2 * Only give sleepers 50% of their service deficit. This allows
3 * them to run sooner, but does not allow tons of sleepers to
4 * rip the spread apart.
6 SCHED_FEAT(GENTLE_FAIR_SLEEPERS, 1)
9 * Place new tasks ahead so that they do not starve already running
12 SCHED_FEAT(START_DEBIT, 1)
15 * Should wakeups try to preempt running tasks.
17 SCHED_FEAT(WAKEUP_PREEMPT, 1)
20 * When converting the wakeup granularity to virtual time, do it such
21 * that heavier tasks preempting a lighter task have an edge.
23 SCHED_FEAT(ASYM_GRAN, 1)
26 * Always wakeup-preempt SYNC wakeups, see SYNC_WAKEUPS.
28 SCHED_FEAT(WAKEUP_SYNC, 0)
31 * Use the SYNC wakeup hint, pipes and the likes use this to indicate
32 * the remote end is likely to consume the data we just wrote, and
33 * therefore has cache benefit from being placed on the same cpu, see
34 * also AFFINE_WAKEUPS.
36 SCHED_FEAT(SYNC_WAKEUPS, 1)
39 * Based on load and program behaviour, see if it makes sense to place
40 * a newly woken task on the same cpu as the task that woke it --
41 * improve cache locality. Typically used with SYNC wakeups as
42 * generated by pipes and the like, see also SYNC_WAKEUPS.
44 SCHED_FEAT(AFFINE_WAKEUPS, 1)
47 * Prefer to schedule the task we woke last (assuming it failed
48 * wakeup-preemption), since its likely going to consume data we
49 * touched, increases cache locality.
51 SCHED_FEAT(NEXT_BUDDY, 0)
54 * Prefer to schedule the task that ran last (when we did
55 * wake-preempt) as that likely will touch the same data, increases
58 SCHED_FEAT(LAST_BUDDY, 1)
61 * Consider buddies to be cache hot, decreases the likelyness of a
62 * cache buddy being migrated away, increases cache locality.
64 SCHED_FEAT(CACHE_HOT_BUDDY, 1)
67 * Use arch dependent cpu power functions
69 SCHED_FEAT(ARCH_POWER, 0)
72 SCHED_FEAT(DOUBLE_TICK, 0)
73 SCHED_FEAT(LB_BIAS, 1)
74 SCHED_FEAT(LB_SHARES_UPDATE, 1)
75 SCHED_FEAT(ASYM_EFF_LOAD, 1)
78 * Spin-wait on mutex acquisition when the mutex owner is running on
79 * another cpu -- assumes that when the owner is running, it will soon
80 * release the lock. Decreases scheduling overhead.
82 SCHED_FEAT(OWNER_SPIN, 1)