[PATCH] elevator_t lifetime rules and sysfs fixes
[pandora-kernel.git] / block / cfq-iosched.c
index ee0bb41..6dc156e 100644 (file)
@@ -6,34 +6,26 @@
  *
  *  Copyright (C) 2003 Jens Axboe <axboe@suse.de>
  */
-#include <linux/kernel.h>
-#include <linux/fs.h>
-#include <linux/blkdev.h>
-#include <linux/elevator.h>
-#include <linux/bio.h>
 #include <linux/config.h>
 #include <linux/module.h>
-#include <linux/slab.h>
-#include <linux/init.h>
-#include <linux/compiler.h>
+#include <linux/blkdev.h>
+#include <linux/elevator.h>
 #include <linux/hash.h>
 #include <linux/rbtree.h>
-#include <linux/mempool.h>
 #include <linux/ioprio.h>
-#include <linux/writeback.h>
 
 /*
  * tunables
  */
-static int cfq_quantum = 4;            /* max queue in one round of service */
-static int cfq_queued = 8;             /* minimum rq allocate limit per-queue*/
-static int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
-static int cfq_back_max = 16 * 1024;   /* maximum backwards seek, in KiB */
-static int cfq_back_penalty = 2;       /* penalty of a backwards seek */
+static const int cfq_quantum = 4;              /* max queue in one round of service */
+static const int cfq_queued = 8;               /* minimum rq allocate limit per-queue*/
+static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 };
+static const int cfq_back_max = 16 * 1024;     /* maximum backwards seek, in KiB */
+static const int cfq_back_penalty = 2;         /* penalty of a backwards seek */
 
-static int cfq_slice_sync = HZ / 10;
+static const int cfq_slice_sync = HZ / 10;
 static int cfq_slice_async = HZ / 25;
-static int cfq_slice_async_rq = 2;
+static const int cfq_slice_async_rq = 2;
 static int cfq_slice_idle = HZ / 100;
 
 #define CFQ_IDLE_GRACE         (HZ / 10)
@@ -45,7 +37,9 @@ static int cfq_slice_idle = HZ / 100;
 /*
  * disable queueing at the driver/hardware level
  */
-static int cfq_max_depth = 2;
+static const int cfq_max_depth = 2;
+
+static DEFINE_RWLOCK(cfq_exit_lock);
 
 /*
  * for the hash of cfqq inside the cfqd
@@ -89,6 +83,9 @@ static kmem_cache_t *crq_pool;
 static kmem_cache_t *cfq_pool;
 static kmem_cache_t *cfq_ioc_pool;
 
+static atomic_t ioc_count = ATOMIC_INIT(0);
+static struct completion *ioc_gone;
+
 #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)
@@ -109,7 +106,6 @@ static kmem_cache_t *cfq_ioc_pool;
  * Per block device queue structure
  */
 struct cfq_data {
-       atomic_t ref;
        request_queue_t *queue;
 
        /*
@@ -175,6 +171,8 @@ struct cfq_data {
        unsigned int cfq_slice_async_rq;
        unsigned int cfq_slice_idle;
        unsigned int cfq_max_depth;
+
+       struct list_head cic_list;
 };
 
 /*
@@ -239,7 +237,6 @@ enum cfqq_state_flags {
        CFQ_CFQQ_FLAG_fifo_expire,
        CFQ_CFQQ_FLAG_idle_window,
        CFQ_CFQQ_FLAG_prio_changed,
-       CFQ_CFQQ_FLAG_expired,
 };
 
 #define CFQ_CFQQ_FNS(name)                                             \
@@ -264,7 +261,6 @@ CFQ_CFQQ_FNS(must_dispatch);
 CFQ_CFQQ_FNS(fifo_expire);
 CFQ_CFQQ_FNS(idle_window);
 CFQ_CFQQ_FNS(prio_changed);
-CFQ_CFQQ_FNS(expired);
 #undef CFQ_CFQQ_FNS
 
 enum cfq_rq_state_flags {
@@ -290,7 +286,7 @@ CFQ_CRQ_FNS(is_sync);
 
 static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short);
 static void cfq_dispatch_insert(request_queue_t *, struct cfq_rq *);
-static void cfq_put_cfqd(struct cfq_data *cfqd);
+static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk, gfp_t gfp_mask);
 
 #define process_sync(tsk)      ((tsk)->flags & PF_SYNCWRITE)
 
@@ -336,7 +332,7 @@ static struct request *cfq_find_rq_hash(struct cfq_data *cfqd, sector_t offset)
  */
 static inline void cfq_schedule_dispatch(struct cfq_data *cfqd)
 {
-       if (!cfqd->rq_in_driver && cfqd->busy_queues)
+       if (cfqd->busy_queues)
                kblockd_schedule_work(&cfqd->unplug_work);
 }
 
@@ -736,12 +732,62 @@ __cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
                cfqq->slice_left = 0;
                cfq_clear_cfqq_must_alloc_slice(cfqq);
                cfq_clear_cfqq_fifo_expire(cfqq);
-               cfq_clear_cfqq_expired(cfqq);
        }
 
        cfqd->active_queue = cfqq;
 }
 
+/*
+ * current cfqq expired its slice (or was too idle), select new one
+ */
+static void
+__cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
+                   int preempted)
+{
+       unsigned long now = jiffies;
+
+       if (cfq_cfqq_wait_request(cfqq))
+               del_timer(&cfqd->idle_slice_timer);
+
+       if (!preempted && !cfq_cfqq_dispatched(cfqq)) {
+               cfqq->service_last = now;
+               cfq_schedule_dispatch(cfqd);
+       }
+
+       cfq_clear_cfqq_must_dispatch(cfqq);
+       cfq_clear_cfqq_wait_request(cfqq);
+
+       /*
+        * store what was left of this slice, if the queue idled out
+        * or was preempted
+        */
+       if (time_after(cfqq->slice_end, now))
+               cfqq->slice_left = cfqq->slice_end - now;
+       else
+               cfqq->slice_left = 0;
+
+       if (cfq_cfqq_on_rr(cfqq))
+               cfq_resort_rr_list(cfqq, preempted);
+
+       if (cfqq == cfqd->active_queue)
+               cfqd->active_queue = NULL;
+
+       if (cfqd->active_cic) {
+               put_io_context(cfqd->active_cic->ioc);
+               cfqd->active_cic = NULL;
+       }
+
+       cfqd->dispatch_slice = 0;
+}
+
+static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted)
+{
+       struct cfq_queue *cfqq = cfqd->active_queue;
+
+       if (cfqq)
+               __cfq_slice_expired(cfqd, cfqq, preempted);
+}
+
 /*
  * 0
  * 0,1
@@ -801,16 +847,7 @@ static int cfq_get_next_prio_level(struct cfq_data *cfqd)
 
 static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd)
 {
-       struct cfq_queue *cfqq;
-
-       /*
-        * if current queue is expired but not done with its requests yet,
-        * wait for that to happen
-        */
-       if ((cfqq = cfqd->active_queue) != NULL) {
-               if (cfq_cfqq_expired(cfqq) && cfq_cfqq_dispatched(cfqq))
-                       return NULL;
-       }
+       struct cfq_queue *cfqq = NULL;
 
        /*
         * if current list is non-empty, grab first entry. if it is empty,
@@ -837,66 +874,11 @@ static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd)
        return cfqq;
 }
 
-/*
- * current cfqq expired its slice (or was too idle), select new one
- */
-static void
-__cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
-                   int preempted)
-{
-       unsigned long now = jiffies;
-
-       if (cfq_cfqq_wait_request(cfqq))
-               del_timer(&cfqd->idle_slice_timer);
-
-       if (!preempted && !cfq_cfqq_dispatched(cfqq))
-               cfqq->service_last = now;
-
-       cfq_clear_cfqq_must_dispatch(cfqq);
-       cfq_clear_cfqq_wait_request(cfqq);
-
-       /*
-        * store what was left of this slice, if the queue idled out
-        * or was preempted
-        */
-       if (time_after(cfqq->slice_end, now))
-               cfqq->slice_left = cfqq->slice_end - now;
-       else
-               cfqq->slice_left = 0;
-
-       if (cfq_cfqq_on_rr(cfqq))
-               cfq_resort_rr_list(cfqq, preempted);
-
-       if (cfqq == cfqd->active_queue)
-               cfqd->active_queue = NULL;
-
-       if (cfqd->active_cic) {
-               put_io_context(cfqd->active_cic->ioc);
-               cfqd->active_cic = NULL;
-       }
-
-       cfqd->dispatch_slice = 0;
-}
-
-static inline void cfq_slice_expired(struct cfq_data *cfqd, int preempted)
-{
-       struct cfq_queue *cfqq = cfqd->active_queue;
-
-       if (cfqq) {
-               /*
-                * use deferred expiry, if there are requests in progress as
-                * not to disturb the slice of the next queue
-                */
-               if (cfq_cfqq_dispatched(cfqq))
-                       cfq_mark_cfqq_expired(cfqq);
-               else
-                       __cfq_slice_expired(cfqd, cfqq, preempted);
-       }
-}
-
 static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 
 {
+       unsigned long sl;
+
        WARN_ON(!RB_EMPTY(&cfqq->sort_list));
        WARN_ON(cfqq != cfqd->active_queue);
 
@@ -916,13 +898,8 @@ static int cfq_arm_slice_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq)
        cfq_mark_cfqq_must_dispatch(cfqq);
        cfq_mark_cfqq_wait_request(cfqq);
 
-       if (!timer_pending(&cfqd->idle_slice_timer)) {
-               unsigned long slice_left = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle);
-
-               cfqd->idle_slice_timer.expires = jiffies + slice_left;
-               add_timer(&cfqd->idle_slice_timer);
-       }
-
+       sl = min(cfqq->slice_end - 1, (unsigned long) cfqd->cfq_slice_idle);
+       mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
        return 1;
 }
 
@@ -1006,9 +983,6 @@ static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
        if (!cfqq)
                goto new_queue;
 
-       if (cfq_cfqq_expired(cfqq))
-               goto new_queue;
-
        /*
         * slice has expired
         */
@@ -1181,12 +1155,8 @@ static void cfq_put_queue(struct cfq_queue *cfqq)
        BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]);
        BUG_ON(cfq_cfqq_on_rr(cfqq));
 
-       if (unlikely(cfqd->active_queue == cfqq)) {
+       if (unlikely(cfqd->active_queue == cfqq))
                __cfq_slice_expired(cfqd, cfqq, 0);
-               cfq_schedule_dispatch(cfqd);
-       }
-
-       cfq_put_cfqd(cfqq->cfqd);
 
        /*
         * it's on the empty list and still hashed
@@ -1205,7 +1175,7 @@ __cfq_find_cfq_hash(struct cfq_data *cfqd, unsigned int key, unsigned int prio,
 
        hlist_for_each_safe(entry, next, hash_list) {
                struct cfq_queue *__cfqq = list_entry_qhash(entry);
-               const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->ioprio_class, __cfqq->ioprio);
+               const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio);
 
                if (__cfqq->key == key && (__p == prio || prio == CFQ_KEY_ANY))
                        return __cfqq;
@@ -1224,13 +1194,24 @@ static void cfq_free_io_context(struct cfq_io_context *cic)
 {
        struct cfq_io_context *__cic;
        struct list_head *entry, *next;
+       int freed = 1;
 
        list_for_each_safe(entry, next, &cic->list) {
                __cic = list_entry(entry, struct cfq_io_context, list);
                kmem_cache_free(cfq_ioc_pool, __cic);
+               freed++;
        }
 
        kmem_cache_free(cfq_ioc_pool, cic);
+       if (atomic_sub_and_test(freed, &ioc_count) && ioc_gone)
+               complete(ioc_gone);
+}
+
+static void cfq_trim(struct io_context *ioc)
+{
+       ioc->set_ioprio = NULL;
+       if (ioc->cic)
+               cfq_free_io_context(ioc->cic);
 }
 
 /*
@@ -1238,27 +1219,37 @@ static void cfq_free_io_context(struct cfq_io_context *cic)
  */
 static void cfq_exit_single_io_context(struct cfq_io_context *cic)
 {
-       struct cfq_data *cfqd = cic->cfqq->cfqd;
-       request_queue_t *q = cfqd->queue;
+       struct cfq_data *cfqd = cic->key;
+       request_queue_t *q;
+
+       if (!cfqd)
+               return;
+
+       q = cfqd->queue;
 
        WARN_ON(!irqs_disabled());
 
        spin_lock(q->queue_lock);
 
-       if (unlikely(cic->cfqq == cfqd->active_queue)) {
-               __cfq_slice_expired(cfqd, cic->cfqq, 0);
-               cfq_schedule_dispatch(cfqd);
+       if (cic->cfqq[ASYNC]) {
+               if (unlikely(cic->cfqq[ASYNC] == cfqd->active_queue))
+                       __cfq_slice_expired(cfqd, cic->cfqq[ASYNC], 0);
+               cfq_put_queue(cic->cfqq[ASYNC]);
+               cic->cfqq[ASYNC] = NULL;
+       }
+
+       if (cic->cfqq[SYNC]) {
+               if (unlikely(cic->cfqq[SYNC] == cfqd->active_queue))
+                       __cfq_slice_expired(cfqd, cic->cfqq[SYNC], 0);
+               cfq_put_queue(cic->cfqq[SYNC]);
+               cic->cfqq[SYNC] = NULL;
        }
 
-       cfq_put_queue(cic->cfqq);
-       cic->cfqq = NULL;
+       cic->key = NULL;
+       list_del_init(&cic->queue_list);
        spin_unlock(q->queue_lock);
 }
 
-/*
- * Another task may update the task cic list, if it is doing a queue lookup
- * on its behalf. cfq_cic_lock excludes such concurrent updates
- */
 static void cfq_exit_io_context(struct cfq_io_context *cic)
 {
        struct cfq_io_context *__cic;
@@ -1270,12 +1261,14 @@ static void cfq_exit_io_context(struct cfq_io_context *cic)
        /*
         * put the reference this task is holding to the various queues
         */
+       read_lock(&cfq_exit_lock);
        list_for_each(entry, &cic->list) {
                __cic = list_entry(entry, struct cfq_io_context, list);
                cfq_exit_single_io_context(__cic);
        }
 
        cfq_exit_single_io_context(cic);
+       read_unlock(&cfq_exit_lock);
        local_irq_restore(flags);
 }
 
@@ -1286,7 +1279,8 @@ cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
 
        if (cic) {
                INIT_LIST_HEAD(&cic->list);
-               cic->cfqq = NULL;
+               cic->cfqq[ASYNC] = NULL;
+               cic->cfqq[SYNC] = NULL;
                cic->key = NULL;
                cic->last_end_request = jiffies;
                cic->ttime_total = 0;
@@ -1294,6 +1288,8 @@ cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
                cic->ttime_mean = 0;
                cic->dtor = cfq_free_io_context;
                cic->exit = cfq_exit_io_context;
+               INIT_LIST_HEAD(&cic->queue_list);
+               atomic_inc(&ioc_count);
        }
 
        return cic;
@@ -1346,14 +1342,27 @@ static void cfq_init_prio_data(struct cfq_queue *cfqq)
        cfq_clear_cfqq_prio_changed(cfqq);
 }
 
-static inline void changed_ioprio(struct cfq_queue *cfqq)
+static inline void changed_ioprio(struct cfq_io_context *cic)
 {
-       if (cfqq) {
-               struct cfq_data *cfqd = cfqq->cfqd;
-
+       struct cfq_data *cfqd = cic->key;
+       struct cfq_queue *cfqq;
+       if (cfqd) {
                spin_lock(cfqd->queue->queue_lock);
-               cfq_mark_cfqq_prio_changed(cfqq);
-               cfq_init_prio_data(cfqq);
+               cfqq = cic->cfqq[ASYNC];
+               if (cfqq) {
+                       struct cfq_queue *new_cfqq;
+                       new_cfqq = cfq_get_queue(cfqd, CFQ_KEY_ASYNC,
+                                               cic->ioc->task, GFP_ATOMIC);
+                       if (new_cfqq) {
+                               cic->cfqq[ASYNC] = new_cfqq;
+                               cfq_put_queue(cfqq);
+                       }
+               }
+               cfqq = cic->cfqq[SYNC];
+               if (cfqq) {
+                       cfq_mark_cfqq_prio_changed(cfqq);
+                       cfq_init_prio_data(cfqq);
+               }
                spin_unlock(cfqd->queue->queue_lock);
        }
 }
@@ -1363,24 +1372,32 @@ static inline void changed_ioprio(struct cfq_queue *cfqq)
  */
 static int cfq_ioc_set_ioprio(struct io_context *ioc, unsigned int ioprio)
 {
-       struct cfq_io_context *cic = ioc->cic;
+       struct cfq_io_context *cic;
+
+       write_lock(&cfq_exit_lock);
 
-       changed_ioprio(cic->cfqq);
+       cic = ioc->cic;
+
+       changed_ioprio(cic);
 
        list_for_each_entry(cic, &cic->list, list)
-               changed_ioprio(cic->cfqq);
+               changed_ioprio(cic);
+
+       write_unlock(&cfq_exit_lock);
 
        return 0;
 }
 
 static struct cfq_queue *
-cfq_get_queue(struct cfq_data *cfqd, unsigned int key, unsigned short ioprio,
+cfq_get_queue(struct cfq_data *cfqd, unsigned int key, struct task_struct *tsk,
              gfp_t gfp_mask)
 {
        const int hashval = hash_long(key, CFQ_QHASH_SHIFT);
        struct cfq_queue *cfqq, *new_cfqq = NULL;
+       unsigned short ioprio;
 
 retry:
+       ioprio = tsk->ioprio;
        cfqq = __cfq_find_cfq_hash(cfqd, key, ioprio, hashval);
 
        if (!cfqq) {
@@ -1409,7 +1426,6 @@ retry:
                hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]);
                atomic_set(&cfqq->ref, 0);
                cfqq->cfqd = cfqd;
-               atomic_inc(&cfqd->ref);
                cfqq->service_last = 0;
                /*
                 * set ->slice_left to allow preemption for a new process
@@ -1447,6 +1463,7 @@ cfq_get_io_context(struct cfq_data *cfqd, pid_t pid, gfp_t gfp_mask)
        if (!ioc)
                return NULL;
 
+restart:
        if ((cic = ioc->cic) == NULL) {
                cic = cfq_alloc_io_context(cfqd, gfp_mask);
 
@@ -1457,11 +1474,13 @@ cfq_get_io_context(struct cfq_data *cfqd, pid_t pid, gfp_t gfp_mask)
                 * manually increment generic io_context usage count, it
                 * cannot go away since we are already holding one ref to it
                 */
-               ioc->cic = cic;
-               ioc->set_ioprio = cfq_ioc_set_ioprio;
                cic->ioc = ioc;
                cic->key = cfqd;
-               atomic_inc(&cfqd->ref);
+               read_lock(&cfq_exit_lock);
+               ioc->set_ioprio = cfq_ioc_set_ioprio;
+               ioc->cic = cic;
+               list_add(&cic->queue_list, &cfqd->cic_list);
+               read_unlock(&cfq_exit_lock);
        } else {
                struct cfq_io_context *__cic;
 
@@ -1471,6 +1490,20 @@ cfq_get_io_context(struct cfq_data *cfqd, pid_t pid, gfp_t gfp_mask)
                if (cic->key == cfqd)
                        goto out;
 
+               if (unlikely(!cic->key)) {
+                       read_lock(&cfq_exit_lock);
+                       if (list_empty(&cic->list))
+                               ioc->cic = NULL;
+                       else
+                               ioc->cic = list_entry(cic->list.next,
+                                                     struct cfq_io_context,
+                                                     list);
+                       read_unlock(&cfq_exit_lock);
+                       kmem_cache_free(cfq_ioc_pool, cic);
+                       atomic_dec(&ioc_count);
+                       goto restart;
+               }
+
                /*
                 * cic exists, check if we already are there. linear search
                 * should be ok here, the list will usually not be more than
@@ -1485,6 +1518,14 @@ cfq_get_io_context(struct cfq_data *cfqd, pid_t pid, gfp_t gfp_mask)
                                cic = __cic;
                                goto out;
                        }
+                       if (unlikely(!__cic->key)) {
+                               read_lock(&cfq_exit_lock);
+                               list_del(&__cic->list);
+                               read_unlock(&cfq_exit_lock);
+                               kmem_cache_free(cfq_ioc_pool, __cic);
+                               atomic_dec(&ioc_count);
+                               goto restart;
+                       }
                }
 
                /*
@@ -1497,8 +1538,10 @@ cfq_get_io_context(struct cfq_data *cfqd, pid_t pid, gfp_t gfp_mask)
 
                __cic->ioc = ioc;
                __cic->key = cfqd;
-               atomic_inc(&cfqd->ref);
+               read_lock(&cfq_exit_lock);
                list_add(&__cic->list, &cic->list);
+               list_add(&__cic->queue_list, &cfqd->cic_list);
+               read_unlock(&cfq_exit_lock);
                cic = __cic;
        }
 
@@ -1715,10 +1758,7 @@ static void cfq_completed_request(request_queue_t *q, struct request *rq)
                        cfqq->service_last = now;
                        cfq_resort_rr_list(cfqq, 0);
                }
-               if (cfq_cfqq_expired(cfqq)) {
-                       __cfq_slice_expired(cfqd, cfqq, 0);
-                       cfq_schedule_dispatch(cfqd);
-               }
+               cfq_schedule_dispatch(cfqd);
        }
 
        if (cfq_crq_is_sync(crq))
@@ -1921,6 +1961,7 @@ cfq_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
        struct cfq_queue *cfqq;
        struct cfq_rq *crq;
        unsigned long flags;
+       int is_sync = key != CFQ_KEY_ASYNC;
 
        might_sleep_if(gfp_mask & __GFP_WAIT);
 
@@ -1931,14 +1972,14 @@ cfq_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
        if (!cic)
                goto queue_fail;
 
-       if (!cic->cfqq) {
-               cfqq = cfq_get_queue(cfqd, key, tsk->ioprio, gfp_mask);
+       if (!cic->cfqq[is_sync]) {
+               cfqq = cfq_get_queue(cfqd, key, tsk, gfp_mask);
                if (!cfqq)
                        goto queue_fail;
 
-               cic->cfqq = cfqq;
+               cic->cfqq[is_sync] = cfqq;
        } else
-               cfqq = cic->cfqq;
+               cfqq = cic->cfqq[is_sync];
 
        cfqq->allocated[rw]++;
        cfq_clear_cfqq_must_alloc(cfqq);
@@ -1955,7 +1996,7 @@ cfq_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
                crq->cfq_queue = cfqq;
                crq->io_context = cic;
 
-               if (rw == READ || process_sync(tsk))
+               if (is_sync)
                        cfq_mark_crq_is_sync(crq);
                else
                        cfq_clear_crq_is_sync(crq);
@@ -2086,15 +2127,35 @@ static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
        blk_sync_queue(cfqd->queue);
 }
 
-static void cfq_put_cfqd(struct cfq_data *cfqd)
+static void cfq_exit_queue(elevator_t *e)
 {
+       struct cfq_data *cfqd = e->elevator_data;
        request_queue_t *q = cfqd->queue;
 
-       if (!atomic_dec_and_test(&cfqd->ref))
-               return;
+       cfq_shutdown_timer_wq(cfqd);
+       write_lock(&cfq_exit_lock);
+       spin_lock_irq(q->queue_lock);
+       if (cfqd->active_queue)
+               __cfq_slice_expired(cfqd, cfqd->active_queue, 0);
+       while(!list_empty(&cfqd->cic_list)) {
+               struct cfq_io_context *cic = list_entry(cfqd->cic_list.next,
+                                                       struct cfq_io_context,
+                                                       queue_list);
+               if (cic->cfqq[ASYNC]) {
+                       cfq_put_queue(cic->cfqq[ASYNC]);
+                       cic->cfqq[ASYNC] = NULL;
+               }
+               if (cic->cfqq[SYNC]) {
+                       cfq_put_queue(cic->cfqq[SYNC]);
+                       cic->cfqq[SYNC] = NULL;
+               }
+               cic->key = NULL;
+               list_del_init(&cic->queue_list);
+       }
+       spin_unlock_irq(q->queue_lock);
+       write_unlock(&cfq_exit_lock);
 
        cfq_shutdown_timer_wq(cfqd);
-       blk_put_queue(q);
 
        mempool_destroy(cfqd->crq_pool);
        kfree(cfqd->crq_hash);
@@ -2102,14 +2163,6 @@ static void cfq_put_cfqd(struct cfq_data *cfqd)
        kfree(cfqd);
 }
 
-static void cfq_exit_queue(elevator_t *e)
-{
-       struct cfq_data *cfqd = e->elevator_data;
-
-       cfq_shutdown_timer_wq(cfqd);
-       cfq_put_cfqd(cfqd);
-}
-
 static int cfq_init_queue(request_queue_t *q, elevator_t *e)
 {
        struct cfq_data *cfqd;
@@ -2128,6 +2181,7 @@ static int cfq_init_queue(request_queue_t *q, elevator_t *e)
        INIT_LIST_HEAD(&cfqd->cur_rr);
        INIT_LIST_HEAD(&cfqd->idle_rr);
        INIT_LIST_HEAD(&cfqd->empty_list);
+       INIT_LIST_HEAD(&cfqd->cic_list);
 
        cfqd->crq_hash = kmalloc(sizeof(struct hlist_head) * CFQ_MHASH_ENTRIES, GFP_KERNEL);
        if (!cfqd->crq_hash)
@@ -2149,7 +2203,6 @@ static int cfq_init_queue(request_queue_t *q, elevator_t *e)
        e->elevator_data = cfqd;
 
        cfqd->queue = q;
-       atomic_inc(&q->refcnt);
 
        cfqd->max_queued = q->nr_requests / 4;
        q->nr_batching = cfq_queued;
@@ -2164,8 +2217,6 @@ static int cfq_init_queue(request_queue_t *q, elevator_t *e)
 
        INIT_WORK(&cfqd->unplug_work, cfq_kick_queue, q);
 
-       atomic_set(&cfqd->ref, 1);
-
        cfqd->cfq_queued = cfq_queued;
        cfqd->cfq_quantum = cfq_quantum;
        cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0];
@@ -2224,11 +2275,6 @@ fail:
 /*
  * sysfs parts below -->
  */
-struct cfq_fs_entry {
-       struct attribute attr;
-       ssize_t (*show)(struct cfq_data *, char *);
-       ssize_t (*store)(struct cfq_data *, const char *, size_t);
-};
 
 static ssize_t
 cfq_var_show(unsigned int var, char *page)
@@ -2246,8 +2292,9 @@ cfq_var_store(unsigned int *var, const char *page, size_t count)
 }
 
 #define SHOW_FUNCTION(__FUNC, __VAR, __CONV)                           \
-static ssize_t __FUNC(struct cfq_data *cfqd, char *page)               \
+static ssize_t __FUNC(elevator_t *e, char *page)                       \
 {                                                                      \
+       struct cfq_data *cfqd = e->elevator_data;                       \
        unsigned int __data = __VAR;                                    \
        if (__CONV)                                                     \
                __data = jiffies_to_msecs(__data);                      \
@@ -2267,8 +2314,9 @@ SHOW_FUNCTION(cfq_max_depth_show, cfqd->cfq_max_depth, 0);
 #undef SHOW_FUNCTION
 
 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV)                        \
-static ssize_t __FUNC(struct cfq_data *cfqd, const char *page, size_t count)   \
+static ssize_t __FUNC(elevator_t *e, const char *page, size_t count)   \
 {                                                                      \
+       struct cfq_data *cfqd = e->elevator_data;                       \
        unsigned int __data;                                            \
        int ret = cfq_var_store(&__data, (page), count);                \
        if (__data < (MIN))                                             \
@@ -2294,63 +2342,63 @@ STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, UINT_MAX,
 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 = {
+static struct elv_fs_entry cfq_quantum_entry = {
        .attr = {.name = "quantum", .mode = S_IRUGO | S_IWUSR },
        .show = cfq_quantum_show,
        .store = cfq_quantum_store,
 };
-static struct cfq_fs_entry cfq_queued_entry = {
+static struct elv_fs_entry cfq_queued_entry = {
        .attr = {.name = "queued", .mode = S_IRUGO | S_IWUSR },
        .show = cfq_queued_show,
        .store = cfq_queued_store,
 };
-static struct cfq_fs_entry cfq_fifo_expire_sync_entry = {
+static struct elv_fs_entry cfq_fifo_expire_sync_entry = {
        .attr = {.name = "fifo_expire_sync", .mode = S_IRUGO | S_IWUSR },
        .show = cfq_fifo_expire_sync_show,
        .store = cfq_fifo_expire_sync_store,
 };
-static struct cfq_fs_entry cfq_fifo_expire_async_entry = {
+static struct elv_fs_entry cfq_fifo_expire_async_entry = {
        .attr = {.name = "fifo_expire_async", .mode = S_IRUGO | S_IWUSR },
        .show = cfq_fifo_expire_async_show,
        .store = cfq_fifo_expire_async_store,
 };
-static struct cfq_fs_entry cfq_back_max_entry = {
+static struct elv_fs_entry cfq_back_max_entry = {
        .attr = {.name = "back_seek_max", .mode = S_IRUGO | S_IWUSR },
        .show = cfq_back_max_show,
        .store = cfq_back_max_store,
 };
-static struct cfq_fs_entry cfq_back_penalty_entry = {
+static struct elv_fs_entry cfq_back_penalty_entry = {
        .attr = {.name = "back_seek_penalty", .mode = S_IRUGO | S_IWUSR },
        .show = cfq_back_penalty_show,
        .store = cfq_back_penalty_store,
 };
-static struct cfq_fs_entry cfq_slice_sync_entry = {
+static struct elv_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 = {
+static struct elv_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 = {
+static struct elv_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 = {
+static struct elv_fs_entry cfq_slice_idle_entry = {
        .attr = {.name = "slice_idle", .mode = S_IRUGO | S_IWUSR },
        .show = cfq_slice_idle_show,
        .store = cfq_slice_idle_store,
 };
-static struct cfq_fs_entry cfq_max_depth_entry = {
+static struct elv_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[] = {
+static struct attribute *cfq_attrs[] = {
        &cfq_quantum_entry.attr,
        &cfq_queued_entry.attr,
        &cfq_fifo_expire_sync_entry.attr,
@@ -2365,43 +2413,6 @@ static struct attribute *default_attrs[] = {
        NULL,
 };
 
-#define to_cfq(atr) container_of((atr), struct cfq_fs_entry, attr)
-
-static ssize_t
-cfq_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
-{
-       elevator_t *e = container_of(kobj, elevator_t, kobj);
-       struct cfq_fs_entry *entry = to_cfq(attr);
-
-       if (!entry->show)
-               return -EIO;
-
-       return entry->show(e->elevator_data, page);
-}
-
-static ssize_t
-cfq_attr_store(struct kobject *kobj, struct attribute *attr,
-              const char *page, size_t length)
-{
-       elevator_t *e = container_of(kobj, elevator_t, kobj);
-       struct cfq_fs_entry *entry = to_cfq(attr);
-
-       if (!entry->store)
-               return -EIO;
-
-       return entry->store(e->elevator_data, page, length);
-}
-
-static struct sysfs_ops cfq_sysfs_ops = {
-       .show   = cfq_attr_show,
-       .store  = cfq_attr_store,
-};
-
-static struct kobj_type cfq_ktype = {
-       .sysfs_ops      = &cfq_sysfs_ops,
-       .default_attrs  = default_attrs,
-};
-
 static struct elevator_type iosched_cfq = {
        .ops = {
                .elevator_merge_fn =            cfq_merge,
@@ -2420,8 +2431,9 @@ static struct elevator_type iosched_cfq = {
                .elevator_may_queue_fn =        cfq_may_queue,
                .elevator_init_fn =             cfq_init_queue,
                .elevator_exit_fn =             cfq_exit_queue,
+               .trim =                         cfq_trim,
        },
-       .elevator_ktype =       &cfq_ktype,
+       .elevator_attrs =       cfq_attrs,
        .elevator_name =        "cfq",
        .elevator_owner =       THIS_MODULE,
 };
@@ -2450,7 +2462,13 @@ static int __init cfq_init(void)
 
 static void __exit cfq_exit(void)
 {
+       DECLARE_COMPLETION(all_gone);
        elv_unregister(&iosched_cfq);
+       ioc_gone = &all_gone;
+       barrier();
+       if (atomic_read(&ioc_count))
+               complete(ioc_gone);
+       synchronize_rcu();
        cfq_slab_kill();
 }