cfq-iosched.c

Linux块设备驱动源码

		if (cic == NULL)
			goto err;

		/*
		 * manually increment generic io_context usage count, it
		 * cannot go away since we are already holding one ref to it
		 */
		ioc->cic = cic;
		ioc->set_ioprio = cfq_ioc_set_ioprio;
		cic->ioc = ioc;
		cic->key = cfqd;
		atomic_inc(&cfqd->ref);
	} else {
		struct cfq_io_context *__cic;

		/*
		 * the first cic on the list is actually the head itself
		 */
		if (cic->key == cfqd)
			goto out;

		/*
		 * cic exists, check if we already are there. linear search
		 * should be ok here, the list will usually not be more than
		 * 1 or a few entries long
		 */
		list_for_each_entry(__cic, &cic->list, list) {
			/*
			 * this process is already holding a reference to
			 * this queue, so no need to get one more
			 */
			if (__cic->key == cfqd) {
				cic = __cic;
				goto out;
			}
		}

		/*
		 * nope, process doesn't have a cic assoicated with this
		 * cfqq yet. get a new one and add to list
		 */
		__cic = cfq_alloc_io_context(cfqd, gfp_mask);
		if (__cic == NULL)
			goto err;

		__cic->ioc = ioc;
		__cic->key = cfqd;
		atomic_inc(&cfqd->ref);
		list_add(&__cic->list, &cic->list);
		cic = __cic;
	}

out:
	return cic;
err:
	put_io_context(ioc);
	return NULL;
}

static void
cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
{
	unsigned long elapsed, ttime;

	/*
	 * if this context already has stuff queued, thinktime is from
	 * last queue not last end
	 */
#if 0
	if (time_after(cic->last_end_request, cic->last_queue))
		elapsed = jiffies - cic->last_end_request;
	else
		elapsed = jiffies - cic->last_queue;
#else
		elapsed = jiffies - cic->last_end_request;
#endif

	ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle);

	cic->ttime_samples = (7*cic->ttime_samples + 256) / 8;
	cic->ttime_total = (7*cic->ttime_total + 256*ttime) / 8;
	cic->ttime_mean = (cic->ttime_total + 128) / cic->ttime_samples;
}

#define sample_valid(samples)	((samples) > 80)

/*
 * Disable idle window if the process thinks too long or seeks so much that
 * it doesn't matter
 */
static void
cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
		       struct cfq_io_context *cic)
{
	int enable_idle = cfq_cfqq_idle_window(cfqq);

	if (!cic->ioc->task || !cfqd->cfq_slice_idle)
		enable_idle = 0;
	else if (sample_valid(cic->ttime_samples)) {
		if (cic->ttime_mean > cfqd->cfq_slice_idle)
			enable_idle = 0;
		else
			enable_idle = 1;
	}

	if (enable_idle)
		cfq_mark_cfqq_idle_window(cfqq);
	else
		cfq_clear_cfqq_idle_window(cfqq);
}


/*
 * Check if new_cfqq should preempt the currently active queue. Return 0 for
 * no or if we aren't sure, a 1 will cause a preempt.
 */
static int
cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
		   struct cfq_rq *crq)
{
	struct cfq_queue *cfqq = cfqd->active_queue;

	if (cfq_class_idle(new_cfqq))
		return 0;

	if (!cfqq)
		return 1;

	if (cfq_class_idle(cfqq))
		return 1;
	if (!cfq_cfqq_wait_request(new_cfqq))
		return 0;
	/*
	 * if it doesn't have slice left, forget it
	 */
	if (new_cfqq->slice_left < cfqd->cfq_slice_idle)
		return 0;
	if (cfq_crq_is_sync(crq) && !cfq_cfqq_sync(cfqq))
		return 1;

	return 0;
}

/*
 * cfqq preempts the active queue. if we allowed preempt with no slice left,
 * let it have half of its nominal slice.
 */
static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	struct cfq_queue *__cfqq, *next;

	list_for_each_entry_safe(__cfqq, next, &cfqd->cur_rr, cfq_list)
		cfq_resort_rr_list(__cfqq, 1);

	if (!cfqq->slice_left)
		cfqq->slice_left = cfq_prio_to_slice(cfqd, cfqq) / 2;

	cfqq->slice_end = cfqq->slice_left + jiffies;
	__cfq_slice_expired(cfqd, cfqq, 1);
	__cfq_set_active_queue(cfqd, cfqq);
}

/*
 * should really be a ll_rw_blk.c helper
 */
static void cfq_start_queueing(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
	request_queue_t *q = cfqd->queue;

	if (!blk_queue_plugged(q))
		q->request_fn(q);
	else
		__generic_unplug_device(q);
}

/*
 * Called when a new fs request (crq) is added (to cfqq). Check if there's
 * something we should do about it
 */
static void
cfq_crq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
		 struct cfq_rq *crq)
{
	struct cfq_io_context *cic;

	cfqq->next_crq = cfq_choose_req(cfqd, cfqq->next_crq, crq);

	/*
	 * we never wait for an async request and we don't allow preemption
	 * of an async request. so just return early
	 */
	if (!cfq_crq_is_sync(crq))
		return;

	cic = crq->io_context;

	cfq_update_io_thinktime(cfqd, cic);
	cfq_update_idle_window(cfqd, cfqq, cic);

	cic->last_queue = jiffies;

	if (cfqq == cfqd->active_queue) {
		/*
		 * if we are waiting for a request for this queue, let it rip
		 * immediately and flag that we must not expire this queue
		 * just now
		 */
		if (cfq_cfqq_wait_request(cfqq)) {
			cfq_mark_cfqq_must_dispatch(cfqq);
			del_timer(&cfqd->idle_slice_timer);
			cfq_start_queueing(cfqd, cfqq);
		}
	} else if (cfq_should_preempt(cfqd, cfqq, crq)) {
		/*
		 * not the active queue - expire current slice if it is
		 * idle and has expired it's mean thinktime or this new queue
		 * has some old slice time left and is of higher priority
		 */
		cfq_preempt_queue(cfqd, cfqq);
		cfq_mark_cfqq_must_dispatch(cfqq);
		cfq_start_queueing(cfqd, cfqq);
	}
}

static void cfq_enqueue(struct cfq_data *cfqd, struct request *rq)
{
	struct cfq_rq *crq = RQ_DATA(rq);
	struct cfq_queue *cfqq = crq->cfq_queue;

	cfq_init_prio_data(cfqq);

	cfq_add_crq_rb(crq);

	list_add_tail(&rq->queuelist, &cfqq->fifo);

	if (rq_mergeable(rq)) {
		cfq_add_crq_hash(cfqd, crq);

		if (!cfqd->queue->last_merge)
			cfqd->queue->last_merge = rq;
	}

	cfq_crq_enqueued(cfqd, cfqq, crq);
}

static void
cfq_insert_request(request_queue_t *q, struct request *rq, int where)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;

	switch (where) {
		case ELEVATOR_INSERT_BACK:
			while (cfq_dispatch_requests(q, INT_MAX, 1))
				;
			list_add_tail(&rq->queuelist, &q->queue_head);
			/*
			 * If we were idling with pending requests on
			 * inactive cfqqs, force dispatching will
			 * remove the idle timer and the queue won't
			 * be kicked by __make_request() afterward.
			 * Kick it here.
			 */
			cfq_schedule_dispatch(cfqd);
			break;
		case ELEVATOR_INSERT_FRONT:
			list_add(&rq->queuelist, &q->queue_head);
			break;
		case ELEVATOR_INSERT_SORT:
			BUG_ON(!blk_fs_request(rq));
			cfq_enqueue(cfqd, rq);
			break;
		default:
			printk("%s: bad insert point %d\n", __FUNCTION__,where);
			return;
	}
}

static void cfq_completed_request(request_queue_t *q, struct request *rq)
{
	struct cfq_rq *crq = RQ_DATA(rq);
	struct cfq_queue *cfqq;

	if (unlikely(!blk_fs_request(rq)))
		return;

	cfqq = crq->cfq_queue;

	if (cfq_crq_in_flight(crq)) {
		const int sync = cfq_crq_is_sync(crq);

		WARN_ON(!cfqq->on_dispatch[sync]);
		cfqq->on_dispatch[sync]--;
	}

	cfq_account_completion(cfqq, crq);
}

static struct request *
cfq_former_request(request_queue_t *q, struct request *rq)
{
	struct cfq_rq *crq = RQ_DATA(rq);
	struct rb_node *rbprev = rb_prev(&crq->rb_node);

	if (rbprev)
		return rb_entry_crq(rbprev)->request;

	return NULL;
}

static struct request *
cfq_latter_request(request_queue_t *q, struct request *rq)
{
	struct cfq_rq *crq = RQ_DATA(rq);
	struct rb_node *rbnext = rb_next(&crq->rb_node);

	if (rbnext)
		return rb_entry_crq(rbnext)->request;

	return NULL;
}

/*
 * we temporarily boost lower priority queues if they are holding fs exclusive
 * resources. they are boosted to normal prio (CLASS_BE/4)
 */
static void cfq_prio_boost(struct cfq_queue *cfqq)
{
	const int ioprio_class = cfqq->ioprio_class;
	const int ioprio = cfqq->ioprio;

	if (has_fs_excl()) {
		/*
		 * boost idle prio on transactions that would lock out other
		 * users of the filesystem
		 */
		if (cfq_class_idle(cfqq))
			cfqq->ioprio_class = IOPRIO_CLASS_BE;
		if (cfqq->ioprio > IOPRIO_NORM)
			cfqq->ioprio = IOPRIO_NORM;
	} else {
		/*
		 * check if we need to unboost the queue
		 */
		if (cfqq->ioprio_class != cfqq->org_ioprio_class)
			cfqq->ioprio_class = cfqq->org_ioprio_class;
		if (cfqq->ioprio != cfqq->org_ioprio)
			cfqq->ioprio = cfqq->org_ioprio;
	}

	/*
	 * refile between round-robin lists if we moved the priority class
	 */
	if ((ioprio_class != cfqq->ioprio_class || ioprio != cfqq->ioprio) &&
	    cfq_cfqq_on_rr(cfqq))
		cfq_resort_rr_list(cfqq, 0);
}

static inline pid_t cfq_queue_pid(struct task_struct *task, int rw)
{
	if (rw == READ || process_sync(task))
		return task->pid;

	return CFQ_KEY_ASYNC;
}

static inline int
__cfq_may_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq,
		struct task_struct *task, int rw)
{
#if 1
	if ((cfq_cfqq_wait_request(cfqq) || cfq_cfqq_must_alloc(cfqq)) &&
	    !cfq_cfqq_must_alloc_slice(cfqq)) {
		cfq_mark_cfqq_must_alloc_slice(cfqq);
		return ELV_MQUEUE_MUST;
	}

	return ELV_MQUEUE_MAY;
#else
	if (!cfqq || task->flags & PF_MEMALLOC)
		return ELV_MQUEUE_MAY;
	if (!cfqq->allocated[rw] || cfq_cfqq_must_alloc(cfqq)) {
		if (cfq_cfqq_wait_request(cfqq))
			return ELV_MQUEUE_MUST;

		/*
		 * only allow 1 ELV_MQUEUE_MUST per slice, otherwise we
		 * can quickly flood the queue with writes from a single task
		 */
		if (rw == READ || !cfq_cfqq_must_alloc_slice(cfqq)) {
			cfq_mark_cfqq_must_alloc_slice(cfqq);
			return ELV_MQUEUE_MUST;
		}

		return ELV_MQUEUE_MAY;
	}
	if (cfq_class_idle(cfqq))
		return ELV_MQUEUE_NO;
	if (cfqq->allocated[rw] >= cfqd->max_queued) {
		struct io_context *ioc = get_io_context(GFP_ATOMIC);
		int ret = ELV_MQUEUE_NO;

		if (ioc && ioc->nr_batch_requests)
			ret = ELV_MQUEUE_MAY;

		put_io_context(ioc);
		return ret;
	}

	return ELV_MQUEUE_MAY;
#endif
}

static int cfq_may_queue(request_queue_t *q, int rw, struct bio *bio)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct task_struct *tsk = current;
	struct cfq_queue *cfqq;

	/*
	 * don't force setup of a queue from here, as a call to may_queue
	 * does not necessarily imply that a request actually will be queued.
	 * so just lookup a possibly existing queue, or return 'may queue'
	 * if that fails
	 */
	cfqq = cfq_find_cfq_hash(cfqd, cfq_queue_pid(tsk, rw), tsk->ioprio);
	if (cfqq) {
		cfq_init_prio_data(cfqq);
		cfq_prio_boost(cfqq);

		return __cfq_may_queue(cfqd, cfqq, tsk, rw);
	}

	return ELV_MQUEUE_MAY;
}

static void cfq_check_waiters(request_queue_t *q, struct cfq_queue *cfqq)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct request_list *rl = &q->rq;

	if (cfqq->allocated[READ] <= cfqd->max_queued || cfqd->rq_starved) {
		smp_mb();
		if (waitqueue_active(&rl->wait[READ]))
			wake_up(&rl->wait[READ]);
	}

	if (cfqq->allocated[WRITE] <= cfqd->max_queued || cfqd->rq_starved) {
		smp_mb();
		if (waitqueue_active(&rl->wait[WRITE]))
			wake_up(&rl->wait[WRITE]);
	}
}

/*
 * queue lock held here
 */
static void cfq_put_request(request_queue_t *q, struct request *rq)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct cfq_rq *crq = RQ_DATA(rq);

	if (crq) {
		struct cfq_queue *cfqq = crq->cfq_queue;
		const int rw = rq_data_dir(rq);

		BUG_ON(!cfqq->allocated[rw]);
		cfqq->allocated[rw]--;

		put_io_context(crq->io_context->ioc);

		mempool_free(crq, cfqd->crq_pool);
		rq->elevator_private = NULL;

		cfq_check_waiters(q, cfqq);
		cfq_put_queue(cfqq);
	}
}

/*
 * Allocate cfq data structures associated with this request.
 */
static int
cfq_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
		int gfp_mask)
{
	struct cfq_data *cfqd = q->elevator->elevator_data;
	struct task_struct *tsk = current;
	struct cfq_io_context *cic;
	const int rw = rq_data_dir(rq);
	pid_t key = cfq_queue_pid(tsk, rw);
	struct cfq_queue *cfqq;
	struct cfq_rq *crq;
	unsigned long flags;

	might_sleep_if(gfp_mask & __GFP_WAIT);

	cic = cfq_get_io_context(cfqd, key, gfp_mask);

	spin_lock_irqsave(q->queue_lock, flags);

	if (!cic)
		goto queue_fail;

	if (!cic->cfqq) {
		cfqq = cfq_get_queue(cfqd, key, tsk->ioprio, gfp_mask);
		if (!cfqq)
			goto queue_fail;

		cic->cfqq = cfqq;
	} else
		cfqq = cic->cfqq;

	cfqq->allocated[rw]++;
	cfq_clear_cfqq_must_alloc(cfqq);
	cfqd->rq_starved = 0;
	atomic_inc(&cfqq->ref);
	spin_unlock_irqrestore(q->queue_lock, flags);

	crq = mempool_alloc(cfqd->crq_pool, gfp_mask);
	if (crq) {
		RB_CLEAR(&crq->rb_node);
		crq->rb_key = 0;
		crq->request = rq;
		INIT_HLIST_NODE(&crq->hash);
		crq->cfq_queue = cfqq;
		crq->io_context = cic;
		cfq_clear_crq_in_flight(crq);
		cfq_clear_crq_in_driver(crq);
		cfq_clear_crq_requeued(crq);