Merge branch 'cfq-ioc-share' of git://git.kernel.dk/linux-2.6-block

* 'cfq-ioc-share' of git://git.kernel.dk/linux-2.6-block:
  cfq-iosched: kill some big inlines
  cfq-iosched: relax IOPRIO_CLASS_IDLE restrictions
  kernel: add CLONE_IO to specifically request sharing of IO contexts
  io_context sharing - anticipatory changes
  block: cfq: make the io contect sharing lockless
  io_context sharing - cfq changes
  io context sharing: preliminary support
  ioprio: move io priority from task_struct to io_context

block/as-iosched.c

@@ -170,9 +170,11 @@ static void free_as_io_context(struct as_io_context *aic)
 
 static void as_trim(struct io_context *ioc)
 {
+	spin_lock(&ioc->lock);
 	if (ioc->aic)
 		free_as_io_context(ioc->aic);
 	ioc->aic = NULL;
+	spin_unlock(&ioc->lock);
 }
 
 /* Called when the task exits */
@@ -462,7 +464,9 @@ static void as_antic_timeout(unsigned long data)
 	spin_lock_irqsave(q->queue_lock, flags);
 	if (ad->antic_status == ANTIC_WAIT_REQ
 			|| ad->antic_status == ANTIC_WAIT_NEXT) {
-		struct as_io_context *aic = ad->io_context->aic;
+		struct as_io_context *aic;
+		spin_lock(&ad->io_context->lock);
+		aic = ad->io_context->aic;
 
 		ad->antic_status = ANTIC_FINISHED;
 		kblockd_schedule_work(&ad->antic_work);
@@ -475,6 +479,7 @@ static void as_antic_timeout(unsigned long data)
 			/* process not "saved" by a cooperating request */
 			ad->exit_no_coop = (7*ad->exit_no_coop + 256)/8;
 		}
+		spin_unlock(&ad->io_context->lock);
 	}
 	spin_unlock_irqrestore(q->queue_lock, flags);
 }
@@ -635,9 +640,11 @@ static int as_can_break_anticipation(struct as_data *ad, struct request *rq)
 
 	ioc = ad->io_context;
 	BUG_ON(!ioc);
+	spin_lock(&ioc->lock);
 
 	if (rq && ioc == RQ_IOC(rq)) {
 		/* request from same process */
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
 
@@ -646,20 +653,25 @@ static int as_can_break_anticipation(struct as_data *ad, struct request *rq)
 		 * In this situation status should really be FINISHED,
 		 * however the timer hasn't had the chance to run yet.
 		 */
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
 
 	aic = ioc->aic;
-	if (!aic)
+	if (!aic) {
+		spin_unlock(&ioc->lock);
 		return 0;
+	}
 
 	if (atomic_read(&aic->nr_queued) > 0) {
 		/* process has more requests queued */
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
 
 	if (atomic_read(&aic->nr_dispatched) > 0) {
 		/* process has more requests dispatched */
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
 
@@ -680,6 +692,7 @@ static int as_can_break_anticipation(struct as_data *ad, struct request *rq)
 		}
 
 		as_update_iohist(ad, aic, rq);
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
 
@@ -688,20 +701,27 @@ static int as_can_break_anticipation(struct as_data *ad, struct request *rq)
 		if (aic->ttime_samples == 0)
 			ad->exit_prob = (7*ad->exit_prob + 256)/8;
 
-		if (ad->exit_no_coop > 128)
+		if (ad->exit_no_coop > 128) {
+			spin_unlock(&ioc->lock);
 			return 1;
 		}
+	}
 
 	if (aic->ttime_samples == 0) {
-		if (ad->new_ttime_mean > ad->antic_expire)
+		if (ad->new_ttime_mean > ad->antic_expire) {
+			spin_unlock(&ioc->lock);
 			return 1;
-		if (ad->exit_prob * ad->exit_no_coop > 128*256)
+		}
+		if (ad->exit_prob * ad->exit_no_coop > 128*256) {
+			spin_unlock(&ioc->lock);
 			return 1;
+		}
 	} else if (aic->ttime_mean > ad->antic_expire) {
 		/* the process thinks too much between requests */
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
-
+	spin_unlock(&ioc->lock);
 	return 0;
 }
 
@@ -1255,7 +1275,9 @@ static void as_merged_requests(struct request_queue *q, struct request *req,
 			 * Don't copy here but swap, because when anext is
 			 * removed below, it must contain the unused context
 			 */
+			double_spin_lock(&rioc->lock, &nioc->lock, rioc < nioc);
 			swap_io_context(&rioc, &nioc);
+			double_spin_unlock(&rioc->lock, &nioc->lock, rioc < nioc);
 		}
 	}
 

block/ll_rw_blk.c

@@ -3853,55 +3853,100 @@ int __init blk_dev_init(void)
 	return 0;
 }
 
+static void cfq_dtor(struct io_context *ioc)
+{
+	struct cfq_io_context *cic[1];
+	int r;
+
+	/*
+	 * We don't have a specific key to lookup with, so use the gang
+	 * lookup to just retrieve the first item stored. The cfq exit
+	 * function will iterate the full tree, so any member will do.
+	 */
+	r = radix_tree_gang_lookup(&ioc->radix_root, (void **) cic, 0, 1);
+	if (r > 0)
+		cic[0]->dtor(ioc);
+}
+
 /*
- * IO Context helper functions
+ * IO Context helper functions. put_io_context() returns 1 if there are no
+ * more users of this io context, 0 otherwise.
  */
-void put_io_context(struct io_context *ioc)
+int put_io_context(struct io_context *ioc)
 {
 	if (ioc == NULL)
-		return;
+		return 1;
 
 	BUG_ON(atomic_read(&ioc->refcount) == 0);
 
 	if (atomic_dec_and_test(&ioc->refcount)) {
-		struct cfq_io_context *cic;
-
 		rcu_read_lock();
 		if (ioc->aic && ioc->aic->dtor)
 			ioc->aic->dtor(ioc->aic);
-		if (ioc->cic_root.rb_node != NULL) {
-			struct rb_node *n = rb_first(&ioc->cic_root);
-
-			cic = rb_entry(n, struct cfq_io_context, rb_node);
-			cic->dtor(ioc);
-		}
 		rcu_read_unlock();
+		cfq_dtor(ioc);
 
 		kmem_cache_free(iocontext_cachep, ioc);
+		return 1;
 	}
+	return 0;
 }
 EXPORT_SYMBOL(put_io_context);
 
+static void cfq_exit(struct io_context *ioc)
+{
+	struct cfq_io_context *cic[1];
+	int r;
+
+	rcu_read_lock();
+	/*
+	 * See comment for cfq_dtor()
+	 */
+	r = radix_tree_gang_lookup(&ioc->radix_root, (void **) cic, 0, 1);
+	rcu_read_unlock();
+
+	if (r > 0)
+		cic[0]->exit(ioc);
+}
+
 /* Called by the exitting task */
 void exit_io_context(void)
 {
 	struct io_context *ioc;
-	struct cfq_io_context *cic;
 
 	task_lock(current);
 	ioc = current->io_context;
 	current->io_context = NULL;
 	task_unlock(current);
 
-	ioc->task = NULL;
+	if (atomic_dec_and_test(&ioc->nr_tasks)) {
 		if (ioc->aic && ioc->aic->exit)
 			ioc->aic->exit(ioc->aic);
-	if (ioc->cic_root.rb_node != NULL) {
-		cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
-		cic->exit(ioc);
-	}
+		cfq_exit(ioc);
 
 		put_io_context(ioc);
+	}
+}
+
+struct io_context *alloc_io_context(gfp_t gfp_flags, int node)
+{
+	struct io_context *ret;
+
+	ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
+	if (ret) {
+		atomic_set(&ret->refcount, 1);
+		atomic_set(&ret->nr_tasks, 1);
+		spin_lock_init(&ret->lock);
+		ret->ioprio_changed = 0;
+		ret->ioprio = 0;
+		ret->last_waited = jiffies; /* doesn't matter... */
+		ret->nr_batch_requests = 0; /* because this is 0 */
+		ret->aic = NULL;
+		INIT_RADIX_TREE(&ret->radix_root, GFP_ATOMIC | __GFP_HIGH);
+		ret->ioc_data = NULL;
+	}
+
+	return ret;
 }
 
 /*
@@ -3921,16 +3966,8 @@ static struct io_context *current_io_context(gfp_t gfp_flags, int node)
 	if (likely(ret))
 		return ret;
 
-	ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
+	ret = alloc_io_context(gfp_flags, node);
 	if (ret) {
-		atomic_set(&ret->refcount, 1);
-		ret->task = current;
-		ret->ioprio_changed = 0;
-		ret->last_waited = jiffies; /* doesn't matter... */
-		ret->nr_batch_requests = 0; /* because this is 0 */
-		ret->aic = NULL;
-		ret->cic_root.rb_node = NULL;
-		ret->ioc_data = NULL;
 		/* make sure set_task_ioprio() sees the settings above */
 		smp_wmb();
 		tsk->io_context = ret;
@@ -3947,10 +3984,18 @@ static struct io_context *current_io_context(gfp_t gfp_flags, int node)
  */
 struct io_context *get_io_context(gfp_t gfp_flags, int node)
 {
-	struct io_context *ret;
+	struct io_context *ret = NULL;
+
+	/*
+	 * Check for unlikely race with exiting task. ioc ref count is
+	 * zero when ioc is being detached.
+	 */
+	do {
 		ret = current_io_context(gfp_flags, node);
-	if (likely(ret))
-		atomic_inc(&ret->refcount);
+		if (unlikely(!ret))
+			break;
+	} while (!atomic_inc_not_zero(&ret->refcount));
+
 	return ret;
 }
 EXPORT_SYMBOL(get_io_context);

fs/ioprio.c

@@ -41,18 +41,28 @@ static int set_task_ioprio(struct task_struct *task, int ioprio)
 		return err;
 
 	task_lock(task);
-
-	task->ioprio = ioprio;
-
+	do {
 		ioc = task->io_context;
 		/* see wmb() in current_io_context() */
 		smp_read_barrier_depends();
-
 		if (ioc)
+			break;
+
+		ioc = alloc_io_context(GFP_ATOMIC, -1);
+		if (!ioc) {
+			err = -ENOMEM;
+			break;
+		}
+		task->io_context = ioc;
+	} while (1);
+
+	if (!err) {
+		ioc->ioprio = ioprio;
 		ioc->ioprio_changed = 1;
+	}
 
 	task_unlock(task);
-	return 0;
+	return err;
 }
 
 asmlinkage long sys_ioprio_set(int which, int who, int ioprio)
@@ -75,8 +85,6 @@ asmlinkage long sys_ioprio_set(int which, int who, int ioprio)
 
 			break;
 		case IOPRIO_CLASS_IDLE:
-			if (!capable(CAP_SYS_ADMIN))
-				return -EPERM;
 			break;
 		case IOPRIO_CLASS_NONE:
 			if (data)
@@ -148,7 +156,9 @@ static int get_task_ioprio(struct task_struct *p)
 	ret = security_task_getioprio(p);
 	if (ret)
 		goto out;
-	ret = p->ioprio;
+	ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
+	if (p->io_context)
+		ret = p->io_context->ioprio;
 out:
 	return ret;
 }

include/linux/blkdev.h

@@ -34,83 +34,10 @@ struct sg_io_hdr;
 #define BLKDEV_MIN_RQ	4
 #define BLKDEV_MAX_RQ	128	/* Default maximum */
 
-/*
- * This is the per-process anticipatory I/O scheduler state.
- */
-struct as_io_context {
-	spinlock_t lock;
-
-	void (*dtor)(struct as_io_context *aic); /* destructor */
-	void (*exit)(struct as_io_context *aic); /* called on task exit */
-
-	unsigned long state;
-	atomic_t nr_queued; /* queued reads & sync writes */
-	atomic_t nr_dispatched; /* number of requests gone to the drivers */
-
-	/* IO History tracking */
-	/* Thinktime */
-	unsigned long last_end_request;
-	unsigned long ttime_total;
-	unsigned long ttime_samples;
-	unsigned long ttime_mean;
-	/* Layout pattern */
-	unsigned int seek_samples;
-	sector_t last_request_pos;
-	u64 seek_total;
-	sector_t seek_mean;
-};
-
-struct cfq_queue;
-struct cfq_io_context {
-	struct rb_node rb_node;
-	void *key;
-
-	struct cfq_queue *cfqq[2];
-
-	struct io_context *ioc;
-
-	unsigned long last_end_request;
-	sector_t last_request_pos;
-
-	unsigned long ttime_total;
-	unsigned long ttime_samples;
-	unsigned long ttime_mean;
-
-	unsigned int seek_samples;
-	u64 seek_total;
-	sector_t seek_mean;
-
-	struct list_head queue_list;
-
-	void (*dtor)(struct io_context *); /* destructor */
-	void (*exit)(struct io_context *); /* called on task exit */
-};
-
-/*
- * This is the per-process I/O subsystem state.  It is refcounted and
- * kmalloc'ed. Currently all fields are modified in process io context
- * (apart from the atomic refcount), so require no locking.
- */
-struct io_context {
-	atomic_t refcount;
-	struct task_struct *task;
-
-	unsigned int ioprio_changed;
-
-	/*
-	 * For request batching
-	 */
-	unsigned long last_waited; /* Time last woken after wait for request */
-	int nr_batch_requests;     /* Number of requests left in the batch */
-
-	struct as_io_context *aic;
-	struct rb_root cic_root;
-	void *ioc_data;
-};
-
-void put_io_context(struct io_context *ioc);
+int put_io_context(struct io_context *ioc);
 void exit_io_context(void);
 struct io_context *get_io_context(gfp_t gfp_flags, int node);
+struct io_context *alloc_io_context(gfp_t gfp_flags, int node);
 void copy_io_context(struct io_context **pdst, struct io_context **psrc);
 void swap_io_context(struct io_context **ioc1, struct io_context **ioc2);
 
@@ -894,6 +821,12 @@ static inline void exit_io_context(void)
 {
 }
 
+static inline int put_io_context(struct io_context *ioc)
+{
+	return 1;
+}
+
+
 #endif /* CONFIG_BLOCK */
 
 #endif

include/linux/init_task.h

@@ -137,7 +137,6 @@ extern struct group_info init_groups;
 		.time_slice	= HZ, 					\
 		.nr_cpus_allowed = NR_CPUS,				\
 	},								\
-	.ioprio		= 0,						\
 	.tasks		= LIST_HEAD_INIT(tsk.tasks),			\
 	.ptrace_children= LIST_HEAD_INIT(tsk.ptrace_children),		\
 	.ptrace_list	= LIST_HEAD_INIT(tsk.ptrace_list),		\

include/linux/iocontext.h

+#ifndef IOCONTEXT_H
+#define IOCONTEXT_H
+
+#include <linux/radix-tree.h>
+
+/*
+ * This is the per-process anticipatory I/O scheduler state.
+ */
+struct as_io_context {
+	spinlock_t lock;
+
+	void (*dtor)(struct as_io_context *aic); /* destructor */
+	void (*exit)(struct as_io_context *aic); /* called on task exit */
+
+	unsigned long state;
+	atomic_t nr_queued; /* queued reads & sync writes */
+	atomic_t nr_dispatched; /* number of requests gone to the drivers */
+
+	/* IO History tracking */
+	/* Thinktime */
+	unsigned long last_end_request;
+	unsigned long ttime_total;
+	unsigned long ttime_samples;
+	unsigned long ttime_mean;
+	/* Layout pattern */
+	unsigned int seek_samples;
+	sector_t last_request_pos;
+	u64 seek_total;
+	sector_t seek_mean;
+};
+
+struct cfq_queue;
+struct cfq_io_context {
+	void *key;
+	unsigned long dead_key;
+
+	struct cfq_queue *cfqq[2];
+
+	struct io_context *ioc;
+
+	unsigned long last_end_request;
+	sector_t last_request_pos;
+
+	unsigned long ttime_total;
+	unsigned long ttime_samples;
+	unsigned long ttime_mean;
+
+	unsigned int seek_samples;
+	u64 seek_total;
+	sector_t seek_mean;
+
+	struct list_head queue_list;
+
+	void (*dtor)(struct io_context *); /* destructor */
+	void (*exit)(struct io_context *); /* called on task exit */
+};
+
+/*
+ * I/O subsystem state of the associated processes.  It is refcounted
+ * and kmalloc'ed. These could be shared between processes.
+ */
+struct io_context {
+	atomic_t refcount;
+	atomic_t nr_tasks;
+
+	/* all the fields below are protected by this lock */
+	spinlock_t lock;
+
+	unsigned short ioprio;
+	unsigned short ioprio_changed;
+
+	/*
+	 * For request batching
+	 */
+	unsigned long last_waited; /* Time last woken after wait for request */
+	int nr_batch_requests;     /* Number of requests left in the batch */
+
+	struct as_io_context *aic;
+	struct radix_tree_root radix_root;
+	void *ioc_data;
+};
+
+static inline struct io_context *ioc_task_link(struct io_context *ioc)
+{
+	/*
+	 * if ref count is zero, don't allow sharing (ioc is going away, it's
+	 * a race).
+	 */
+	if (ioc && atomic_inc_not_zero(&ioc->refcount))
+		return ioc;
+
+	return NULL;
+}
+
+#endif

include/linux/ioprio.h

@@ -2,6 +2,7 @@
 #define IOPRIO_H
 
 #include <linux/sched.h>
+#include <linux/iocontext.h>
 
 /*
  * Gives us 8 prio classes with 13-bits of data for each class
@@ -45,18 +46,18 @@ enum {
  * the cpu scheduler nice value to an io priority
  */
 #define IOPRIO_NORM	(4)
-static inline int task_ioprio(struct task_struct *task)
+static inline int task_ioprio(struct io_context *ioc)
 {
-	if (ioprio_valid(task->ioprio))
-		return IOPRIO_PRIO_DATA(task->ioprio);
+	if (ioprio_valid(ioc->ioprio))
+		return IOPRIO_PRIO_DATA(ioc->ioprio);
 
 	return IOPRIO_NORM;
 }
 
-static inline int task_ioprio_class(struct task_struct *task)
+static inline int task_ioprio_class(struct io_context *ioc)
 {
-	if (ioprio_valid(task->ioprio))
-		return IOPRIO_PRIO_CLASS(task->ioprio);
+	if (ioprio_valid(ioc->ioprio))
+		return IOPRIO_PRIO_CLASS(ioc->ioprio);
 
 	return IOPRIO_CLASS_BE;
 }

include/linux/sched.h

@@ -27,6 +27,7 @@
 #define CLONE_NEWUSER		0x10000000	/* New user namespace */
 #define CLONE_NEWPID		0x20000000	/* New pid namespace */
 #define CLONE_NEWNET		0x40000000	/* New network namespace */
+#define CLONE_IO		0x80000000	/* Clone io context */
 
 /*
  * Scheduling policies
@@ -975,7 +976,6 @@ struct task_struct {
 	struct hlist_head preempt_notifiers;
 #endif
 
-	unsigned short ioprio;
 	/*
 	 * fpu_counter contains the number of consecutive context switches
 	 * that the FPU is used. If this is over a threshold, the lazy fpu

kernel/fork.c

@@ -51,6 +51,7 @@
 #include <linux/random.h>
 #include <linux/tty.h>
 #include <linux/proc_fs.h>
+#include <linux/blkdev.h>
 
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
@@ -791,6 +792,31 @@ static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 	return error;
 }
 
+static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
+{
+#ifdef CONFIG_BLOCK
+	struct io_context *ioc = current->io_context;
+
+	if (!ioc)
+		return 0;
+	/*
+	 * Share io context with parent, if CLONE_IO is set
+	 */
+	if (clone_flags & CLONE_IO) {
+		tsk->io_context = ioc_task_link(ioc);
+		if (unlikely(!tsk->io_context))
+			return -ENOMEM;
+	} else if (ioprio_valid(ioc->ioprio)) {
+		tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
+		if (unlikely(!tsk->io_context))
+			return -ENOMEM;
+
+		tsk->io_context->ioprio = ioc->ioprio;
+	}
+#endif
+	return 0;
+}
+
 /*
  *	Helper to unshare the files of the current task.
  *	We don't want to expose copy_files internals to
@@ -1156,15 +1182,17 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 		goto bad_fork_cleanup_mm;
 	if ((retval = copy_namespaces(clone_flags, p)))
 		goto bad_fork_cleanup_keys;
+	if ((retval = copy_io(clone_flags, p)))
+		goto bad_fork_cleanup_namespaces;
 	retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
 	if (retval)
-		goto bad_fork_cleanup_namespaces;
+		goto bad_fork_cleanup_io;
 
 	if (pid != &init_struct_pid) {
 		retval = -ENOMEM;
 		pid = alloc_pid(task_active_pid_ns(p));
 		if (!pid)
-			goto bad_fork_cleanup_namespaces;
+			goto bad_fork_cleanup_io;
 
 		if (clone_flags & CLONE_NEWPID) {
 			retval = pid_ns_prepare_proc(task_active_pid_ns(p));
@@ -1234,9 +1262,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	/* Need tasklist lock for parent etc handling! */
 	write_lock_irq(&tasklist_lock);
 
-	/* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
-	p->ioprio = current->ioprio;
-
 	/*
 	 * The task hasn't been attached yet, so its cpus_allowed mask will
 	 * not be changed, nor will its assigned CPU.
@@ -1328,6 +1353,8 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 bad_fork_free_pid:
 	if (pid != &init_struct_pid)
 		free_pid(pid);
+bad_fork_cleanup_io:
+	put_io_context(p->io_context);
 bad_fork_cleanup_namespaces:
 	exit_task_namespaces(p);
 bad_fork_cleanup_keys: