq)
-		kmem_cache_free(cfq_icq_pool, icq);
-	return ret;
+	return icq;
 }
 
 /**
@@ -3022,7 +3011,6 @@ static struct cfq_io_cq *cfq_get_cic(struct cfq_data *cfqd, gfp_t gfp_mask)
 	struct request_queue *q = cfqd->queue;
 	struct cfq_io_cq *cic = NULL;
 	struct io_context *ioc;
-	int err;
 
 	lockdep_assert_held(q->queue_lock);
 
@@ -3037,9 +3025,9 @@ static struct cfq_io_cq *cfq_get_cic(struct cfq_data *cfqd, gfp_t gfp_mask)
 
 		/* slow path - unlock, create missing ones and retry */
 		spin_unlock_irq(q->queue_lock);
-		err = cfq_create_cic(cfqd, gfp_mask);
+		cic = icq_to_cic(ioc_create_icq(q, gfp_mask));
 		spin_lock_irq(q->queue_lock);
-		if (err)
+		if (!cic)
 			return NULL;
 	}
 
@@ -3975,6 +3963,7 @@ static struct elevator_type iosched_cfq = {
 		.elevator_completed_req_fn =	cfq_completed_request,
 		.elevator_former_req_fn =	elv_rb_former_request,
 		.elevator_latter_req_fn =	elv_rb_latter_request,
+		.elevator_init_icq_fn =		cfq_init_icq,
 		.elevator_exit_icq_fn =		cfq_exit_icq,
 		.elevator_set_req_fn =		cfq_set_request,
 		.elevator_put_req_fn =		cfq_put_request,
@@ -4028,7 +4017,6 @@ static int __init cfq_init(void)
 		kmem_cache_destroy(cfq_pool);
 		return ret;
 	}
-	cfq_icq_pool = iosched_cfq.icq_cache;
 
 	blkio_policy_register(&blkio_policy_cfq);
 
include/linux/elevator.h +2 -0
index 06e4dd56..c8f1e67 100644
@@ -26,6 +26,7 @@ typedef struct request *(elevator_request_list_fn) (struct request_queue *, stru
 typedef void (elevator_completed_req_fn) (struct request_queue *, struct request *);
 typedef int (elevator_may_queue_fn) (struct request_queue *, int);
 
+typedef void (elevator_init_icq_fn) (struct io_cq *);
 typedef void (elevator_exit_icq_fn) (struct io_cq *);
 typedef int (elevator_set_req_fn) (struct request_queue *, struct request *, gfp_t);
 typedef void (elevator_put_req_fn) (struct request *);
@@ -59,6 +60,7 @@ struct elevator_ops
 	elevator_request_list_fn *elevator_former_req_fn;
 	elevator_request_list_fn *elevator_latter_req_fn;
 
+	elevator_init_icq_fn *elevator_init_icq_fn;
 	elevator_exit_icq_fn *elevator_exit_icq_fn;
 
 	elevator_set_req_fn *elevator_set_req_fn;
Signature · 1 line
1.7.3.1
author
#10 lore raw

PATCH 10/10 block, cfq: move icq creation and rq->elv.icq association to block core

In reply to Tejun Heo

Now block layer knows everything necessary to create and associate icq's with requests. Move ioc_create_icq() to blk-ioc.c and update get_request() such that, if elevator_type->icq_size is set, requests are automatically associated with their matching icq's before elv_set_request(). io_context reference is also managed by block core on request alloc/free.

* Only ioprio/cgroup changed handling remains from cfq_get_cic(). Collapsed into cfq_set_request().

* This removes queue kicking on icq allocation failure (for now). As icq allocation failure is rare and the only effect of queue kicking achieved was possibily accelerating queue processing, this change shouldn't be noticeable.

There is a larger underlying problem. Unlike request allocation, icq allocation is not guaranteed to succeed eventually after retries. The number of icq is unbound and thus mempool can't be the solution either. This effectively adds allocation dependency on memory free path and thus possibility of deadlock.

This usually wouldn't happen because icq allocation is not a hot path and, even when the condition triggers, it's highly unlikely that none of the writeback workers already has icq.

However, this is still possible especially if elevator is being switched under high memory pressure, so we better get it fixed. Probably the only solution is just bypassing elevator and appending to dispatch queue on any elevator allocation failure.

* Comment added to explain how icq's are managed and synchronized.

This completes cleanup of io_context interface.

---
 block/blk-core.c          |   46 +++++++++++++--
 block/blk-ioc.c           |   60 ++++++++++++++++++++-
 block/blk.h               |    1 +
 block/cfq-iosched.c       |  135 ++++-----------------------------------------
 include/linux/elevator.h  |    8 +-
 include/linux/iocontext.h |   59 ++++++++++++++++++++
 6 files changed, 173 insertions(+), 136 deletions(-)
block/blk-core.c +39 -7
index 5315f09..2da3cd6 100644
@@ -635,13 +635,18 @@ EXPORT_SYMBOL(blk_get_queue);
 
 static inline void blk_free_request(struct request_queue *q, struct request *rq)
 {
-	if (rq->cmd_flags & REQ_ELVPRIV)
+	if (rq->cmd_flags & REQ_ELVPRIV) {
 		elv_put_request(q, rq);
+		if (rq->elv.icq)
+			put_io_context(rq->elv.icq->ioc, q);
+	}
+
 	mempool_free(rq, q->rq.rq_pool);
 }
 
 static struct request *
-blk_alloc_request(struct request_queue *q, unsigned int flags, gfp_t gfp_mask)
+blk_alloc_request(struct request_queue *q, struct io_cq *icq,
+		  unsigned int flags, gfp_t gfp_mask)
 {
 	struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
 
@@ -652,10 +657,15 @@ blk_alloc_request(struct request_queue *q, unsigned int flags, gfp_t gfp_mask)
 
 	rq->cmd_flags = flags | REQ_ALLOCED;
 
-	if ((flags & REQ_ELVPRIV) &&
-	    unlikely(elv_set_request(q, rq, gfp_mask))) {
-		mempool_free(rq, q->rq.rq_pool);
-		return NULL;
+	if (flags & REQ_ELVPRIV) {
+		rq->elv.icq = icq;
+		if (unlikely(elv_set_request(q, rq, gfp_mask))) {
+			mempool_free(rq, q->rq.rq_pool);
+			return NULL;
+		}
+		/* @rq->elv.icq holds on to io_context until @rq is freed */
+		if (icq)
+			get_io_context(icq->ioc);
 	}
 
 	return rq;
@@ -767,11 +777,14 @@ static struct request *get_request(struct request_queue *q, int rw_flags,
 {
 	struct request *rq = NULL;
 	struct request_list *rl = &q->rq;
+	struct elevator_type *et;
 	struct io_context *ioc;
+	struct io_cq *icq = NULL;
 	const bool is_sync = rw_is_sync(rw_flags) != 0;
 	bool retried = false;
 	int may_queue;
 retry:
+	et = q->elevator->type;
 	ioc = current->io_context;
 
 	if (unlikely(blk_queue_dead(q)))
@@ -832,17 +845,36 @@ retry:
 	rl->count[is_sync]++;
 	rl->starved[is_sync] = 0;
 
+	/*
+	 * Decide whether the new request will be managed by elevator.  If
+	 * so, mark @rw_flags and increment elvpriv.  Non-zero elvpriv will
+	 * prevent the current elevator from being destroyed until the new
+	 * request is freed.  This guarantees icq's won't be destroyed and
+	 * makes creating new ones safe.
+	 *
+	 * Also, lookup icq while holding queue_lock.  If it doesn't exist,
+	 * it will be created after releasing queue_lock.
+	 */
 	if (blk_rq_should_init_elevator(bio) &&
 	    !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags)) {
 		rw_flags |= REQ_ELVPRIV;
 		rl->elvpriv++;
+		if (et->icq_cache && ioc)
+			icq = ioc_lookup_icq(ioc, q);
 	}
 
 	if (blk_queue_io_stat(q))
 		rw_flags |= REQ_IO_STAT;
 	spin_unlock_irq(q->queue_lock);
 
-	rq = blk_alloc_request(q, rw_flags, gfp_mask);
+	/* create icq if missing */
+	if (unlikely(et->icq_cache && !icq))
+		icq = ioc_create_icq(q, gfp_mask);
+
+	/* rqs are guaranteed to have icq on elv_set_request() if requested */
+	if (likely(!et->icq_cache || icq))
+		rq = blk_alloc_request(q, icq, rw_flags, gfp_mask);
+
 	if (unlikely(!rq)) {
 		/*
 		 * Allocation failed presumably due to memory. Undo anything
block/blk-ioc.c +59 -1

72 lines not on this page · raw message

block/blk.h +1 -0
index ed4d9bf..7efd772 100644
@@ -200,6 +200,7 @@ static inline int blk_do_io_stat(struct request *rq)
  */
 void get_io_context(struct io_context *ioc);
 struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
+struct io_cq *ioc_create_icq(struct request_queue *q, gfp_t gfp_mask);
 void ioc_clear_queue(struct request_queue *q);
 
 void create_io_context_slowpath(struct task_struct *task, gfp_t gfp_mask,
block/cfq-iosched.c +11 -124

169 lines not on this page · raw message

include/linux/elevator.h +4 -4
index c8f1e67..c24f3d7 100644
@@ -60,8 +60,8 @@ struct elevator_ops
 	elevator_request_list_fn *elevator_former_req_fn;
 	elevator_request_list_fn *elevator_latter_req_fn;
 
-	elevator_init_icq_fn *elevator_init_icq_fn;
-	elevator_exit_icq_fn *elevator_exit_icq_fn;
+	elevator_init_icq_fn *elevator_init_icq_fn;	/* see iocontext.h */
+	elevator_exit_icq_fn *elevator_exit_icq_fn;	/* ditto */
 
 	elevator_set_req_fn *elevator_set_req_fn;
 	elevator_put_req_fn *elevator_put_req_fn;
@@ -90,8 +90,8 @@ struct elevator_type
 
 	/* fields provided by elevator implementation */
 	struct elevator_ops ops;
-	size_t icq_size;
-	size_t icq_align;
+	size_t icq_size;	/* see iocontext.h */
+	size_t icq_align;	/* ditto */
 	struct elv_fs_entry *elevator_attrs;
 	char elevator_name[ELV_NAME_MAX];
 	struct module *elevator_owner;
include/linux/iocontext.h +59 -0

65 lines not on this page · raw message

Signature · 1 line
1.7.3.1
maintainer
#12 lore raw
Quoted 1 line · Tejun Heo

Jens, ping?

I'll queue this up as soon as the current stuff has been shipped off, it's just too close (and inside of) the merge window to slide in at this point.

Signature · 1 line
Jens Axboe
author
#13 lore raw

Hello,

Quoted 6 lines · Jens Axboe
Quoted 1 line · Tejun Heo

Jens, ping?

I'll queue this up as soon as the current stuff has been shipped off, it's just too close (and inside of) the merge window to slide in at this point.

Oh, all these definitely are not for this window. Was just wondering as there was no response. Alright, will wait for the merge window to close.

Thanks.

Signature · 1 line
tejun
maintainer
#14 lore raw
Quoted 6 lines · Tejun Heo

Hello,

Quoted 3 lines · Jens Axboe
Quoted 1 line · Tejun Heo

Jens, ping?

Quoted 7 lines
Quoted 3 lines

I'll queue this up as soon as the current stuff has been shipped off, it's just too close (and inside of) the merge window to slide in at this point.

Oh, all these definitely are not for this window. Was just wondering as there was no response. Alright, will wait for the merge window to close.

That matches my expectations as well. They will be queued up shortly, pushed to for-next in a week or so. I'll have time to go over these early next week!

Signature · 1 line
Jens Axboe
author
#15 lore raw

Hey, Jens.

Quoted 3 lines · Jens Axboe

That matches my expectations as well. They will be queued up shortly, pushed to for-next in a week or so. I'll have time to go over these early next week!

Any progress?

Thanks.

Signature · 1 line
tejun
maintainer
#16 lore raw
Quoted 8 lines · Tejun Heo

Hey, Jens.

Quoted 3 lines · Jens Axboe

That matches my expectations as well. They will be queued up shortly, pushed to for-next in a week or so. I'll have time to go over these early next week!

Any progress?

We agreed to queue these for 3.3, right?

Signature · 1 line
Jens Axboe
author
#17 lore raw

Hello, Jens.

Quoted 1 line · Jens Axboe

We agreed to queue these for 3.3, right?

Ah... I'm getting confused about version numbers. This is 3.2-rc1, so yes this is for 3.3-rc1. Just wondering whether/when you are gonna pull this into -next. If you agree with the series, I'll be going ahead and making similar changes to blkcg.

Thank you.

Signature · 1 line
tejun