diff --git a/drivers/block/elevator.c b/drivers/block/elevator.c index 3b8099ccedff8be73d1169c13579d7bf7b98b801..55621d5c577403e3024a6dac9b580185c6c6b9cf 100644 --- a/drivers/block/elevator.c +++ b/drivers/block/elevator.c @@ -34,6 +34,7 @@ #include #include #include +#include #include @@ -131,11 +132,7 @@ static int elevator_attach(request_queue_t *q, struct elevator_type *e, eq->ops = &e->ops; eq->elevator_type = e; - INIT_LIST_HEAD(&q->queue_head); - q->last_merge = NULL; q->elevator = eq; - q->end_sector = 0; - q->boundary_rq = NULL; if (eq->ops->elevator_init_fn) ret = eq->ops->elevator_init_fn(q, eq); @@ -184,6 +181,11 @@ int elevator_init(request_queue_t *q, char *name) struct elevator_queue *eq; int ret = 0; + INIT_LIST_HEAD(&q->queue_head); + q->last_merge = NULL; + q->end_sector = 0; + q->boundary_rq = NULL; + elevator_setup_default(); if (!name) @@ -336,23 +338,14 @@ void __elv_add_request(request_queue_t *q, struct request *rq, int where, q->end_sector = rq_end_sector(rq); q->boundary_rq = rq; } - } + } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT) + where = ELEVATOR_INSERT_BACK; if (plug) blk_plug_device(q); rq->q = q; - if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) { - /* - * if drain is set, store the request "locally". when the drain - * is finished, the requests will be handed ordered to the io - * scheduler - */ - list_add_tail(&rq->queuelist, &q->drain_list); - return; - } - switch (where) { case ELEVATOR_INSERT_FRONT: rq->flags |= REQ_SOFTBARRIER; @@ -659,25 +652,36 @@ EXPORT_SYMBOL_GPL(elv_unregister); * switch to new_e io scheduler. be careful not to introduce deadlocks - * we don't free the old io scheduler, before we have allocated what we * need for the new one. this way we have a chance of going back to the old - * one, if the new one fails init for some reason. we also do an intermediate - * switch to noop to ensure safety with stack-allocated requests, since they - * don't originate from the block layer allocator. noop is safe here, because - * it never needs to touch the elevator itself for completion events. DRAIN - * flags will make sure we don't touch it for additions either. + * one, if the new one fails init for some reason. */ static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) { - elevator_t *e = kmalloc(sizeof(elevator_t), GFP_KERNEL); - struct elevator_type *noop_elevator = NULL; - elevator_t *old_elevator; + elevator_t *old_elevator, *e; + /* + * Allocate new elevator + */ + e = kmalloc(sizeof(elevator_t), GFP_KERNEL); if (!e) goto error; /* - * first step, drain requests from the block freelist + * Turn on BYPASS and drain all requests w/ elevator private data */ - blk_wait_queue_drained(q, 0); + spin_lock_irq(q->queue_lock); + + set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); + + while (q->elevator->ops->elevator_dispatch_fn(q, 1)) + ; + + while (q->rq.elvpriv) { + spin_unlock_irq(q->queue_lock); + msleep(10); + spin_lock_irq(q->queue_lock); + } + + spin_unlock_irq(q->queue_lock); /* * unregister old elevator data @@ -685,18 +689,6 @@ static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) elv_unregister_queue(q); old_elevator = q->elevator; - /* - * next step, switch to noop since it uses no private rq structures - * and doesn't allocate any memory for anything. then wait for any - * non-fs requests in-flight - */ - noop_elevator = elevator_get("noop"); - spin_lock_irq(q->queue_lock); - elevator_attach(q, noop_elevator, e); - spin_unlock_irq(q->queue_lock); - - blk_wait_queue_drained(q, 1); - /* * attach and start new elevator */ @@ -707,11 +699,10 @@ static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) goto fail_register; /* - * finally exit old elevator and start queue again + * finally exit old elevator and turn off BYPASS. */ elevator_exit(old_elevator); - blk_finish_queue_drain(q); - elevator_put(noop_elevator); + clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); return; fail_register: @@ -720,13 +711,13 @@ static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) * one again (along with re-adding the sysfs dir) */ elevator_exit(e); + e = NULL; fail: q->elevator = old_elevator; elv_register_queue(q); - blk_finish_queue_drain(q); + clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); + kfree(e); error: - if (noop_elevator) - elevator_put(noop_elevator); elevator_put(new_e); printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); } diff --git a/drivers/block/ll_rw_blk.c b/drivers/block/ll_rw_blk.c index c4f186223bc56459f83f9face3ccd2998a65b010..0af73512b9a8c358e0bf4ae07922590c4700011d 100644 --- a/drivers/block/ll_rw_blk.c +++ b/drivers/block/ll_rw_blk.c @@ -263,8 +263,6 @@ void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn) blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); blk_queue_activity_fn(q, NULL, NULL); - - INIT_LIST_HEAD(&q->drain_list); } EXPORT_SYMBOL(blk_queue_make_request); @@ -1050,6 +1048,7 @@ static char *rq_flags[] = { "REQ_STARTED", "REQ_DONTPREP", "REQ_QUEUED", + "REQ_ELVPRIV", "REQ_PC", "REQ_BLOCK_PC", "REQ_SENSE", @@ -1640,9 +1639,9 @@ static int blk_init_free_list(request_queue_t *q) rl->count[READ] = rl->count[WRITE] = 0; rl->starved[READ] = rl->starved[WRITE] = 0; + rl->elvpriv = 0; init_waitqueue_head(&rl->wait[READ]); init_waitqueue_head(&rl->wait[WRITE]); - init_waitqueue_head(&rl->drain); rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, mempool_free_slab, request_cachep, q->node); @@ -1785,12 +1784,14 @@ EXPORT_SYMBOL(blk_get_queue); static inline void blk_free_request(request_queue_t *q, struct request *rq) { - elv_put_request(q, rq); + if (rq->flags & REQ_ELVPRIV) + elv_put_request(q, rq); mempool_free(rq, q->rq.rq_pool); } static inline struct request * -blk_alloc_request(request_queue_t *q, int rw, struct bio *bio, gfp_t gfp_mask) +blk_alloc_request(request_queue_t *q, int rw, struct bio *bio, + int priv, gfp_t gfp_mask) { struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); @@ -1803,11 +1804,15 @@ blk_alloc_request(request_queue_t *q, int rw, struct bio *bio, gfp_t gfp_mask) */ rq->flags = rw; - if (!elv_set_request(q, rq, bio, gfp_mask)) - return rq; + if (priv) { + if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) { + mempool_free(rq, q->rq.rq_pool); + return NULL; + } + rq->flags |= REQ_ELVPRIV; + } - mempool_free(rq, q->rq.rq_pool); - return NULL; + return rq; } /* @@ -1863,22 +1868,18 @@ static void __freed_request(request_queue_t *q, int rw) * A request has just been released. Account for it, update the full and * congestion status, wake up any waiters. Called under q->queue_lock. */ -static void freed_request(request_queue_t *q, int rw) +static void freed_request(request_queue_t *q, int rw, int priv) { struct request_list *rl = &q->rq; rl->count[rw]--; + if (priv) + rl->elvpriv--; __freed_request(q, rw); if (unlikely(rl->starved[rw ^ 1])) __freed_request(q, rw ^ 1); - - if (!rl->count[READ] && !rl->count[WRITE]) { - smp_mb(); - if (unlikely(waitqueue_active(&rl->drain))) - wake_up(&rl->drain); - } } #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) @@ -1893,9 +1894,7 @@ static struct request *get_request(request_queue_t *q, int rw, struct bio *bio, struct request *rq = NULL; struct request_list *rl = &q->rq; struct io_context *ioc = current_io_context(GFP_ATOMIC); - - if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) - goto out; + int priv; if (rl->count[rw]+1 >= q->nr_requests) { /* @@ -1940,9 +1939,14 @@ static struct request *get_request(request_queue_t *q, int rw, struct bio *bio, rl->starved[rw] = 0; if (rl->count[rw] >= queue_congestion_on_threshold(q)) set_queue_congested(q, rw); + + priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); + if (priv) + rl->elvpriv++; + spin_unlock_irq(q->queue_lock); - rq = blk_alloc_request(q, rw, bio, gfp_mask); + rq = blk_alloc_request(q, rw, bio, priv, gfp_mask); if (!rq) { /* * Allocation failed presumably due to memory. Undo anything @@ -1952,7 +1956,7 @@ static struct request *get_request(request_queue_t *q, int rw, struct bio *bio, * wait queue, but this is pretty rare. */ spin_lock_irq(q->queue_lock); - freed_request(q, rw); + freed_request(q, rw, priv); /* * in the very unlikely event that allocation failed and no @@ -2470,11 +2474,12 @@ static void __blk_put_request(request_queue_t *q, struct request *req) */ if (rl) { int rw = rq_data_dir(req); + int priv = req->flags & REQ_ELVPRIV; BUG_ON(!list_empty(&req->queuelist)); blk_free_request(q, req); - freed_request(q, rw); + freed_request(q, rw, priv); } } @@ -2802,97 +2807,6 @@ static inline void blk_partition_remap(struct bio *bio) } } -void blk_finish_queue_drain(request_queue_t *q) -{ - 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); - - while (!list_empty(&q->drain_list)) { - rq = list_entry_rq(q->drain_list.next); - - list_del_init(&rq->queuelist); - elv_requeue_request(q, rq); - requeued++; - } - - if (requeued) - q->request_fn(q); - - spin_unlock_irq(q->queue_lock); - - wake_up(&rl->wait[0]); - wake_up(&rl->wait[1]); - wake_up(&rl->drain); -} - -static int wait_drain(request_queue_t *q, struct request_list *rl, int dispatch) -{ - int wait = rl->count[READ] + rl->count[WRITE]; - - if (dispatch) - wait += !list_empty(&q->queue_head); - - return wait; -} - -/* - * We rely on the fact that only requests allocated through blk_alloc_request() - * have io scheduler private data structures associated with them. Any other - * type of request (allocated on stack or through kmalloc()) should not go - * to the io scheduler core, but be attached to the queue head instead. - */ -void blk_wait_queue_drained(request_queue_t *q, int wait_dispatch) -{ - struct request_list *rl = &q->rq; - DEFINE_WAIT(wait); - - spin_lock_irq(q->queue_lock); - set_bit(QUEUE_FLAG_DRAIN, &q->queue_flags); - - while (wait_drain(q, rl, wait_dispatch)) { - prepare_to_wait(&rl->drain, &wait, TASK_UNINTERRUPTIBLE); - - if (wait_drain(q, rl, wait_dispatch)) { - __generic_unplug_device(q); - spin_unlock_irq(q->queue_lock); - io_schedule(); - spin_lock_irq(q->queue_lock); - } - - finish_wait(&rl->drain, &wait); - } - - spin_unlock_irq(q->queue_lock); -} - -/* - * block waiting for the io scheduler being started again. - */ -static inline void block_wait_queue_running(request_queue_t *q) -{ - DEFINE_WAIT(wait); - - while (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) { - struct request_list *rl = &q->rq; - - prepare_to_wait_exclusive(&rl->drain, &wait, - TASK_UNINTERRUPTIBLE); - - /* - * re-check the condition. avoids using prepare_to_wait() - * in the fast path (queue is running) - */ - if (test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags)) - io_schedule(); - - finish_wait(&rl->drain, &wait); - } -} - static void handle_bad_sector(struct bio *bio) { char b[BDEVNAME_SIZE]; @@ -2988,8 +2902,6 @@ void generic_make_request(struct bio *bio) if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) goto end_io; - block_wait_queue_running(q); - /* * If this device has partitions, remap block n * of partition p to block n+start(p) of the disk. diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h index fffe163316aa4b87c239b2b744d218456fd685fc..025a7f084dbd82b2caf7a290736287bc3f012d37 100644 --- a/include/linux/blkdev.h +++ b/include/linux/blkdev.h @@ -107,9 +107,9 @@ typedef void (rq_end_io_fn)(struct request *); struct request_list { int count[2]; int starved[2]; + int elvpriv; mempool_t *rq_pool; wait_queue_head_t wait[2]; - wait_queue_head_t drain; }; #define BLK_MAX_CDB 16 @@ -211,6 +211,7 @@ enum rq_flag_bits { __REQ_STARTED, /* drive already may have started this one */ __REQ_DONTPREP, /* don't call prep for this one */ __REQ_QUEUED, /* uses queueing */ + __REQ_ELVPRIV, /* elevator private data attached */ /* * for ATA/ATAPI devices */ @@ -244,6 +245,7 @@ enum rq_flag_bits { #define REQ_STARTED (1 << __REQ_STARTED) #define REQ_DONTPREP (1 << __REQ_DONTPREP) #define REQ_QUEUED (1 << __REQ_QUEUED) +#define REQ_ELVPRIV (1 << __REQ_ELVPRIV) #define REQ_PC (1 << __REQ_PC) #define REQ_BLOCK_PC (1 << __REQ_BLOCK_PC) #define REQ_SENSE (1 << __REQ_SENSE) @@ -413,8 +415,6 @@ struct request_queue unsigned int sg_reserved_size; int node; - struct list_head drain_list; - /* * reserved for flush operations */ @@ -442,7 +442,7 @@ enum { #define QUEUE_FLAG_DEAD 5 /* queue being torn down */ #define QUEUE_FLAG_REENTER 6 /* Re-entrancy avoidance */ #define QUEUE_FLAG_PLUGGED 7 /* queue is plugged */ -#define QUEUE_FLAG_DRAIN 8 /* draining queue for sched switch */ +#define QUEUE_FLAG_ELVSWITCH 8 /* don't use elevator, just do FIFO */ #define QUEUE_FLAG_FLUSH 9 /* doing barrier flush sequence */ #define blk_queue_plugged(q) test_bit(QUEUE_FLAG_PLUGGED, &(q)->queue_flags) @@ -668,8 +668,6 @@ extern void blk_dump_rq_flags(struct request *, char *); extern void generic_unplug_device(request_queue_t *); extern void __generic_unplug_device(request_queue_t *); extern long nr_blockdev_pages(void); -extern void blk_wait_queue_drained(request_queue_t *, int); -extern void blk_finish_queue_drain(request_queue_t *); int blk_get_queue(request_queue_t *); request_queue_t *blk_alloc_queue(gfp_t);