*/
static struct kmem_cache *iocontext_cachep;
-static void cfq_dtor(struct io_context *ioc)
+/**
+ * get_io_context - increment reference count to io_context
+ * @ioc: io_context to get
+ *
+ * Increment reference count to @ioc.
+ */
+void get_io_context(struct io_context *ioc)
{
- if (!hlist_empty(&ioc->cic_list)) {
- struct cfq_io_context *cic;
-
- cic = hlist_entry(ioc->cic_list.first, struct cfq_io_context,
- cic_list);
- cic->dtor(ioc);
- }
+ BUG_ON(atomic_long_read(&ioc->refcount) <= 0);
+ atomic_long_inc(&ioc->refcount);
}
+EXPORT_SYMBOL(get_io_context);
+
+/*
+ * Releasing ioc may nest into another put_io_context() leading to nested
+ * fast path release. As the ioc's can't be the same, this is okay but
+ * makes lockdep whine. Keep track of nesting and use it as subclass.
+ */
+#ifdef CONFIG_LOCKDEP
+#define ioc_release_depth(q) ((q) ? (q)->ioc_release_depth : 0)
+#define ioc_release_depth_inc(q) (q)->ioc_release_depth++
+#define ioc_release_depth_dec(q) (q)->ioc_release_depth--
+#else
+#define ioc_release_depth(q) 0
+#define ioc_release_depth_inc(q) do { } while (0)
+#define ioc_release_depth_dec(q) do { } while (0)
+#endif
/*
- * IO Context helper functions. put_io_context() returns 1 if there are no
- * more users of this io context, 0 otherwise.
+ * Slow path for ioc release in put_io_context(). Performs double-lock
+ * dancing to unlink all cic's and then frees ioc.
*/
-int put_io_context(struct io_context *ioc)
+static void ioc_release_fn(struct work_struct *work)
{
- if (ioc == NULL)
- return 1;
+ struct io_context *ioc = container_of(work, struct io_context,
+ release_work);
+ struct request_queue *last_q = NULL;
- BUG_ON(atomic_long_read(&ioc->refcount) == 0);
+ spin_lock_irq(&ioc->lock);
- if (atomic_long_dec_and_test(&ioc->refcount)) {
- rcu_read_lock();
- cfq_dtor(ioc);
- rcu_read_unlock();
+ 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;
- kmem_cache_free(iocontext_cachep, ioc);
- return 1;
+ if (this_q != last_q) {
+ /*
+ * Need to switch to @this_q. Once we release
+ * @ioc->lock, it can go away along with @cic.
+ * Hold on to it.
+ */
+ __blk_get_queue(this_q);
+
+ /*
+ * blk_put_queue() might sleep thanks to kobject
+ * idiocy. Always release both locks, put and
+ * restart.
+ */
+ if (last_q) {
+ spin_unlock(last_q->queue_lock);
+ spin_unlock_irq(&ioc->lock);
+ blk_put_queue(last_q);
+ } else {
+ spin_unlock_irq(&ioc->lock);
+ }
+
+ last_q = this_q;
+ spin_lock_irq(this_q->queue_lock);
+ spin_lock(&ioc->lock);
+ continue;
+ }
+ ioc_release_depth_inc(this_q);
+ cic->exit(cic);
+ cic->release(cic);
+ ioc_release_depth_dec(this_q);
}
- return 0;
+
+ if (last_q) {
+ spin_unlock(last_q->queue_lock);
+ spin_unlock_irq(&ioc->lock);
+ blk_put_queue(last_q);
+ } else {
+ spin_unlock_irq(&ioc->lock);
+ }
+
+ kmem_cache_free(iocontext_cachep, ioc);
}
-EXPORT_SYMBOL(put_io_context);
-static void cfq_exit(struct io_context *ioc)
+/**
+ * put_io_context - put a reference of io_context
+ * @ioc: io_context to put
+ * @locked_q: request_queue the caller is holding queue_lock of (hint)
+ *
+ * Decrement reference count of @ioc and release it if the count reaches
+ * zero. If the caller is holding queue_lock of a queue, it can indicate
+ * that with @locked_q. This is an optimization hint and the caller is
+ * allowed to pass in %NULL even when it's holding a queue_lock.
+ */
+void put_io_context(struct io_context *ioc, struct request_queue *locked_q)
{
- rcu_read_lock();
+ struct request_queue *last_q = locked_q;
+ unsigned long flags;
+
+ if (ioc == NULL)
+ return;
+
+ BUG_ON(atomic_long_read(&ioc->refcount) <= 0);
+ if (locked_q)
+ lockdep_assert_held(locked_q->queue_lock);
+
+ if (!atomic_long_dec_and_test(&ioc->refcount))
+ return;
+
+ /*
+ * Destroy @ioc. This is a bit messy because cic'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.
+ * ie. We need to do reverse-order double lock dancing.
+ *
+ * Another twist is that we are often called with one of the
+ * matching queue_locks held as indicated by @locked_q, which
+ * prevents performing double-lock dance for other queues.
+ *
+ * So, we do it in two stages. The fast path uses the queue_lock
+ * the caller is holding and, if other queues need to be accessed,
+ * uses trylock to avoid introducing locking dependency. This can
+ * handle most cases, especially if @ioc was performing IO on only
+ * single device.
+ *
+ * If trylock doesn't cut it, we defer to @ioc->release_work which
+ * can do all the double-locking dancing.
+ */
+ spin_lock_irqsave_nested(&ioc->lock, flags,
+ ioc_release_depth(locked_q));
- if (!hlist_empty(&ioc->cic_list)) {
- struct cfq_io_context *cic;
+ 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;
- cic = hlist_entry(ioc->cic_list.first, struct cfq_io_context,
- cic_list);
- cic->exit(ioc);
+ if (this_q != last_q) {
+ if (last_q && last_q != locked_q)
+ spin_unlock(last_q->queue_lock);
+ last_q = NULL;
+
+ if (!spin_trylock(this_q->queue_lock))
+ break;
+ last_q = this_q;
+ continue;
+ }
+ ioc_release_depth_inc(this_q);
+ cic->exit(cic);
+ cic->release(cic);
+ ioc_release_depth_dec(this_q);
}
- rcu_read_unlock();
+
+ if (last_q && last_q != locked_q)
+ spin_unlock(last_q->queue_lock);
+
+ spin_unlock_irqrestore(&ioc->lock, flags);
+
+ /* if no cic's left, we're done; otherwise, kick release_work */
+ if (hlist_empty(&ioc->cic_list))
+ kmem_cache_free(iocontext_cachep, ioc);
+ else
+ schedule_work(&ioc->release_work);
}
+EXPORT_SYMBOL(put_io_context);
/* Called by the exiting task */
void exit_io_context(struct task_struct *task)
{
struct io_context *ioc;
+ /* PF_EXITING prevents new io_context from being attached to @task */
+ WARN_ON_ONCE(!(current->flags & PF_EXITING));
+
task_lock(task);
ioc = task->io_context;
task->io_context = NULL;
task_unlock(task);
- if (atomic_dec_and_test(&ioc->nr_tasks))
- cfq_exit(ioc);
-
- put_io_context(ioc);
+ atomic_dec(&ioc->nr_tasks);
+ put_io_context(ioc, NULL);
}
-struct io_context *alloc_io_context(gfp_t gfp_flags, int node)
+static struct io_context *create_task_io_context(struct task_struct *task,
+ gfp_t gfp_flags, int node,
+ bool take_ref)
{
struct io_context *ioc;
- ioc = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
- if (ioc) {
- atomic_long_set(&ioc->refcount, 1);
- atomic_set(&ioc->nr_tasks, 1);
- spin_lock_init(&ioc->lock);
- ioc->ioprio_changed = 0;
- ioc->ioprio = 0;
- ioc->last_waited = 0; /* doesn't matter... */
- ioc->nr_batch_requests = 0; /* because this is 0 */
- INIT_RADIX_TREE(&ioc->radix_root, GFP_ATOMIC | __GFP_HIGH);
- INIT_HLIST_HEAD(&ioc->cic_list);
- ioc->ioc_data = NULL;
-#if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
- ioc->cgroup_changed = 0;
-#endif
+ ioc = kmem_cache_alloc_node(iocontext_cachep, gfp_flags | __GFP_ZERO,
+ node);
+ if (unlikely(!ioc))
+ return NULL;
+
+ /* initialize */
+ 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_WORK(&ioc->release_work, ioc_release_fn);
+
+ /* try to install, somebody might already have beaten us to it */
+ task_lock(task);
+
+ if (!task->io_context && !(task->flags & PF_EXITING)) {
+ task->io_context = ioc;
+ } else {
+ kmem_cache_free(iocontext_cachep, ioc);
+ ioc = task->io_context;
}
+ if (ioc && take_ref)
+ get_io_context(ioc);
+
+ task_unlock(task);
return ioc;
}
-/*
- * If the current task has no IO context then create one and initialise it.
- * Otherwise, return its existing IO context.
+/**
+ * current_io_context - get io_context of %current
+ * @gfp_flags: allocation flags, used if allocation is necessary
+ * @node: allocation node, used if allocation is necessary
*
- * This returned IO context doesn't have a specifically elevated refcount,
- * but since the current task itself holds a reference, the context can be
- * used in general code, so long as it stays within `current` context.
+ * Return io_context of %current. If it doesn't exist, it is created with
+ * @gfp_flags and @node. The returned io_context does NOT have its
+ * reference count incremented. Because io_context is exited only on task
+ * exit, %current can be sure that the returned io_context is valid and
+ * alive as long as it is executing.
*/
struct io_context *current_io_context(gfp_t gfp_flags, int node)
{
- struct task_struct *tsk = current;
- struct io_context *ret;
-
- ret = tsk->io_context;
- if (likely(ret))
- return ret;
-
- ret = alloc_io_context(gfp_flags, node);
- if (ret) {
- /* make sure set_task_ioprio() sees the settings above */
- smp_wmb();
- tsk->io_context = ret;
+ might_sleep_if(gfp_flags & __GFP_WAIT);
+
+ if (current->io_context)
+ return current->io_context;
+
+ return create_task_io_context(current, gfp_flags, node, false);
+}
+EXPORT_SYMBOL(current_io_context);
+
+/**
+ * get_task_io_context - get io_context of a task
+ * @task: task of interest
+ * @gfp_flags: allocation flags, used if allocation is necessary
+ * @node: allocation node, used if allocation is necessary
+ *
+ * Return io_context of @task. If it doesn't exist, it is created with
+ * @gfp_flags and @node. The returned io_context has its reference count
+ * incremented.
+ *
+ * This function always goes through task_lock() and it's better to use
+ * current_io_context() + get_io_context() for %current.
+ */
+struct io_context *get_task_io_context(struct task_struct *task,
+ gfp_t gfp_flags, int node)
+{
+ struct io_context *ioc;
+
+ might_sleep_if(gfp_flags & __GFP_WAIT);
+
+ task_lock(task);
+ ioc = task->io_context;
+ if (likely(ioc)) {
+ get_io_context(ioc);
+ task_unlock(task);
+ return ioc;
}
+ task_unlock(task);
- return ret;
+ return create_task_io_context(task, gfp_flags, node, true);
}
+EXPORT_SYMBOL(get_task_io_context);
-/*
- * If the current task has no IO context then create one and initialise it.
- * If it does have a context, take a ref on it.
+void ioc_set_changed(struct io_context *ioc, int which)
+{
+ struct cfq_io_context *cic;
+ struct hlist_node *n;
+
+ hlist_for_each_entry(cic, n, &ioc->cic_list, cic_list)
+ set_bit(which, &cic->changed);
+}
+
+/**
+ * ioc_ioprio_changed - notify ioprio change
+ * @ioc: io_context of interest
+ * @ioprio: new ioprio
*
- * This is always called in the context of the task which submitted the I/O.
+ * @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
+ * request issue path.
*/
-struct io_context *get_io_context(gfp_t gfp_flags, int node)
+void ioc_ioprio_changed(struct io_context *ioc, int ioprio)
{
- struct io_context *ioc = NULL;
+ unsigned long flags;
- /*
- * Check for unlikely race with exiting task. ioc ref count is
- * zero when ioc is being detached.
- */
- do {
- ioc = current_io_context(gfp_flags, node);
- if (unlikely(!ioc))
- break;
- } while (!atomic_long_inc_not_zero(&ioc->refcount));
+ spin_lock_irqsave(&ioc->lock, flags);
+ ioc->ioprio = ioprio;
+ ioc_set_changed(ioc, CIC_IOPRIO_CHANGED);
+ spin_unlock_irqrestore(&ioc->lock, flags);
+}
- return ioc;
+/**
+ * ioc_cgroup_changed - notify cgroup change
+ * @ioc: io_context of interest
+ *
+ * @ioc's cgroup has changed. Set %CIC_CGROUP_CHANGED for all cic's.
+ * iosched is responsible for checking the bit and applying it on request
+ * issue path.
+ */
+void ioc_cgroup_changed(struct io_context *ioc)
+{
+ unsigned long flags;
+
+ spin_lock_irqsave(&ioc->lock, flags);
+ ioc_set_changed(ioc, CIC_CGROUP_CHANGED);
+ spin_unlock_irqrestore(&ioc->lock, flags);
}
-EXPORT_SYMBOL(get_io_context);
static int __init blk_ioc_init(void)
{