BTW, don't enable CONFIG_IOPRIO_WRITE yet, it's broken so far. The patch is against current BK. ===== arch/i386/kernel/entry.S 1.88 vs edited ===== Index: linux-2.6.10-ck2/arch/i386/kernel/entry.S =================================================================== --- linux-2.6.10-ck2.orig/arch/i386/kernel/entry.S 2005-01-04 09:30:09.042779624 +1100 +++ linux-2.6.10-ck2/arch/i386/kernel/entry.S 2005-01-04 09:31:32.258128968 +1100 @@ -864,5 +864,7 @@ .long sys_add_key .long sys_request_key .long sys_keyctl + .long sys_ioprio_set + .long sys_ioprio_get /* 290 */ syscall_table_size=(.-sys_call_table) Index: linux-2.6.10-ck2/arch/ia64/kernel/entry.S =================================================================== --- linux-2.6.10-ck2.orig/arch/ia64/kernel/entry.S 2004-12-25 10:14:46.000000000 +1100 +++ linux-2.6.10-ck2/arch/ia64/kernel/entry.S 2005-01-04 09:31:32.259128816 +1100 @@ -1531,8 +1531,8 @@ data8 sys_add_key data8 sys_request_key data8 sys_keyctl - data8 sys_ni_syscall - data8 sys_ni_syscall // 1275 + data8 sys_ioprio_set + data8 sys_ioprio_get // 1275 data8 sys_ni_syscall data8 sys_ni_syscall data8 sys_ni_syscall Index: linux-2.6.10-ck2/arch/ppc/kernel/misc.S =================================================================== --- linux-2.6.10-ck2.orig/arch/ppc/kernel/misc.S 2004-12-25 10:14:47.000000000 +1100 +++ linux-2.6.10-ck2/arch/ppc/kernel/misc.S 2005-01-04 09:31:32.260128664 +1100 @@ -1450,3 +1450,5 @@ .long sys_add_key .long sys_request_key /* 270 */ .long sys_keyctl + .long sys_ioprio_set + .long sys_ioprio_get Index: linux-2.6.10-ck2/drivers/block/as-iosched.c =================================================================== --- linux-2.6.10-ck2.orig/drivers/block/as-iosched.c 2004-12-25 10:14:47.000000000 +1100 +++ linux-2.6.10-ck2/drivers/block/as-iosched.c 2005-01-04 09:31:32.261128512 +1100 @@ -1805,7 +1805,8 @@ rq->elevator_private = NULL; } -static int as_set_request(request_queue_t *q, struct request *rq, int gfp_mask) +static int as_set_request(request_queue_t *q, struct request *rq, int gfp_mask, + struct bio *bio) { struct as_data *ad = q->elevator->elevator_data; struct as_rq *arq = mempool_alloc(ad->arq_pool, gfp_mask); @@ -1826,7 +1827,7 @@ return 1; } -static int as_may_queue(request_queue_t *q, int rw) +static int as_may_queue(request_queue_t *q, int rw, struct bio *bio) { int ret = ELV_MQUEUE_MAY; struct as_data *ad = q->elevator->elevator_data; Index: linux-2.6.10-ck2/drivers/block/cfq-iosched.c =================================================================== --- linux-2.6.10-ck2.orig/drivers/block/cfq-iosched.c 2004-12-25 10:14:47.000000000 +1100 +++ linux-2.6.10-ck2/drivers/block/cfq-iosched.c 2005-01-04 09:31:53.319927088 +1100 @@ -21,22 +21,32 @@ #include #include #include - -static unsigned long max_elapsed_crq; -static unsigned long max_elapsed_dispatch; +#include /* * 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_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 / 45; +static int cfq_slice_async = HZ / 175; +static int cfq_slice_async_rq = 16; +static int cfq_slice_idle = HZ / 100; + +#define CFQ_MAX_SEEK_MEAN ((sector_t) -1) +#define CFQ_IDLE_GRACE (HZ / 10) + +/* + * disable queueing at the driver/hardware level + */ +static int cfq_max_depth = 1; + /* * for the hash of cfqq inside the cfqd */ @@ -55,6 +65,7 @@ #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 @@ -76,50 +87,75 @@ #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_PID, CFQ_KEY_UID, CFQ_KEY_GID, CFQ_KEY_LAST, }; -static char *cfq_key_types[] = { "pgid", "tgid", "uid", "gid", NULL }; +static char *cfq_key_types[] = { "pgid", "tgid", "pid", "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) + +/* + * 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 cur_rr; + struct list_head idle_rr; + unsigned int busy_queues; + struct list_head empty_list; struct hlist_head *cfq_hash; struct hlist_head *crq_hash; - /* queues on rr_list (ie they have pending requests */ - unsigned int busy_queues; - unsigned int max_queued; - atomic_t ref; - int key_type; mempool_t *crq_pool; - request_queue_t *queue; + int rq_in_driver; - sector_t last_sector; + /* + * schedule slice state info + */ + /* + * idle window management + */ + struct timer_list idle_slice_timer; + struct work_struct unplug_work; + unsigned long idle_start; + + struct cfq_queue *active_queue; + struct cfq_io_context *active_cic; + int cur_prio, cur_end_prio, end_prio; + unsigned int dispatch_slice; - int rq_in_driver; + struct timer_list idle_class_timer; + + sector_t last_sector; + unsigned long last_end_request; /* * tunables, see top of file @@ -131,11 +167,15 @@ unsigned int cfq_fifo_batch_expire; 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; @@ -145,8 +185,6 @@ struct hlist_node cfq_hash; /* hash key */ unsigned long key; - /* whether queue is on rr (or empty) list */ - int on_rr; /* on either rr or empty list of cfqd */ struct list_head cfq_list; /* sorted list of pending requests */ @@ -164,15 +202,34 @@ int key_type; - unsigned long service_start; - unsigned long service_used; - - unsigned int max_rate; + unsigned long slice_start; + unsigned long slice_end; + unsigned long slice_left; + unsigned long service_last; /* number of requests that have been handed to the driver */ int in_flight; - /* number of currently allocated requests */ - int alloc_limit[2]; + + /* io prio of this group */ + int ioprio; + int org_ioprio; + short ioprio_class; + short org_ioprio_class; + + unsigned int idle_hit, idle_miss; + + /* whether queue is on rr (or empty) list */ + unsigned int on_rr : 1; + /* idle slice, waiting for new request submission */ + unsigned int wait_request : 1; + /* set when wait_request gets set, reset on first rq alloc */ + unsigned int must_alloc : 1; + /* idle slice, request added, now waiting to dispatch it */ + unsigned int must_dispatch : 1; + + unsigned int idle_window : 1; + unsigned int doing_sync : 1; + unsigned int prio_changed : 1; }; struct cfq_rq { @@ -184,27 +241,37 @@ struct cfq_queue *cfq_queue; struct cfq_io_context *io_context; - unsigned long service_start; - unsigned long queue_start; + sector_t end_pos; unsigned int in_flight : 1; unsigned int accounted : 1; unsigned int is_sync : 1; - unsigned int is_write : 1; + unsigned int requeued : 1; }; static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned long); 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); +static struct cfq_io_context *cfq_get_io_context(struct cfq_queue **, int, struct bio *); + +#define process_sync(tsk) ((tsk)->flags & PF_SYNCWRITE) /* * 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) +cfq_hash_key(struct cfq_data *cfqd, struct task_struct *tsk, struct bio *bio) { + if (bio && bio_data_dir(bio) == WRITE && !process_sync(current)) { + struct page *p = bio_iovec_idx(bio, 0)->bv_page; + unsigned long key; + + if ((key = get_page_dirty_pid(p)) != 0) + return key; + } + /* * optimize this so that ->key_type is the offset into the struct */ @@ -214,6 +281,8 @@ default: case CFQ_KEY_TGID: return tsk->tgid; + case CFQ_KEY_PID: + return tsk->pid; case CFQ_KEY_UID: return tsk->uid; case CFQ_KEY_GID: @@ -287,6 +356,10 @@ return crq2; if (crq2 == NULL) return crq1; + if (crq1->requeued) + return crq1; + if (crq2->requeued) + return crq2; s1 = crq1->request->sector; s2 = crq2->request->sector; @@ -401,67 +474,40 @@ 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; + struct list_head *list, *entry; - /* - * list might still be ordered - */ - next = cfqq->cfq_list.next; - if (next != head) { - struct cfq_queue *cnext = list_entry_cfqq(next); - - if (cfqq->service_used > cnext->service_used) - return 1; - } + BUG_ON(!cfqq->on_rr); - prev = cfqq->cfq_list.prev; - if (prev != head) { - struct cfq_queue *cprev = list_entry_cfqq(prev); - - if (cfqq->service_used < cprev->service_used) - return 1; - } - - return 0; -} + list_del(&cfqq->cfq_list); -static void cfq_sort_rr_list(struct cfq_queue *cfqq, int new_queue) -{ - struct list_head *entry = &cfqq->cfqd->rr_list; + if (cfq_class_rt(cfqq)) + list = &cfqq->cfqd->cur_rr; + else if (cfq_class_idle(cfqq)) + list = &cfqq->cfqd->idle_rr; + else + list = &cfqq->cfqd->rr_list[cfqq->ioprio]; - 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 + */ + if (unlikely(preempted)) { + 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); @@ -469,21 +515,24 @@ /* * 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 - */ + BUG_ON(cfqq->on_rr); cfqq->on_rr = 1; cfqd->busy_queues++; - if (time_after(jiffies, cfqq->service_start + cfq_service)) - cfqq->service_used >>= 3; - - cfq_sort_rr_list(cfqq, 1); + /* + * idle io should not end up on cur_rr and get premium service + */ + if (!requeue || cfq_class_idle(cfqq)) + cfq_resort_rr_list(cfqq, 0); + else { + list_del(&cfqq->cfq_list); + list_add_tail(&cfqq->cfq_list, &cfqd->cur_rr); + } } static inline void @@ -507,10 +556,10 @@ struct cfq_data *cfqd = cfqq->cfqd; BUG_ON(!cfqq->queued[crq->is_sync]); + cfqq->queued[crq->is_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); @@ -562,7 +611,7 @@ rb_insert_color(&crq->rb_node, &cfqq->sort_list); if (!cfqq->on_rr) - cfq_add_cfqq_rr(cfqd, cfqq); + cfq_add_cfqq_rr(cfqd, cfqq, crq->requeued); /* * check if this request is a better next-serve candidate @@ -582,9 +631,9 @@ } static struct request * -cfq_find_rq_rb(struct cfq_data *cfqd, sector_t sector) +cfq_find_rq_rb(struct cfq_data *cfqd, sector_t sector, struct bio *bio) { - const unsigned long key = cfq_hash_key(cfqd, current); + const unsigned long key = cfq_hash_key(cfqd, current, bio); struct cfq_queue *cfqq = cfq_find_cfq_hash(cfqd, key); struct rb_node *n; @@ -610,22 +659,33 @@ /* * make sure the service time gets corrected on reissue of this request */ +static void cfq_enqueue(struct cfq_data *cfqd, struct request *rq); static void cfq_requeue_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 (crq->accounted) { + crq->accounted = 0; + WARN_ON(!cfqq->cfqd->rq_in_driver); + cfqq->cfqd->rq_in_driver--; + } + if (crq->in_flight) { + crq->in_flight = 0; + WARN_ON(!cfqq->in_flight); + cfqq->in_flight--; } - - crq->accounted = 0; - cfqq->cfqd->rq_in_driver--; } - list_add(&rq->queuelist, &q->queue_head); + + if (blk_fs_request(rq)) { + crq->cfq_queue->next_crq = crq; + crq->requeued = 1; + cfq_enqueue(cfqd, rq); + } else + list_add(&rq->queuelist, &q->queue_head); } static void cfq_remove_request(request_queue_t *q, struct request *rq) @@ -635,9 +695,8 @@ if (crq) { cfq_remove_merge_hints(q, crq); list_del_init(&rq->queuelist); + cfq_del_crq_rb(crq); - if (crq->cfq_queue) - cfq_del_crq_rb(crq); } } @@ -664,7 +723,7 @@ } } - __rq = cfq_find_rq_rb(cfqd, bio->bi_sector + bio_sectors(bio)); + __rq = cfq_find_rq_rb(cfqd, bio->bi_sector + bio_sectors(bio), bio); if (__rq) { if (elv_rq_merge_ok(__rq, bio)) { ret = ELEVATOR_FRONT_MERGE; @@ -702,22 +761,189 @@ 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_update_next_crq(cnext); cfq_remove_request(q, next); } +static inline void +__cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) +{ + if (cfqq) { + cfqq->slice_start = jiffies; + cfqq->slice_end = 0; + cfqq->slice_left = 0; + del_timer(&cfqd->idle_class_timer); + } + + 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 > cfqd->end_prio) + cfqd->end_prio = cfqd->cur_end_prio; + if (cfqd->end_prio == CFQ_PRIO_LISTS) { + cfqd->cur_prio = 0; + cfqd->cur_end_prio = 0; + cfqd->end_prio = 0; + } + + return 0; +} + +static void cfq_set_active_queue(struct cfq_data *cfqd) +{ + struct cfq_queue *cfqq = 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)) + 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); +} + +/* + * current cfqq expired its slice (or was too idle), select new one + */ +static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted) +{ + struct cfq_queue *cfqq = cfqd->active_queue; + + if (cfqq) { + unsigned long now = jiffies; + + if (cfqq->wait_request) + del_timer(&cfqd->idle_slice_timer); + + if (!preempted) + cfqq->service_last = now; + + cfqq->must_dispatch = 0; + cfqq->wait_request = 0; + + /* + * store what was left of this slice, if the queue idled out + * or was preempted + */ + if (time_before(now, cfqq->slice_end)) + cfqq->slice_left = now - cfqq->slice_end; + else + cfqq->slice_left = 0; + + if (cfqq->on_rr) + cfq_resort_rr_list(cfqq, preempted); + + cfqd->active_queue = NULL; + + if (cfqd->active_cic) { + put_io_context(cfqd->active_cic->ioc); + cfqd->active_cic = NULL; + } + } + + cfqd->dispatch_slice = 0; +} + +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); + + if (!cfqd->cfq_slice_idle) + return 0; + if (!cfqq->idle_window) + return 0; + + cfqq->wait_request = 1; + cfqq->must_alloc = 1; + + if (!timer_pending(&cfqd->idle_slice_timer)) { + unsigned long slice_left = cfqq->slice_end - 1; + + cfqd->idle_start = jiffies; + cfqd->idle_slice_timer.expires = min(jiffies + cfqd->cfq_slice_idle, slice_left); + add_timer(&cfqd->idle_slice_timer); + } + + return 1; +} + /* * we dispatch cfqd->cfq_quantum requests in total from the rr_list queues, * this function sector sorts the selected request to minimize seeks. we start @@ -731,31 +957,34 @@ 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); - - if (blk_barrier_rq(crq->request)) + list_for_each_entry_reverse(__rq, head, queuelist) { + if (blk_barrier_rq(__rq)) break; - if (!blk_fs_request(crq->request)) + if (!blk_fs_request(__rq)) 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; + + cfq_del_crq_rb(crq); + cfq_remove_merge_hints(q, crq); + + crq->end_pos = crq->request->sector + crq->request->nr_sectors; crq->in_flight = 1; + crq->requeued = 0; cfqq->in_flight++; - list_add(&crq->request->queuelist, entry); + list_add_tail(&crq->request->queuelist, entry); } /* @@ -767,102 +996,163 @@ 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)) 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 (reads) { + crq = RQ_DATA(list_entry_fifo(cfqq->fifo[READ].next)); + rq = crq->request; + if (time_after(now, rq->start_time + cfqd->cfq_fifo_expire_r)) { + cfqq->last_fifo_expire = now; + return crq; + } } - 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; + if (writes) { + crq = RQ_DATA(list_entry_fifo(cfqq->fifo[WRITE].next)); + rq = crq->request; + if (time_after(now, rq->start_time + cfqd->cfq_fifo_expire_w)) { + cfqq->last_fifo_expire = now; + return crq; + } } return NULL; } /* - * dispatch a single request from given queue + * Scale schedule slice based on io priority */ -static inline void -cfq_dispatch_request(request_queue_t *q, struct cfq_data *cfqd, - struct cfq_queue *cfqq) +static inline int cfq_prio_to_slice(struct cfq_data *cfqd, int prio, int sync) { - struct cfq_rq *crq; + int base_slice = cfqd->cfq_slice[sync]; + + return base_slice + base_slice * (CFQ_PRIO_LISTS - 1 - prio); +} + +/* + * get next queue for service + */ +static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) +{ + unsigned long now = jiffies; + struct cfq_queue *cfqq; + + cfqq = cfqd->active_queue; + if (!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 (!cfqq->must_dispatch && time_after(jiffies, cfqq->slice_end)) + goto new_queue; /* - * 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 (time_before(now, cfqq->slice_end)) { + if (cfq_arm_slice_timer(cfqd, cfqq)) + return NULL; + } + +new_queue: + cfq_slice_expired(cfqd, 0); + cfq_set_active_queue(cfqd); +keep_queue: + return cfqd->active_queue; } -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, sync = 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 list */ - if (first_round && cfqq->in_flight) - continue; + cfq_dispatch_sort(cfqd->queue, crq); - cfq_dispatch_request(q, cfqd, cfqq); + cfqd->dispatch_slice++; + dispatched++; + sync += crq->is_sync; + + if (!cfqd->active_cic) { + atomic_inc(&crq->io_context->ioc->refcount); + cfqd->active_cic = crq->io_context; + } + + if (RB_EMPTY(&cfqq->sort_list)) + break; - if (!RB_EMPTY(&cfqq->sort_list)) - busy_queues++; + } while (dispatched < max_dispatch); - queued++; - } + /* + * 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) + cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq->ioprio, !!sync) + jiffies; + + /* + * expire an async queue immediately if it has used up its slice. idle + * queue always expire after 1 dispatch round. + */ + if ((!sync && cfqd->dispatch_slice >= cfqd->cfq_slice_async_rq) || + cfq_class_idle(cfqq)) + cfq_slice_expired(cfqd, 0); + + cfqq->doing_sync = !!sync; + return dispatched; +} + +static int cfq_dispatch_requests(request_queue_t *q, int max_dispatch) +{ + struct cfq_data *cfqd = q->elevator->elevator_data; + struct cfq_queue *cfqq; + + if (!cfqd->busy_queues) + return 0; + + cfqq = cfq_select_queue(cfqd); + if (cfqq) { + cfqq->wait_request = 0; + cfqq->must_dispatch = 0; + del_timer(&cfqd->idle_slice_timer); - if ((queued < max_dispatch) && (busy_queues || first_round)) { - first_round = 0; - goto restart; + if (cfq_class_idle(cfqq)) + max_dispatch = 1; + + 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; /* @@ -872,37 +1162,8 @@ if (crq->accounted) 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); - } + cfqd->rq_in_driver++; } static inline void @@ -916,20 +1177,12 @@ 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 (time_after(now, cfqq->service_start + cfq_service)) { - cfqq->service_start = now; - cfqq->service_used >>= 3; - } - - cfqq->service_used += duration; - cfq_sort_rr_list(cfqq, 0); + if (!cfq_class_idle(cfqq)) + cfqd->last_end_request = jiffies; - if (duration > max_elapsed_crq) - max_elapsed_crq = duration; + if (crq->is_sync) { + crq->io_context->last_end_request = jiffies; + crq->io_context->last_request_pos = crq->end_pos; } } @@ -943,7 +1196,15 @@ dispatch: rq = list_entry_rq(q->queue_head.next); - if ((crq = RQ_DATA(rq)) != NULL) { + crq = RQ_DATA(rq); + if (crq) { + /* + * if idle window is disabled, allow queue buildup + */ + if (!crq->in_flight && !crq->cfq_queue->idle_window && + cfqd->rq_in_driver >= cfqd->cfq_max_depth) + return NULL; + cfq_remove_merge_hints(q, crq); cfq_account_dispatch(crq); } @@ -965,6 +1226,8 @@ */ static void cfq_put_queue(struct cfq_queue *cfqq) { + struct cfq_data *cfqd = cfqq->cfqd; + BUG_ON(!atomic_read(&cfqq->ref)); if (!atomic_dec_and_test(&cfqq->ref)) @@ -973,6 +1236,9 @@ BUG_ON(rb_first(&cfqq->sort_list)); BUG_ON(cfqq->on_rr); + if (unlikely(cfqd->active_queue == cfqq)) + cfq_slice_expired(cfqd, 0); + cfq_put_cfqd(cfqq->cfqd); /* @@ -1007,9 +1273,9 @@ static inline void cfq_rehash_cfqq(struct cfq_data *cfqd, struct cfq_queue **cfqq, - struct cfq_io_context *cic) + struct cfq_io_context *cic, struct bio *bio) { - unsigned long hashkey = cfq_hash_key(cfqd, current); + unsigned long hashkey = cfq_hash_key(cfqd, current, bio); unsigned long hashval = hash_long(hashkey, CFQ_QHASH_SHIFT); struct cfq_queue *__cfqq; unsigned long flags; @@ -1039,55 +1305,142 @@ kmem_cache_free(cfq_ioc_pool, cic); } -/* - * locking hierarchy is: io_context lock -> queue locks - */ -static void cfq_exit_io_context(struct cfq_io_context *cic) +static void cfq_exit_single_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_data *cfqd = cic->cfqq->cfqd; + request_queue_t *q = cfqd->queue; unsigned long 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; + spin_lock_irqsave(q->queue_lock, flags); - __cic = list_entry(entry, struct cfq_io_context, list); - list_del(entry); + if (cic->cfqq == cfqd->active_queue) { + cfq_slice_expired(cfqd, 0); - q = __cic->cfqq->cfqd->queue; - spin_lock(q->queue_lock); - cfq_put_queue(__cic->cfqq); - spin_unlock(q->queue_lock); + if (cfqd->busy_queues) + kblockd_schedule_work(&cfqd->unplug_work); } - q = cfqq->cfqd->queue; - spin_lock(q->queue_lock); - cfq_put_queue(cfqq); - spin_unlock(q->queue_lock); - + cfq_put_queue(cic->cfqq); cic->cfqq = NULL; - spin_unlock_irqrestore(&cic->ioc->lock, flags); + spin_unlock_irqrestore(q->queue_lock, flags); + put_io_context(cic->ioc); +} + +/* + * Only the task itself ever browses or updates the cic list, so no locking + * is needed + */ +static void cfq_exit_io_context(struct cfq_io_context *cic) +{ + struct cfq_io_context *__cic, *nxt; + + /* + * put the reference this task is holding to the various queues + */ + list_for_each_entry_safe(__cic, nxt, &cic->list, list) { + list_del(&__cic->list); + cfq_exit_single_io_context(__cic); + } + + cfq_exit_single_io_context(cic); } -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_flags) { struct cfq_io_context *cic = kmem_cache_alloc(cfq_ioc_pool, gfp_flags); if (cic) { - cic->dtor = cfq_free_io_context; - cic->exit = cfq_exit_io_context; INIT_LIST_HEAD(&cic->list); cic->cfqq = NULL; + cic->last_end_request = jiffies; + cic->ttime_total = 0; + cic->ttime_samples = 0; + cic->ttime_mean = 0; + cic->seek_samples = 0; + cic->last_request_pos = cfqd->last_sector; + cic->seek_total = 0; + cic->seek_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) +{ + int ioprio_class; + + if (!cfqq->prio_changed) + return; + + ioprio_class = IOPRIO_PRIO_CLASS(current->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_ioprio(current); + cfqq->ioprio_class = IOPRIO_CLASS_BE; + break; + case IOPRIO_CLASS_RT: + cfqq->ioprio_class = IOPRIO_CLASS_RT; + cfqq->ioprio = task_ioprio(current); + break; + case IOPRIO_CLASS_BE: + cfqq->ioprio_class = IOPRIO_CLASS_BE; + cfqq->ioprio = task_ioprio(current); + break; + case IOPRIO_CLASS_IDLE: + cfqq->ioprio_class = IOPRIO_CLASS_IDLE; + cfqq->ioprio = 7; + cfqq->idle_window = 0; + 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 (cfqq->on_rr) + cfq_resort_rr_list(cfqq, 0); + + cfqq->prio_changed = 0; +} + +static inline void changed_ioprio(struct cfq_queue *cfqq) +{ + if (cfqq) { + struct cfq_data *cfqd = cfqq->cfqd; + + spin_lock(cfqd->queue->queue_lock); + cfqq->prio_changed = 1; + 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; +} + /* * 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 @@ -1095,7 +1448,7 @@ * 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_queue **cfqq, int gfp_flags, struct bio *bio) { struct cfq_data *cfqd = (*cfqq)->cfqd; struct cfq_queue *__cfqq = *cfqq; @@ -1109,18 +1462,25 @@ return NULL; if ((cic = ioc->cic) == NULL) { - cic = cfq_alloc_io_context(gfp_flags); + cic = cfq_alloc_io_context(cfqd, gfp_flags); 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; + atomic_inc(&ioc->refcount); + cic->cfqq = __cfqq; atomic_inc(&__cfqq->ref); + atomic_inc(&cfqd->ref); } else { struct cfq_io_context *__cic; - unsigned long flags; /* * since the first cic on the list is actually the head @@ -1135,35 +1495,37 @@ * 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) { + /* + * move hot entries to the front + */ + list_del(&__cic->list); + list_add(&__cic->list, &cic->list); 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_flags); if (__cic == NULL) goto err; __cic->ioc = ioc; + atomic_inc(&ioc->refcount); + __cic->cfqq = __cfqq; atomic_inc(&__cfqq->ref); - spin_lock_irqsave(&ioc->lock, flags); + atomic_inc(&cfqd->ref); list_add(&__cic->list, &cic->list); - spin_unlock_irqrestore(&ioc->lock, flags); - cic = __cic; *cfqq = __cfqq; } @@ -1174,7 +1536,7 @@ * each queue at lookup time */ if ((*cfqq)->key_type != cfqd->key_type) - cfq_rehash_cfqq(cfqd, cfqq, cic); + cfq_rehash_cfqq(cfqd, cfqq, cic, bio); return cic; err: @@ -1183,7 +1545,7 @@ } static struct cfq_queue * -__cfq_get_queue(struct cfq_data *cfqd, unsigned long key, int gfp_mask) +cfq_get_queue(struct cfq_data *cfqd, unsigned long key, int gfp_mask) { const int hashval = hash_long(key, CFQ_QHASH_SHIFT); struct cfq_queue *cfqq, *new_cfqq = NULL; @@ -1200,8 +1562,11 @@ new_cfqq = kmem_cache_alloc(cfq_pool, gfp_mask); spin_lock_irq(cfqd->queue->queue_lock); goto retry; - } else - goto out; + } else { + cfqq = kmem_cache_alloc(cfq_pool, gfp_mask); + if (!cfqq) + goto out; + } memset(cfqq, 0, sizeof(*cfqq)); @@ -1217,7 +1582,15 @@ cfqq->cfqd = cfqd; atomic_inc(&cfqd->ref); cfqq->key_type = cfqd->key_type; - cfqq->service_start = ~0UL; + cfqq->service_last = 0; + /* + * set ->slice_left to allow preemption for a new process + */ + cfqq->slice_left = 2 * cfqd->cfq_slice_idle; + cfqq->idle_window = 1; + cfqq->ioprio = -1; + cfqq->ioprio_class = -1; + cfqq->prio_changed = 1; } if (new_cfqq) @@ -1229,23 +1602,225 @@ return cfqq; } -static void cfq_enqueue(struct cfq_data *cfqd, struct cfq_rq *crq) +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; +} + +static void +cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_rq *crq, + struct cfq_io_context *cic) +{ + sector_t seek; + u64 total; + + if (!cic->last_request_pos) + seek = 0; + else if (cic->last_request_pos < crq->request->sector) + seek = crq->request->sector - cic->last_request_pos; + else + seek = cic->last_request_pos - crq->request->sector; + + /* + * Don't allow the seek distance to get too large from the + * odd fragment, pagein, etc + */ + if (cic->seek_samples <= 60) /* second&third seek */ + seek = min(seek, (cic->seek_mean * 4) + 2*1024*1024); + else + seek = min(seek, (cic->seek_mean * 4) + 2*1024*64); + + cic->seek_samples = (7*cic->seek_samples + 256) / 8; + cic->seek_total = (7*cic->seek_total + (u64)256*seek) / 8; + total = cic->seek_total + (cic->seek_samples/2); + do_div(total, cic->seek_samples); + cic->seek_mean = total; +} + +#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 = 0; + + if (!cic->ioc->task) + goto disable; + + if (sample_valid(cic->seek_samples) && + cic->seek_mean > CFQ_MAX_SEEK_MEAN) + goto disable; + + if (sample_valid(cic->ttime_samples) && + cic->ttime_mean > cfqd->cfq_slice_idle) + goto disable; + + if (!cfqd->cfq_slice_idle) + goto disable; + + enable_idle = 1; +disable: + cfqq->idle_window = enable_idle; +} + + +/* + * 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 (!new_cfqq->wait_request) + return 0; + + if (new_cfqq->ioprio > cfqq->ioprio) + return 0; + if (crq->is_sync && !cfqq->doing_sync) + return 1; + if (new_cfqq->ioprio == cfqq->ioprio) + return 0; + + /* + * if it doesn't have slice left, forget it + */ + if (new_cfqq->slice_left >= 2 * cfqd->cfq_slice_idle) + return 1; + + return 0; +} + +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->ioprio, 1)/2; + + cfqq->slice_end = cfqq->slice_left + jiffies; + cfq_slice_expired(cfqd, 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) +{ + if (crq->is_sync) { + struct cfq_io_context *cic = crq->io_context; + + cfq_update_io_thinktime(cfqd, cic); + cfq_update_io_seektime(cfqd, crq, 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 (cfqq->wait_request) { + cfqq->idle_hit++; + cfqq->must_dispatch = 1; + 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); + cfqq->must_dispatch = 1; + cfq_start_queueing(cfqd, cfqq); + } +} + +static void cfq_enqueue(struct cfq_data *cfqd, struct request *rq) { - crq->is_sync = 0; - if (rq_data_dir(crq->request) == READ || current->flags & PF_SYNCWRITE) - crq->is_sync = 1; + 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[crq->is_sync]); + + 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: @@ -1258,26 +1833,19 @@ 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); + return list_empty(&q->queue_head) && !cfqd->busy_queues; } static void cfq_completed_request(request_queue_t *q, struct request *rq) @@ -1322,50 +1890,84 @@ 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) +{ + 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 > 4) + cfqq->ioprio = 4; + } 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) && + cfqq->on_rr) + cfq_resort_rr_list(cfqq, 0); +} + +static int cfq_may_queue(request_queue_t *q, int rw, struct bio *bio) { struct cfq_data *cfqd = q->elevator->elevator_data; struct cfq_queue *cfqq; int ret = ELV_MQUEUE_MAY; - if (current->flags & PF_MEMALLOC) - return ELV_MQUEUE_MAY; + cfqq = cfq_find_cfq_hash(cfqd, cfq_hash_key(cfqd, current, bio)); + if (!cfqq && rw == WRITE && process_sync(current)) + cfqq = cfq_find_cfq_hash(cfqd, cfq_hash_key(cfqd, current,NULL)); - cfqq = cfq_find_cfq_hash(cfqd, cfq_hash_key(cfqd, current)); if (cfqq) { - int limit = cfqd->max_queued; + int idle; - if (cfqq->allocated[rw] < cfqd->cfq_queued) - return ELV_MQUEUE_MUST; + cfq_prio_boost(cfqq); - if (cfqd->busy_queues) - limit = q->nr_requests / cfqd->busy_queues; + idle = cfq_class_idle(cfqq); - if (limit < cfqd->cfq_queued) - limit = cfqd->cfq_queued; - else if (limit > cfqd->max_queued) - limit = cfqd->max_queued; - - if (cfqq->allocated[rw] >= limit) { - if (limit > cfqq->alloc_limit[rw]) - cfqq->alloc_limit[rw] = limit; + if (!idle) + ret = ELV_MQUEUE_MUST; + if (cfqq->must_alloc) + goto out; + if (cfqq->allocated[rw] < cfqd->cfq_queued) + goto out; + if (cfqq->allocated[rw] >= cfqd->max_queued || idle) ret = ELV_MQUEUE_NO; - } } +out: return ret; } static void cfq_check_waiters(request_queue_t *q, struct cfq_queue *cfqq) { struct request_list *rl = &q->rq; - const int write = waitqueue_active(&rl->wait[WRITE]); - const int read = waitqueue_active(&rl->wait[READ]); + const int writes = waitqueue_active(&rl->wait[WRITE]); + const int reads = waitqueue_active(&rl->wait[READ]); + struct cfq_data *cfqd = q->elevator->elevator_data; - if (read && cfqq->allocated[READ] < cfqq->alloc_limit[READ]) + if (reads && cfqq->allocated[READ] == cfqd->max_queued) wake_up(&rl->wait[READ]); - if (write && cfqq->allocated[WRITE] < cfqq->alloc_limit[WRITE]) + if (writes && cfqq->allocated[WRITE] == cfqd->max_queued) wake_up(&rl->wait[WRITE]); } @@ -1379,15 +1981,15 @@ 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; @@ -1401,12 +2003,14 @@ /* * Allocate cfq data structures associated with this request. A queue and */ -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, int gfp_mask, + struct bio *bio) { struct cfq_data *cfqd = q->elevator->elevator_data; - struct cfq_io_context *cic; const int rw = rq_data_dir(rq); struct cfq_queue *cfqq, *saved_cfqq; + struct cfq_io_context *cic; struct cfq_rq *crq; unsigned long flags; @@ -1414,22 +2018,24 @@ spin_lock_irqsave(q->queue_lock, flags); - cfqq = __cfq_get_queue(cfqd, cfq_hash_key(cfqd, current), gfp_mask); - if (!cfqq) - goto out_lock; + cfqq = cfq_get_queue(cfqd, cfq_hash_key(cfqd, current, bio), gfp_mask); + if (!cfqq) { + if (rw == WRITE && process_sync(current)) + cfqq = cfq_get_queue(cfqd, cfq_hash_key(cfqd, current, NULL), gfp_mask); + if (!cfqq) + goto out_lock; + } repeat: - if (cfqq->allocated[rw] >= cfqd->max_queued) - goto out_lock; - cfqq->allocated[rw]++; + cfqq->must_alloc = 0; 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); + cic = cfq_get_io_context(&saved_cfqq, gfp_mask, bio); if (!cic) goto err; @@ -1437,6 +2043,7 @@ * repeat allocation checks on queue change */ if (unlikely(saved_cfqq != cfqq)) { + put_io_context(cic->ioc); spin_lock_irqsave(q->queue_lock, flags); saved_cfqq->allocated[rw]--; goto repeat; @@ -1450,11 +2057,10 @@ 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; + crq->in_flight = crq->accounted = 0; + crq->is_sync = (rw == READ || process_sync(current)); + crq->requeued = 0; rq->elevator_private = crq; - cfqq->alloc_limit[rw] = 0; return 0; } @@ -1462,12 +2068,91 @@ err: spin_lock_irqsave(q->queue_lock, flags); cfqq->allocated[rw]--; + cfqq->must_alloc = 1; cfq_put_queue(cfqq); out_lock: spin_unlock_irqrestore(q->queue_lock, flags); return 1; } +static void cfq_kick_queue(void *data) +{ + request_queue_t *q = data; + + blk_run_queue(q); +} + +/* + * 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) { + /* + * expired + */ + if (time_after(jiffies, cfqq->slice_end)) + goto miss; + + /* + * only expire and reinvoke request handler, if there are + * other queues with pending requests + */ + if (!cfqd->busy_queues) { + cfqd->idle_slice_timer.expires = min(jiffies + 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)) { + cfqq->must_dispatch = 1; + goto out_kick; + } +miss: + cfqq->idle_miss++; + } + + cfq_slice_expired(cfqd, 0); +out_kick: + if (cfqd->busy_queues) + kblockd_schedule_work(&cfqd->unplug_work); +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 + kblockd_schedule_work(&cfqd->unplug_work); + + spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); +} + + static void cfq_put_cfqd(struct cfq_data *cfqd) { request_queue_t *q = cfqd->queue; @@ -1475,6 +2160,8 @@ if (!atomic_dec_and_test(&cfqd->ref)) return; + blk_sync_queue(q); + blk_put_queue(q); mempool_destroy(cfqd->crq_pool); @@ -1485,7 +2172,11 @@ static void cfq_exit_queue(elevator_t *e) { - cfq_put_cfqd(e->elevator_data); + struct cfq_data *cfqd = e->elevator_data; + + del_timer_sync(&cfqd->idle_slice_timer); + del_timer_sync(&cfqd->idle_class_timer); + cfq_put_cfqd(cfqd); } static int cfq_init_queue(request_queue_t *q, elevator_t *e) @@ -1498,7 +2189,12 @@ 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->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); @@ -1523,15 +2219,20 @@ 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; + cfqd->key_type = CFQ_KEY_PID; + + 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; @@ -1541,6 +2242,11 @@ cfqd->cfq_fifo_batch_expire = cfq_fifo_rate; 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: @@ -1585,7 +2291,6 @@ return -ENOMEM; } - /* * sysfs parts below --> */ @@ -1611,13 +2316,6 @@ } 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); @@ -1625,6 +2323,8 @@ cfqd->key_type = CFQ_KEY_PGID; else if (!strncmp(page, "tgid", 4)) cfqd->key_type = CFQ_KEY_TGID; + else if (!strncmp(page, "pid", 3)) + cfqd->key_type = CFQ_KEY_PID; else if (!strncmp(page, "uid", 3)) cfqd->key_type = CFQ_KEY_UID; else if (!strncmp(page, "gid", 3)) @@ -1662,9 +2362,13 @@ 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_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) \ @@ -1687,9 +2391,13 @@ 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_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 = { @@ -1717,11 +2425,6 @@ .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, -}; static struct cfq_fs_entry cfq_back_max_entry = { .attr = {.name = "back_seek_max", .mode = S_IRUGO | S_IWUSR }, .show = cfq_back_max_show, @@ -1732,15 +2435,36 @@ .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_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_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_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, @@ -1749,10 +2473,13 @@ &cfq_fifo_expire_w_entry.attr, &cfq_fifo_batch_expire_entry.attr, &cfq_key_type_entry.attr, - &cfq_find_best_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, }; @@ -1817,25 +2544,50 @@ .elevator_owner = THIS_MODULE, }; -int cfq_init(void) +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); } Index: linux-2.6.10-ck2/drivers/block/deadline-iosched.c =================================================================== --- linux-2.6.10-ck2.orig/drivers/block/deadline-iosched.c 2004-12-25 10:14:47.000000000 +1100 +++ linux-2.6.10-ck2/drivers/block/deadline-iosched.c 2005-01-04 09:31:32.265127904 +1100 @@ -758,7 +758,8 @@ } static int -deadline_set_request(request_queue_t *q, struct request *rq, int gfp_mask) +deadline_set_request(request_queue_t *q, struct request *rq, int gfp_mask, + struct bio *bio) { struct deadline_data *dd = q->elevator->elevator_data; struct deadline_rq *drq; Index: linux-2.6.10-ck2/drivers/block/elevator.c =================================================================== --- linux-2.6.10-ck2.orig/drivers/block/elevator.c 2005-01-04 09:30:08.241901376 +1100 +++ linux-2.6.10-ck2/drivers/block/elevator.c 2005-01-04 09:31:32.265127904 +1100 @@ -440,12 +440,13 @@ return NULL; } -int elv_set_request(request_queue_t *q, struct request *rq, int gfp_mask) +int elv_set_request(request_queue_t *q, struct request *rq, int gfp_mask, + struct bio *bio) { elevator_t *e = q->elevator; if (e->ops->elevator_set_req_fn) - return e->ops->elevator_set_req_fn(q, rq, gfp_mask); + return e->ops->elevator_set_req_fn(q, rq, gfp_mask, bio); rq->elevator_private = NULL; return 0; @@ -459,12 +460,12 @@ e->ops->elevator_put_req_fn(q, rq); } -int elv_may_queue(request_queue_t *q, int rw) +int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) { elevator_t *e = q->elevator; if (e->ops->elevator_may_queue_fn) - return e->ops->elevator_may_queue_fn(q, rw); + return e->ops->elevator_may_queue_fn(q, rw, bio); return ELV_MQUEUE_MAY; } Index: linux-2.6.10-ck2/drivers/block/Kconfig.iosched =================================================================== --- linux-2.6.10-ck2.orig/drivers/block/Kconfig.iosched 2004-12-25 10:14:47.000000000 +1100 +++ linux-2.6.10-ck2/drivers/block/Kconfig.iosched 2005-01-04 09:31:32.265127904 +1100 @@ -38,4 +38,17 @@ among all processes in the system. It should provide a fair working environment, suitable for desktop systems. +config IOPRIO_WRITE + bool "Support for full write io fairness" + depends on IOSCHED_CFQ + ---help--- + Most data writeout that isn't direct or raw happen asyncronously + and thus not in the context of the process that originally dirtied + the data. This type of writeback is typically handled by the pdflush + kernel threads. The result is that a process dirtying a lot of pages + gets proportionally more bandwidth than it should. + + Say Y here it you want to apply full fairness for file system + asyncronous writes at the expense of using 4 bytes in struct page. + endmenu Index: linux-2.6.10-ck2/drivers/block/ll_rw_blk.c =================================================================== --- linux-2.6.10-ck2.orig/drivers/block/ll_rw_blk.c 2004-12-25 10:14:47.000000000 +1100 +++ linux-2.6.10-ck2/drivers/block/ll_rw_blk.c 2005-01-04 09:31:32.282125320 +1100 @@ -1257,11 +1257,7 @@ if (!blk_remove_plug(q)) return; - /* - * was plugged, fire request_fn if queue has stuff to do - */ - if (elv_next_request(q)) - q->request_fn(q); + q->request_fn(q); } EXPORT_SYMBOL(__generic_unplug_device); @@ -1563,8 +1559,8 @@ mempool_free(rq, q->rq.rq_pool); } -static inline struct request *blk_alloc_request(request_queue_t *q, int rw, - int gfp_mask) +static inline struct request * +blk_alloc_request(request_queue_t *q, int rw, int gfp_mask, struct bio *bio) { struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); @@ -1577,7 +1573,7 @@ */ rq->flags = rw; - if (!elv_set_request(q, rq, gfp_mask)) + if (!elv_set_request(q, rq, gfp_mask, bio)) return rq; mempool_free(rq, q->rq.rq_pool); @@ -1644,7 +1640,8 @@ /* * Get a free request, queue_lock must not be held */ -static struct request *get_request(request_queue_t *q, int rw, int gfp_mask) +static struct request *get_request(request_queue_t *q, int rw, int gfp_mask, + struct bio *bio) { struct request *rq = NULL; struct request_list *rl = &q->rq; @@ -1667,7 +1664,7 @@ } } - switch (elv_may_queue(q, rw)) { + switch (elv_may_queue(q, rw, bio)) { case ELV_MQUEUE_NO: spin_unlock_irq(q->queue_lock); goto out; @@ -1692,7 +1689,7 @@ set_queue_congested(q, rw); spin_unlock_irq(q->queue_lock); - rq = blk_alloc_request(q, rw, gfp_mask); + rq = blk_alloc_request(q, rw, gfp_mask, bio); if (!rq) { /* * Allocation failed presumably due to memory. Undo anything @@ -1734,7 +1731,8 @@ * No available requests for this queue, unplug the device and wait for some * requests to become available. */ -static struct request *get_request_wait(request_queue_t *q, int rw) +static struct request * +get_request_wait(request_queue_t *q, int rw, struct bio *bio) { DEFINE_WAIT(wait); struct request *rq; @@ -1746,7 +1744,7 @@ prepare_to_wait_exclusive(&rl->wait[rw], &wait, TASK_UNINTERRUPTIBLE); - rq = get_request(q, rw, GFP_NOIO); + rq = get_request(q, rw, GFP_NOIO, bio); if (!rq) { struct io_context *ioc; @@ -1776,9 +1774,9 @@ BUG_ON(rw != READ && rw != WRITE); if (gfp_mask & __GFP_WAIT) - rq = get_request_wait(q, rw); + rq = get_request_wait(q, rw, NULL); else - rq = get_request(q, rw, gfp_mask); + rq = get_request(q, rw, gfp_mask, NULL); return rq; } @@ -2152,7 +2150,6 @@ return; req->rq_status = RQ_INACTIVE; - req->q = NULL; req->rl = NULL; /* @@ -2419,7 +2416,7 @@ freereq = NULL; } else { spin_unlock_irq(q->queue_lock); - if ((freereq = get_request(q, rw, GFP_ATOMIC)) == NULL) { + if ((freereq = get_request(q, rw, GFP_ATOMIC, bio)) == NULL) { /* * READA bit set */ @@ -2427,7 +2424,7 @@ if (bio_rw_ahead(bio)) goto end_io; - freereq = get_request_wait(q, rw); + freereq = get_request_wait(q, rw, bio); } goto again; } @@ -2502,6 +2499,7 @@ { struct request_list *rl = &q->rq; struct request *rq; + int requeued = 0; spin_lock_irq(q->queue_lock); clear_bit(QUEUE_FLAG_DRAIN, &q->queue_flags); @@ -2510,9 +2508,13 @@ rq = list_entry_rq(q->drain_list.next); list_del_init(&rq->queuelist); - __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 1); + elv_requeue_request(q, rq); + requeued++; } + if (requeued) + q->request_fn(q); + spin_unlock_irq(q->queue_lock); wake_up(&rl->wait[0]); @@ -3069,6 +3071,7 @@ local_irq_save(flags); ioc = current->io_context; current->io_context = NULL; + ioc->task = NULL; local_irq_restore(flags); if (ioc->aic && ioc->aic->exit) @@ -3103,12 +3106,12 @@ ret = kmem_cache_alloc(iocontext_cachep, gfp_flags); if (ret) { atomic_set(&ret->refcount, 1); - ret->pid = tsk->pid; + ret->task = current; + ret->set_ioprio = NULL; ret->last_waited = jiffies; /* doesn't matter... */ ret->nr_batch_requests = 0; /* because this is 0 */ ret->aic = NULL; ret->cic = NULL; - spin_lock_init(&ret->lock); local_irq_save(flags); @@ -3405,3 +3408,4 @@ kobject_put(&disk->kobj); } } + Index: linux-2.6.10-ck2/fs/buffer.c =================================================================== --- linux-2.6.10-ck2.orig/fs/buffer.c 2004-12-25 10:14:50.000000000 +1100 +++ linux-2.6.10-ck2/fs/buffer.c 2005-01-04 09:31:32.284125016 +1100 @@ -39,6 +39,7 @@ #include #include #include +#include static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); static void invalidate_bh_lrus(void); @@ -858,6 +859,7 @@ spin_unlock(&mapping->private_lock); if (!TestSetPageDirty(page)) { + set_page_dirty_pid(page, current->pid); spin_lock_irq(&mapping->tree_lock); if (page->mapping) { /* Race with truncate? */ if (!mapping->backing_dev_info->memory_backed) Index: linux-2.6.10-ck2/fs/ioprio.c =================================================================== --- linux-2.6.10-ck2.orig/fs/ioprio.c 2004-11-30 03:07:23.000000000 +1100 +++ linux-2.6.10-ck2/fs/ioprio.c 2005-01-04 09:31:32.284125016 +1100 @@ -0,0 +1,46 @@ +#include +#include +#include + +asmlinkage int sys_ioprio_set(int ioprio) +{ + int class = IOPRIO_PRIO_CLASS(ioprio); + int data = IOPRIO_PRIO_DATA(ioprio); + struct io_context *ioc; + unsigned long flags; + + switch (class) { + case IOPRIO_CLASS_RT: + if (!capable(CAP_SYS_ADMIN)) + return -EPERM; + /* fall through, rt has prio field too */ + case IOPRIO_CLASS_BE: + if (data >= IOPRIO_BE_NR) + return -EINVAL; + + break; + case IOPRIO_CLASS_IDLE: + break; + default: + return -EINVAL; + } + + ioc = get_io_context(GFP_KERNEL); + if (!ioc) + return -ENOMEM; + + local_irq_save(flags); + current->ioprio = ioprio; + if (ioc->set_ioprio) + ioc->set_ioprio(ioc, ioprio); + local_irq_restore(flags); + + put_io_context(ioc); + return 0; +} + +asmlinkage int sys_ioprio_get(void) +{ + return current->ioprio; +} + Index: linux-2.6.10-ck2/fs/Makefile =================================================================== --- linux-2.6.10-ck2.orig/fs/Makefile 2004-12-25 10:14:50.000000000 +1100 +++ linux-2.6.10-ck2/fs/Makefile 2005-01-04 09:31:32.284125016 +1100 @@ -10,6 +10,7 @@ ioctl.o readdir.o select.o fifo.o locks.o dcache.o inode.o \ attr.o bad_inode.o file.o filesystems.o namespace.o aio.o \ seq_file.o xattr.o libfs.o fs-writeback.o mpage.o direct-io.o \ + ioprio.o obj-$(CONFIG_EPOLL) += eventpoll.o obj-$(CONFIG_COMPAT) += compat.o Index: linux-2.6.10-ck2/fs/reiserfs/journal.c =================================================================== --- linux-2.6.10-ck2.orig/fs/reiserfs/journal.c 2004-12-25 10:14:50.000000000 +1100 +++ linux-2.6.10-ck2/fs/reiserfs/journal.c 2005-01-04 09:31:32.286124712 +1100 @@ -659,18 +659,22 @@ static void write_chunk(struct buffer_chunk *chunk) { int i; + get_fs_excl(); for (i = 0; i < chunk->nr ; i++) { submit_logged_buffer(chunk->bh[i]) ; } chunk->nr = 0; + put_fs_excl(); } static void write_ordered_chunk(struct buffer_chunk *chunk) { int i; + get_fs_excl(); for (i = 0; i < chunk->nr ; i++) { submit_ordered_buffer(chunk->bh[i]) ; } chunk->nr = 0; + put_fs_excl(); } static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh, @@ -932,6 +936,8 @@ return 0 ; } + get_fs_excl(); + /* before we can put our commit blocks on disk, we have to make sure everyone older than ** us is on disk too */ @@ -1069,6 +1075,7 @@ if (retval) reiserfs_abort (s, retval, "Journal write error in %s", __FUNCTION__); + put_fs_excl(); return retval; } @@ -1265,6 +1272,8 @@ return 0 ; } + get_fs_excl(); + /* if all the work is already done, get out of here */ if (atomic_read(&(jl->j_nonzerolen)) <= 0 && atomic_read(&(jl->j_commit_left)) <= 0) { @@ -1464,6 +1473,7 @@ put_journal_list(s, jl); if (flushall) up(&journal->j_flush_sem); + put_fs_excl(); return err ; } @@ -2733,6 +2743,7 @@ th->t_trans_id = journal->j_trans_id ; unlock_journal(p_s_sb) ; INIT_LIST_HEAD (&th->t_list); + get_fs_excl(); return 0 ; out_fail: @@ -3536,6 +3547,7 @@ BUG_ON (th->t_refcount > 1); BUG_ON (!th->t_trans_id); + put_fs_excl(); current->journal_info = th->t_handle_save; reiserfs_check_lock_depth(p_s_sb, "journal end"); if (journal->j_len == 0) { Index: linux-2.6.10-ck2/include/asm-i386/unistd.h =================================================================== --- linux-2.6.10-ck2.orig/include/asm-i386/unistd.h 2004-12-25 10:14:50.000000000 +1100 +++ linux-2.6.10-ck2/include/asm-i386/unistd.h 2005-01-04 09:31:32.286124712 +1100 @@ -294,8 +294,10 @@ #define __NR_add_key 286 #define __NR_request_key 287 #define __NR_keyctl 288 +#define __NR_ioprio_set 289 +#define __NR_ioprio_get 290 -#define NR_syscalls 289 +#define NR_syscalls 291 /* * user-visible error numbers are in the range -1 - -128: see Index: linux-2.6.10-ck2/include/asm-ia64/unistd.h =================================================================== --- linux-2.6.10-ck2.orig/include/asm-ia64/unistd.h 2004-12-25 10:14:51.000000000 +1100 +++ linux-2.6.10-ck2/include/asm-ia64/unistd.h 2005-01-04 09:31:32.287124560 +1100 @@ -263,6 +263,8 @@ #define __NR_add_key 1271 #define __NR_request_key 1272 #define __NR_keyctl 1273 +#define __NR_ioprio_set 1274 +#define __NR_ioprio_get 1275 #ifdef __KERNEL__ Index: linux-2.6.10-ck2/include/asm-ppc/unistd.h =================================================================== --- linux-2.6.10-ck2.orig/include/asm-ppc/unistd.h 2004-12-25 10:14:51.000000000 +1100 +++ linux-2.6.10-ck2/include/asm-ppc/unistd.h 2005-01-04 09:31:32.287124560 +1100 @@ -276,8 +276,10 @@ #define __NR_add_key 269 #define __NR_request_key 270 #define __NR_keyctl 271 +#define __NR_ioprio_set 272 +#define __NR_ioprio_get 273 -#define __NR_syscalls 272 +#define __NR_syscalls 274 #define __NR(n) #n Index: linux-2.6.10-ck2/include/asm-x86_64/unistd.h =================================================================== --- linux-2.6.10-ck2.orig/include/asm-x86_64/unistd.h 2004-10-19 08:57:12.000000000 +1000 +++ linux-2.6.10-ck2/include/asm-x86_64/unistd.h 2005-01-04 09:31:32.288124408 +1100 @@ -556,8 +556,12 @@ __SYSCALL(__NR_kexec_load, sys_ni_syscall) #define __NR_waitid 247 __SYSCALL(__NR_waitid, sys_waitid) +#define __NR_ioprio_set 248 +__SYSCALL(__NR_ioprio_set, sys_ioprio_set) +#define __NR_ioprio_get 249 +__SYSCALL(__NR_ioprio_get, sys_ioprio_get) -#define __NR_syscall_max __NR_waitid +#define __NR_syscall_max __NR_ioprio_get #ifndef __NO_STUBS /* user-visible error numbers are in the range -1 - -4095 */ Index: linux-2.6.10-ck2/include/linux/blkdev.h =================================================================== --- linux-2.6.10-ck2.orig/include/linux/blkdev.h 2004-12-25 10:14:51.000000000 +1100 +++ linux-2.6.10-ck2/include/linux/blkdev.h 2005-01-04 09:31:32.288124408 +1100 @@ -54,16 +54,27 @@ struct cfq_queue; struct cfq_io_context { - void (*dtor)(struct cfq_io_context *); - void (*exit)(struct cfq_io_context *); - - struct io_context *ioc; - /* * circular list of cfq_io_contexts belonging to a process io context */ struct list_head list; struct cfq_queue *cfqq; + + struct io_context *ioc; + + unsigned long last_end_request; + unsigned long last_queue; + unsigned long ttime_total; + unsigned long ttime_samples; + unsigned long ttime_mean; + + unsigned int seek_samples; + sector_t last_request_pos; + u64 seek_total; + sector_t seek_mean; + + void (*dtor)(struct cfq_io_context *); + void (*exit)(struct cfq_io_context *); }; /* @@ -73,7 +84,9 @@ */ struct io_context { atomic_t refcount; - pid_t pid; + struct task_struct *task; + + int (*set_ioprio)(struct io_context *, unsigned int); /* * For request batching @@ -81,8 +94,6 @@ unsigned long last_waited; /* Time last woken after wait for request */ int nr_batch_requests; /* Number of requests left in the batch */ - spinlock_t lock; - struct as_io_context *aic; struct cfq_io_context *cic; }; @@ -723,5 +734,4 @@ #define MODULE_ALIAS_BLOCKDEV_MAJOR(major) \ MODULE_ALIAS("block-major-" __stringify(major) "-*") - #endif Index: linux-2.6.10-ck2/include/linux/elevator.h =================================================================== --- linux-2.6.10-ck2.orig/include/linux/elevator.h 2004-12-25 10:14:51.000000000 +1100 +++ linux-2.6.10-ck2/include/linux/elevator.h 2005-01-04 09:31:32.289124256 +1100 @@ -16,9 +16,9 @@ typedef void (elevator_requeue_req_fn) (request_queue_t *, struct request *); typedef struct request *(elevator_request_list_fn) (request_queue_t *, struct request *); typedef void (elevator_completed_req_fn) (request_queue_t *, struct request *); -typedef int (elevator_may_queue_fn) (request_queue_t *, int); +typedef int (elevator_may_queue_fn) (request_queue_t *, int, struct bio *); -typedef int (elevator_set_req_fn) (request_queue_t *, struct request *, int); +typedef int (elevator_set_req_fn) (request_queue_t *, struct request *, int, struct bio *); typedef void (elevator_put_req_fn) (request_queue_t *, struct request *); typedef int (elevator_init_fn) (request_queue_t *, elevator_t *); @@ -93,9 +93,9 @@ extern struct request *elv_latter_request(request_queue_t *, struct request *); extern int elv_register_queue(request_queue_t *q); extern void elv_unregister_queue(request_queue_t *q); -extern int elv_may_queue(request_queue_t *, int); +extern int elv_may_queue(request_queue_t *, int, struct bio *); extern void elv_completed_request(request_queue_t *, struct request *); -extern int elv_set_request(request_queue_t *, struct request *, int); +extern int elv_set_request(request_queue_t *, struct request *, int, struct bio *); extern void elv_put_request(request_queue_t *, struct request *); /* Index: linux-2.6.10-ck2/include/linux/fs.h =================================================================== --- linux-2.6.10-ck2.orig/include/linux/fs.h 2004-12-25 10:14:51.000000000 +1100 +++ linux-2.6.10-ck2/include/linux/fs.h 2005-01-04 09:31:32.290124104 +1100 @@ -18,6 +18,7 @@ #include #include #include +#include #include struct iovec; @@ -811,16 +812,34 @@ #define vfs_check_frozen(sb, level) \ wait_event((sb)->s_wait_unfrozen, ((sb)->s_frozen < (level))) +static inline void get_fs_excl(void) +{ + atomic_inc(¤t->fs_excl); +} + +static inline void put_fs_excl(void) +{ + atomic_dec(¤t->fs_excl); +} + +static inline int has_fs_excl(void) +{ + return atomic_read(¤t->fs_excl); +} + + /* * Superblock locking. */ static inline void lock_super(struct super_block * sb) { + get_fs_excl(); down(&sb->s_lock); } static inline void unlock_super(struct super_block * sb) { + put_fs_excl(); up(&sb->s_lock); } Index: linux-2.6.10-ck2/include/linux/init_task.h =================================================================== --- linux-2.6.10-ck2.orig/include/linux/init_task.h 2005-01-04 09:30:05.624299312 +1100 +++ linux-2.6.10-ck2/include/linux/init_task.h 2005-01-04 09:31:32.290124104 +1100 @@ -79,6 +79,7 @@ .mm = NULL, \ .active_mm = &init_mm, \ .run_list = LIST_HEAD_INIT(tsk.run_list), \ + .ioprio = 0, \ .time_slice = HZ, \ .tasks = LIST_HEAD_INIT(tsk.tasks), \ .ptrace_children= LIST_HEAD_INIT(tsk.ptrace_children), \ @@ -112,6 +113,7 @@ .proc_lock = SPIN_LOCK_UNLOCKED, \ .switch_lock = SPIN_LOCK_UNLOCKED, \ .journal_info = NULL, \ + .fs_excl = ATOMIC_INIT(0), \ } Index: linux-2.6.10-ck2/include/linux/ioprio.h =================================================================== --- linux-2.6.10-ck2.orig/include/linux/ioprio.h 2004-11-30 03:07:23.000000000 +1100 +++ linux-2.6.10-ck2/include/linux/ioprio.h 2005-01-04 09:31:32.290124104 +1100 @@ -0,0 +1,50 @@ +#ifndef IOPRIO_H +#define IOPRIO_H + +#include + +/* + * Gives us 8 prio classes with 24-bits of data for each class + */ +#define IOPRIO_CLASS_SHIFT (24) +#define IOPRIO_PRIO_MASK ((1UL << IOPRIO_CLASS_SHIFT) - 1) + +#define IOPRIO_PRIO_CLASS(mask) ((mask) >> IOPRIO_CLASS_SHIFT) +#define IOPRIO_PRIO_DATA(mask) ((mask) & IOPRIO_PRIO_MASK) + +enum { + IOPRIO_CLASS_NONE, + IOPRIO_CLASS_RT, + IOPRIO_CLASS_BE, + IOPRIO_CLASS_IDLE, +}; + +/* + * 8 best effort priority levels are supported + */ +#define IOPRIO_BE_NR (8) + +asmlinkage int sys_ioprio_set(int); +asmlinkage int sys_ioprio_get(void); + +/* + * if process has set io priority explicitly, use that. if not, convert + * the cpu scheduler nice value to an io priority + */ +static inline int task_ioprio(struct task_struct *task) +{ + if (!task->ioprio) + return (task_nice(task) + 20) / 5; + else + return IOPRIO_PRIO_DATA(task->ioprio); +} + +#if defined(CONFIG_IOPRIO_WRITE) +#define set_page_dirty_pid(page, pid) ((page)->dirty_pid = (pid)) +#define get_page_dirty_pid(page) ((page)->dirty_pid) +#else +#define set_page_dirty_pid(page, pid) +#define get_page_dirty_pid(page) (0) +#endif + +#endif Index: linux-2.6.10-ck2/include/linux/mm.h =================================================================== --- linux-2.6.10-ck2.orig/include/linux/mm.h 2004-12-25 10:14:51.000000000 +1100 +++ linux-2.6.10-ck2/include/linux/mm.h 2005-01-04 09:31:32.291123952 +1100 @@ -237,6 +237,12 @@ void *virtual; /* Kernel virtual address (NULL if not kmapped, ie. highmem) */ #endif /* WANT_PAGE_VIRTUAL */ + /* + * Keep track of who dirtied this page + */ +#if defined(CONFIG_IOPRIO_WRITE) + pid_t dirty_pid; +#endif }; /* Index: linux-2.6.10-ck2/include/linux/sched.h =================================================================== --- linux-2.6.10-ck2.orig/include/linux/sched.h 2005-01-04 09:30:09.071775216 +1100 +++ linux-2.6.10-ck2/include/linux/sched.h 2005-01-04 09:31:32.292123800 +1100 @@ -536,7 +536,7 @@ unsigned long long timestamp; unsigned long runtime, totalrun, ns_debit; - unsigned int burst; + unsigned int burst, ioprio; unsigned long policy; cpumask_t cpus_allowed; @@ -674,6 +674,7 @@ struct mempolicy *mempolicy; short il_next; /* could be shared with used_math */ #endif + atomic_t fs_excl; }; static inline pid_t process_group(struct task_struct *tsk) Index: linux-2.6.10-ck2/kernel/exit.c =================================================================== --- linux-2.6.10-ck2.orig/kernel/exit.c 2005-01-04 09:30:04.045539320 +1100 +++ linux-2.6.10-ck2/kernel/exit.c 2005-01-04 09:31:32.292123800 +1100 @@ -784,6 +784,8 @@ profile_task_exit(tsk); + WARN_ON(atomic_read(&tsk->fs_excl)); + if (unlikely(in_interrupt())) panic("Aiee, killing interrupt handler!"); if (unlikely(!tsk->pid)) Index: linux-2.6.10-ck2/kernel/fork.c =================================================================== --- linux-2.6.10-ck2.orig/kernel/fork.c 2004-12-25 10:14:51.000000000 +1100 +++ linux-2.6.10-ck2/kernel/fork.c 2005-01-04 09:31:32.293123648 +1100 @@ -1006,6 +1006,11 @@ spin_unlock(¤t->sighand->siglock); } + /* + * inherit ioprio + */ + p->ioprio = current->ioprio; + SET_LINKS(p); if (unlikely(p->ptrace & PT_PTRACED)) __ptrace_link(p, current->parent); Index: linux-2.6.10-ck2/mm/page_alloc.c =================================================================== --- linux-2.6.10-ck2.orig/mm/page_alloc.c 2005-01-04 09:30:06.794121472 +1100 +++ linux-2.6.10-ck2/mm/page_alloc.c 2005-01-04 09:31:32.294123496 +1100 @@ -32,6 +32,7 @@ #include #include #include +#include #include @@ -1418,6 +1419,7 @@ reset_page_mapcount(page); SetPageReserved(page); INIT_LIST_HEAD(&page->lru); + set_page_dirty_pid(page, 0); #ifdef WANT_PAGE_VIRTUAL /* The shift won't overflow because ZONE_NORMAL is below 4G. */ if (!is_highmem_idx(zone)) Index: linux-2.6.10-ck2/mm/page-writeback.c =================================================================== --- linux-2.6.10-ck2.orig/mm/page-writeback.c 2005-01-04 09:30:10.691528976 +1100 +++ linux-2.6.10-ck2/mm/page-writeback.c 2005-01-04 09:31:32.295123344 +1100 @@ -29,6 +29,7 @@ #include #include #include +#include /* * The maximum number of pages to writeout in a single bdflush/kupdate @@ -609,6 +610,8 @@ struct address_space *mapping = page_mapping(page); struct address_space *mapping2; + set_page_dirty_pid(page, current->pid); + if (mapping) { spin_lock_irq(&mapping->tree_lock); mapping2 = page_mapping(page); @@ -651,6 +654,8 @@ { struct address_space *mapping = page_mapping(page); + set_page_dirty_pid(page, current->pid); + if (likely(mapping)) { int (*spd)(struct page *) = mapping->a_ops->set_page_dirty; if (spd)