2 * Disregards a certain amount of sleep time (sched_latency_ns) and
3 * considers the task to be running during that period. This gives it
4 * a service deficit on wakeup, allowing it to run sooner.
6 SCHED_FEAT(FAIR_SLEEPERS, 1)
9 * Only give sleepers 50% of their service deficit. This allows
10 * them to run sooner, but does not allow tons of sleepers to
11 * rip the spread apart.
13 SCHED_FEAT(GENTLE_FAIR_SLEEPERS, 1)
16 * By not normalizing the sleep time, heavy tasks get an effective
17 * longer period, and lighter task an effective shorter period they
18 * are considered running.
20 SCHED_FEAT(NORMALIZED_SLEEPER, 0)
23 * Place new tasks ahead so that they do not starve already running
26 SCHED_FEAT(START_DEBIT, 1)
29 * Should wakeups try to preempt running tasks.
31 SCHED_FEAT(WAKEUP_PREEMPT, 1)
34 * Compute wakeup_gran based on task behaviour, clipped to
35 * [0, sched_wakeup_gran_ns]
37 SCHED_FEAT(ADAPTIVE_GRAN, 1)
40 * When converting the wakeup granularity to virtual time, do it such
41 * that heavier tasks preempting a lighter task have an edge.
43 SCHED_FEAT(ASYM_GRAN, 1)
46 * Always wakeup-preempt SYNC wakeups, see SYNC_WAKEUPS.
48 SCHED_FEAT(WAKEUP_SYNC, 0)
51 * Wakeup preempt based on task behaviour. Tasks that do not overlap
52 * don't get preempted.
54 SCHED_FEAT(WAKEUP_OVERLAP, 0)
57 * Use the SYNC wakeup hint, pipes and the likes use this to indicate
58 * the remote end is likely to consume the data we just wrote, and
59 * therefore has cache benefit from being placed on the same cpu, see
60 * also AFFINE_WAKEUPS.
62 SCHED_FEAT(SYNC_WAKEUPS, 1)
65 * Based on load and program behaviour, see if it makes sense to place
66 * a newly woken task on the same cpu as the task that woke it --
67 * improve cache locality. Typically used with SYNC wakeups as
68 * generated by pipes and the like, see also SYNC_WAKEUPS.
70 SCHED_FEAT(AFFINE_WAKEUPS, 1)
73 * Weaken SYNC hint based on overlap
75 SCHED_FEAT(SYNC_LESS, 1)
78 * Add SYNC hint based on overlap
80 SCHED_FEAT(SYNC_MORE, 0)
83 * Prefer to schedule the task we woke last (assuming it failed
84 * wakeup-preemption), since its likely going to consume data we
85 * touched, increases cache locality.
87 SCHED_FEAT(NEXT_BUDDY, 0)
90 * Prefer to schedule the task that ran last (when we did
91 * wake-preempt) as that likely will touch the same data, increases
94 SCHED_FEAT(LAST_BUDDY, 1)
97 * Consider buddies to be cache hot, decreases the likelyness of a
98 * cache buddy being migrated away, increases cache locality.
100 SCHED_FEAT(CACHE_HOT_BUDDY, 1)
103 * Use arch dependent cpu power functions
105 SCHED_FEAT(ARCH_POWER, 0)
107 SCHED_FEAT(HRTICK, 0)
108 SCHED_FEAT(DOUBLE_TICK, 0)
109 SCHED_FEAT(LB_BIAS, 1)
110 SCHED_FEAT(LB_SHARES_UPDATE, 1)
111 SCHED_FEAT(ASYM_EFF_LOAD, 1)
114 * Spin-wait on mutex acquisition when the mutex owner is running on
115 * another cpu -- assumes that when the owner is running, it will soon
116 * release the lock. Decreases scheduling overhead.
118 SCHED_FEAT(OWNER_SPIN, 1)