diff options
Diffstat (limited to 'drivers/block/cfq-iosched.c')
-rw-r--r-- | drivers/block/cfq-iosched.c | 2085 |
1 files changed, 1433 insertions, 652 deletions
diff --git a/drivers/block/cfq-iosched.c b/drivers/block/cfq-iosched.c index 2210bacad56a..de5746e38af9 100644 --- a/drivers/block/cfq-iosched.c +++ b/drivers/block/cfq-iosched.c @@ -21,22 +21,34 @@ #include <linux/hash.h> #include <linux/rbtree.h> #include <linux/mempool.h> - -static unsigned long max_elapsed_crq; -static unsigned long max_elapsed_dispatch; +#include <linux/ioprio.h> +#include <linux/writeback.h> /* * tunables */ static int cfq_quantum = 4; /* max queue in one round of service */ static int cfq_queued = 8; /* minimum rq allocate limit per-queue*/ -static int cfq_service = HZ; /* period over which service is avg */ -static int cfq_fifo_expire_r = HZ / 2; /* fifo timeout for sync requests */ -static int cfq_fifo_expire_w = 5 * HZ; /* fifo timeout for async requests */ -static int cfq_fifo_rate = HZ / 8; /* fifo expiry rate */ +static int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; static int cfq_back_max = 16 * 1024; /* maximum backwards seek, in KiB */ static int cfq_back_penalty = 2; /* penalty of a backwards seek */ +static int cfq_slice_sync = HZ / 10; +static int cfq_slice_async = HZ / 25; +static int cfq_slice_async_rq = 2; +static int cfq_slice_idle = HZ / 100; + +#define CFQ_IDLE_GRACE (HZ / 10) +#define CFQ_SLICE_SCALE (5) + +#define CFQ_KEY_ASYNC (0) +#define CFQ_KEY_ANY (0xffff) + +/* + * disable queueing at the driver/hardware level + */ +static int cfq_max_depth = 1; + /* * for the hash of cfqq inside the cfqd */ @@ -55,6 +67,7 @@ static int cfq_back_penalty = 2; /* penalty of a backwards seek */ #define list_entry_hash(ptr) hlist_entry((ptr), struct cfq_rq, hash) #define list_entry_cfqq(ptr) list_entry((ptr), struct cfq_queue, cfq_list) +#define list_entry_fifo(ptr) list_entry((ptr), struct request, queuelist) #define RQ_DATA(rq) (rq)->elevator_private @@ -75,78 +88,110 @@ static int cfq_back_penalty = 2; /* penalty of a backwards seek */ #define rb_entry_crq(node) rb_entry((node), struct cfq_rq, rb_node) #define rq_rb_key(rq) (rq)->sector -/* - * threshold for switching off non-tag accounting - */ -#define CFQ_MAX_TAG (4) - -/* - * sort key types and names - */ -enum { - CFQ_KEY_PGID, - CFQ_KEY_TGID, - CFQ_KEY_UID, - CFQ_KEY_GID, - CFQ_KEY_LAST, -}; - -static char *cfq_key_types[] = { "pgid", "tgid", "uid", "gid", NULL }; - static kmem_cache_t *crq_pool; static kmem_cache_t *cfq_pool; static kmem_cache_t *cfq_ioc_pool; +#define CFQ_PRIO_LISTS IOPRIO_BE_NR +#define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) +#define cfq_class_be(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_BE) +#define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) + +#define ASYNC (0) +#define SYNC (1) + +#define cfq_cfqq_dispatched(cfqq) \ + ((cfqq)->on_dispatch[ASYNC] + (cfqq)->on_dispatch[SYNC]) + +#define cfq_cfqq_class_sync(cfqq) ((cfqq)->key != CFQ_KEY_ASYNC) + +#define cfq_cfqq_sync(cfqq) \ + (cfq_cfqq_class_sync(cfqq) || (cfqq)->on_dispatch[SYNC]) + +/* + * Per block device queue structure + */ struct cfq_data { - struct list_head rr_list; + atomic_t ref; + request_queue_t *queue; + + /* + * rr list of queues with requests and the count of them + */ + struct list_head rr_list[CFQ_PRIO_LISTS]; + struct list_head busy_rr; + struct list_head cur_rr; + struct list_head idle_rr; + unsigned int busy_queues; + + /* + * non-ordered list of empty cfqq's + */ struct list_head empty_list; + /* + * cfqq lookup hash + */ struct hlist_head *cfq_hash; - struct hlist_head *crq_hash; - /* queues on rr_list (ie they have pending requests */ - unsigned int busy_queues; + /* + * global crq hash for all queues + */ + struct hlist_head *crq_hash; unsigned int max_queued; - atomic_t ref; + mempool_t *crq_pool; - int key_type; + int rq_in_driver; - mempool_t *crq_pool; + /* + * schedule slice state info + */ + /* + * idle window management + */ + struct timer_list idle_slice_timer; + struct work_struct unplug_work; - request_queue_t *queue; + struct cfq_queue *active_queue; + struct cfq_io_context *active_cic; + int cur_prio, cur_end_prio; + unsigned int dispatch_slice; + + struct timer_list idle_class_timer; sector_t last_sector; + unsigned long last_end_request; - int rq_in_driver; + unsigned int rq_starved; /* * tunables, see top of file */ unsigned int cfq_quantum; unsigned int cfq_queued; - unsigned int cfq_fifo_expire_r; - unsigned int cfq_fifo_expire_w; - unsigned int cfq_fifo_batch_expire; + unsigned int cfq_fifo_expire[2]; unsigned int cfq_back_penalty; unsigned int cfq_back_max; - unsigned int find_best_crq; - - unsigned int cfq_tagged; + unsigned int cfq_slice[2]; + unsigned int cfq_slice_async_rq; + unsigned int cfq_slice_idle; + unsigned int cfq_max_depth; }; +/* + * Per process-grouping structure + */ struct cfq_queue { /* reference count */ atomic_t ref; /* parent cfq_data */ struct cfq_data *cfqd; - /* hash of mergeable requests */ + /* cfqq lookup hash */ struct hlist_node cfq_hash; /* hash key */ - unsigned long key; - /* whether queue is on rr (or empty) list */ - int on_rr; + unsigned int key; /* on either rr or empty list of cfqd */ struct list_head cfq_list; /* sorted list of pending requests */ @@ -158,21 +203,22 @@ struct cfq_queue { /* currently allocated requests */ int allocated[2]; /* fifo list of requests in sort_list */ - struct list_head fifo[2]; - /* last time fifo expired */ - unsigned long last_fifo_expire; + struct list_head fifo; - int key_type; + unsigned long slice_start; + unsigned long slice_end; + unsigned long slice_left; + unsigned long service_last; - unsigned long service_start; - unsigned long service_used; + /* number of requests that are on the dispatch list */ + int on_dispatch[2]; - unsigned int max_rate; + /* io prio of this group */ + unsigned short ioprio, org_ioprio; + unsigned short ioprio_class, org_ioprio_class; - /* number of requests that have been handed to the driver */ - int in_flight; - /* number of currently allocated requests */ - int alloc_limit[2]; + /* various state flags, see below */ + unsigned int flags; }; struct cfq_rq { @@ -184,42 +230,78 @@ struct cfq_rq { struct cfq_queue *cfq_queue; struct cfq_io_context *io_context; - unsigned long service_start; - unsigned long queue_start; + unsigned int crq_flags; +}; + +enum cfqq_state_flags { + CFQ_CFQQ_FLAG_on_rr = 0, + CFQ_CFQQ_FLAG_wait_request, + CFQ_CFQQ_FLAG_must_alloc, + CFQ_CFQQ_FLAG_must_alloc_slice, + CFQ_CFQQ_FLAG_must_dispatch, + CFQ_CFQQ_FLAG_fifo_expire, + CFQ_CFQQ_FLAG_idle_window, + CFQ_CFQQ_FLAG_prio_changed, + CFQ_CFQQ_FLAG_expired, +}; - unsigned int in_flight : 1; - unsigned int accounted : 1; - unsigned int is_sync : 1; - unsigned int is_write : 1; +#define CFQ_CFQQ_FNS(name) \ +static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ +{ \ + cfqq->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ +} \ +static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ +{ \ + cfqq->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ +} \ +static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ +{ \ + return (cfqq->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ +} + +CFQ_CFQQ_FNS(on_rr); +CFQ_CFQQ_FNS(wait_request); +CFQ_CFQQ_FNS(must_alloc); +CFQ_CFQQ_FNS(must_alloc_slice); +CFQ_CFQQ_FNS(must_dispatch); +CFQ_CFQQ_FNS(fifo_expire); +CFQ_CFQQ_FNS(idle_window); +CFQ_CFQQ_FNS(prio_changed); +CFQ_CFQQ_FNS(expired); +#undef CFQ_CFQQ_FNS + +enum cfq_rq_state_flags { + CFQ_CRQ_FLAG_in_flight = 0, + CFQ_CRQ_FLAG_in_driver, + CFQ_CRQ_FLAG_is_sync, + CFQ_CRQ_FLAG_requeued, }; -static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned long); +#define CFQ_CRQ_FNS(name) \ +static inline void cfq_mark_crq_##name(struct cfq_rq *crq) \ +{ \ + crq->crq_flags |= (1 << CFQ_CRQ_FLAG_##name); \ +} \ +static inline void cfq_clear_crq_##name(struct cfq_rq *crq) \ +{ \ + crq->crq_flags &= ~(1 << CFQ_CRQ_FLAG_##name); \ +} \ +static inline int cfq_crq_##name(const struct cfq_rq *crq) \ +{ \ + return (crq->crq_flags & (1 << CFQ_CRQ_FLAG_##name)) != 0; \ +} + +CFQ_CRQ_FNS(in_flight); +CFQ_CRQ_FNS(in_driver); +CFQ_CRQ_FNS(is_sync); +CFQ_CRQ_FNS(requeued); +#undef CFQ_CRQ_FNS + +static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short); static void cfq_dispatch_sort(request_queue_t *, struct cfq_rq *); -static void cfq_update_next_crq(struct cfq_rq *); static void cfq_put_cfqd(struct cfq_data *cfqd); -/* - * what the fairness is based on (ie how processes are grouped and - * differentiated) - */ -static inline unsigned long -cfq_hash_key(struct cfq_data *cfqd, struct task_struct *tsk) -{ - /* - * optimize this so that ->key_type is the offset into the struct - */ - switch (cfqd->key_type) { - case CFQ_KEY_PGID: - return process_group(tsk); - default: - case CFQ_KEY_TGID: - return tsk->tgid; - case CFQ_KEY_UID: - return tsk->uid; - case CFQ_KEY_GID: - return tsk->gid; - } -} +#define process_sync(tsk) ((tsk)->flags & PF_SYNCWRITE) /* * lots of deadline iosched dupes, can be abstracted later... @@ -235,16 +317,12 @@ static void cfq_remove_merge_hints(request_queue_t *q, struct cfq_rq *crq) if (q->last_merge == crq->request) q->last_merge = NULL; - - cfq_update_next_crq(crq); } static inline void cfq_add_crq_hash(struct cfq_data *cfqd, struct cfq_rq *crq) { const int hash_idx = CFQ_MHASH_FN(rq_hash_key(crq->request)); - BUG_ON(!hlist_unhashed(&crq->hash)); - hlist_add_head(&crq->hash, &cfqd->crq_hash[hash_idx]); } @@ -257,8 +335,6 @@ static struct request *cfq_find_rq_hash(struct cfq_data *cfqd, sector_t offset) struct cfq_rq *crq = list_entry_hash(entry); struct request *__rq = crq->request; - BUG_ON(hlist_unhashed(&crq->hash)); - if (!rq_mergeable(__rq)) { cfq_del_crq_hash(crq); continue; @@ -271,6 +347,28 @@ static struct request *cfq_find_rq_hash(struct cfq_data *cfqd, sector_t offset) return NULL; } +static inline int cfq_pending_requests(struct cfq_data *cfqd) +{ + return !list_empty(&cfqd->queue->queue_head) || cfqd->busy_queues; +} + +/* + * scheduler run of queue, if there are requests pending and no one in the + * driver that will restart queueing + */ +static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) +{ + if (!cfqd->rq_in_driver && cfq_pending_requests(cfqd)) + kblockd_schedule_work(&cfqd->unplug_work); +} + +static int cfq_queue_empty(request_queue_t *q) +{ + struct cfq_data *cfqd = q->elevator->elevator_data; + + return !cfq_pending_requests(cfqd); +} + /* * Lifted from AS - choose which of crq1 and crq2 that is best served now. * We choose the request that is closest to the head right now. Distance @@ -287,36 +385,16 @@ cfq_choose_req(struct cfq_data *cfqd, struct cfq_rq *crq1, struct cfq_rq *crq2) return crq2; if (crq2 == NULL) return crq1; + if (cfq_crq_requeued(crq1)) + return crq1; + if (cfq_crq_requeued(crq2)) + return crq2; s1 = crq1->request->sector; s2 = crq2->request->sector; last = cfqd->last_sector; -#if 0 - if (!list_empty(&cfqd->queue->queue_head)) { - struct list_head *entry = &cfqd->queue->queue_head; - unsigned long distance = ~0UL; - struct request *rq; - - while ((entry = entry->prev) != &cfqd->queue->queue_head) { - rq = list_entry_rq(entry); - - if (blk_barrier_rq(rq)) - break; - - if (distance < abs(s1 - rq->sector + rq->nr_sectors)) { - distance = abs(s1 - rq->sector +rq->nr_sectors); - last = rq->sector + rq->nr_sectors; - } - if (distance < abs(s2 - rq->sector + rq->nr_sectors)) { - distance = abs(s2 - rq->sector +rq->nr_sectors); - last = rq->sector + rq->nr_sectors; - } - } - } -#endif - /* * by definition, 1KiB is 2 sectors */ @@ -377,11 +455,14 @@ cfq_find_next_crq(struct cfq_data *cfqd, struct cfq_queue *cfqq, struct cfq_rq *crq_next = NULL, *crq_prev = NULL; struct rb_node *rbnext, *rbprev; - if (!ON_RB(&last->rb_node)) - return NULL; - - if ((rbnext = rb_next(&last->rb_node)) == NULL) + rbnext = NULL; + if (ON_RB(&last->rb_node)) + rbnext = rb_next(&last->rb_node); + if (!rbnext) { rbnext = rb_first(&cfqq->sort_list); + if (rbnext == &last->rb_node) + rbnext = NULL; + } rbprev = rb_prev(&last->rb_node); @@ -401,67 +482,53 @@ static void cfq_update_next_crq(struct cfq_rq *crq) cfqq->next_crq = cfq_find_next_crq(cfqq->cfqd, cfqq, crq); } -static int cfq_check_sort_rr_list(struct cfq_queue *cfqq) +static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted) { - struct list_head *head = &cfqq->cfqd->rr_list; - struct list_head *next, *prev; - - /* - * list might still be ordered - */ - next = cfqq->cfq_list.next; - if (next != head) { - struct cfq_queue *cnext = list_entry_cfqq(next); + struct cfq_data *cfqd = cfqq->cfqd; + struct list_head *list, *entry; - if (cfqq->service_used > cnext->service_used) - return 1; - } + BUG_ON(!cfq_cfqq_on_rr(cfqq)); - prev = cfqq->cfq_list.prev; - if (prev != head) { - struct cfq_queue *cprev = list_entry_cfqq(prev); + list_del(&cfqq->cfq_list); - if (cfqq->service_used < cprev->service_used) - return 1; + if (cfq_class_rt(cfqq)) + list = &cfqd->cur_rr; + else if (cfq_class_idle(cfqq)) + list = &cfqd->idle_rr; + else { + /* + * if cfqq has requests in flight, don't allow it to be + * found in cfq_set_active_queue before it has finished them. + * this is done to increase fairness between a process that + * has lots of io pending vs one that only generates one + * sporadically or synchronously + */ + if (cfq_cfqq_dispatched(cfqq)) + list = &cfqd->busy_rr; + else + list = &cfqd->rr_list[cfqq->ioprio]; } - return 0; -} - -static void cfq_sort_rr_list(struct cfq_queue *cfqq, int new_queue) -{ - struct list_head *entry = &cfqq->cfqd->rr_list; - - if (!cfqq->on_rr) - return; - if (!new_queue && !cfq_check_sort_rr_list(cfqq)) + /* + * if queue was preempted, just add to front to be fair. busy_rr + * isn't sorted. + */ + if (preempted || list == &cfqd->busy_rr) { + list_add(&cfqq->cfq_list, list); return; - - list_del(&cfqq->cfq_list); + } /* - * sort by our mean service_used, sub-sort by in-flight requests + * sort by when queue was last serviced */ - while ((entry = entry->prev) != &cfqq->cfqd->rr_list) { + entry = list; + while ((entry = entry->prev) != list) { struct cfq_queue *__cfqq = list_entry_cfqq(entry); - if (cfqq->service_used > __cfqq->service_used) + if (!__cfqq->service_last) + break; + if (time_before(__cfqq->service_last, cfqq->service_last)) break; - else if (cfqq->service_used == __cfqq->service_used) { - struct list_head *prv; - - while ((prv = entry->prev) != &cfqq->cfqd->rr_list) { - __cfqq = list_entry_cfqq(prv); - - WARN_ON(__cfqq->service_used > cfqq->service_used); - if (cfqq->service_used != __cfqq->service_used) - break; - if (cfqq->in_flight > __cfqq->in_flight) - break; - - entry = prv; - } - } } list_add(&cfqq->cfq_list, entry); @@ -469,28 +536,24 @@ static void cfq_sort_rr_list(struct cfq_queue *cfqq, int new_queue) /* * add to busy list of queues for service, trying to be fair in ordering - * the pending list according to requests serviced + * the pending list according to last request service */ static inline void -cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) +cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq, int requeue) { - /* - * it's currently on the empty list - */ - cfqq->on_rr = 1; + BUG_ON(cfq_cfqq_on_rr(cfqq)); + cfq_mark_cfqq_on_rr(cfqq); cfqd->busy_queues++; - if (time_after(jiffies, cfqq->service_start + cfq_service)) - cfqq->service_used >>= 3; - - cfq_sort_rr_list(cfqq, 1); + cfq_resort_rr_list(cfqq, requeue); } static inline void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) { + BUG_ON(!cfq_cfqq_on_rr(cfqq)); + cfq_clear_cfqq_on_rr(cfqq); list_move(&cfqq->cfq_list, &cfqd->empty_list); - cfqq->on_rr = 0; BUG_ON(!cfqd->busy_queues); cfqd->busy_queues--; @@ -505,16 +568,17 @@ static inline void cfq_del_crq_rb(struct cfq_rq *crq) if (ON_RB(&crq->rb_node)) { struct cfq_data *cfqd = cfqq->cfqd; + const int sync = cfq_crq_is_sync(crq); - BUG_ON(!cfqq->queued[crq->is_sync]); + BUG_ON(!cfqq->queued[sync]); + cfqq->queued[sync]--; cfq_update_next_crq(crq); - cfqq->queued[crq->is_sync]--; rb_erase(&crq->rb_node, &cfqq->sort_list); RB_CLEAR_COLOR(&crq->rb_node); - if (RB_EMPTY(&cfqq->sort_list) && cfqq->on_rr) + if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY(&cfqq->sort_list)) cfq_del_cfqq_rr(cfqd, cfqq); } } @@ -550,7 +614,7 @@ static void cfq_add_crq_rb(struct cfq_rq *crq) struct cfq_rq *__alias; crq->rb_key = rq_rb_key(rq); - cfqq->queued[crq->is_sync]++; + cfqq->queued[cfq_crq_is_sync(crq)]++; /* * looks a little odd, but the first insert might return an alias. @@ -561,8 +625,8 @@ static void cfq_add_crq_rb(struct cfq_rq *crq) rb_insert_color(&crq->rb_node, &cfqq->sort_list); - if (!cfqq->on_rr) - cfq_add_cfqq_rr(cfqd, cfqq); + if (!cfq_cfqq_on_rr(cfqq)) + cfq_add_cfqq_rr(cfqd, cfqq, cfq_crq_requeued(crq)); /* * check if this request is a better next-serve candidate @@ -575,17 +639,16 @@ cfq_reposition_crq_rb(struct cfq_queue *cfqq, struct cfq_rq *crq) { if (ON_RB(&crq->rb_node)) { rb_erase(&crq->rb_node, &cfqq->sort_list); - cfqq->queued[crq->is_sync]--; + cfqq->queued[cfq_crq_is_sync(crq)]--; } cfq_add_crq_rb(crq); } -static struct request * -cfq_find_rq_rb(struct cfq_data *cfqd, sector_t sector) +static struct request *cfq_find_rq_rb(struct cfq_data *cfqd, sector_t sector) + { - const unsigned long key = cfq_hash_key(cfqd, current); - struct cfq_queue *cfqq = cfq_find_cfq_hash(cfqd, key); + struct cfq_queue *cfqq = cfq_find_cfq_hash(cfqd, current->pid, CFQ_KEY_ANY); struct rb_node *n; if (!cfqq) @@ -609,20 +672,25 @@ out: static void cfq_deactivate_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; - if (cfqq->cfqd->cfq_tagged) { - cfqq->service_used--; - cfq_sort_rr_list(cfqq, 0); + if (cfq_crq_in_driver(crq)) { + cfq_clear_crq_in_driver(crq); + WARN_ON(!cfqd->rq_in_driver); + cfqd->rq_in_driver--; } + if (cfq_crq_in_flight(crq)) { + const int sync = cfq_crq_is_sync(crq); - if (crq->accounted) { - crq->accounted = 0; - cfqq->cfqd->rq_in_driver--; + cfq_clear_crq_in_flight(crq); + WARN_ON(!cfqq->on_dispatch[sync]); + cfqq->on_dispatch[sync]--; } + cfq_mark_crq_requeued(crq); } } @@ -640,11 +708,10 @@ static void cfq_remove_request(request_queue_t *q, struct request *rq) struct cfq_rq *crq = RQ_DATA(rq); if (crq) { - cfq_remove_merge_hints(q, crq); list_del_init(&rq->queuelist); + cfq_del_crq_rb(crq); + cfq_remove_merge_hints(q, crq); - if (crq->cfq_queue) - cfq_del_crq_rb(crq); } } @@ -662,21 +729,15 @@ cfq_merge(request_queue_t *q, struct request **req, struct bio *bio) } __rq = cfq_find_rq_hash(cfqd, bio->bi_sector); - if (__rq) { - BUG_ON(__rq->sector + __rq->nr_sectors != bio->bi_sector); - - if (elv_rq_merge_ok(__rq, bio)) { - ret = ELEVATOR_BACK_MERGE; - goto out; - } + if (__rq && elv_rq_merge_ok(__rq, bio)) { + ret = ELEVATOR_BACK_MERGE; + goto out; } __rq = cfq_find_rq_rb(cfqd, bio->bi_sector + bio_sectors(bio)); - if (__rq) { - if (elv_rq_merge_ok(__rq, bio)) { - ret = ELEVATOR_FRONT_MERGE; - goto out; - } + if (__rq && elv_rq_merge_ok(__rq, bio)) { + ret = ELEVATOR_FRONT_MERGE; + goto out; } return ELEVATOR_NO_MERGE; @@ -709,20 +770,220 @@ static void cfq_merged_requests(request_queue_t *q, struct request *rq, struct request *next) { - struct cfq_rq *crq = RQ_DATA(rq); - struct cfq_rq *cnext = RQ_DATA(next); - cfq_merged_request(q, rq); - if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist)) { - if (time_before(cnext->queue_start, crq->queue_start)) { - list_move(&rq->queuelist, &next->queuelist); - crq->queue_start = cnext->queue_start; + /* + * reposition in fifo if next is older than rq + */ + if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && + time_before(next->start_time, rq->start_time)) + list_move(&rq->queuelist, &next->queuelist); + + cfq_remove_request(q, next); +} + +static inline void +__cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) +{ + if (cfqq) { + /* + * stop potential idle class queues waiting service + */ + del_timer(&cfqd->idle_class_timer); + + cfqq->slice_start = jiffies; + cfqq->slice_end = 0; + cfqq->slice_left = 0; + cfq_clear_cfqq_must_alloc_slice(cfqq); + cfq_clear_cfqq_fifo_expire(cfqq); + cfq_clear_cfqq_expired(cfqq); + } + + cfqd->active_queue = cfqq; +} + +/* + * 0 + * 0,1 + * 0,1,2 + * 0,1,2,3 + * 0,1,2,3,4 + * 0,1,2,3,4,5 + * 0,1,2,3,4,5,6 + * 0,1,2,3,4,5,6,7 + */ +static int cfq_get_next_prio_level(struct cfq_data *cfqd) +{ + int prio, wrap; + + prio = -1; + wrap = 0; + do { + int p; + + for (p = cfqd->cur_prio; p <= cfqd->cur_end_prio; p++) { + if (!list_empty(&cfqd->rr_list[p])) { + prio = p; + break; + } + } + + if (prio != -1) + break; + cfqd->cur_prio = 0; + if (++cfqd->cur_end_prio == CFQ_PRIO_LISTS) { + cfqd->cur_end_prio = 0; + if (wrap) + break; + wrap = 1; } + } while (1); + + if (unlikely(prio == -1)) + return -1; + + BUG_ON(prio >= CFQ_PRIO_LISTS); + + list_splice_init(&cfqd->rr_list[prio], &cfqd->cur_rr); + + cfqd->cur_prio = prio + 1; + if (cfqd->cur_prio > cfqd->cur_end_prio) { + cfqd->cur_end_prio = cfqd->cur_prio; + cfqd->cur_prio = 0; + } + if (cfqd->cur_end_prio == CFQ_PRIO_LISTS) { + cfqd->cur_prio = 0; + cfqd->cur_end_prio = 0; } - cfq_update_next_crq(cnext); - cfq_remove_request(q, next); + return prio; +} + +static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd) +{ + struct cfq_queue *cfqq; + + /* + * if current queue is expired but not done with its requests yet, + * wait for that to happen + */ + if ((cfqq = cfqd->active_queue) != NULL) { + if (cfq_cfqq_expired(cfqq) && cfq_cfqq_dispatched(cfqq)) + return NULL; + } + + /* + * if current list is non-empty, grab first entry. if it is empty, + * get next prio level and grab first entry then if any are spliced + */ + if (!list_empty(&cfqd->cur_rr) || cfq_get_next_prio_level(cfqd) != -1) + cfqq = list_entry_cfqq(cfqd->cur_rr.next); + + /* + * if we have idle queues and no rt or be queues had pending + * requests, either allow immediate service if the grace period + * has passed or arm the idle grace timer + */ + if (!cfqq && !list_empty(&cfqd->idle_rr)) { + unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE; + + if (time_after_eq(jiffies, end)) + cfqq = list_entry_cfqq(cfqd->idle_rr.next); + else + mod_timer(&cfqd->idle_class_timer, end); + } + + __cfq_set_active_queue(cfqd, cfqq); + return cfqq; +} + +/* + * current cfqq expired its slice (or was too idle), select new one + */ +static void +__cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, + int preempted) +{ + unsigned long now = jiffies; + + if (cfq_cfqq_wait_request(cfqq)) + del_timer(&cfqd->idle_slice_timer); + + if (!preempted && !cfq_cfqq_dispatched(cfqq)) + cfqq->service_last = now; + + cfq_clear_cfqq_must_dispatch(cfqq); + cfq_clear_cfqq_wait_request(cfqq); + + /* + * store what was left of this slice, if the queue idled out + * or was preempted + */ + if (time_after(now, cfqq->slice_end)) + cfqq->slice_left = now - cfqq->slice_end; + else + cfqq->slice_left = 0; + + if (cfq_cfqq_on_rr(cfqq)) + cfq_resort_rr_list(cfqq, preempted); + + if (cfqq == cfqd->active_queue) + cfqd->active_queue = NULL; + + if (cfqd->active_cic) { + put_io_context(cfqd->active_cic->ioc); + cfqd->active_cic = NULL; + } + + cfqd->dispatch_slice = 0; +} + +static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted) +{ + struct cfq_queue *cfqq = cfqd->active_queue; + + if (cfqq) { + /* + * use deferred expiry, if there are requests in progress as + * not to disturb the slice of the next queue + */ + if (cfq_cfqq_dispatched(cfqq)) + cfq_mark_cfqq_expired(cfqq); + else + __cfq_slice_expired(cfqd, cfqq, preempted); + } +} + +static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) + +{ + WARN_ON(!RB_EMPTY(&cfqq->sort_list)); + WARN_ON(cfqq != cfqd->active_queue); + + /* + * idle is disabled, either manually or by past process history + */ + if (!cfqd->cfq_slice_idle) + return 0; + if (!cfq_cfqq_idle_window(cfqq)) + return 0; + /* + * task has exited, don't wait + */ + if (cfqd->active_cic && !cfqd->active_cic->ioc->task) + return 0; + + cfq_mark_cfqq_must_dispatch(cfqq); + cfq_mark_cfqq_wait_request(cfqq); + + if (!timer_pending(&cfqd->idle_slice_timer)) { + unsigned long slice_left = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle); + + cfqd->idle_slice_timer.expires = jiffies + slice_left; + add_timer(&cfqd->idle_slice_timer); + } + + return 1; } /* @@ -738,31 +999,40 @@ static void cfq_dispatch_sort(request_queue_t *q, struct cfq_rq *crq) struct request *__rq; sector_t last; - cfq_del_crq_rb(crq); - cfq_remove_merge_hints(q, crq); list_del(&crq->request->queuelist); last = cfqd->last_sector; - while ((entry = entry->prev) != head) { - __rq = list_entry_rq(entry); + list_for_each_entry_reverse(__rq, head, queuelist) { + struct cfq_rq *__crq = RQ_DATA(__rq); - if (blk_barrier_rq(crq->request)) + if (blk_barrier_rq(__rq)) + break; + if (!blk_fs_request(__rq)) break; - if (!blk_fs_request(crq->request)) + if (cfq_crq_requeued(__crq)) break; - if (crq->request->sector > __rq->sector) + if (__rq->sector <= crq->request->sector) break; if (__rq->sector > last && crq->request->sector < last) { - last = crq->request->sector; + last = crq->request->sector + crq->request->nr_sectors; break; } + entry = &__rq->queuelist; } cfqd->last_sector = last; - crq->in_flight = 1; - cfqq->in_flight++; - list_add(&crq->request->queuelist, entry); + + cfqq->next_crq = cfq_find_next_crq(cfqd, cfqq, crq); + + cfq_del_crq_rb(crq); + cfq_remove_merge_hints(q, crq); + + cfq_mark_crq_in_flight(crq); + cfq_clear_crq_requeued(crq); + + cfqq->on_dispatch[cfq_crq_is_sync(crq)]++; + list_add_tail(&crq->request->queuelist, entry); } /* @@ -771,173 +1041,225 @@ static void cfq_dispatch_sort(request_queue_t *q, struct cfq_rq *crq) static inline struct cfq_rq *cfq_check_fifo(struct cfq_queue *cfqq) { struct cfq_data *cfqd = cfqq->cfqd; - const int reads = !list_empty(&cfqq->fifo[0]); - const int writes = !list_empty(&cfqq->fifo[1]); - unsigned long now = jiffies; + struct request *rq; struct cfq_rq *crq; - if (time_before(now, cfqq->last_fifo_expire + cfqd->cfq_fifo_batch_expire)) + if (cfq_cfqq_fifo_expire(cfqq)) return NULL; - crq = RQ_DATA(list_entry(cfqq->fifo[0].next, struct request, queuelist)); - if (reads && time_after(now, crq->queue_start + cfqd->cfq_fifo_expire_r)) { - cfqq->last_fifo_expire = now; - return crq; - } + if (!list_empty(&cfqq->fifo)) { + int fifo = cfq_cfqq_class_sync(cfqq); - crq = RQ_DATA(list_entry(cfqq->fifo[1].next, struct request, queuelist)); - if (writes && time_after(now, crq->queue_start + cfqd->cfq_fifo_expire_w)) { - cfqq->last_fifo_expire = now; - return crq; + crq = RQ_DATA(list_entry_fifo(cfqq->fifo.next)); + rq = crq->request; + if (time_after(jiffies, rq->start_time + cfqd->cfq_fifo_expire[fifo])) { + cfq_mark_cfqq_fifo_expire(cfqq); + return crq; + } } return NULL; } /* - * dispatch a single request from given queue + * Scale schedule slice based on io priority. Use the sync time slice only + * if a queue is marked sync and has sync io queued. A sync queue with async + * io only, should not get full sync slice length. */ +static inline int +cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) +{ + const int base_slice = cfqd->cfq_slice[cfq_cfqq_sync(cfqq)]; + + WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); + + return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - cfqq->ioprio)); +} + static inline void -cfq_dispatch_request(request_queue_t *q, struct cfq_data *cfqd, - struct cfq_queue *cfqq) +cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) { - struct cfq_rq *crq; + cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies; +} + +static inline int +cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) +{ + const int base_rq = cfqd->cfq_slice_async_rq; + + WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); + + return 2 * (base_rq + base_rq * (CFQ_PRIO_LISTS - 1 - cfqq->ioprio)); +} + +/* + * get next queue for service + */ +static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd, int force) +{ + unsigned long now = jiffies; + struct cfq_queue *cfqq; + + cfqq = cfqd->active_queue; + if (!cfqq) + goto new_queue; + + if (cfq_cfqq_expired(cfqq)) + goto new_queue; /* - * follow expired path, else get first next available + * slice has expired */ - if ((crq = cfq_check_fifo(cfqq)) == NULL) { - if (cfqd->find_best_crq) - crq = cfqq->next_crq; - else - crq = rb_entry_crq(rb_first(&cfqq->sort_list)); - } - - cfqd->last_sector = crq->request->sector + crq->request->nr_sectors; + if (!cfq_cfqq_must_dispatch(cfqq) && time_after(now, cfqq->slice_end)) + goto expire; /* - * finally, insert request into driver list + * if queue has requests, dispatch one. if not, check if + * enough slice is left to wait for one */ - cfq_dispatch_sort(q, crq); + if (!RB_EMPTY(&cfqq->sort_list)) + goto keep_queue; + else if (!force && cfq_cfqq_class_sync(cfqq) && + time_before(now, cfqq->slice_end)) { + if (cfq_arm_slice_timer(cfqd, cfqq)) + return NULL; + } + +expire: + cfq_slice_expired(cfqd, 0); +new_queue: + cfqq = cfq_set_active_queue(cfqd); +keep_queue: + return cfqq; } -static int cfq_dispatch_requests(request_queue_t *q, int max_dispatch) +static int +__cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq, + int max_dispatch) { - struct cfq_data *cfqd = q->elevator->elevator_data; - struct cfq_queue *cfqq; - struct list_head *entry, *tmp; - int queued, busy_queues, first_round; + int dispatched = 0; - if (list_empty(&cfqd->rr_list)) - return 0; + BUG_ON(RB_EMPTY(&cfqq->sort_list)); - queued = 0; - first_round = 1; -restart: - busy_queues = 0; - list_for_each_safe(entry, tmp, &cfqd->rr_list) { - cfqq = list_entry_cfqq(entry); + do { + struct cfq_rq *crq; - BUG_ON(RB_EMPTY(&cfqq->sort_list)); + /* + * follow expired path, else get first next available + */ + if ((crq = cfq_check_fifo(cfqq)) == NULL) + crq = cfqq->next_crq; /* - * first round of queueing, only select from queues that - * don't already have io in-flight + * finally, insert request into driver dispatch list */ - if (first_round && cfqq->in_flight) - continue; + cfq_dispatch_sort(cfqd->queue, crq); + + cfqd->dispatch_slice++; + dispatched++; - cfq_dispatch_request(q, cfqd, cfqq); + if (!cfqd->active_cic) { + atomic_inc(&crq->io_context->ioc->refcount); + cfqd->active_cic = crq->io_context; + } - if (!RB_EMPTY(&cfqq->sort_list)) - busy_queues++; + if (RB_EMPTY(&cfqq->sort_list)) + break; - queued++; - } + } 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 (!cfqq->slice_end) + cfq_set_prio_slice(cfqd, cfqq); + + /* + * expire an async queue immediately if it has used up its slice. idle + * queue always expire after 1 dispatch round. + */ + if ((!cfq_cfqq_sync(cfqq) && + cfqd->dispatch_slice >= cfq_prio_to_maxrq(cfqd, cfqq)) || + cfq_class_idle(cfqq)) + cfq_slice_expired(cfqd, 0); + + return dispatched; +} + +static int +cfq_dispatch_requests(request_queue_t *q, int max_dispatch, int force) +{ + struct cfq_data *cfqd = q->elevator->elevator_data; + struct cfq_queue *cfqq; + + if (!cfqd->busy_queues) + return 0; + + cfqq = cfq_select_queue(cfqd, force); + if (cfqq) { + 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; - if ((queued < max_dispatch) && (busy_queues || first_round)) { - first_round = 0; - goto restart; + return __cfq_dispatch_requests(cfqd, cfqq, max_dispatch); } - return queued; + return 0; } static inline void cfq_account_dispatch(struct cfq_rq *crq) { struct cfq_queue *cfqq = crq->cfq_queue; struct cfq_data *cfqd = cfqq->cfqd; - unsigned long now, elapsed; - if (!blk_fs_request(crq->request)) + if (unlikely(!blk_fs_request(crq->request))) return; /* * accounted bit is necessary since some drivers will call * elv_next_request() many times for the same request (eg ide) */ - if (crq->accounted) + if (cfq_crq_in_driver(crq)) return; - now = jiffies; - if (cfqq->service_start == ~0UL) - cfqq->service_start = now; - - /* - * on drives with tagged command queueing, command turn-around time - * doesn't necessarily reflect the time spent processing this very - * command inside the drive. so do the accounting differently there, - * by just sorting on the number of requests - */ - if (cfqd->cfq_tagged) { - if (time_after(now, cfqq->service_start + cfq_service)) { - cfqq->service_start = now; - cfqq->service_used /= 10; - } - - cfqq->service_used++; - cfq_sort_rr_list(cfqq, 0); - } - - elapsed = now - crq->queue_start; - if (elapsed > max_elapsed_dispatch) - max_elapsed_dispatch = elapsed; - - crq->accounted = 1; - crq->service_start = now; - - if (++cfqd->rq_in_driver >= CFQ_MAX_TAG && !cfqd->cfq_tagged) { - cfqq->cfqd->cfq_tagged = 1; - printk("cfq: depth %d reached, tagging now on\n", CFQ_MAX_TAG); - } + cfq_mark_crq_in_driver(crq); + cfqd->rq_in_driver++; } static inline void cfq_account_completion(struct cfq_queue *cfqq, struct cfq_rq *crq) { struct cfq_data *cfqd = cfqq->cfqd; + unsigned long now; - if (!crq->accounted) + if (!cfq_crq_in_driver(crq)) return; + now = jiffies; + WARN_ON(!cfqd->rq_in_driver); cfqd->rq_in_driver--; - if (!cfqd->cfq_tagged) { - unsigned long now = jiffies; - unsigned long duration = now - crq->service_start; + if (!cfq_class_idle(cfqq)) + cfqd->last_end_request = now; - if (time_after(now, cfqq->service_start + cfq_service)) { - cfqq->service_start = now; - cfqq->service_used >>= 3; + if (!cfq_cfqq_dispatched(cfqq)) { + if (cfq_cfqq_on_rr(cfqq)) { + cfqq->service_last = now; + cfq_resort_rr_list(cfqq, 0); + } + if (cfq_cfqq_expired(cfqq)) { + __cfq_slice_expired(cfqd, cfqq, 0); + cfq_schedule_dispatch(cfqd); } - - cfqq->service_used += duration; - cfq_sort_rr_list(cfqq, 0); - - if (duration > max_elapsed_crq) - max_elapsed_crq = duration; } + + if (cfq_crq_is_sync(crq)) + crq->io_context->last_end_request = now; } static struct request *cfq_next_request(request_queue_t *q) @@ -950,7 +1272,18 @@ static struct request *cfq_next_request(request_queue_t *q) dispatch: rq = list_entry_rq(q->queue_head.next); - if ((crq = RQ_DATA(rq)) != NULL) { + crq = RQ_DATA(rq); + if (crq) { + struct cfq_queue *cfqq = crq->cfq_queue; + + /* + * if idle window is disabled, allow queue buildup + */ + if (!cfq_crq_in_driver(crq) && + !cfq_cfqq_idle_window(cfqq) && + cfqd->rq_in_driver >= cfqd->cfq_max_depth) + return NULL; + cfq_remove_merge_hints(q, crq); cfq_account_dispatch(crq); } @@ -958,7 +1291,7 @@ dispatch: return rq; } - if (cfq_dispatch_requests(q, cfqd->cfq_quantum)) + if (cfq_dispatch_requests(q, cfqd->cfq_quantum, 0)) goto dispatch; return NULL; @@ -972,13 +1305,21 @@ dispatch: */ static void cfq_put_queue(struct cfq_queue *cfqq) { - BUG_ON(!atomic_read(&cfqq->ref)); + struct cfq_data *cfqd = cfqq->cfqd; + + BUG_ON(atomic_read(&cfqq->ref) <= 0); if (!atomic_dec_and_test(&cfqq->ref)) return; BUG_ON(rb_first(&cfqq->sort_list)); - BUG_ON(cfqq->on_rr); + BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); + BUG_ON(cfq_cfqq_on_rr(cfqq)); + + if (unlikely(cfqd->active_queue == cfqq)) { + __cfq_slice_expired(cfqd, cfqq, 0); + cfq_schedule_dispatch(cfqd); + } cfq_put_cfqd(cfqq->cfqd); @@ -991,15 +1332,17 @@ static void cfq_put_queue(struct cfq_queue *cfqq) } static inline struct cfq_queue * -__cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned long key, const int hashval) +__cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio, + const int hashval) { struct hlist_head *hash_list = &cfqd->cfq_hash[hashval]; struct hlist_node *entry, *next; hlist_for_each_safe(entry, next, hash_list) { struct cfq_queue *__cfqq = list_entry_qhash(entry); + const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->ioprio_class, __cfqq->ioprio); - if (__cfqq->key == key) + if (__cfqq->key == key && (__p == prio || prio == CFQ_KEY_ANY)) return __cfqq; } @@ -1007,94 +1350,220 @@ __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned long key, const int hashval) } static struct cfq_queue * -cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned long key) +cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned short prio) { - return __cfq_find_cfq_hash(cfqd, key, hash_long(key, CFQ_QHASH_SHIFT)); + return __cfq_find_cfq_hash(cfqd, key, prio, hash_long(key, CFQ_QHASH_SHIFT)); } -static inline void -cfq_rehash_cfqq(struct cfq_data *cfqd, struct cfq_queue **cfqq, - struct cfq_io_context *cic) +static void cfq_free_io_context(struct cfq_io_context *cic) { - unsigned long hashkey = cfq_hash_key(cfqd, current); - unsigned long hashval = hash_long(hashkey, CFQ_QHASH_SHIFT); - struct cfq_queue *__cfqq; - unsigned long flags; + struct cfq_io_context *__cic; + struct list_head *entry, *next; - spin_lock_irqsave(cfqd->queue->queue_lock, flags); - - hlist_del(&(*cfqq)->cfq_hash); - - __cfqq = __cfq_find_cfq_hash(cfqd, hashkey, hashval); - if (!__cfqq || __cfqq == *cfqq) { - __cfqq = *cfqq; - hlist_add_head(&__cfqq->cfq_hash, &cfqd->cfq_hash[hashval]); - __cfqq->key_type = cfqd->key_type; - } else { - atomic_inc(&__cfqq->ref); - cic->cfqq = __cfqq; - cfq_put_queue(*cfqq); - *cfqq = __cfqq; + list_for_each_safe(entry, next, &cic->list) { + __cic = list_entry(entry, struct cfq_io_context, list); + kmem_cache_free(cfq_ioc_pool, __cic); } - cic->cfqq = __cfqq; - spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); + kmem_cache_free(cfq_ioc_pool, cic); } -static void cfq_free_io_context(struct cfq_io_context *cic) +/* + * Called with interrupts disabled + */ +static void cfq_exit_single_io_context(struct cfq_io_context *cic) { - kmem_cache_free(cfq_ioc_pool, cic); + struct cfq_data *cfqd = cic->cfqq->cfqd; + request_queue_t *q = cfqd->queue; + + WARN_ON(!irqs_disabled()); + + spin_lock(q->queue_lock); + + if (unlikely(cic->cfqq == cfqd->active_queue)) { + __cfq_slice_expired(cfqd, cic->cfqq, 0); + cfq_schedule_dispatch(cfqd); + } + + cfq_put_queue(cic->cfqq); + cic->cfqq = NULL; + spin_unlock(q->queue_lock); } /* - * locking hierarchy is: io_context lock -> queue locks + * Another task may update the task cic list, if it is doing a queue lookup + * on its behalf. cfq_cic_lock excludes such concurrent updates */ static void cfq_exit_io_context(struct cfq_io_context *cic) { - struct cfq_queue *cfqq = cic->cfqq; - struct list_head *entry = &cic->list; - request_queue_t *q; + struct cfq_io_context *__cic; + struct list_head *entry; unsigned long flags; + local_irq_save(flags); + /* * put the reference this task is holding to the various queues */ - spin_lock_irqsave(&cic->ioc->lock, flags); - while ((entry = cic->list.next) != &cic->list) { - struct cfq_io_context *__cic; - + list_for_each(entry, &cic->list) { __cic = list_entry(entry, struct cfq_io_context, list); - list_del(entry); - - q = __cic->cfqq->cfqd->queue; - spin_lock(q->queue_lock); - cfq_put_queue(__cic->cfqq); - spin_unlock(q->queue_lock); + cfq_exit_single_io_context(__cic); } - q = cfqq->cfqd->queue; - spin_lock(q->queue_lock); - cfq_put_queue(cfqq); - spin_unlock(q->queue_lock); - - cic->cfqq = NULL; - spin_unlock_irqrestore(&cic->ioc->lock, flags); + cfq_exit_single_io_context(cic); + local_irq_restore(flags); } -static struct cfq_io_context *cfq_alloc_io_context(int gfp_flags) +static struct cfq_io_context * +cfq_alloc_io_context(struct cfq_data *cfqd, int gfp_mask) { - struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_flags); + struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_mask); if (cic) { - cic->dtor = cfq_free_io_context; - cic->exit = cfq_exit_io_context; INIT_LIST_HEAD(&cic->list); cic->cfqq = NULL; + cic->key = NULL; + cic->last_end_request = jiffies; + cic->ttime_total = 0; + cic->ttime_samples = 0; + cic->ttime_mean = 0; + cic->dtor = cfq_free_io_context; + cic->exit = cfq_exit_io_context; } return cic; } +static void cfq_init_prio_data(struct cfq_queue *cfqq) +{ + struct task_struct *tsk = current; + int ioprio_class; + + if (!cfq_cfqq_prio_changed(cfqq)) + return; + + ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio); + switch (ioprio_class) { + default: + printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); + case IOPRIO_CLASS_NONE: + /* + * no prio set, place us in the middle of the BE classes + */ + cfqq->ioprio = task_nice_ioprio(tsk); + cfqq->ioprio_class = IOPRIO_CLASS_BE; + break; + case IOPRIO_CLASS_RT: + cfqq->ioprio = task_ioprio(tsk); + cfqq->ioprio_class = IOPRIO_CLASS_RT; + break; + case IOPRIO_CLASS_BE: + cfqq->ioprio = task_ioprio(tsk); + cfqq->ioprio_class = IOPRIO_CLASS_BE; + break; + case IOPRIO_CLASS_IDLE: + cfqq->ioprio_class = IOPRIO_CLASS_IDLE; + cfqq->ioprio = 7; + cfq_clear_cfqq_idle_window(cfqq); + break; + } + + /* + * keep track of original prio settings in case we have to temporarily + * elevate the priority of this queue + */ + cfqq->org_ioprio = cfqq->ioprio; + cfqq->org_ioprio_class = cfqq->ioprio_class; + + if (cfq_cfqq_on_rr(cfqq)) + cfq_resort_rr_list(cfqq, 0); + + cfq_clear_cfqq_prio_changed(cfqq); +} + +static inline void changed_ioprio(struct cfq_queue *cfqq) +{ + if (cfqq) { + struct cfq_data *cfqd = cfqq->cfqd; + + spin_lock(cfqd->queue->queue_lock); + cfq_mark_cfqq_prio_changed(cfqq); + cfq_init_prio_data(cfqq); + spin_unlock(cfqd->queue->queue_lock); + } +} + +/* + * callback from sys_ioprio_set, irqs are disabled + */ +static int cfq_ioc_set_ioprio(struct io_context *ioc, unsigned int ioprio) +{ + struct cfq_io_context *cic = ioc->cic; + + changed_ioprio(cic->cfqq); + + list_for_each_entry(cic, &cic->list, list) + changed_ioprio(cic->cfqq); + + return 0; +} + +static struct cfq_queue * +cfq_get_queue(struct cfq_data *cfqd, unsigned int key, unsigned short ioprio, + int gfp_mask) +{ + const int hashval = hash_long(key, CFQ_QHASH_SHIFT); + struct cfq_queue *cfqq, *new_cfqq = NULL; + +retry: + cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval); + + if (!cfqq) { + if (new_cfqq) { + cfqq = new_cfqq; + new_cfqq = NULL; + } else if (gfp_mask & __GFP_WAIT) { + spin_unlock_irq(cfqd->queue->queue_lock); + new_cfqq = kmem_cache_alloc(cfq_pool, gfp_mask); + spin_lock_irq(cfqd->queue->queue_lock); + goto retry; + } else { + cfqq = kmem_cache_alloc(cfq_pool, gfp_mask); + if (!cfqq) + goto out; + } + + memset(cfqq, 0, sizeof(*cfqq)); + + 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; + hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]); + atomic_set(&cfqq->ref, 0); + cfqq->cfqd = cfqd; + atomic_inc(&cfqd->ref); + cfqq->service_last = 0; + /* + * set ->slice_left to allow preemption for a new process + */ + cfqq->slice_left = 2 * cfqd->cfq_slice_idle; + cfq_mark_cfqq_idle_window(cfqq); + cfq_mark_cfqq_prio_changed(cfqq); + cfq_init_prio_data(cfqq); + } + + if (new_cfqq) + kmem_cache_free(cfq_pool, new_cfqq); + + atomic_inc(&cfqq->ref); +out: + WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); + return cfqq; +} + /* * Setup general io context and cfq io context. There can be several cfq * io contexts per general io context, if this process is doing io to more @@ -1102,39 +1571,39 @@ static struct cfq_io_context *cfq_alloc_io_context(int gfp_flags) * cfqq, so we don't need to worry about it disappearing */ static struct cfq_io_context * -cfq_get_io_context(struct cfq_queue **cfqq, int gfp_flags) +cfq_get_io_context(struct cfq_data *cfqd, pid_t pid, int gfp_mask) { - struct cfq_data *cfqd = (*cfqq)->cfqd; - struct cfq_queue *__cfqq = *cfqq; + struct io_context *ioc = NULL; struct cfq_io_context *cic; - struct io_context *ioc; - might_sleep_if(gfp_flags & __GFP_WAIT); + might_sleep_if(gfp_mask & __GFP_WAIT); - ioc = get_io_context(gfp_flags); + ioc = get_io_context(gfp_mask); if (!ioc) return NULL; if ((cic = ioc->cic) == NULL) { - cic = cfq_alloc_io_context(gfp_flags); + cic = cfq_alloc_io_context(cfqd, gfp_mask); 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->cfqq = __cfqq; - atomic_inc(&__cfqq->ref); + cic->key = cfqd; + atomic_inc(&cfqd->ref); } else { struct cfq_io_context *__cic; - unsigned long flags; /* - * since the first cic on the list is actually the head - * itself, need to check this here or we'll duplicate an - * cic per ioc for no reason + * the first cic on the list is actually the head itself */ - if (cic->cfqq == __cfqq) + if (cic->key == cfqd) goto out; /* @@ -1142,152 +1611,250 @@ cfq_get_io_context(struct cfq_queue **cfqq, int gfp_flags) * should be ok here, the list will usually not be more than * 1 or a few entries long */ - spin_lock_irqsave(&ioc->lock, flags); 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->cfqq == __cfqq) { + if (__cic->key == cfqd) { cic = __cic; - spin_unlock_irqrestore(&ioc->lock, flags); goto out; } } - spin_unlock_irqrestore(&ioc->lock, flags); /* * 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(gfp_flags); + __cic = cfq_alloc_io_context(cfqd, gfp_mask); if (__cic == NULL) goto err; __cic->ioc = ioc; - __cic->cfqq = __cfqq; - atomic_inc(&__cfqq->ref); - spin_lock_irqsave(&ioc->lock, flags); + __cic->key = cfqd; + atomic_inc(&cfqd->ref); list_add(&__cic->list, &cic->list); - spin_unlock_irqrestore(&ioc->lock, flags); - cic = __cic; - *cfqq = __cfqq; } out: - /* - * if key_type has been changed on the fly, we lazily rehash - * each queue at lookup time - */ - if ((*cfqq)->key_type != cfqd->key_type) - cfq_rehash_cfqq(cfqd, cfqq, cic); - return cic; err: put_io_context(ioc); return NULL; } -static struct cfq_queue * -__cfq_get_queue(struct cfq_data *cfqd, unsigned long key, int gfp_mask) +static void +cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic) { - const int hashval = hash_long(key, CFQ_QHASH_SHIFT); - struct cfq_queue *cfqq, *new_cfqq = NULL; + unsigned long elapsed, ttime; -retry: - cfqq = __cfq_find_cfq_hash(cfqd, key, hashval); - - if (!cfqq) { - if (new_cfqq) { - cfqq = new_cfqq; - new_cfqq = NULL; - } else { - spin_unlock_irq(cfqd->queue->queue_lock); - new_cfqq = kmem_cache_alloc(cfq_pool, gfp_mask); - spin_lock_irq(cfqd->queue->queue_lock); + /* + * 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 - if (!new_cfqq && !(gfp_mask & __GFP_WAIT)) - goto out; + ttime = min(elapsed, 2UL * cfqd->cfq_slice_idle); - goto retry; - } + 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; +} - memset(cfqq, 0, sizeof(*cfqq)); +#define sample_valid(samples) ((samples) > 80) - INIT_HLIST_NODE(&cfqq->cfq_hash); - INIT_LIST_HEAD(&cfqq->cfq_list); - RB_CLEAR_ROOT(&cfqq->sort_list); - INIT_LIST_HEAD(&cfqq->fifo[0]); - INIT_LIST_HEAD(&cfqq->fifo[1]); +/* + * 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); - cfqq->key = key; - hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]); - atomic_set(&cfqq->ref, 0); - cfqq->cfqd = cfqd; - atomic_inc(&cfqd->ref); - cfqq->key_type = cfqd->key_type; - cfqq->service_start = ~0UL; + 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 (new_cfqq) - kmem_cache_free(cfq_pool, new_cfqq); + if (enable_idle) + cfq_mark_cfqq_idle_window(cfqq); + else + cfq_clear_cfqq_idle_window(cfqq); +} - atomic_inc(&cfqq->ref); -out: - WARN_ON((gfp_mask & __GFP_WAIT) && !cfqq); - return 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); } -static void cfq_enqueue(struct cfq_data *cfqd, struct cfq_rq *crq) +/* + * 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) { - crq->is_sync = 0; - if (rq_data_dir(crq->request) == READ || current->flags & PF_SYNCWRITE) - crq->is_sync = 1; + const int sync = cfq_crq_is_sync(crq); + + cfqq->next_crq = cfq_choose_req(cfqd, cfqq->next_crq, crq); + + if (sync) { + struct cfq_io_context *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); - crq->queue_start = jiffies; - list_add_tail(&crq->request->queuelist, &crq->cfq_queue->fifo[crq->is_sync]); + 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; - struct cfq_rq *crq = RQ_DATA(rq); switch (where) { case ELEVATOR_INSERT_BACK: - while (cfq_dispatch_requests(q, cfqd->cfq_quantum)) + 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, crq); + cfq_enqueue(cfqd, rq); break; default: printk("%s: bad insert point %d\n", __FUNCTION__,where); return; } - - if (rq_mergeable(rq)) { - cfq_add_crq_hash(cfqd, crq); - - if (!q->last_merge) - q->last_merge = rq; - } -} - -static int cfq_queue_empty(request_queue_t *q) -{ - struct cfq_data *cfqd = q->elevator->elevator_data; - - return list_empty(&q->queue_head) && list_empty(&cfqd->rr_list); } static void cfq_completed_request(request_queue_t *q, struct request *rq) @@ -1300,9 +1867,11 @@ static void cfq_completed_request(request_queue_t *q, struct request *rq) cfqq = crq->cfq_queue; - if (crq->in_flight) { - WARN_ON(!cfqq->in_flight); - cfqq->in_flight--; + 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); @@ -1332,51 +1901,136 @@ cfq_latter_request(request_queue_t *q, struct request *rq) return NULL; } -static int cfq_may_queue(request_queue_t *q, int rw) +/* + * 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) { - struct cfq_data *cfqd = q->elevator->elevator_data; - struct cfq_queue *cfqq; - int ret = ELV_MQUEUE_MAY; + const int ioprio_class = cfqq->ioprio_class; + const int ioprio = cfqq->ioprio; - if (current->flags & PF_MEMALLOC) - return ELV_MQUEUE_MAY; + 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; + } - cfqq = cfq_find_cfq_hash(cfqd, cfq_hash_key(cfqd, current)); - if (cfqq) { - int limit = cfqd->max_queued; + /* + * 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); +} - if (cfqq->allocated[rw] < cfqd->cfq_queued) - return ELV_MQUEUE_MUST; +static inline pid_t cfq_queue_pid(struct task_struct *task, int rw) +{ + if (rw == READ || process_sync(task)) + return task->pid; - if (cfqd->busy_queues) - limit = q->nr_requests / cfqd->busy_queues; + return CFQ_KEY_ASYNC; +} - if (limit < cfqd->cfq_queued) - limit = cfqd->cfq_queued; - else if (limit > cfqd->max_queued) - limit = cfqd->max_queued; +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; + } - if (cfqq->allocated[rw] >= limit) { - if (limit > cfqq->alloc_limit[rw]) - cfqq->alloc_limit[rw] = limit; + 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; - ret = ELV_MQUEUE_NO; + /* + * 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; - return ret; + 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; - const int write = waitqueue_active(&rl->wait[WRITE]); - const int read = waitqueue_active(&rl->wait[READ]); - if (read && cfqq->allocated[READ] < cfqq->alloc_limit[READ]) - wake_up(&rl->wait[READ]); - if (write && cfqq->allocated[WRITE] < cfqq->alloc_limit[WRITE]) - wake_up(&rl->wait[WRITE]); + 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]); + } } /* @@ -1389,69 +2043,61 @@ static void cfq_put_request(request_queue_t *q, struct request *rq) if (crq) { struct cfq_queue *cfqq = crq->cfq_queue; + const int rw = rq_data_dir(rq); - BUG_ON(q->last_merge == rq); - BUG_ON(!hlist_unhashed(&crq->hash)); - - if (crq->io_context) - put_io_context(crq->io_context->ioc); + BUG_ON(!cfqq->allocated[rw]); + cfqq->allocated[rw]--; - BUG_ON(!cfqq->allocated[crq->is_write]); - cfqq->allocated[crq->is_write]--; + put_io_context(crq->io_context->ioc); mempool_free(crq, cfqd->crq_pool); rq->elevator_private = NULL; - smp_mb(); cfq_check_waiters(q, cfqq); cfq_put_queue(cfqq); } } /* - * Allocate cfq data structures associated with this request. A queue and + * Allocate cfq data structures associated with this request. */ -static int cfq_set_request(request_queue_t *q, struct request *rq, int gfp_mask) +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); - struct cfq_queue *cfqq, *saved_cfqq; + 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); - cfqq = __cfq_get_queue(cfqd, cfq_hash_key(cfqd, current), gfp_mask); - if (!cfqq) - goto out_lock; + if (!cic) + goto queue_fail; -repeat: - if (cfqq->allocated[rw] >= cfqd->max_queued) - goto out_lock; + 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); - /* - * if hashing type has changed, the cfq_queue might change here. - */ - saved_cfqq = cfqq; - cic = cfq_get_io_context(&cfqq, gfp_mask); - if (!cic) - goto err; - - /* - * repeat allocation checks on queue change - */ - if (unlikely(saved_cfqq != cfqq)) { - spin_lock_irqsave(q->queue_lock, flags); - saved_cfqq->allocated[rw]--; - goto repeat; - } - crq = mempool_alloc(cfqd->crq_pool, gfp_mask); if (crq) { RB_CLEAR(&crq->rb_node); @@ -1460,24 +2106,141 @@ repeat: INIT_HLIST_NODE(&crq->hash); crq->cfq_queue = cfqq; crq->io_context = cic; - crq->service_start = crq->queue_start = 0; - crq->in_flight = crq->accounted = crq->is_sync = 0; - crq->is_write = rw; + cfq_clear_crq_in_flight(crq); + cfq_clear_crq_in_driver(crq); + cfq_clear_crq_requeued(crq); + + if (rw == READ || process_sync(tsk)) + cfq_mark_crq_is_sync(crq); + else + cfq_clear_crq_is_sync(crq); + rq->elevator_private = crq; - cfqq->alloc_limit[rw] = 0; return 0; } - put_io_context(cic->ioc); -err: spin_lock_irqsave(q->queue_lock, flags); cfqq->allocated[rw]--; + if (!(cfqq->allocated[0] + cfqq->allocated[1])) + cfq_mark_cfqq_must_alloc(cfqq); cfq_put_queue(cfqq); -out_lock: +queue_fail: + if (cic) + put_io_context(cic->ioc); + /* + * mark us rq allocation starved. we need to kickstart the process + * ourselves if there are no pending requests that can do it for us. + * that would be an extremely rare OOM situation + */ + cfqd->rq_starved = 1; + cfq_schedule_dispatch(cfqd); spin_unlock_irqrestore(q->queue_lock, flags); return 1; } +static void cfq_kick_queue(void *data) +{ + request_queue_t *q = data; + struct cfq_data *cfqd = q->elevator->elevator_data; + unsigned long flags; + + spin_lock_irqsave(q->queue_lock, flags); + + if (cfqd->rq_starved) { + struct request_list *rl = &q->rq; + + /* + * we aren't guaranteed to get a request after this, but we + * have to be opportunistic + */ + smp_mb(); + if (waitqueue_active(&rl->wait[READ])) + wake_up(&rl->wait[READ]); + if (waitqueue_active(&rl->wait[WRITE])) + wake_up(&rl->wait[WRITE]); + } + + blk_remove_plug(q); + q->request_fn(q); + spin_unlock_irqrestore(q->queue_lock, flags); +} + +/* + * Timer running if the active_queue is currently idling inside its time slice + */ +static void cfq_idle_slice_timer(unsigned long data) +{ + struct cfq_data *cfqd = (struct cfq_data *) data; + struct cfq_queue *cfqq; + unsigned long flags; + + spin_lock_irqsave(cfqd->queue->queue_lock, flags); + + if ((cfqq = cfqd->active_queue) != NULL) { + unsigned long now = jiffies; + + /* + * expired + */ + if (time_after(now, cfqq->slice_end)) + goto expire; + + /* + * only expire and reinvoke request handler, if there are + * other queues with pending requests + */ + if (!cfq_pending_requests(cfqd)) { + cfqd->idle_slice_timer.expires = min(now + cfqd->cfq_slice_idle, cfqq->slice_end); + add_timer(&cfqd->idle_slice_timer); + goto out_cont; + } + + /* + * not expired and it has a request pending, let it dispatch + */ + if (!RB_EMPTY(&cfqq->sort_list)) { + cfq_mark_cfqq_must_dispatch(cfqq); + goto out_kick; + } + } +expire: + cfq_slice_expired(cfqd, 0); +out_kick: + cfq_schedule_dispatch(cfqd); +out_cont: + spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); +} + +/* + * Timer running if an idle class queue is waiting for service + */ +static void cfq_idle_class_timer(unsigned long data) +{ + struct cfq_data *cfqd = (struct cfq_data *) data; + unsigned long flags, end; + + spin_lock_irqsave(cfqd->queue->queue_lock, flags); + + /* + * 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 + cfq_schedule_dispatch(cfqd); + + spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); +} + +static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) +{ + del_timer_sync(&cfqd->idle_slice_timer); + del_timer_sync(&cfqd->idle_class_timer); + blk_sync_queue(cfqd->queue); +} + static void cfq_put_cfqd(struct cfq_data *cfqd) { request_queue_t *q = cfqd->queue; @@ -1487,6 +2250,9 @@ static void cfq_put_cfqd(struct cfq_data *cfqd) blk_put_queue(q); + cfq_shutdown_timer_wq(cfqd); + q->elevator->elevator_data = NULL; + mempool_destroy(cfqd->crq_pool); kfree(cfqd->crq_hash); kfree(cfqd->cfq_hash); @@ -1495,7 +2261,10 @@ static void cfq_put_cfqd(struct cfq_data *cfqd) static void cfq_exit_queue(elevator_t *e) { - cfq_put_cfqd(e->elevator_data); + struct cfq_data *cfqd = e->elevator_data; + + cfq_shutdown_timer_wq(cfqd); + cfq_put_cfqd(cfqd); } static int cfq_init_queue(request_queue_t *q, elevator_t *e) @@ -1508,7 +2277,13 @@ static int cfq_init_queue(request_queue_t *q, elevator_t *e) return -ENOMEM; memset(cfqd, 0, sizeof(*cfqd)); - INIT_LIST_HEAD(&cfqd->rr_list); + + for (i = 0; i < CFQ_PRIO_LISTS; i++) + INIT_LIST_HEAD(&cfqd->rr_list[i]); + + INIT_LIST_HEAD(&cfqd->busy_rr); + INIT_LIST_HEAD(&cfqd->cur_rr); + INIT_LIST_HEAD(&cfqd->idle_rr); INIT_LIST_HEAD(&cfqd->empty_list); cfqd->crq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_MHASH_ENTRIES, GFP_KERNEL); @@ -1533,24 +2308,32 @@ static int cfq_init_queue(request_queue_t *q, elevator_t *e) cfqd->queue = q; atomic_inc(&q->refcnt); - /* - * just set it to some high value, we want anyone to be able to queue - * some requests. fairness is handled differently - */ - q->nr_requests = 1024; - cfqd->max_queued = q->nr_requests / 16; + cfqd->max_queued = q->nr_requests / 4; q->nr_batching = cfq_queued; - cfqd->key_type = CFQ_KEY_TGID; - cfqd->find_best_crq = 1; + + init_timer(&cfqd->idle_slice_timer); + cfqd->idle_slice_timer.function = cfq_idle_slice_timer; + cfqd->idle_slice_timer.data = (unsigned long) cfqd; + + init_timer(&cfqd->idle_class_timer); + cfqd->idle_class_timer.function = cfq_idle_class_timer; + cfqd->idle_class_timer.data = (unsigned long) cfqd; + + INIT_WORK(&cfqd->unplug_work, cfq_kick_queue, q); + atomic_set(&cfqd->ref, 1); cfqd->cfq_queued = cfq_queued; cfqd->cfq_quantum = cfq_quantum; - cfqd->cfq_fifo_expire_r = cfq_fifo_expire_r; - cfqd->cfq_fifo_expire_w = cfq_fifo_expire_w; - cfqd->cfq_fifo_batch_expire = cfq_fifo_rate; + cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; + cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; cfqd->cfq_back_max = cfq_back_max; cfqd->cfq_back_penalty = cfq_back_penalty; + cfqd->cfq_slice[0] = cfq_slice_async; + cfqd->cfq_slice[1] = cfq_slice_sync; + cfqd->cfq_slice_async_rq = cfq_slice_async_rq; + cfqd->cfq_slice_idle = cfq_slice_idle; + cfqd->cfq_max_depth = cfq_max_depth; return 0; out_crqpool: @@ -1595,7 +2378,6 @@ fail: return -ENOMEM; } - /* * sysfs parts below --> */ @@ -1620,45 +2402,6 @@ cfq_var_store(unsigned int *var, const char *page, size_t count) return count; } -static ssize_t -cfq_clear_elapsed(struct cfq_data *cfqd, const char *page, size_t count) -{ - max_elapsed_dispatch = max_elapsed_crq = 0; - return count; -} - -static ssize_t -cfq_set_key_type(struct cfq_data *cfqd, const char *page, size_t count) -{ - spin_lock_irq(cfqd->queue->queue_lock); - if (!strncmp(page, "pgid", 4)) - cfqd->key_type = CFQ_KEY_PGID; - else if (!strncmp(page, "tgid", 4)) - cfqd->key_type = CFQ_KEY_TGID; - else if (!strncmp(page, "uid", 3)) - cfqd->key_type = CFQ_KEY_UID; - else if (!strncmp(page, "gid", 3)) - cfqd->key_type = CFQ_KEY_GID; - spin_unlock_irq(cfqd->queue->queue_lock); - return count; -} - -static ssize_t -cfq_read_key_type(struct cfq_data *cfqd, char *page) -{ - ssize_t len = 0; - int i; - - for (i = CFQ_KEY_PGID; i < CFQ_KEY_LAST; i++) { - if (cfqd->key_type == i) - len += sprintf(page+len, "[%s] ", cfq_key_types[i]); - else - len += sprintf(page+len, "%s ", cfq_key_types[i]); - } - len += sprintf(page+len, "\n"); - return len; -} - #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ static ssize_t __FUNC(struct cfq_data *cfqd, char *page) \ { \ @@ -1669,12 +2412,15 @@ static ssize_t __FUNC(struct cfq_data *cfqd, char *page) \ } SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); SHOW_FUNCTION(cfq_queued_show, cfqd->cfq_queued, 0); -SHOW_FUNCTION(cfq_fifo_expire_r_show, cfqd->cfq_fifo_expire_r, 1); -SHOW_FUNCTION(cfq_fifo_expire_w_show, cfqd->cfq_fifo_expire_w, 1); -SHOW_FUNCTION(cfq_fifo_batch_expire_show, cfqd->cfq_fifo_batch_expire, 1); -SHOW_FUNCTION(cfq_find_best_show, cfqd->find_best_crq, 0); +SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); +SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); SHOW_FUNCTION(cfq_back_max_show, cfqd->cfq_back_max, 0); SHOW_FUNCTION(cfq_back_penalty_show, cfqd->cfq_back_penalty, 0); +SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); +SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); +SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); +SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); +SHOW_FUNCTION(cfq_max_depth_show, cfqd->cfq_max_depth, 0); #undef SHOW_FUNCTION #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ @@ -1694,12 +2440,15 @@ static ssize_t __FUNC(struct cfq_data *cfqd, const char *page, size_t count) \ } STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); STORE_FUNCTION(cfq_queued_store, &cfqd->cfq_queued, 1, UINT_MAX, 0); -STORE_FUNCTION(cfq_fifo_expire_r_store, &cfqd->cfq_fifo_expire_r, 1, UINT_MAX, 1); -STORE_FUNCTION(cfq_fifo_expire_w_store, &cfqd->cfq_fifo_expire_w, 1, UINT_MAX, 1); -STORE_FUNCTION(cfq_fifo_batch_expire_store, &cfqd->cfq_fifo_batch_expire, 0, UINT_MAX, 1); -STORE_FUNCTION(cfq_find_best_store, &cfqd->find_best_crq, 0, 1, 0); +STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, UINT_MAX, 1); +STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, UINT_MAX, 1); STORE_FUNCTION(cfq_back_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); STORE_FUNCTION(cfq_back_penalty_store, &cfqd->cfq_back_penalty, 1, UINT_MAX, 0); +STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); +STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); +STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); +STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX, 0); +STORE_FUNCTION(cfq_max_depth_store, &cfqd->cfq_max_depth, 1, UINT_MAX, 0); #undef STORE_FUNCTION static struct cfq_fs_entry cfq_quantum_entry = { @@ -1712,25 +2461,15 @@ static struct cfq_fs_entry cfq_queued_entry = { .show = cfq_queued_show, .store = cfq_queued_store, }; -static struct cfq_fs_entry cfq_fifo_expire_r_entry = { +static struct cfq_fs_entry cfq_fifo_expire_sync_entry = { .attr = {.name = "fifo_expire_sync", .mode = S_IRUGO | S_IWUSR }, - .show = cfq_fifo_expire_r_show, - .store = cfq_fifo_expire_r_store, + .show = cfq_fifo_expire_sync_show, + .store = cfq_fifo_expire_sync_store, }; -static struct cfq_fs_entry cfq_fifo_expire_w_entry = { +static struct cfq_fs_entry cfq_fifo_expire_async_entry = { .attr = {.name = "fifo_expire_async", .mode = S_IRUGO | S_IWUSR }, - .show = cfq_fifo_expire_w_show, - .store = cfq_fifo_expire_w_store, -}; -static struct cfq_fs_entry cfq_fifo_batch_expire_entry = { - .attr = {.name = "fifo_batch_expire", .mode = S_IRUGO | S_IWUSR }, - .show = cfq_fifo_batch_expire_show, - .store = cfq_fifo_batch_expire_store, -}; -static struct cfq_fs_entry cfq_find_best_entry = { - .attr = {.name = "find_best_crq", .mode = S_IRUGO | S_IWUSR }, - .show = cfq_find_best_show, - .store = cfq_find_best_store, + .show = cfq_fifo_expire_async_show, + .store = cfq_fifo_expire_async_store, }; static struct cfq_fs_entry cfq_back_max_entry = { .attr = {.name = "back_seek_max", .mode = S_IRUGO | S_IWUSR }, @@ -1742,27 +2481,44 @@ static struct cfq_fs_entry cfq_back_penalty_entry = { .show = cfq_back_penalty_show, .store = cfq_back_penalty_store, }; -static struct cfq_fs_entry cfq_clear_elapsed_entry = { - .attr = {.name = "clear_elapsed", .mode = S_IWUSR }, - .store = cfq_clear_elapsed, +static struct cfq_fs_entry cfq_slice_sync_entry = { + .attr = {.name = "slice_sync", .mode = S_IRUGO | S_IWUSR }, + .show = cfq_slice_sync_show, + .store = cfq_slice_sync_store, +}; +static struct cfq_fs_entry cfq_slice_async_entry = { + .attr = {.name = "slice_async", .mode = S_IRUGO | S_IWUSR }, + .show = cfq_slice_async_show, + .store = cfq_slice_async_store, +}; +static struct cfq_fs_entry cfq_slice_async_rq_entry = { + .attr = {.name = "slice_async_rq", .mode = S_IRUGO | S_IWUSR }, + .show = cfq_slice_async_rq_show, + .store = cfq_slice_async_rq_store, }; -static struct cfq_fs_entry cfq_key_type_entry = { - .attr = {.name = "key_type", .mode = S_IRUGO | S_IWUSR }, - .show = cfq_read_key_type, - .store = cfq_set_key_type, +static struct cfq_fs_entry cfq_slice_idle_entry = { + .attr = {.name = "slice_idle", .mode = S_IRUGO | S_IWUSR }, + .show = cfq_slice_idle_show, + .store = cfq_slice_idle_store, +}; +static struct cfq_fs_entry cfq_max_depth_entry = { + .attr = {.name = "max_depth", .mode = S_IRUGO | S_IWUSR }, + .show = cfq_max_depth_show, + .store = cfq_max_depth_store, }; static struct attribute *default_attrs[] = { &cfq_quantum_entry.attr, &cfq_queued_entry.attr, - &cfq_fifo_expire_r_entry.attr, - &cfq_fifo_expire_w_entry.attr, - &cfq_fifo_batch_expire_entry.attr, - &cfq_key_type_entry.attr, - &cfq_find_best_entry.attr, + &cfq_fifo_expire_sync_entry.attr, + &cfq_fifo_expire_async_entry.attr, &cfq_back_max_entry.attr, &cfq_back_penalty_entry.attr, - &cfq_clear_elapsed_entry.attr, + &cfq_slice_sync_entry.attr, + &cfq_slice_async_entry.attr, + &cfq_slice_async_rq_entry.attr, + &cfq_slice_idle_entry.attr, + &cfq_max_depth_entry.attr, NULL, }; @@ -1775,7 +2531,7 @@ cfq_attr_show(struct kobject *kobj, struct attribute *attr, char *page) struct cfq_fs_entry *entry = to_cfq(attr); if (!entry->show) - return 0; + return -EIO; return entry->show(e->elevator_data, page); } @@ -1788,7 +2544,7 @@ cfq_attr_store(struct kobject *kobj, struct attribute *attr, struct cfq_fs_entry *entry = to_cfq(attr); if (!entry->store) - return -EINVAL; + return -EIO; return entry->store(e->elevator_data, page, length); } @@ -1832,21 +2588,46 @@ static int __init cfq_init(void) { int ret; + /* + * could be 0 on HZ < 1000 setups + */ + if (!cfq_slice_async) + cfq_slice_async = 1; + if (!cfq_slice_idle) + cfq_slice_idle = 1; + if (cfq_slab_setup()) return -ENOMEM; ret = elv_register(&iosched_cfq); - if (!ret) { - __module_get(THIS_MODULE); - return 0; - } + if (ret) + cfq_slab_kill(); - cfq_slab_kill(); return ret; } static void __exit cfq_exit(void) { + struct task_struct *g, *p; + unsigned long flags; + + read_lock_irqsave(&tasklist_lock, flags); + + /* + * iterate each process in the system, removing our io_context + */ + do_each_thread(g, p) { + struct io_context *ioc = p->io_context; + + if (ioc && ioc->cic) { + ioc->cic->exit(ioc->cic); + cfq_free_io_context(ioc->cic); + ioc->cic = NULL; + } + } while_each_thread(g, p); + + read_unlock_irqrestore(&tasklist_lock, flags); + cfq_slab_kill(); elv_unregister(&iosched_cfq); } |