Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 434dab22 authored by Abhijeet Dharmapurikar's avatar Abhijeet Dharmapurikar
Browse files

sched/fair: do not reset target_capacity 



Currently, we reset target_capacity to ULONG_MAX in placement boost.
This causes us to bypass that capacity check in the cpu loop thereby
allowing us to visit higher capacity cluster.

Instead choose to keep target_capacity intact and delete the capacity
check in cpu loop.

The capacity check was introduced by
commit cf9d0b63 ("sched/fair: fix
incorrect CPU selection for non latency sensitive tasks")
and it is safe to delete it because once we find a cpu in a group we
exit the group traversing loop. We will never have an situation
when we have found a cpu and are visiting higher capacity cluster,
except for placement boost when we intentionally want to visit higher
capacity cpus and were using the target_capacity = ULONG_MAX trick.

Change-Id: I73d25a1e7b30c0420e175371bd9f5911c71ddee0
Signed-off-by: default avatarAbhijeet Dharmapurikar <adharmap@codeaurora.org>
parent 7b2a034c

kernel/sched/fair.c

+9 −29
Original line number Diff line number Diff line
@@ -7286,13 +7286,6 @@ static inline int find_best_target(struct task_struct *p, int *backup_cpu, 
				continue;

			}



			/*

			 * Favor CPUs with smaller capacity for Non latency

			 * sensitive tasks.

			 */

			if (capacity_orig > target_capacity)

				continue;



			/*

			 * Case B) Non latency sensitive tasks on IDLE CPUs.

			 *
@@ -7383,29 +7376,16 @@ static inline int find_best_target(struct task_struct *p, int *backup_cpu, 
		}



		/*

		 * For placement boost (or otherwise), we start with group

		 * where the task should be placed. When

		 * boost is active, and we are not at the highest

		 * capacity group reset the target_capacity to keep

		 * traversing to other higher clusters.

		 * If we already are at the highest capacity cluster we skip

		 * going around to the lower capacity cluster if we've found

		 * a cpu.

		 */

		if (fbt_env->placement_boost == SCHED_BOOST_ON_BIG) {

			if (capacity_orig_of(group_first_cpu(sg)) <

				capacity_orig_of(group_first_cpu(sg->next)))

				target_capacity = ULONG_MAX;

			else

				if (target_cpu != -1 || best_idle_cpu != -1)

					break;

		}



		/*

		 * if we have found a target cpu within a group, don't bother

		 * checking other groups, provided we are not in placement boost

		 * If we've found a cpu, but the boost is ON_ALL we continue

		 * visiting other clusters. If the boost is ON_BIG we visit

		 * next cluster if they are higher in capacity. If we are

		 * not in any kind of boost, we break.

		 */

		if (target_capacity != ULONG_MAX)

		if ((target_cpu != -1 || best_idle_cpu != -1) &&

			(fbt_env->placement_boost == SCHED_BOOST_NONE ||

			(fbt_env->placement_boost == SCHED_BOOST_ON_BIG &&

				(capacity_orig_of(group_first_cpu(sg)) >

				capacity_orig_of(group_first_cpu(sg->next))))))

			break;



	} while (sg = sg->next, sg != sd->groups);