[PATCH try 5] CFS: Add hierarchical tree-based penalty.

[William Pitcock <nenolod@dereferenced.org>]

Inspired by the recent change to BFS by Con Kolivas, this patch causes
vruntime to be penalized based on parent depth from their root task
group.

I have, for the moment, decided to make it a default feature since the
design of CFS ensures that broken applications depending on task enqueue
behaviour behaving traditionally will continue to work.

Signed-off-by: William Pitcock <nenolod@dereferenced.org>
---
 include/linux/sched.h   |    2 ++
 kernel/sched.c          |    4 ++++
 kernel/sched_fair.c     |    8 ++++++++
 kernel/sched_features.h |   12 ++++++++++++
 4 files changed, 26 insertions(+), 0 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 1e2a6db..3f0cff8 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1494,6 +1494,8 @@ struct task_struct {
 		unsigned long memsw_bytes; /* uncharged mem+swap usage */
 	} memcg_batch;
 #endif
+
+	int fork_depth;
 };
 
 /* Future-safe accessor for struct task_struct's cpus_allowed. */
diff --git a/kernel/sched.c b/kernel/sched.c
index dc85ceb..3fc3ebc 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -2621,6 +2621,10 @@ void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
 #endif
 
 	rq = task_rq_lock(p, &flags);
+
+	if (!(clone_flags & CLONE_THREAD))
+		p->fork_depth++;
+
 	activate_task(rq, p, 0);
 	trace_sched_wakeup_new(p, 1);
 	check_preempt_curr(rq, p, WF_FORK);
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index db3f674..a81acc5 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -737,6 +737,14 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
 	if (initial && sched_feat(START_DEBIT))
 		vruntime += sched_vslice(cfs_rq, se);
 
+	if (sched_feat(HIERARCHICAL_PENALTY) &&
+	    likely(entity_is_task(se))) {
+		struct task_struct *tsk = task_of(se);
+
+		if (tsk->fork_depth > 1)
+			vruntime *= tsk->fork_depth;
+	}
+
 	/* sleeps up to a single latency don't count. */
 	if (!initial) {
 		unsigned long thresh = sysctl_sched_latency;
diff --git a/kernel/sched_features.h b/kernel/sched_features.h
index 83c66e8..cf17f97 100644
--- a/kernel/sched_features.h
+++ b/kernel/sched_features.h
@@ -45,6 +45,18 @@ SCHED_FEAT(LAST_BUDDY, 1)
 SCHED_FEAT(CACHE_HOT_BUDDY, 1)
 
 /*
+ * Hierarchical tree-based penalty: penalize service deficit by
+ * an order of magnitude for each parent process in the process
+ * tree.  This has the natural effect of forcing preference towards
+ * processes that are not fork()-hungry, like make(1), which helps
+ * to preserve good latency.
+ *
+ * This also has the side effect of providing in a limited way,
+ * per-user CPU entitlement partitioning.
+ */
+SCHED_FEAT(HIERARCHICAL_PENALTY, 1)
+
+/*
  * Use arch dependent cpu power functions
  */
 SCHED_FEAT(ARCH_POWER, 0)
-- 
1.7.2.1