/* End of sysctl-exported parameters */
 
+unsigned long global_dirty_limit;
 
 /*
  * Scale the writeback cache size proportional to the relative writeout speeds.
        bdi->avg_write_bandwidth = avg;
 }
 
+/*
+ * The global dirtyable memory and dirty threshold could be suddenly knocked
+ * down by a large amount (eg. on the startup of KVM in a swapless system).
+ * This may throw the system into deep dirty exceeded state and throttle
+ * heavy/light dirtiers alike. To retain good responsiveness, maintain
+ * global_dirty_limit for tracking slowly down to the knocked down dirty
+ * threshold.
+ */
+static void update_dirty_limit(unsigned long thresh, unsigned long dirty)
+{
+       unsigned long limit = global_dirty_limit;
+
+       /*
+        * Follow up in one step.
+        */
+       if (limit < thresh) {
+               limit = thresh;
+               goto update;
+       }
+
+       /*
+        * Follow down slowly. Use the higher one as the target, because thresh
+        * may drop below dirty. This is exactly the reason to introduce
+        * global_dirty_limit which is guaranteed to lie above the dirty pages.
+        */
+       thresh = max(thresh, dirty);
+       if (limit > thresh) {
+               limit -= (limit - thresh) >> 5;
+               goto update;
+       }
+       return;
+update:
+       global_dirty_limit = limit;
+}
+
+static void global_update_bandwidth(unsigned long thresh,
+                                   unsigned long dirty,
+                                   unsigned long now)
+{
+       static DEFINE_SPINLOCK(dirty_lock);
+       static unsigned long update_time;
+
+       /*
+        * check locklessly first to optimize away locking for the most time
+        */
+       if (time_before(now, update_time + BANDWIDTH_INTERVAL))
+               return;
+
+       spin_lock(&dirty_lock);
+       if (time_after_eq(now, update_time + BANDWIDTH_INTERVAL)) {
+               update_dirty_limit(thresh, dirty);
+               update_time = now;
+       }
+       spin_unlock(&dirty_lock);
+}
+
 void __bdi_update_bandwidth(struct backing_dev_info *bdi,
+                           unsigned long thresh,
+                           unsigned long dirty,
+                           unsigned long bdi_thresh,
+                           unsigned long bdi_dirty,
                            unsigned long start_time)
 {
        unsigned long now = jiffies;
        if (elapsed > HZ && time_before(bdi->bw_time_stamp, start_time))
                goto snapshot;
 
+       if (thresh)
+               global_update_bandwidth(thresh, dirty, now);
+
        bdi_update_write_bandwidth(bdi, elapsed, written);
 
 snapshot:
 }
 
 static void bdi_update_bandwidth(struct backing_dev_info *bdi,
+                                unsigned long thresh,
+                                unsigned long dirty,
+                                unsigned long bdi_thresh,
+                                unsigned long bdi_dirty,
                                 unsigned long start_time)
 {
        if (time_is_after_eq_jiffies(bdi->bw_time_stamp + BANDWIDTH_INTERVAL))
                return;
        spin_lock(&bdi->wb.list_lock);
-       __bdi_update_bandwidth(bdi, start_time);
+       __bdi_update_bandwidth(bdi, thresh, dirty, bdi_thresh, bdi_dirty,
+                              start_time);
        spin_unlock(&bdi->wb.list_lock);
 }
 
                if (!bdi->dirty_exceeded)
                        bdi->dirty_exceeded = 1;
 
-               bdi_update_bandwidth(bdi, start_time);
+               bdi_update_bandwidth(bdi, dirty_thresh, nr_dirty,
+                                    bdi_thresh, bdi_dirty, start_time);
 
                /* Note: nr_reclaimable denotes nr_dirty + nr_unstable.
                 * Unstable writes are a feature of certain networked