Skip to content

Commit

Permalink
sched/core: Optimize the process of picking the max prio task for the…
Browse files Browse the repository at this point in the history
… core

When we pick the max prio task for the core in the case of
sched_core_enabled(), if there's a task with a higher prio sched
class in the runqueue of a SMT, it's not necessary for other SMTs
to traverse lower prio sched classes. So we can change the traversal
order: for each sched class, if there exists a max prio task among
the core, pick it and break the loop.

To compare the prio of the tasks with the same sched class, we
introduce sched_class::prio_less().

Signed-off-by: Cruz Zhao <CruzZhao@linux.alibaba.com>
  • Loading branch information
Cruz Zhao authored and intel-lab-lkp committed Sep 29, 2022
1 parent 5aec788 commit 606244f
Show file tree
Hide file tree
Showing 7 changed files with 70 additions and 14 deletions.
33 changes: 22 additions & 11 deletions kernel/sched/core.c
Expand Up @@ -188,7 +188,7 @@ static inline bool prio_less(struct task_struct *a, struct task_struct *b, bool
return !dl_time_before(a->dl.deadline, b->dl.deadline);

if (pa == MAX_RT_PRIO + MAX_NICE) /* fair */
return cfs_prio_less(a, b, in_fi);
return fair_sched_class.prio_less(a, b, in_fi);

return false;
}
Expand Down Expand Up @@ -5869,6 +5869,7 @@ pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
int i, cpu, occ = 0;
struct rq *rq_i;
bool need_sync;
struct sched_class *class;

if (!sched_core_enabled(rq))
return __pick_next_task(rq, prev, rf);
Expand Down Expand Up @@ -5961,12 +5962,6 @@ pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
}
}

/*
* For each thread: do the regular task pick and find the max prio task
* amongst them.
*
* Tie-break prio towards the current CPU
*/
for_each_cpu_wrap(i, smt_mask, cpu) {
rq_i = cpu_rq(i);

Expand All @@ -5977,12 +5972,28 @@ pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
*/
if (i != cpu && (rq_i != rq->core || !core_clock_updated))
update_rq_clock(rq_i);
}

p = rq_i->core_pick = pick_task(rq_i);
if (!max || prio_less(max, p, fi_before))
max = p;
/*
* For each thread: do the regular task pick and find the max prio task
* amongst them.
*
* Tie-break prio towards the current CPU
*/
for_each_class(class) {
for_each_cpu_wrap(i, smt_mask, cpu) {
rq_i = cpu_rq(i);
p = rq_i->core_pick = class->pick_task(rq_i);
if (!max || (p && class->prio_less(max, p, fi_before)))
max = p;
}
if (max)
break;
}

if (!max)
BUG();

cookie = rq->core->core_cookie = max->core_cookie;

/*
Expand All @@ -5993,7 +6004,7 @@ pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
rq_i = cpu_rq(i);
p = rq_i->core_pick;

if (!cookie_equals(p, cookie)) {
if (!p || !cookie_equals(p, cookie)) {
p = NULL;
if (cookie)
p = sched_core_find(rq_i, cookie);
Expand Down
11 changes: 11 additions & 0 deletions kernel/sched/deadline.c
Expand Up @@ -1942,6 +1942,7 @@ static int balance_dl(struct rq *rq, struct task_struct *p, struct rq_flags *rf)

return sched_stop_runnable(rq) || sched_dl_runnable(rq);
}

#endif /* CONFIG_SMP */

/*
Expand Down Expand Up @@ -2054,6 +2055,13 @@ static void put_prev_task_dl(struct rq *rq, struct task_struct *p)
enqueue_pushable_dl_task(rq, p);
}

#ifdef CONFIG_SCHED_CORE
static bool prio_less_dl(struct task_struct *a, struct task_struct *b, bool in_fi)
{
return !dl_time_before(a->dl.deadline, b->dl.deadline);
}
#endif

/*
* scheduler tick hitting a task of our scheduling class.
*
Expand Down Expand Up @@ -2704,6 +2712,9 @@ DEFINE_SCHED_CLASS(dl) = {
.pick_next_task = pick_next_task_dl,
.put_prev_task = put_prev_task_dl,
.set_next_task = set_next_task_dl,
#ifdef CONFIG_SCHED_CORE
.prio_less = prio_less_dl,
#endif

#ifdef CONFIG_SMP
.balance = balance_dl,
Expand Down
5 changes: 4 additions & 1 deletion kernel/sched/fair.c
Expand Up @@ -11347,7 +11347,7 @@ void task_vruntime_update(struct rq *rq, struct task_struct *p, bool in_fi)
se_fi_update(se, rq->core->core_forceidle_seq, in_fi);
}

bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool in_fi)
static bool prio_less_fair(struct task_struct *a, struct task_struct *b, bool in_fi)
{
struct rq *rq = task_rq(a);
struct sched_entity *sea = &a->se;
Expand Down Expand Up @@ -11990,6 +11990,9 @@ DEFINE_SCHED_CLASS(fair) = {
.pick_next_task = __pick_next_task_fair,
.put_prev_task = put_prev_task_fair,
.set_next_task = set_next_task_fair,
#ifdef CONFIG_SCHED_CORE
.prio_less = prio_less_fair,
#endif

#ifdef CONFIG_SMP
.balance = balance_fair,
Expand Down
10 changes: 10 additions & 0 deletions kernel/sched/idle.c
Expand Up @@ -436,6 +436,13 @@ static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool fir
schedstat_inc(rq->sched_goidle);
}

#ifdef CONFIG_SCHED_CORE
static bool prio_less_idle(struct task_struct *a, struct task_struct *b, bool in_fi)
{
return false;
}
#endif

#ifdef CONFIG_SMP
static struct task_struct *pick_task_idle(struct rq *rq)
{
Expand Down Expand Up @@ -507,6 +514,9 @@ DEFINE_SCHED_CLASS(idle) = {
.pick_next_task = pick_next_task_idle,
.put_prev_task = put_prev_task_idle,
.set_next_task = set_next_task_idle,
#ifdef CONFIG_SCHED_CORE
.prio_less = prio_less_idle,
#endif

#ifdef CONFIG_SMP
.balance = balance_idle,
Expand Down
10 changes: 10 additions & 0 deletions kernel/sched/rt.c
Expand Up @@ -1838,6 +1838,13 @@ static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
enqueue_pushable_task(rq, p);
}

static bool prio_less_rt(struct task_struct *a, struct task_struct *b, bool in_fi)
{
int pa = rt_prio(a->prio), pb = rt_prio(b->prio);

return -pa < -pb;
}

#ifdef CONFIG_SMP

/* Only try algorithms three times */
Expand Down Expand Up @@ -2685,6 +2692,9 @@ DEFINE_SCHED_CLASS(rt) = {
.pick_next_task = pick_next_task_rt,
.put_prev_task = put_prev_task_rt,
.set_next_task = set_next_task_rt,
#ifdef CONFIG_SCHED_CORE
.prio_less = prio_less_rt,
#endif

#ifdef CONFIG_SMP
.balance = balance_rt,
Expand Down
5 changes: 3 additions & 2 deletions kernel/sched/sched.h
Expand Up @@ -1218,8 +1218,6 @@ static inline raw_spinlock_t *__rq_lockp(struct rq *rq)
return &rq->__lock;
}

bool cfs_prio_less(struct task_struct *a, struct task_struct *b, bool fi);

/*
* Helpers to check if the CPU's core cookie matches with the task's cookie
* when core scheduling is enabled.
Expand Down Expand Up @@ -2155,6 +2153,9 @@ struct sched_class {

void (*put_prev_task)(struct rq *rq, struct task_struct *p);
void (*set_next_task)(struct rq *rq, struct task_struct *p, bool first);
#ifdef CONFIG_SCHED_CORE
bool prio_less(struct task_struct *a, struct task_struct *b, bool in_fi);
#endif

#ifdef CONFIG_SMP
int (*balance)(struct rq *rq, struct task_struct *prev, struct rq_flags *rf);
Expand Down
10 changes: 10 additions & 0 deletions kernel/sched/stop_task.c
Expand Up @@ -84,6 +84,13 @@ static void put_prev_task_stop(struct rq *rq, struct task_struct *prev)
update_current_exec_runtime(curr, now, delta_exec);
}

#ifdef CONFIG_SCHED_CORE
static bool prio_less_stop(struct task_struct *a, struct task_struct *b, bool in_fi)
{
return false;
}
#endif

/*
* scheduler tick hitting a task of our scheduling class.
*
Expand Down Expand Up @@ -125,6 +132,9 @@ DEFINE_SCHED_CLASS(stop) = {
.pick_next_task = pick_next_task_stop,
.put_prev_task = put_prev_task_stop,
.set_next_task = set_next_task_stop,
#ifdef CONFIG_SCHED_CORE
.prio_less = prio_less_stop,
#endif

#ifdef CONFIG_SMP
.balance = balance_stop,
Expand Down

0 comments on commit 606244f

Please sign in to comment.