在 2026/2/27 14:25, Chen Jinghuang 写道:

Offering: HULK hulk inclusion category: perf bugzilla: https://atomgit.com/openeuler/kernel/issues/8555

--------------------------------

Implement fast load balancing with steal-task mechanism within the LLC domain; if task pulling fails, fall back to the native newidle_balance mechanism on LLC and upper scheduling domains.

This optimization prioritizes task scheduling within the LLC domain to improve cache hit rate, while mitigating system load imbalance issues.

Signed-off-by: Chen Jinghuang <chenjinghuang2@huawei.com> --- kernel/sched/fair.c | 15 +++++++++++++-- kernel/sched/features.h | 1 + 2 files changed, 14 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c index e60f19cb0fee..fd37f7648c86 100644 --- a/kernel/sched/fair.c +++ b/kernel/sched/fair.c @@ -10343,9 +10343,15 @@ done: __maybe_unused; */ rq_idle_stamp_update(rq);

- new_tasks = newidle_balance(rq, rf); - if (new_tasks == 0) + if (sched_feat(STEAL_FAIL_NI)) { new_tasks = try_steal(rq, rf); + if (new_tasks == 0) + new_tasks = newidle_balance(rq, rf); + } else { + new_tasks = newidle_balance(rq, rf); + if (new_tasks == 0) + new_tasks = try_steal(rq, rf); + } schedstat_end_time(rq, time);

if (new_tasks) @@ -14313,6 +14319,11 @@ static int newidle_balance(struct rq *this_rq, struct rq_flags *rf) int continue_balancing = 1; u64 domain_cost;

+ if (sched_feat(STEAL_FAIL_NI)) {

关了CONFIG_SCHED_STEAL会有编译错误？

+ if (sd->flags & SD_SHARE_PKG_RESOURCES) + continue; + } + update_next_balance(sd, &next_balance);

if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) diff --git a/kernel/sched/features.h b/kernel/sched/features.h index 1f665fbf0137..52ea0097c513 100644 --- a/kernel/sched/features.h +++ b/kernel/sched/features.h @@ -63,6 +63,7 @@ SCHED_FEAT(SIS_UTIL, true) * Improves CPU utilization. */ SCHED_FEAT(STEAL, false) +SCHED_FEAT(STEAL_FAIL_NI, false) #endif

#ifdef CONFIG_SCHED_PARAL