linux-3.19.8 patch for optimizing the kernel scheduling performance for data-intensive programs

mypatch

diff --color -uprN linux-3.19.8/include/linux/sched.h linux-3.19.8.changed/include/linux/sched.h
--- linux-3.19.8/include/linux/sched.h	2015-05-11 12:34:10.000000000 +0000
+++ linux-3.19.8.changed/include/linux/sched.h	2023-04-30 11:07:41.226266056 +0000
@@ -1164,6 +1164,8 @@ struct sched_entity {
 	struct list_head	group_node;
 	unsigned int		on_rq;
 
+	unsigned int		data_intensive;
+
 	u64			exec_start;
 	u64			sum_exec_runtime;
 	u64			vruntime;
diff --color -uprN linux-3.19.8/kernel/sched/core.c linux-3.19.8.changed/kernel/sched/core.c
--- linux-3.19.8/kernel/sched/core.c	2015-05-11 12:34:10.000000000 +0000
+++ linux-3.19.8.changed/kernel/sched/core.c	2023-04-30 11:07:41.338267408 +0000
@@ -1835,6 +1835,8 @@ static void __sched_fork(unsigned long c
 {
 	p->on_rq			= 0;
 
+	p->se.data_intensive   = 0;
+	
 	p->se.on_rq			= 0;
 	p->se.exec_start		= 0;
 	p->se.sum_exec_runtime		= 0;
diff --color -uprN linux-3.19.8/kernel/sched/fair.c linux-3.19.8.changed/kernel/sched/fair.c
--- linux-3.19.8/kernel/sched/fair.c	2015-05-11 12:34:10.000000000 +0000
+++ linux-3.19.8.changed/kernel/sched/fair.c	2023-04-30 11:07:41.338267408 +0000
@@ -30,6 +30,7 @@
 #include <linux/mempolicy.h>
 #include <linux/migrate.h>
 #include <linux/task_work.h>
+#include <linux/math64.h>
 
 #include <trace/events/sched.h>
 
@@ -688,6 +689,43 @@ void init_task_runnable_average(struct t
 }
 #endif
 
+static int check_data_intensive(struct cfs_rq *cfs_rq, struct sched_entity *se)
+{
+	if(se->statistics.iowait_count < 10)
+		return 0;
+	if(se->statistics.block_max > cfs_rq->block_max)
+		cfs_rq->block_max = se->statistics.block_max;
+	else
+		cfs_rq->block_max = cfs_rq->block_max >> 1;
+	if (se->statistics.block_max > cfs_rq->block_max){
+		cfs_rq->nr_data_intensive++;
+		se->data_intensive = 1;
+		return 1;
+	} else if (se->data_intensive == 1){
+		cfs_rq->nr_data_intensive--;
+		se->data_intensive = 0;
+		return 0;
+	}
+}
+
+static u64 sched_slice_io(struct cfs_rq *cfs_rq, struct sched_entity *se)
+{
+	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
+	for_each_sched_entity(se) {
+		u64 sum_iowait_time;
+		cfs_rq = cfs_rq_of(se);
+		sum_iowait_time = &cfs_rq->sum_iowait_time;
+		if (unlikely(!se->on_rq)) {
+			cfs_rq->sum_iowait_time += se->statistics.iowait_sum;
+		}
+		if(se->statistics.iowait_sum > cfs_rq->sum_iowait_time)
+			cfs_rq->sum_iowait_time = se->statistics.iowait_sum;
+		if(cfs_rq->sum_iowait_time)
+			slice = div_u64(slice * se->statistics.iowait_sum,cfs_rq->sum_iowait_time);
+	}
+	return slice;
+}
+
 /*
  * Update the current task's runtime statistics.
  */
@@ -713,6 +751,14 @@ static void update_curr(struct cfs_rq *c
 	schedstat_add(cfs_rq, exec_clock, delta_exec);
 
 	curr->vruntime += calc_delta_fair(delta_exec, curr);
+
+	if(check_data_intensive(cfs_rq, curr)){
+		u64 slice_io = sched_slice_io(cfs_rq, curr);
+		curr->vruntime -= slice_io;
+	}else{
+		curr->vruntime += cfs_rq->nr_data_intensive * sysctl_sched_min_granularity;
+	}
+
 	update_min_vruntime(cfs_rq);
 
 	if (entity_is_task(curr)) {
@@ -2926,6 +2972,7 @@ static void enqueue_sleeper(struct cfs_r
 			if (tsk->in_iowait) {
 				se->statistics.iowait_sum += delta;
 				se->statistics.iowait_count++;
+				cfs_rq->sum_iowait_time += delta;
 				trace_sched_stat_iowait(tsk, delta);
 			}
 
diff --color -uprN linux-3.19.8/kernel/sched/features.h linux-3.19.8.changed/kernel/sched/features.h
--- linux-3.19.8/kernel/sched/features.h	2015-05-11 12:34:10.000000000 +0000
+++ linux-3.19.8.changed/kernel/sched/features.h	2023-04-30 11:09:01.903224771 +0000
@@ -9,14 +9,14 @@ SCHED_FEAT(GENTLE_FAIR_SLEEPERS, true)
  * Place new tasks ahead so that they do not starve already running
  * tasks
  */
-SCHED_FEAT(START_DEBIT, true)
+SCHED_FEAT(START_DEBIT, false)
 
 /*
  * Prefer to schedule the task we woke last (assuming it failed
  * wakeup-preemption), since its likely going to consume data we
  * touched, increases cache locality.
  */
-SCHED_FEAT(NEXT_BUDDY, false)
+SCHED_FEAT(NEXT_BUDDY, true)
 
 /*
  * Prefer to schedule the task that ran last (when we did
diff --color -uprN linux-3.19.8/kernel/sched/sched.h linux-3.19.8.changed/kernel/sched/sched.h
--- linux-3.19.8/kernel/sched/sched.h	2015-05-11 12:34:10.000000000 +0000
+++ linux-3.19.8.changed/kernel/sched/sched.h	2023-04-30 11:07:41.338267408 +0000
@@ -337,6 +337,10 @@ struct cfs_rq {
 	struct load_weight load;
 	unsigned int nr_running, h_nr_running;
 
+    unsigned int nr_data_intensive;
+    u64 block_max;
+    u64 sum_iowait_time;
+
 	u64 exec_clock;
 	u64 min_vruntime;
 #ifndef CONFIG_64BIT