BFQ_BFQQ_FNS(wait_request);
 BFQ_BFQQ_FNS(non_blocking_wait_rq);
 BFQ_BFQQ_FNS(fifo_expire);
-BFQ_BFQQ_FNS(idle_window);
+BFQ_BFQQ_FNS(has_short_ttime);
 BFQ_BFQQ_FNS(sync);
 BFQ_BFQQ_FNS(IO_bound);
 BFQ_BFQQ_FNS(in_large_burst);
        unsigned int old_wr_coeff = bfqq->wr_coeff;
        bool busy = bfq_already_existing && bfq_bfqq_busy(bfqq);
 
-       if (bic->saved_idle_window)
-               bfq_mark_bfqq_idle_window(bfqq);
+       if (bic->saved_has_short_ttime)
+               bfq_mark_bfqq_has_short_ttime(bfqq);
        else
-               bfq_clear_bfqq_idle_window(bfqq);
+               bfq_clear_bfqq_has_short_ttime(bfqq);
 
        if (bic->saved_IO_bound)
                bfq_mark_bfqq_IO_bound(bfqq);
                return;
 
        bic->saved_ttime = bfqq->ttime;
-       bic->saved_idle_window = bfq_bfqq_idle_window(bfqq);
+       bic->saved_has_short_ttime = bfq_bfqq_has_short_ttime(bfqq);
        bic->saved_IO_bound = bfq_bfqq_IO_bound(bfqq);
        bic->saved_in_large_burst = bfq_bfqq_in_large_burst(bfqq);
        bic->was_in_burst_list = !hlist_unhashed(&bfqq->burst_list_node);
        }
 
        bfq_log_bfqq(bfqd, bfqq,
-               "expire (%d, slow %d, num_disp %d, idle_win %d)", reason,
-               slow, bfqq->dispatched, bfq_bfqq_idle_window(bfqq));
+               "expire (%d, slow %d, num_disp %d, short_ttime %d)", reason,
+               slow, bfqq->dispatched, bfq_bfqq_has_short_ttime(bfqq));
 
        /*
         * Increase, decrease or leave budget unchanged according to
        if (bfqd->strict_guarantees)
                return true;
 
+       /*
+        * Idling is performed only if slice_idle > 0. In addition, we
+        * do not idle if
+        * (a) bfqq is async
+        * (b) bfqq is in the idle io prio class: in this case we do
+        * not idle because we want to minimize the bandwidth that
+        * queues in this class can steal to higher-priority queues
+        */
+       if (bfqd->bfq_slice_idle == 0 || !bfq_bfqq_sync(bfqq) ||
+           bfq_class_idle(bfqq))
+               return false;
+
        /*
         * The next variable takes into account the cases where idling
         * boosts the throughput.
         */
        idling_boosts_thr = !bfqd->hw_tag ||
                (!blk_queue_nonrot(bfqd->queue) && bfq_bfqq_IO_bound(bfqq) &&
-                bfq_bfqq_idle_window(bfqq));
+                bfq_bfqq_has_short_ttime(bfqq));
 
        /*
         * The value of the next variable,
                asymmetric_scenario && !bfq_bfqq_in_large_burst(bfqq);
 
        /*
-        * We have now all the components we need to compute the return
-        * value of the function, which is true only if both the following
-        * conditions hold:
-        * 1) bfqq is sync, because idling make sense only for sync queues;
-        * 2) idling either boosts the throughput (without issues), or
-        *    is necessary to preserve service guarantees.
+        * We have now all the components we need to compute the
+        * return value of the function, which is true only if idling
+        * either boosts the throughput (without issues), or is
+        * necessary to preserve service guarantees.
         */
-       return bfq_bfqq_sync(bfqq) &&
-               (idling_boosts_thr_without_issues ||
-                idling_needed_for_service_guarantees);
+       return idling_boosts_thr_without_issues ||
+               idling_needed_for_service_guarantees;
 }
 
 /*
  */
 static bool bfq_bfqq_must_idle(struct bfq_queue *bfqq)
 {
-       struct bfq_data *bfqd = bfqq->bfqd;
-
-       return RB_EMPTY_ROOT(&bfqq->sort_list) && bfqd->bfq_slice_idle != 0 &&
-              bfq_bfqq_may_idle(bfqq);
+       return RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_may_idle(bfqq);
 }
 
 /*
        case IOPRIO_CLASS_IDLE:
                bfqq->new_ioprio_class = IOPRIO_CLASS_IDLE;
                bfqq->new_ioprio = 7;
-               bfq_clear_bfqq_idle_window(bfqq);
                break;
        }
 
                bfq_set_next_ioprio_data(bfqq, bic);
 
        if (is_sync) {
+               /*
+                * No need to mark as has_short_ttime if in
+                * idle_class, because no device idling is performed
+                * for queues in idle class
+                */
                if (!bfq_class_idle(bfqq))
-                       bfq_mark_bfqq_idle_window(bfqq);
+                       /* tentatively mark as has_short_ttime */
+                       bfq_mark_bfqq_has_short_ttime(bfqq);
                bfq_mark_bfqq_sync(bfqq);
                bfq_mark_bfqq_just_created(bfqq);
        } else
                 blk_rq_sectors(rq) < BFQQ_SECT_THR_NONROT);
 }
 
-/*
- * Disable idle window if the process thinks too long or seeks so much that
- * it doesn't matter.
- */
-static void bfq_update_idle_window(struct bfq_data *bfqd,
-                                  struct bfq_queue *bfqq,
-                                  struct bfq_io_cq *bic)
+static void bfq_update_has_short_ttime(struct bfq_data *bfqd,
+                                      struct bfq_queue *bfqq,
+                                      struct bfq_io_cq *bic)
 {
-       int enable_idle;
+       bool has_short_ttime = true;
 
-       /* Don't idle for async or idle io prio class. */
-       if (!bfq_bfqq_sync(bfqq) || bfq_class_idle(bfqq))
+       /*
+        * No need to update has_short_ttime if bfqq is async or in
+        * idle io prio class, or if bfq_slice_idle is zero, because
+        * no device idling is performed for bfqq in this case.
+        */
+       if (!bfq_bfqq_sync(bfqq) || bfq_class_idle(bfqq) ||
+           bfqd->bfq_slice_idle == 0)
                return;
 
        /* Idle window just restored, statistics are meaningless. */
                                     bfqd->bfq_wr_min_idle_time))
                return;
 
-       enable_idle = bfq_bfqq_idle_window(bfqq);
-
+       /* Think time is infinite if no process is linked to
+        * bfqq. Otherwise check average think time to
+        * decide whether to mark as has_short_ttime
+        */
        if (atomic_read(&bic->icq.ioc->active_ref) == 0 ||
-           bfqd->bfq_slice_idle == 0 ||
-               (bfqd->hw_tag && BFQQ_SEEKY(bfqq) &&
-                       bfqq->wr_coeff == 1))
-               enable_idle = 0;
-       else if (bfq_sample_valid(bfqq->ttime.ttime_samples)) {
-               if (bfqq->ttime.ttime_mean > bfqd->bfq_slice_idle &&
-                       bfqq->wr_coeff == 1)
-                       enable_idle = 0;
-               else
-                       enable_idle = 1;
-       }
-       bfq_log_bfqq(bfqd, bfqq, "update_idle_window: enable_idle %d",
-               enable_idle);
+           (bfq_sample_valid(bfqq->ttime.ttime_samples) &&
+            bfqq->ttime.ttime_mean > bfqd->bfq_slice_idle))
+               has_short_ttime = false;
+
+       bfq_log_bfqq(bfqd, bfqq, "update_has_short_ttime: has_short_ttime %d",
+                    has_short_ttime);
 
-       if (enable_idle)
-               bfq_mark_bfqq_idle_window(bfqq);
+       if (has_short_ttime)
+               bfq_mark_bfqq_has_short_ttime(bfqq);
        else
-               bfq_clear_bfqq_idle_window(bfqq);
+               bfq_clear_bfqq_has_short_ttime(bfqq);
 }
 
 /*
                bfqq->meta_pending++;
 
        bfq_update_io_thinktime(bfqd, bfqq);
+       bfq_update_has_short_ttime(bfqd, bfqq, bic);
        bfq_update_io_seektime(bfqd, bfqq, rq);
-       if (bfqq->entity.service > bfq_max_budget(bfqd) / 8 ||
-           !BFQQ_SEEKY(bfqq))
-               bfq_update_idle_window(bfqd, bfqq, bic);
 
        bfq_log_bfqq(bfqd, bfqq,
-                    "rq_enqueued: idle_window=%d (seeky %d)",
-                    bfq_bfqq_idle_window(bfqq), BFQQ_SEEKY(bfqq));
+                    "rq_enqueued: has_short_ttime=%d (seeky %d)",
+                    bfq_bfqq_has_short_ttime(bfqq), BFQQ_SEEKY(bfqq));
 
        bfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq);