[kernel/sched/core.c] Review and Modified of the prio_less() about sched class priority.

[kernel/sched/core.c] Review and Modified of the prio_less() about sched class priority.

[JaeJoon Jung]

The sched_class structure is defined to be sorted by pointer size.
You can see it in the macro definition like this:

kernel/sched/sched.h
#define DEFINE_SCHED_CLASS(name)
const struct sched_class name##_sched_class \
        __aligned(__alignof__(struct sched_class)) \
        __section("__" #name "_sched_class")

include/asm-generic/vmlinux.lds.h
#define SCHED_DATA \
STRUCT_ALIGN(); \
__sched_class_highest = .; \
*(__stop_sched_class) \
*(__dl_sched_class) \
*(__rt_sched_class) \
*(__fair_sched_class) \
*(__idle_sched_class) \
__sched_class_lowest = .;

And in the System.map file,
you can see that they are arranged in memory address order.

System.map
----------------------------------------------------------------
ffffffff8260d520 R __sched_class_highest
ffffffff8260d520 R stop_sched_class
ffffffff8260d5f0 R dl_sched_class
ffffffff8260d6c0 R rt_sched_class
ffffffff8260d790 R fair_sched_class
ffffffff8260d860 R idle_sched_class
ffffffff8260d930 R __sched_class_lowest
----------------------------------------------------------------

This matches the sched class priority.
Therefore, in the prio_less() function in kernel/sched/core.c,
the less value can be determined by pointer operation as follows.

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index f730b6fe94a7..7a64ac8ea3d8 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -151,21 +151,6 @@ __read_mostly int scheduler_running;

 DEFINE_STATIC_KEY_FALSE(__sched_core_enabled);

-/* kernel prio, less is more */
-static inline int __task_prio(struct task_struct *p)
-{
-       if (p->sched_class == &stop_sched_class) /* trumps deadline */
-               return -2;
-
-       if (rt_prio(p->prio)) /* includes deadline */
-               return p->prio; /* [-1, 99] */
-
-       if (p->sched_class == &idle_sched_class)
-               return MAX_RT_PRIO + NICE_WIDTH; /* 140 */
-
-       return MAX_RT_PRIO + MAX_NICE; /* 120, squash fair */
-}
-
 /*
  * l(a,b)
  * le(a,b) := !l(b,a)
@@ -176,22 +161,18 @@ static inline int __task_prio(struct task_struct *p)
 /* real prio, less is less */
 static inline bool prio_less(struct task_struct *a, struct
task_struct *b, bool in_fi)
 {
+        int less = a->sched_class - b->sched_class;

-       int pa = __task_prio(a), pb = __task_prio(b);
+        if (less == 0) {
+                if (a->sched_class == &dl_sched_class)
+                        return !dl_time_before(a->dl.deadline, b->dl.deadline);

-       if (-pa < -pb)
-               return true;
-
-       if (-pb < -pa)
-               return false;
-
-       if (pa == -1) /* dl_prio() doesn't work because of stop_class above */
-               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 false;
+                else if (a->sched_class == &fair_sched_class)
+                        return cfs_prio_less(a, b, in_fi);
+                else
+                        return false;
+        } else
+                return (less > 0) ? true : false;
 }

If the prio_less() function is modified as above, the __task_prio()
function is not required.
Please review.
Thanks,
From JaeJoon Jung.

Re: [kernel/sched/core.c] Review and Modified of the prio_less() about sched class priority.

[Peter Zijlstra]

On Mon, Feb 13, 2023 at 10:11:04AM +0900, JaeJoon Jung wrote:
> The sched_class structure is defined to be sorted by pointer size.
> @@ -176,22 +161,18 @@ static inline int __task_prio(struct task_struct *p)
>  /* real prio, less is less */
>  static inline bool prio_less(struct task_struct *a, struct
> task_struct *b, bool in_fi)
>  {
> +        int less = a->sched_class - b->sched_class;
> 
> +        if (less == 0) {
> +                if (a->sched_class == &dl_sched_class)
> +                        return !dl_time_before(a->dl.deadline, b->dl.deadline);
> 
> +                else if (a->sched_class == &fair_sched_class)
> +                        return cfs_prio_less(a, b, in_fi);
> +                else
> +                        return false;
> +        } else
> +                return (less > 0) ? true : false;
>  }
> 
> If the prio_less() function is modified as above, the __task_prio()
> function is not required.

Yeah, except your patch is whitespace mangled..

Re: [kernel/sched/core.c] Review and Modified of the prio_less() about sched class priority.

[JaeJoon Jung]

Thank you for answering.
Is it going to be reflected in your patch?

JaeJoon Jung.

On Mon, 13 Feb 2023 at 19:08, Peter Zijlstra <peterz@infradead.org> wrote:
>
> On Mon, Feb 13, 2023 at 10:11:04AM +0900, JaeJoon Jung wrote:
> > The sched_class structure is defined to be sorted by pointer size.
> > @@ -176,22 +161,18 @@ static inline int __task_prio(struct task_struct *p)
> >  /* real prio, less is less */
> >  static inline bool prio_less(struct task_struct *a, struct
> > task_struct *b, bool in_fi)
> >  {
> > +        int less = a->sched_class - b->sched_class;
> >
> > +        if (less == 0) {
> > +                if (a->sched_class == &dl_sched_class)
> > +                        return !dl_time_before(a->dl.deadline, b->dl.deadline);
> >
> > +                else if (a->sched_class == &fair_sched_class)
> > +                        return cfs_prio_less(a, b, in_fi);
> > +                else
> > +                        return false;
> > +        } else
> > +                return (less > 0) ? true : false;
> >  }
> >
> > If the prio_less() function is modified as above, the __task_prio()
> > function is not required.
>
> Yeah, except your patch is whitespace mangled..