*
* Copyright (C) 2003 Jens Axboe <axboe@suse.de>
*/
-#include <linux/config.h>
#include <linux/module.h>
#include <linux/blkdev.h>
#include <linux/elevator.h>
static const int cfq_slice_sync = HZ / 10;
static int cfq_slice_async = HZ / 25;
static const int cfq_slice_async_rq = 2;
-static int cfq_slice_idle = HZ / 70;
+static int cfq_slice_idle = HZ / 125;
#define CFQ_IDLE_GRACE (HZ / 10)
#define CFQ_SLICE_SCALE (5)
/*
* rb-tree defines
*/
-#define RB_NONE (2)
-#define RB_EMPTY(node) ((node)->rb_node == NULL)
-#define RB_CLEAR_COLOR(node) (node)->rb_color = RB_NONE
-#define RB_CLEAR(node) do { \
- (node)->rb_parent = NULL; \
- RB_CLEAR_COLOR((node)); \
- (node)->rb_right = NULL; \
- (node)->rb_left = NULL; \
-} while (0)
-#define RB_CLEAR_ROOT(root) ((root)->rb_node = NULL)
#define rb_entry_crq(node) rb_entry((node), struct cfq_rq, rb_node)
#define rq_rb_key(rq) (rq)->sector
*/
struct hlist_head *crq_hash;
- unsigned int max_queued;
-
mempool_t *crq_pool;
int rq_in_driver;
+ int hw_tag;
/*
* schedule slice state info
static void cfq_dispatch_insert(request_queue_t *, struct cfq_rq *);
static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, gfp_t gfp_mask);
-#define process_sync(tsk) ((tsk)->flags & PF_SYNCWRITE)
-
/*
* lots of deadline iosched dupes, can be abstracted later...
*/
static inline pid_t cfq_queue_pid(struct task_struct *task, int rw)
{
- if (rw == READ || process_sync(task))
+ if (rw == READ || rw == WRITE_SYNC)
return task->pid;
return CFQ_KEY_ASYNC;
/*
* if queue was preempted, just add to front to be fair. busy_rr
- * isn't sorted.
+ * isn't sorted, but insert at the back for fairness.
*/
if (preempted || list == &cfqd->busy_rr) {
- list_add(&cfqq->cfq_list, list);
+ if (preempted)
+ list = list->prev;
+
+ list_add_tail(&cfqq->cfq_list, list);
return;
}
cfq_update_next_crq(crq);
rb_erase(&crq->rb_node, &cfqq->sort_list);
- RB_CLEAR_COLOR(&crq->rb_node);
- if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY(&cfqq->sort_list))
+ if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list))
cfq_del_cfqq_rr(cfqd, cfqq);
}
struct cfq_data *cfqd = q->elevator->elevator_data;
cfqd->rq_in_driver++;
+
+ /*
+ * If the depth is larger 1, it really could be queueing. But lets
+ * make the mark a little higher - idling could still be good for
+ * low queueing, and a low queueing number could also just indicate
+ * a SCSI mid layer like behaviour where limit+1 is often seen.
+ */
+ if (!cfqd->hw_tag && cfqd->rq_in_driver > 4)
+ cfqd->hw_tag = 1;
}
static void cfq_deactivate_request(request_queue_t *q, struct request *rq)
return cfqq;
}
+#define CIC_SEEKY(cic) ((cic)->seek_mean > (128 * 1024))
+
static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
{
struct cfq_io_context *cic;
unsigned long sl;
- WARN_ON(!RB_EMPTY(&cfqq->sort_list));
+ WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list));
WARN_ON(cfqq != cfqd->active_queue);
/*
* fair distribution of slice time for a process doing back-to-back
* seeks. so allow a little bit of time for him to submit a new rq
*/
- if (sample_valid(cic->seek_samples) && cic->seek_mean > 131072)
+ if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
sl = 2;
mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
{
struct cfq_data *cfqd = q->elevator->elevator_data;
struct cfq_queue *cfqq = crq->cfq_queue;
+ struct request *rq;
cfqq->next_crq = cfq_find_next_crq(cfqd, cfqq, crq);
cfq_remove_request(crq->request);
cfqq->on_dispatch[cfq_crq_is_sync(crq)]++;
elv_dispatch_sort(q, crq->request);
+
+ rq = list_entry(q->queue_head.prev, struct request, queuelist);
+ cfqd->last_sector = rq->sector + rq->nr_sectors;
}
/*
* if queue has requests, dispatch one. if not, check if
* enough slice is left to wait for one
*/
- if (!RB_EMPTY(&cfqq->sort_list))
+ if (!RB_EMPTY_ROOT(&cfqq->sort_list))
goto keep_queue;
- else if (cfq_cfqq_class_sync(cfqq) &&
- time_before(now, cfqq->slice_end)) {
+ else if (cfq_cfqq_dispatched(cfqq)) {
+ cfqq = NULL;
+ goto keep_queue;
+ } else if (cfq_cfqq_class_sync(cfqq)) {
if (cfq_arm_slice_timer(cfqd, cfqq))
return NULL;
}
{
int dispatched = 0;
- BUG_ON(RB_EMPTY(&cfqq->sort_list));
+ BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list));
do {
struct cfq_rq *crq;
cfqd->active_cic = crq->io_context;
}
- if (RB_EMPTY(&cfqq->sort_list))
+ if (RB_EMPTY_ROOT(&cfqq->sort_list))
break;
} while (dispatched < max_dispatch);
/*
- * if slice end isn't set yet, set it. if at least one request was
- * sync, use the sync time slice value
+ * if slice end isn't set yet, set it.
*/
if (!cfqq->slice_end)
cfq_set_prio_slice(cfqd, cfqq);
*/
if ((!cfq_cfqq_sync(cfqq) &&
cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) ||
- cfq_class_idle(cfqq))
+ cfq_class_idle(cfqq) ||
+ !cfq_cfqq_idle_window(cfqq))
cfq_slice_expired(cfqd, 0);
return dispatched;
static int
cfq_forced_dispatch_cfqqs(struct list_head *list)
{
- int dispatched = 0;
struct cfq_queue *cfqq, *next;
struct cfq_rq *crq;
+ int dispatched;
+ dispatched = 0;
list_for_each_entry_safe(cfqq, next, list, cfq_list) {
while ((crq = cfqq->next_crq)) {
cfq_dispatch_insert(cfqq->cfqd->queue, crq);
}
BUG_ON(!list_empty(&cfqq->fifo));
}
+
return dispatched;
}
cfq_dispatch_requests(request_queue_t *q, int force)
{
struct cfq_data *cfqd = q->elevator->elevator_data;
- struct cfq_queue *cfqq;
+ struct cfq_queue *cfqq, *prev_cfqq;
+ int dispatched;
if (!cfqd->busy_queues)
return 0;
if (unlikely(force))
return cfq_forced_dispatch(cfqd);
- cfqq = cfq_select_queue(cfqd);
- if (cfqq) {
+ dispatched = 0;
+ prev_cfqq = NULL;
+ while ((cfqq = cfq_select_queue(cfqd)) != NULL) {
int max_dispatch;
+ /*
+ * Don't repeat dispatch from the previous queue.
+ */
+ if (prev_cfqq == cfqq)
+ break;
+
cfq_clear_cfqq_must_dispatch(cfqq);
cfq_clear_cfqq_wait_request(cfqq);
del_timer(&cfqd->idle_slice_timer);
if (cfq_class_idle(cfqq))
max_dispatch = 1;
- return __cfq_dispatch_requests(cfqd, cfqq, max_dispatch);
+ dispatched += __cfq_dispatch_requests(cfqd, cfqq, max_dispatch);
+
+ /*
+ * If the dispatch cfqq has idling enabled and is still
+ * the active queue, break out.
+ */
+ if (cfq_cfqq_idle_window(cfqq) && cfqd->active_queue)
+ break;
+
+ prev_cfqq = cfqq;
}
- return 0;
+ return dispatched;
}
/*
struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_mask);
if (cic) {
- RB_CLEAR(&cic->rb_node);
- cic->key = NULL;
- cic->cfqq[ASYNC] = NULL;
- cic->cfqq[SYNC] = NULL;
+ memset(cic, 0, sizeof(*cic));
cic->last_end_request = jiffies;
- cic->ttime_total = 0;
- cic->ttime_samples = 0;
- cic->ttime_mean = 0;
+ INIT_LIST_HEAD(&cic->queue_list);
cic->dtor = cfq_free_io_context;
cic->exit = cfq_exit_io_context;
- INIT_LIST_HEAD(&cic->queue_list);
atomic_inc(&ioc_count);
}
{
struct cfq_data *cfqd = cic->key;
struct cfq_queue *cfqq;
- if (cfqd) {
- spin_lock(cfqd->queue->queue_lock);
- cfqq = cic->cfqq[ASYNC];
- if (cfqq) {
- struct cfq_queue *new_cfqq;
- new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC,
- cic->ioc->task, GFP_ATOMIC);
- if (new_cfqq) {
- cic->cfqq[ASYNC] = new_cfqq;
- cfq_put_queue(cfqq);
- }
- }
- cfqq = cic->cfqq[SYNC];
- if (cfqq) {
- cfq_mark_cfqq_prio_changed(cfqq);
- cfq_init_prio_data(cfqq);
+
+ if (unlikely(!cfqd))
+ return;
+
+ spin_lock(cfqd->queue->queue_lock);
+
+ cfqq = cic->cfqq[ASYNC];
+ if (cfqq) {
+ struct cfq_queue *new_cfqq;
+ new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC, cic->ioc->task,
+ GFP_ATOMIC);
+ if (new_cfqq) {
+ cic->cfqq[ASYNC] = new_cfqq;
+ cfq_put_queue(cfqq);
}
- spin_unlock(cfqd->queue->queue_lock);
}
+
+ cfqq = cic->cfqq[SYNC];
+ if (cfqq)
+ cfq_mark_cfqq_prio_changed(cfqq);
+
+ spin_unlock(cfqd->queue->queue_lock);
}
/*
INIT_HLIST_NODE(&cfqq->cfq_hash);
INIT_LIST_HEAD(&cfqq->cfq_list);
- RB_CLEAR_ROOT(&cfqq->sort_list);
INIT_LIST_HEAD(&cfqq->fifo);
cfqq->key = key;
{
int enable_idle = cfq_cfqq_idle_window(cfqq);
- if (!cic->ioc->task || !cfqd->cfq_slice_idle)
+ if (!cic->ioc->task || !cfqd->cfq_slice_idle ||
+ (cfqd->hw_tag && CIC_SEEKY(cic)))
enable_idle = 0;
else if (sample_valid(cic->ttime_samples)) {
if (cic->ttime_mean > cfqd->cfq_slice_idle)
return 0;
if (!cfqq)
- return 1;
+ return 0;
if (cfq_class_idle(cfqq))
return 1;
cfqq->slice_left = cfq_prio_to_slice(cfqd, cfqq) / 2;
cfqq->slice_end = cfqq->slice_left + jiffies;
- __cfq_slice_expired(cfqd, cfqq, 1);
+ cfq_slice_expired(cfqd, 1);
__cfq_set_active_queue(cfqd, cfqq);
}
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);
-
- cic = crq->io_context;
+ struct cfq_io_context *cic = crq->io_context;
/*
* we never wait for an async request and we don't allow preemption
cfqq->service_last = now;
cfq_resort_rr_list(cfqq, 0);
}
- cfq_schedule_dispatch(cfqd);
}
- if (cfq_crq_is_sync(crq))
+ if (sync)
crq->io_context->last_end_request = now;
+
+ /*
+ * If this is the active queue, check if it needs to be expired,
+ * or if we want to idle in case it has no pending requests.
+ */
+ if (cfqd->active_queue == cfqq) {
+ if (time_after(now, cfqq->slice_end))
+ cfq_slice_expired(cfqd, 0);
+ else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list)) {
+ if (!cfq_arm_slice_timer(cfqd, cfqq))
+ cfq_schedule_dispatch(cfqd);
+ }
+ }
}
static struct request *
__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_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)
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) {
+ if (unlikely(cfqd->rq_starved)) {
+ struct request_list *rl = &q->rq;
+
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]);
}
crq = mempool_alloc(cfqd->crq_pool, gfp_mask);
if (crq) {
- RB_CLEAR(&crq->rb_node);
+ RB_CLEAR_NODE(&crq->rb_node);
crq->rb_key = 0;
crq->request = rq;
INIT_HLIST_NODE(&crq->hash);
* only expire and reinvoke request handler, if there are
* other queues with pending requests
*/
- if (!cfqd->busy_queues) {
- cfqd->idle_slice_timer.expires = min(now + cfqd->cfq_slice_idle, cfqq->slice_end);
- add_timer(&cfqd->idle_slice_timer);
+ if (!cfqd->busy_queues)
goto out_cont;
- }
/*
* not expired and it has a request pending, let it dispatch
*/
- if (!RB_EMPTY(&cfqq->sort_list)) {
+ if (!RB_EMPTY_ROOT(&cfqq->sort_list)) {
cfq_mark_cfqq_must_dispatch(cfqq);
goto out_kick;
}
* race with a non-idle queue, reset timer
*/
end = cfqd->last_end_request + CFQ_IDLE_GRACE;
- if (!time_after_eq(jiffies, end)) {
- cfqd->idle_class_timer.expires = end;
- add_timer(&cfqd->idle_class_timer);
- } else
+ if (!time_after_eq(jiffies, end))
+ mod_timer(&cfqd->idle_class_timer, end);
+ else
cfq_schedule_dispatch(cfqd);
spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
kfree(cfqd);
}
-static int cfq_init_queue(request_queue_t *q, elevator_t *e)
+static void *cfq_init_queue(request_queue_t *q, elevator_t *e)
{
struct cfq_data *cfqd;
int i;
cfqd = kmalloc(sizeof(*cfqd), GFP_KERNEL);
if (!cfqd)
- return -ENOMEM;
+ return NULL;
memset(cfqd, 0, sizeof(*cfqd));
for (i = 0; i < CFQ_QHASH_ENTRIES; i++)
INIT_HLIST_HEAD(&cfqd->cfq_hash[i]);
- e->elevator_data = cfqd;
-
cfqd->queue = q;
- cfqd->max_queued = q->nr_requests / 4;
- q->nr_batching = cfq_queued;
-
init_timer(&cfqd->idle_slice_timer);
cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
cfqd->idle_slice_timer.data = (unsigned long) cfqd;
cfqd->cfq_slice_async_rq = cfq_slice_async_rq;
cfqd->cfq_slice_idle = cfq_slice_idle;
- return 0;
+ return cfqd;
out_crqpool:
kfree(cfqd->cfq_hash);
out_cfqhash:
kfree(cfqd->crq_hash);
out_crqhash:
kfree(cfqd);
- return -ENOMEM;
+ return NULL;
}
static void cfq_slab_kill(void)
Code Review / linux-2.6.git / blobdiff