Commit | Line | Data |
---|---|---|
51e0304c IM |
1 | /* |
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. | |
5 | */ | |
f8b6d1cc | 6 | SCHED_FEAT(GENTLE_FAIR_SLEEPERS, true) |
e26af0e8 | 7 | |
e26af0e8 PZ |
8 | /* |
9 | * Place new tasks ahead so that they do not starve already running | |
10 | * tasks | |
11 | */ | |
f8b6d1cc | 12 | SCHED_FEAT(START_DEBIT, true) |
e26af0e8 | 13 | |
e26af0e8 PZ |
14 | /* |
15 | * Based on load and program behaviour, see if it makes sense to place | |
16 | * a newly woken task on the same cpu as the task that woke it -- | |
17 | * improve cache locality. Typically used with SYNC wakeups as | |
18 | * generated by pipes and the like, see also SYNC_WAKEUPS. | |
19 | */ | |
f8b6d1cc | 20 | SCHED_FEAT(AFFINE_WAKEUPS, true) |
e26af0e8 PZ |
21 | |
22 | /* | |
23 | * Prefer to schedule the task we woke last (assuming it failed | |
24 | * wakeup-preemption), since its likely going to consume data we | |
25 | * touched, increases cache locality. | |
26 | */ | |
f8b6d1cc | 27 | SCHED_FEAT(NEXT_BUDDY, false) |
e26af0e8 PZ |
28 | |
29 | /* | |
30 | * Prefer to schedule the task that ran last (when we did | |
31 | * wake-preempt) as that likely will touch the same data, increases | |
32 | * cache locality. | |
33 | */ | |
f8b6d1cc | 34 | SCHED_FEAT(LAST_BUDDY, true) |
e26af0e8 PZ |
35 | |
36 | /* | |
37 | * Consider buddies to be cache hot, decreases the likelyness of a | |
38 | * cache buddy being migrated away, increases cache locality. | |
39 | */ | |
f8b6d1cc | 40 | SCHED_FEAT(CACHE_HOT_BUDDY, true) |
e26af0e8 | 41 | |
8e6598af PZ |
42 | /* |
43 | * Use arch dependent cpu power functions | |
44 | */ | |
f8b6d1cc | 45 | SCHED_FEAT(ARCH_POWER, false) |
8e6598af | 46 | |
f8b6d1cc PZ |
47 | SCHED_FEAT(HRTICK, false) |
48 | SCHED_FEAT(DOUBLE_TICK, false) | |
49 | SCHED_FEAT(LB_BIAS, true) | |
e26af0e8 PZ |
50 | |
51 | /* | |
52 | * Spin-wait on mutex acquisition when the mutex owner is running on | |
53 | * another cpu -- assumes that when the owner is running, it will soon | |
54 | * release the lock. Decreases scheduling overhead. | |
55 | */ | |
f8b6d1cc | 56 | SCHED_FEAT(OWNER_SPIN, true) |
aa483808 VP |
57 | |
58 | /* | |
095c0aa8 | 59 | * Decrement CPU power based on time not spent running tasks |
aa483808 | 60 | */ |
f8b6d1cc | 61 | SCHED_FEAT(NONTASK_POWER, true) |
317f3941 PZ |
62 | |
63 | /* | |
64 | * Queue remote wakeups on the target CPU and process them | |
65 | * using the scheduler IPI. Reduces rq->lock contention/bounces. | |
66 | */ | |
f8b6d1cc | 67 | SCHED_FEAT(TTWU_QUEUE, true) |
e3589f6c | 68 | |
f8b6d1cc PZ |
69 | SCHED_FEAT(FORCE_SD_OVERLAP, false) |
70 | SCHED_FEAT(RT_RUNTIME_SHARE, true) | |
eb95308e | 71 | SCHED_FEAT(LB_MIN, false) |