Merge branch 'for-linus' into for-2.6.33
[linux-2.6.git] / block / cfq-iosched.c
index aa1e953..13b612f 100644 (file)
@@ -27,6 +27,8 @@ static const int cfq_slice_sync = HZ / 10;
 static int cfq_slice_async = HZ / 25;
 static const int cfq_slice_async_rq = 2;
 static int cfq_slice_idle = HZ / 125;
+static const int cfq_target_latency = HZ * 3/10; /* 300 ms */
+static const int cfq_hist_divisor = 4;
 
 /*
  * offset from end of service tree
@@ -38,6 +40,12 @@ static int cfq_slice_idle = HZ / 125;
  */
 #define CFQ_MIN_TT             (2)
 
+/*
+ * Allow merged cfqqs to perform this amount of seeky I/O before
+ * deciding to break the queues up again.
+ */
+#define CFQQ_COOP_TOUT         (HZ)
+
 #define CFQ_SLICE_SCALE                (5)
 #define CFQ_HW_QUEUE_MIN       (5)
 
@@ -67,8 +75,9 @@ static DEFINE_SPINLOCK(ioc_gone_lock);
 struct cfq_rb_root {
        struct rb_root rb;
        struct rb_node *left;
+       unsigned count;
 };
-#define CFQ_RB_ROOT    (struct cfq_rb_root) { RB_ROOT, NULL, }
+#define CFQ_RB_ROOT    (struct cfq_rb_root) { RB_ROOT, NULL, 0, }
 
 /*
  * Per process-grouping structure
@@ -112,19 +121,56 @@ struct cfq_queue {
        unsigned short ioprio, org_ioprio;
        unsigned short ioprio_class, org_ioprio_class;
 
+       unsigned int seek_samples;
+       u64 seek_total;
+       sector_t seek_mean;
+       sector_t last_request_pos;
+       unsigned long seeky_start;
+
        pid_t pid;
+
+       struct cfq_rb_root *service_tree;
+       struct cfq_queue *new_cfqq;
 };
 
 /*
+ * First index in the service_trees.
+ * IDLE is handled separately, so it has negative index
+ */
+enum wl_prio_t {
+       IDLE_WORKLOAD = -1,
+       BE_WORKLOAD = 0,
+       RT_WORKLOAD = 1
+};
+
+/*
+ * Second index in the service_trees.
+ */
+enum wl_type_t {
+       ASYNC_WORKLOAD = 0,
+       SYNC_NOIDLE_WORKLOAD = 1,
+       SYNC_WORKLOAD = 2
+};
+
+
+/*
  * Per block device queue structure
  */
 struct cfq_data {
        struct request_queue *queue;
 
        /*
-        * rr list of queues with requests and the count of them
+        * rr lists of queues with requests, onle rr for each priority class.
+        * Counts are embedded in the cfq_rb_root
+        */
+       struct cfq_rb_root service_trees[2][3];
+       struct cfq_rb_root service_tree_idle;
+       /*
+        * The priority currently being served
         */
-       struct cfq_rb_root service_tree;
+       enum wl_prio_t serving_prio;
+       enum wl_type_t serving_type;
+       unsigned long workload_expires;
 
        /*
         * Each priority tree is sorted by next_request position.  These
@@ -134,6 +180,7 @@ struct cfq_data {
        struct rb_root prio_trees[CFQ_PRIO_LISTS];
 
        unsigned int busy_queues;
+       unsigned int busy_queues_avg[2];
 
        int rq_in_driver[2];
        int sync_flight;
@@ -185,6 +232,16 @@ struct cfq_data {
        unsigned long last_end_sync_rq;
 };
 
+static struct cfq_rb_root *service_tree_for(enum wl_prio_t prio,
+                                           enum wl_type_t type,
+                                           struct cfq_data *cfqd)
+{
+       if (prio == IDLE_WORKLOAD)
+               return &cfqd->service_tree_idle;
+
+       return &cfqd->service_trees[prio][type];
+}
+
 enum cfqq_state_flags {
        CFQ_CFQQ_FLAG_on_rr = 0,        /* on round-robin busy list */
        CFQ_CFQQ_FLAG_wait_request,     /* waiting for a request */
@@ -195,7 +252,7 @@ enum cfqq_state_flags {
        CFQ_CFQQ_FLAG_prio_changed,     /* task priority has changed */
        CFQ_CFQQ_FLAG_slice_new,        /* no requests dispatched in slice */
        CFQ_CFQQ_FLAG_sync,             /* synchronous queue */
-       CFQ_CFQQ_FLAG_coop,             /* has done a coop jump of the queue */
+       CFQ_CFQQ_FLAG_coop,             /* cfqq is shared */
        CFQ_CFQQ_FLAG_coop_preempt,     /* coop preempt */
 };
 
@@ -231,6 +288,35 @@ CFQ_CFQQ_FNS(coop_preempt);
 #define cfq_log(cfqd, fmt, args...)    \
        blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args)
 
+static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq)
+{
+       if (cfq_class_idle(cfqq))
+               return IDLE_WORKLOAD;
+       if (cfq_class_rt(cfqq))
+               return RT_WORKLOAD;
+       return BE_WORKLOAD;
+}
+
+
+static enum wl_type_t cfqq_type(struct cfq_queue *cfqq)
+{
+       if (!cfq_cfqq_sync(cfqq))
+               return ASYNC_WORKLOAD;
+       if (!cfq_cfqq_idle_window(cfqq))
+               return SYNC_NOIDLE_WORKLOAD;
+       return SYNC_WORKLOAD;
+}
+
+static inline int cfq_busy_queues_wl(enum wl_prio_t wl, struct cfq_data *cfqd)
+{
+       if (wl == IDLE_WORKLOAD)
+               return cfqd->service_tree_idle.count;
+
+       return cfqd->service_trees[wl][ASYNC_WORKLOAD].count
+               + cfqd->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count
+               + cfqd->service_trees[wl][SYNC_WORKLOAD].count;
+}
+
 static void cfq_dispatch_insert(struct request_queue *, struct request *);
 static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool,
                                       struct io_context *, gfp_t);
@@ -303,10 +389,49 @@ cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
        return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio);
 }
 
+/*
+ * get averaged number of queues of RT/BE priority.
+ * average is updated, with a formula that gives more weight to higher numbers,
+ * to quickly follows sudden increases and decrease slowly
+ */
+
+static inline unsigned cfq_get_avg_queues(struct cfq_data *cfqd, bool rt)
+{
+       unsigned min_q, max_q;
+       unsigned mult  = cfq_hist_divisor - 1;
+       unsigned round = cfq_hist_divisor / 2;
+       unsigned busy = cfq_busy_queues_wl(rt, cfqd);
+
+       min_q = min(cfqd->busy_queues_avg[rt], busy);
+       max_q = max(cfqd->busy_queues_avg[rt], busy);
+       cfqd->busy_queues_avg[rt] = (mult * max_q + min_q + round) /
+               cfq_hist_divisor;
+       return cfqd->busy_queues_avg[rt];
+}
+
 static inline void
 cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 {
-       cfqq->slice_end = cfq_prio_to_slice(cfqd, cfqq) + jiffies;
+       unsigned slice = cfq_prio_to_slice(cfqd, cfqq);
+       if (cfqd->cfq_latency) {
+               /* interested queues (we consider only the ones with the same
+                * priority class) */
+               unsigned iq = cfq_get_avg_queues(cfqd, cfq_class_rt(cfqq));
+               unsigned sync_slice = cfqd->cfq_slice[1];
+               unsigned expect_latency = sync_slice * iq;
+               if (expect_latency > cfq_target_latency) {
+                       unsigned base_low_slice = 2 * cfqd->cfq_slice_idle;
+                       /* scale low_slice according to IO priority
+                        * and sync vs async */
+                       unsigned low_slice =
+                               min(slice, base_low_slice * slice / sync_slice);
+                       /* the adapted slice value is scaled to fit all iqs
+                        * into the target latency */
+                       slice = max(slice * cfq_target_latency / expect_latency,
+                                   low_slice);
+               }
+       }
+       cfqq->slice_end = jiffies + slice;
        cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies);
 }
 
@@ -445,6 +570,7 @@ static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
        if (root->left == n)
                root->left = NULL;
        rb_erase_init(n, &root->rb);
+       --root->count;
 }
 
 /*
@@ -485,7 +611,7 @@ static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
 }
 
 /*
- * The cfqd->service_tree holds all pending cfq_queue's that have
+ * The cfqd->service_trees holds all pending cfq_queue's that have
  * requests waiting to be processed. It is sorted in the order that
  * we will service the queues.
  */
@@ -495,11 +621,13 @@ static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
        struct rb_node **p, *parent;
        struct cfq_queue *__cfqq;
        unsigned long rb_key;
+       struct cfq_rb_root *service_tree;
        int left;
 
+       service_tree = service_tree_for(cfqq_prio(cfqq), cfqq_type(cfqq), cfqd);
        if (cfq_class_idle(cfqq)) {
                rb_key = CFQ_IDLE_DELAY;
-               parent = rb_last(&cfqd->service_tree.rb);
+               parent = rb_last(&service_tree->rb);
                if (parent && parent != &cfqq->rb_node) {
                        __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
                        rb_key += __cfqq->rb_key;
@@ -517,7 +645,7 @@ static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
                cfqq->slice_resid = 0;
        } else {
                rb_key = -HZ;
-               __cfqq = cfq_rb_first(&cfqd->service_tree);
+               __cfqq = cfq_rb_first(service_tree);
                rb_key += __cfqq ? __cfqq->rb_key : jiffies;
        }
 
@@ -525,15 +653,18 @@ static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
                /*
                 * same position, nothing more to do
                 */
-               if (rb_key == cfqq->rb_key)
+               if (rb_key == cfqq->rb_key &&
+                   cfqq->service_tree == service_tree)
                        return;
 
-               cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
+               cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
+               cfqq->service_tree = NULL;
        }
 
        left = 1;
        parent = NULL;
-       p = &cfqd->service_tree.rb.rb_node;
+       cfqq->service_tree = service_tree;
+       p = &service_tree->rb.rb_node;
        while (*p) {
                struct rb_node **n;
 
@@ -541,35 +672,25 @@ static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq,
                __cfqq = rb_entry(parent, struct cfq_queue, rb_node);
 
                /*
-                * sort RT queues first, we always want to give
-                * preference to them. IDLE queues goes to the back.
-                * after that, sort on the next service time.
+                * sort by key, that represents service time.
                 */
-               if (cfq_class_rt(cfqq) > cfq_class_rt(__cfqq))
-                       n = &(*p)->rb_left;
-               else if (cfq_class_rt(cfqq) < cfq_class_rt(__cfqq))
-                       n = &(*p)->rb_right;
-               else if (cfq_class_idle(cfqq) < cfq_class_idle(__cfqq))
+               if (time_before(rb_key, __cfqq->rb_key))
                        n = &(*p)->rb_left;
-               else if (cfq_class_idle(cfqq) > cfq_class_idle(__cfqq))
-                       n = &(*p)->rb_right;
-               else if (time_before(rb_key, __cfqq->rb_key))
-                       n = &(*p)->rb_left;
-               else
+               else {
                        n = &(*p)->rb_right;
-
-               if (n == &(*p)->rb_right)
                        left = 0;
+               }
 
                p = n;
        }
 
        if (left)
-               cfqd->service_tree.left = &cfqq->rb_node;
+               service_tree->left = &cfqq->rb_node;
 
        cfqq->rb_key = rb_key;
        rb_link_node(&cfqq->rb_node, parent, p);
-       rb_insert_color(&cfqq->rb_node, &cfqd->service_tree.rb);
+       rb_insert_color(&cfqq->rb_node, &service_tree->rb);
+       service_tree->count++;
 }
 
 static struct cfq_queue *
@@ -671,8 +792,10 @@ static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
        BUG_ON(!cfq_cfqq_on_rr(cfqq));
        cfq_clear_cfqq_on_rr(cfqq);
 
-       if (!RB_EMPTY_NODE(&cfqq->rb_node))
-               cfq_rb_erase(&cfqq->rb_node, &cfqd->service_tree);
+       if (!RB_EMPTY_NODE(&cfqq->rb_node)) {
+               cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree);
+               cfqq->service_tree = NULL;
+       }
        if (cfqq->p_root) {
                rb_erase(&cfqq->p_node, cfqq->p_root);
                cfqq->p_root = NULL;
@@ -933,10 +1056,12 @@ static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out)
  */
 static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
 {
-       if (RB_EMPTY_ROOT(&cfqd->service_tree.rb))
-               return NULL;
+       struct cfq_rb_root *service_tree =
+               service_tree_for(cfqd->serving_prio, cfqd->serving_type, cfqd);
 
-       return cfq_rb_first(&cfqd->service_tree);
+       if (RB_EMPTY_ROOT(&service_tree->rb))
+               return NULL;
+       return cfq_rb_first(service_tree);
 }
 
 /*
@@ -947,6 +1072,7 @@ static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd,
 {
        if (!cfqq) {
                cfqq = cfq_get_next_queue(cfqd);
+
                if (cfqq && !cfq_cfqq_coop_preempt(cfqq))
                        cfq_clear_cfqq_coop(cfqq);
        }
@@ -967,16 +1093,16 @@ static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd,
                return cfqd->last_position - blk_rq_pos(rq);
 }
 
-#define CIC_SEEK_THR   8 * 1024
-#define CIC_SEEKY(cic) ((cic)->seek_mean > CIC_SEEK_THR)
+#define CFQQ_SEEK_THR          8 * 1024
+#define CFQQ_SEEKY(cfqq)       ((cfqq)->seek_mean > CFQQ_SEEK_THR)
 
-static inline int cfq_rq_close(struct cfq_data *cfqd, struct request *rq)
+static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq,
+                              struct request *rq)
 {
-       struct cfq_io_context *cic = cfqd->active_cic;
-       sector_t sdist = cic->seek_mean;
+       sector_t sdist = cfqq->seek_mean;
 
-       if (!sample_valid(cic->seek_samples))
-               sdist = CIC_SEEK_THR;
+       if (!sample_valid(cfqq->seek_samples))
+               sdist = CFQQ_SEEK_THR;
 
        return cfq_dist_from_last(cfqd, rq) <= sdist;
 }
@@ -1005,7 +1131,7 @@ static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
         * will contain the closest sector.
         */
        __cfqq = rb_entry(parent, struct cfq_queue, p_node);
-       if (cfq_rq_close(cfqd, __cfqq->next_rq))
+       if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
                return __cfqq;
 
        if (blk_rq_pos(__cfqq->next_rq) < sector)
@@ -1016,7 +1142,7 @@ static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
                return NULL;
 
        __cfqq = rb_entry(node, struct cfq_queue, p_node);
-       if (cfq_rq_close(cfqd, __cfqq->next_rq))
+       if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq))
                return __cfqq;
 
        return NULL;
@@ -1033,16 +1159,13 @@ static struct cfq_queue *cfqq_close(struct cfq_data *cfqd,
  * assumption.
  */
 static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd,
-                                             struct cfq_queue *cur_cfqq,
-                                             bool probe)
+                                             struct cfq_queue *cur_cfqq)
 {
        struct cfq_queue *cfqq;
 
-       /*
-        * A valid cfq_io_context is necessary to compare requests against
-        * the seek_mean of the current cfqq.
-        */
-       if (!cfqd->active_cic)
+       if (!cfq_cfqq_sync(cur_cfqq))
+               return NULL;
+       if (CFQQ_SEEKY(cur_cfqq))
                return NULL;
 
        /*
@@ -1054,14 +1177,53 @@ static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd,
        if (!cfqq)
                return NULL;
 
-       if (cfq_cfqq_coop(cfqq))
+       /*
+        * It only makes sense to merge sync queues.
+        */
+       if (!cfq_cfqq_sync(cfqq))
+               return NULL;
+       if (CFQQ_SEEKY(cfqq))
+               return NULL;
+
+       /*
+        * Do not merge queues of different priority classes
+        */
+       if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq))
                return NULL;
 
-       if (!probe)
-               cfq_mark_cfqq_coop(cfqq);
        return cfqq;
 }
 
+/*
+ * Determine whether we should enforce idle window for this queue.
+ */
+
+static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq)
+{
+       enum wl_prio_t prio = cfqq_prio(cfqq);
+       struct cfq_rb_root *service_tree = cfqq->service_tree;
+
+       /* We never do for idle class queues. */
+       if (prio == IDLE_WORKLOAD)
+               return false;
+
+       /* We do for queues that were marked with idle window flag. */
+       if (cfq_cfqq_idle_window(cfqq))
+               return true;
+
+       /*
+        * Otherwise, we do only if they are the last ones
+        * in their service tree.
+        */
+       if (!service_tree)
+               service_tree = service_tree_for(prio, cfqq_type(cfqq), cfqd);
+
+       if (service_tree->count == 0)
+               return true;
+
+       return (service_tree->count == 1 && cfq_rb_first(service_tree) == cfqq);
+}
+
 static void cfq_arm_slice_timer(struct cfq_data *cfqd)
 {
        struct cfq_queue *cfqq = cfqd->active_queue;
@@ -1082,7 +1244,7 @@ static void cfq_arm_slice_timer(struct cfq_data *cfqd)
        /*
         * idle is disabled, either manually or by past process history
         */
-       if (!cfqd->cfq_slice_idle || !cfq_cfqq_idle_window(cfqq))
+       if (!cfqd->cfq_slice_idle || !cfq_should_idle(cfqd, cfqq))
                return;
 
        /*
@@ -1109,14 +1271,20 @@ static void cfq_arm_slice_timer(struct cfq_data *cfqd)
 
        cfq_mark_cfqq_wait_request(cfqq);
 
-       /*
-        * we don't want to idle for seeks, but we do want to allow
-        * fair distribution of slice time for a process doing back-to-back
-        * seeks. so allow a little bit of time for him to submit a new rq
-        */
        sl = cfqd->cfq_slice_idle;
-       if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
+       /* are we servicing noidle tree, and there are more queues?
+        * non-rotational or NCQ: no idle
+        * non-NCQ rotational : very small idle, to allow
+        *     fair distribution of slice time for a process doing back-to-back
+        *     seeks.
+        */
+       if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD &&
+           service_tree_for(cfqd->serving_prio, SYNC_NOIDLE_WORKLOAD, cfqd)
+               ->count > 0) {
+               if (blk_queue_nonrot(cfqd->queue) || cfqd->hw_tag)
+                       return;
                sl = min(sl, msecs_to_jiffies(CFQ_MIN_TT));
+       }
 
        mod_timer(&cfqd->idle_slice_timer, jiffies + sl);
        cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu", sl);
@@ -1175,6 +1343,152 @@ cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 }
 
 /*
+ * Must be called with the queue_lock held.
+ */
+static int cfqq_process_refs(struct cfq_queue *cfqq)
+{
+       int process_refs, io_refs;
+
+       io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE];
+       process_refs = atomic_read(&cfqq->ref) - io_refs;
+       BUG_ON(process_refs < 0);
+       return process_refs;
+}
+
+static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq)
+{
+       int process_refs, new_process_refs;
+       struct cfq_queue *__cfqq;
+
+       /* Avoid a circular list and skip interim queue merges */
+       while ((__cfqq = new_cfqq->new_cfqq)) {
+               if (__cfqq == cfqq)
+                       return;
+               new_cfqq = __cfqq;
+       }
+
+       process_refs = cfqq_process_refs(cfqq);
+       /*
+        * If the process for the cfqq has gone away, there is no
+        * sense in merging the queues.
+        */
+       if (process_refs == 0)
+               return;
+
+       /*
+        * Merge in the direction of the lesser amount of work.
+        */
+       new_process_refs = cfqq_process_refs(new_cfqq);
+       if (new_process_refs >= process_refs) {
+               cfqq->new_cfqq = new_cfqq;
+               atomic_add(process_refs, &new_cfqq->ref);
+       } else {
+               new_cfqq->new_cfqq = cfqq;
+               atomic_add(new_process_refs, &cfqq->ref);
+       }
+}
+
+static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, enum wl_prio_t prio,
+                                   bool prio_changed)
+{
+       struct cfq_queue *queue;
+       int i;
+       bool key_valid = false;
+       unsigned long lowest_key = 0;
+       enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD;
+
+       if (prio_changed) {
+               /*
+                * When priorities switched, we prefer starting
+                * from SYNC_NOIDLE (first choice), or just SYNC
+                * over ASYNC
+                */
+               if (service_tree_for(prio, cur_best, cfqd)->count)
+                       return cur_best;
+               cur_best = SYNC_WORKLOAD;
+               if (service_tree_for(prio, cur_best, cfqd)->count)
+                       return cur_best;
+
+               return ASYNC_WORKLOAD;
+       }
+
+       for (i = 0; i < 3; ++i) {
+               /* otherwise, select the one with lowest rb_key */
+               queue = cfq_rb_first(service_tree_for(prio, i, cfqd));
+               if (queue &&
+                   (!key_valid || time_before(queue->rb_key, lowest_key))) {
+                       lowest_key = queue->rb_key;
+                       cur_best = i;
+                       key_valid = true;
+               }
+       }
+
+       return cur_best;
+}
+
+static void choose_service_tree(struct cfq_data *cfqd)
+{
+       enum wl_prio_t previous_prio = cfqd->serving_prio;
+       bool prio_changed;
+       unsigned slice;
+       unsigned count;
+
+       /* Choose next priority. RT > BE > IDLE */
+       if (cfq_busy_queues_wl(RT_WORKLOAD, cfqd))
+               cfqd->serving_prio = RT_WORKLOAD;
+       else if (cfq_busy_queues_wl(BE_WORKLOAD, cfqd))
+               cfqd->serving_prio = BE_WORKLOAD;
+       else {
+               cfqd->serving_prio = IDLE_WORKLOAD;
+               cfqd->workload_expires = jiffies + 1;
+               return;
+       }
+
+       /*
+        * For RT and BE, we have to choose also the type
+        * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload
+        * expiration time
+        */
+       prio_changed = (cfqd->serving_prio != previous_prio);
+       count = service_tree_for(cfqd->serving_prio, cfqd->serving_type, cfqd)
+               ->count;
+
+       /*
+        * If priority didn't change, check workload expiration,
+        * and that we still have other queues ready
+        */
+       if (!prio_changed && count &&
+           !time_after(jiffies, cfqd->workload_expires))
+               return;
+
+       /* otherwise select new workload type */
+       cfqd->serving_type =
+               cfq_choose_wl(cfqd, cfqd->serving_prio, prio_changed);
+       count = service_tree_for(cfqd->serving_prio, cfqd->serving_type, cfqd)
+               ->count;
+
+       /*
+        * the workload slice is computed as a fraction of target latency
+        * proportional to the number of queues in that workload, over
+        * all the queues in the same priority class
+        */
+       slice = cfq_target_latency * count /
+               max_t(unsigned, cfqd->busy_queues_avg[cfqd->serving_prio],
+                     cfq_busy_queues_wl(cfqd->serving_prio, cfqd));
+
+       if (cfqd->serving_type == ASYNC_WORKLOAD)
+               /* async workload slice is scaled down according to
+                * the sync/async slice ratio. */
+               slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1];
+       else
+               /* sync workload slice is at least 2 * cfq_slice_idle */
+               slice = max(slice, 2 * cfqd->cfq_slice_idle);
+
+       slice = max_t(unsigned, slice, CFQ_MIN_TT);
+       cfqd->workload_expires = jiffies + slice;
+}
+
+/*
  * Select a queue for service. If we have a current active queue,
  * check whether to continue servicing it, or retrieve and set a new one.
  */
@@ -1203,11 +1517,14 @@ static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
         * If another queue has a request waiting within our mean seek
         * distance, let it run.  The expire code will check for close
         * cooperators and put the close queue at the front of the service
-        * tree.
+        * tree.  If possible, merge the expiring queue with the new cfqq.
         */
-       new_cfqq = cfq_close_cooperator(cfqd, cfqq, 0);
-       if (new_cfqq)
+       new_cfqq = cfq_close_cooperator(cfqd, cfqq);
+       if (new_cfqq) {
+               if (!cfqq->new_cfqq)
+                       cfq_setup_merge(cfqq, new_cfqq);
                goto expire;
+       }
 
        /*
         * No requests pending. If the active queue still has requests in
@@ -1215,7 +1532,7 @@ static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
         * conditions to happen (or time out) before selecting a new queue.
         */
        if (timer_pending(&cfqd->idle_slice_timer) ||
-           (cfqq->dispatched && cfq_cfqq_idle_window(cfqq))) {
+           (cfqq->dispatched && cfq_should_idle(cfqd, cfqq))) {
                cfqq = NULL;
                goto keep_queue;
        }
@@ -1223,6 +1540,13 @@ static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd)
 expire:
        cfq_slice_expired(cfqd, 0);
 new_queue:
+       /*
+        * Current queue expired. Check if we have to switch to a new
+        * service tree
+        */
+       if (!new_cfqq)
+               choose_service_tree(cfqd);
+
        cfqq = cfq_set_active_queue(cfqd, new_cfqq);
 keep_queue:
        return cfqq;
@@ -1249,8 +1573,14 @@ static int cfq_forced_dispatch(struct cfq_data *cfqd)
 {
        struct cfq_queue *cfqq;
        int dispatched = 0;
-
-       while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL)
+       int i, j;
+       for (i = 0; i < 2; ++i)
+               for (j = 0; j < 3; ++j)
+                       while ((cfqq = cfq_rb_first(&cfqd->service_trees[i][j]))
+                               != NULL)
+                               dispatched += __cfq_forced_dispatch_cfqq(cfqq);
+
+       while ((cfqq = cfq_rb_first(&cfqd->service_tree_idle)) != NULL)
                dispatched += __cfq_forced_dispatch_cfqq(cfqq);
 
        cfq_slice_expired(cfqd, 0);
@@ -1268,7 +1598,7 @@ static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq)
        /*
         * Drain async requests before we start sync IO
         */
-       if (cfq_cfqq_idle_window(cfqq) && cfqd->rq_in_driver[BLK_RW_ASYNC])
+       if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_driver[BLK_RW_ASYNC])
                return false;
 
        /*
@@ -1518,11 +1848,29 @@ static void cfq_free_io_context(struct io_context *ioc)
 
 static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 {
+       struct cfq_queue *__cfqq, *next;
+
        if (unlikely(cfqq == cfqd->active_queue)) {
                __cfq_slice_expired(cfqd, cfqq, 0);
                cfq_schedule_dispatch(cfqd);
        }
 
+       /*
+        * If this queue was scheduled to merge with another queue, be
+        * sure to drop the reference taken on that queue (and others in
+        * the merge chain).  See cfq_setup_merge and cfq_merge_cfqqs.
+        */
+       __cfqq = cfqq->new_cfqq;
+       while (__cfqq) {
+               if (__cfqq == cfqq) {
+                       WARN(1, "cfqq->new_cfqq loop detected\n");
+                       break;
+               }
+               next = __cfqq->new_cfqq;
+               cfq_put_queue(__cfqq);
+               __cfqq = next;
+       }
+
        cfq_put_queue(cfqq);
 }
 
@@ -1952,33 +2300,46 @@ cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_io_context *cic)
 }
 
 static void
-cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_io_context *cic,
+cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq,
                       struct request *rq)
 {
        sector_t sdist;
        u64 total;
 
-       if (!cic->last_request_pos)
+       if (!cfqq->last_request_pos)
                sdist = 0;
-       else if (cic->last_request_pos < blk_rq_pos(rq))
-               sdist = blk_rq_pos(rq) - cic->last_request_pos;
+       else if (cfqq->last_request_pos < blk_rq_pos(rq))
+               sdist = blk_rq_pos(rq) - cfqq->last_request_pos;
        else
-               sdist = cic->last_request_pos - blk_rq_pos(rq);
+               sdist = cfqq->last_request_pos - blk_rq_pos(rq);
 
        /*
         * Don't allow the seek distance to get too large from the
         * odd fragment, pagein, etc
         */
-       if (cic->seek_samples <= 60) /* second&third seek */
-               sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*1024);
+       if (cfqq->seek_samples <= 60) /* second&third seek */
+               sdist = min(sdist, (cfqq->seek_mean * 4) + 2*1024*1024);
        else
-               sdist = min(sdist, (cic->seek_mean * 4) + 2*1024*64);
+               sdist = min(sdist, (cfqq->seek_mean * 4) + 2*1024*64);
 
-       cic->seek_samples = (7*cic->seek_samples + 256) / 8;
-       cic->seek_total = (7*cic->seek_total + (u64)256*sdist) / 8;
-       total = cic->seek_total + (cic->seek_samples/2);
-       do_div(total, cic->seek_samples);
-       cic->seek_mean = (sector_t)total;
+       cfqq->seek_samples = (7*cfqq->seek_samples + 256) / 8;
+       cfqq->seek_total = (7*cfqq->seek_total + (u64)256*sdist) / 8;
+       total = cfqq->seek_total + (cfqq->seek_samples/2);
+       do_div(total, cfqq->seek_samples);
+       cfqq->seek_mean = (sector_t)total;
+
+       /*
+        * If this cfqq is shared between multiple processes, check to
+        * make sure that those processes are still issuing I/Os within
+        * the mean seek distance.  If not, it may be time to break the
+        * queues apart again.
+        */
+       if (cfq_cfqq_coop(cfqq)) {
+               if (CFQQ_SEEKY(cfqq) && !cfqq->seeky_start)
+                       cfqq->seeky_start = jiffies;
+               else if (!CFQQ_SEEKY(cfqq))
+                       cfqq->seeky_start = 0;
+       }
 }
 
 /*
@@ -2000,13 +2361,10 @@ cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
        enable_idle = old_idle = cfq_cfqq_idle_window(cfqq);
 
        if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle ||
-           (!cfqd->cfq_latency && cfqd->hw_tag && CIC_SEEKY(cic)))
+           (sample_valid(cfqq->seek_samples) && CFQQ_SEEKY(cfqq)))
                enable_idle = 0;
        else if (sample_valid(cic->ttime_samples)) {
-               unsigned int slice_idle = cfqd->cfq_slice_idle;
-               if (sample_valid(cic->seek_samples) && CIC_SEEKY(cic))
-                       slice_idle = msecs_to_jiffies(CFQ_MIN_TT);
-               if (cic->ttime_mean > slice_idle)
+               if (cic->ttime_mean > cfqd->cfq_slice_idle)
                        enable_idle = 0;
                else
                        enable_idle = 1;
@@ -2044,6 +2402,10 @@ cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
        if (cfq_class_idle(cfqq))
                return true;
 
+       if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD
+           && new_cfqq->service_tree == cfqq->service_tree)
+               return true;
+
        /*
         * if the new request is sync, but the currently running queue is
         * not, let the sync request have priority.
@@ -2071,7 +2433,8 @@ cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq,
         * if this request is as-good as one we would expect from the
         * current cfqq, let it preempt
         */
-       if (cfq_rq_close(cfqd, rq) && (!cfq_cfqq_coop(new_cfqq) ||
+       if (cfq_rq_close(cfqd, cfqq, rq))
+       if (cfq_rq_close(cfqd, cfqq, rq) && (!cfq_cfqq_coop(new_cfqq) ||
            cfqd->busy_queues == 1)) {
                /*
                 * Mark new queue coop_preempt, so its coop flag will not be
@@ -2121,10 +2484,10 @@ cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq,
                cfqq->meta_pending++;
 
        cfq_update_io_thinktime(cfqd, cic);
-       cfq_update_io_seektime(cfqd, cic, rq);
+       cfq_update_io_seektime(cfqd, cfqq, rq);
        cfq_update_idle_window(cfqd, cfqq, cic);
 
-       cic->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
+       cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);
 
        if (cfqq == cfqd->active_queue) {
                /*
@@ -2165,10 +2528,9 @@ static void cfq_insert_request(struct request_queue *q, struct request *rq)
        cfq_log_cfqq(cfqd, cfqq, "insert_request");
        cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc);
 
-       cfq_add_rq_rb(rq);
-
        rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]);
        list_add_tail(&rq->queuelist, &cfqq->fifo);
+       cfq_add_rq_rb(rq);
 
        cfq_rq_enqueued(cfqd, cfqq, rq);
 }
@@ -2179,6 +2541,8 @@ static void cfq_insert_request(struct request_queue *q, struct request *rq)
  */
 static void cfq_update_hw_tag(struct cfq_data *cfqd)
 {
+       struct cfq_queue *cfqq = cfqd->active_queue;
+
        if (rq_in_driver(cfqd) > cfqd->rq_in_driver_peak)
                cfqd->rq_in_driver_peak = rq_in_driver(cfqd);
 
@@ -2186,6 +2550,16 @@ static void cfq_update_hw_tag(struct cfq_data *cfqd)
            rq_in_driver(cfqd) <= CFQ_HW_QUEUE_MIN)
                return;
 
+       /*
+        * If active queue hasn't enough requests and can idle, cfq might not
+        * dispatch sufficient requests to hardware. Don't zero hw_tag in this
+        * case
+        */
+       if (cfqq && cfq_cfqq_idle_window(cfqq) &&
+           cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] <
+           CFQ_HW_QUEUE_MIN && rq_in_driver(cfqd) < CFQ_HW_QUEUE_MIN)
+               return;
+
        if (cfqd->hw_tag_samples++ < 50)
                return;
 
@@ -2243,7 +2617,7 @@ static void cfq_completed_request(struct request_queue *q, struct request *rq)
                 */
                if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
                        cfq_slice_expired(cfqd, 1);
-               else if (cfqq_empty && !cfq_close_cooperator(cfqd, cfqq, 1) &&
+               else if (cfqq_empty && !cfq_close_cooperator(cfqd, cfqq) &&
                         sync && !rq_noidle(rq))
                        cfq_arm_slice_timer(cfqd);
        }
@@ -2269,12 +2643,10 @@ static void cfq_prio_boost(struct cfq_queue *cfqq)
                        cfqq->ioprio = IOPRIO_NORM;
        } else {
                /*
-                * check if we need to unboost the queue
+                * unboost the queue (if needed)
                 */
-               if (cfqq->ioprio_class != cfqq->org_ioprio_class)
-                       cfqq->ioprio_class = cfqq->org_ioprio_class;
-               if (cfqq->ioprio != cfqq->org_ioprio)
-                       cfqq->ioprio = cfqq->org_ioprio;
+               cfqq->ioprio_class = cfqq->org_ioprio_class;
+               cfqq->ioprio = cfqq->org_ioprio;
        }
 }
 
@@ -2338,6 +2710,43 @@ static void cfq_put_request(struct request *rq)
        }
 }
 
+static struct cfq_queue *
+cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_context *cic,
+               struct cfq_queue *cfqq)
+{
+       cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq);
+       cic_set_cfqq(cic, cfqq->new_cfqq, 1);
+       cfq_mark_cfqq_coop(cfqq->new_cfqq);
+       cfq_put_queue(cfqq);
+       return cic_to_cfqq(cic, 1);
+}
+
+static int should_split_cfqq(struct cfq_queue *cfqq)
+{
+       if (cfqq->seeky_start &&
+           time_after(jiffies, cfqq->seeky_start + CFQQ_COOP_TOUT))
+               return 1;
+       return 0;
+}
+
+/*
+ * Returns NULL if a new cfqq should be allocated, or the old cfqq if this
+ * was the last process referring to said cfqq.
+ */
+static struct cfq_queue *
+split_cfqq(struct cfq_io_context *cic, struct cfq_queue *cfqq)
+{
+       if (cfqq_process_refs(cfqq) == 1) {
+               cfqq->seeky_start = 0;
+               cfqq->pid = current->pid;
+               cfq_clear_cfqq_coop(cfqq);
+               return cfqq;
+       }
+
+       cic_set_cfqq(cic, NULL, 1);
+       cfq_put_queue(cfqq);
+       return NULL;
+}
 /*
  * Allocate cfq data structures associated with this request.
  */
@@ -2360,10 +2769,30 @@ cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
        if (!cic)
                goto queue_fail;
 
+new_queue:
        cfqq = cic_to_cfqq(cic, is_sync);
        if (!cfqq || cfqq == &cfqd->oom_cfqq) {
                cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask);
                cic_set_cfqq(cic, cfqq, is_sync);
+       } else {
+               /*
+                * If the queue was seeky for too long, break it apart.
+                */
+               if (cfq_cfqq_coop(cfqq) && should_split_cfqq(cfqq)) {
+                       cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq");
+                       cfqq = split_cfqq(cic, cfqq);
+                       if (!cfqq)
+                               goto new_queue;
+               }
+
+               /*
+                * Check to see if this queue is scheduled to merge with
+                * another, closely cooperating queue.  The merging of
+                * queues happens here as it must be done in process context.
+                * The reference on new_cfqq was taken in merge_cfqqs.
+                */
+               if (cfqq->new_cfqq)
+                       cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq);
        }
 
        cfqq->allocated[rw]++;
@@ -2500,13 +2929,16 @@ static void cfq_exit_queue(struct elevator_queue *e)
 static void *cfq_init_queue(struct request_queue *q)
 {
        struct cfq_data *cfqd;
-       int i;
+       int i, j;
 
        cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node);
        if (!cfqd)
                return NULL;
 
-       cfqd->service_tree = CFQ_RB_ROOT;
+       for (i = 0; i < 2; ++i)
+               for (j = 0; j < 3; ++j)
+                       cfqd->service_trees[i][j] = CFQ_RB_ROOT;
+       cfqd->service_tree_idle = CFQ_RB_ROOT;
 
        /*
         * Not strictly needed (since RB_ROOT just clears the node and we