According to the original code logic:
cfs_rq->avg.util_avg
sa->util_avg = -------------------- * se->load.weight
cfs_rq->avg.load_avg
but for fair_sched_class in 64bits platform:
se->load.weight = 1024 * sched_prio_to_weight[prio];
cfs_rq->avg.util_avg
so the -------------------- must be extremely small, the
cfs_rq->avg.load_avg
judgment condition "sa->util_avg < cap" could be established.
It's not fair for those tasks who has smaller nice value.
Signed-off-by: Xuewen Yan <[email protected]>
---
changes since V2:
*kernel/sched/fair.c | 6 +++++-
* 1 file changed, 5 insertions(+), 1 deletion(-)
*
*diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
*index 290f9e3..079760b 100644
*--- a/kernel/sched/fair.c
*+++ b/kernel/sched/fair.c
*@@ -794,7 +794,11 @@ void post_init_entity_util_avg(struct task_struct *p)
*
* if (cap > 0) {
* if (cfs_rq->avg.util_avg != 0) {
*- sa->util_avg = cfs_rq->avg.util_avg * se->load.weight;
*+ if (p->sched_class == &fair_sched_class)
*+ sa->util_avg = cfs_rq->avg.util_avg * se_weight(se);
*+ else
*+ sa->util_avg = cfs_rq->avg.util_avg * se->load.weight;
*+
* sa->util_avg /= (cfs_rq->avg.load_avg + 1);
*
* if (sa->util_avg > cap)
*
---
comment from Vincent Guittot <[email protected]>:
>
> According to the original code logic:
> cfs_rq->avg.util_avg
> sa->util_avg = -------------------- * se->load.weight
> cfs_rq->avg.load_avg
this should have been scale_load_down(se->load.weight) from the beginning
> but for fair_sched_class:
> se->load.weight = 1024 * sched_prio_to_weight[prio];
This is only true for 64bits platform otherwise scale_load and
scale_load_down are nop
> cfs_rq->avg.util_avg
> so the -------------------- must be extremely small, the
> cfs_rq->avg.load_avg
> judgment condition "sa->util_avg < cap" could be established.
> It's not fair for those tasks who has smaller nice value.
>
> Signed-off-by: Xuewen Yan <[email protected]>
> ---
> kernel/sched/fair.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 290f9e3..079760b 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -794,7 +794,11 @@ void post_init_entity_util_avg(struct task_struct *p)
>
> if (cap > 0) {
> if (cfs_rq->avg.util_avg != 0) {
We should now use cpu_util() instead of cfs_rq->avg.util_avg which
takes into account other classes
> - sa->util_avg = cfs_rq->avg.util_avg * se->load.weight;
> + if (p->sched_class == &fair_sched_class)
> + sa->util_avg = cfs_rq->avg.util_avg * se_weight(se);
> + else
> + sa->util_avg = cfs_rq->avg.util_avg * se->load.weight;
Why this else keeps using se->load.weight ?
Either we uses sa->util_avg = cfs_rq->avg.util_avg * se_weight(se);
for all classes
Or we want a different init value for other classes. But in this case
se->load.weight is meaningless and we should simply set them to 0
although we could probably compute a value based on bandwidth for
deadline class.
---
kernel/sched/fair.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 290f9e3..c6186cc 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -794,7 +794,7 @@ void post_init_entity_util_avg(struct task_struct *p)
if (cap > 0) {
if (cfs_rq->avg.util_avg != 0) {
- sa->util_avg = cfs_rq->avg.util_avg * se->load.weight;
+ sa->util_avg = cfs_rq->avg.util_avg * se_weight(se);
sa->util_avg /= (cfs_rq->avg.load_avg + 1);
if (sa->util_avg > cap)
--
1.9.1
On Fri, Nov 06, 2020 at 11:22:03AM +0800, Xuewen Yan wrote:
> According to the original code logic:
> cfs_rq->avg.util_avg
> sa->util_avg = -------------------- * se->load.weight
> cfs_rq->avg.load_avg
> but for fair_sched_class in 64bits platform:
> se->load.weight = 1024 * sched_prio_to_weight[prio];
> cfs_rq->avg.util_avg
> so the -------------------- must be extremely small, the
> cfs_rq->avg.load_avg
> judgment condition "sa->util_avg < cap" could be established.
> It's not fair for those tasks who has smaller nice value.
>
> Signed-off-by: Xuewen Yan <[email protected]>
> ---
> changes since V2:
>
> *kernel/sched/fair.c | 6 +++++-
> * 1 file changed, 5 insertions(+), 1 deletion(-)
> *
> *diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> *index 290f9e3..079760b 100644
> *--- a/kernel/sched/fair.c
> *+++ b/kernel/sched/fair.c
> *@@ -794,7 +794,11 @@ void post_init_entity_util_avg(struct task_struct *p)
> *
> * if (cap > 0) {
> * if (cfs_rq->avg.util_avg != 0) {
> *- sa->util_avg = cfs_rq->avg.util_avg * se->load.weight;
> *+ if (p->sched_class == &fair_sched_class)
> *+ sa->util_avg = cfs_rq->avg.util_avg * se_weight(se);
> *+ else
> *+ sa->util_avg = cfs_rq->avg.util_avg * se->load.weight;
> *+
> * sa->util_avg /= (cfs_rq->avg.load_avg + 1);
> *
> * if (sa->util_avg > cap)
> *
> ---
> comment from Vincent Guittot <[email protected]>:
> >
> > According to the original code logic:
> > cfs_rq->avg.util_avg
> > sa->util_avg = -------------------- * se->load.weight
> > cfs_rq->avg.load_avg
>
> this should have been scale_load_down(se->load.weight) from the beginning
>
> > but for fair_sched_class:
> > se->load.weight = 1024 * sched_prio_to_weight[prio];
>
> This is only true for 64bits platform otherwise scale_load and
> scale_load_down are nop
>
> > cfs_rq->avg.util_avg
> > so the -------------------- must be extremely small, the
> > cfs_rq->avg.load_avg
> > judgment condition "sa->util_avg < cap" could be established.
> > It's not fair for those tasks who has smaller nice value.
> >
> > Signed-off-by: Xuewen Yan <[email protected]>
> > ---
> > kernel/sched/fair.c | 6 +++++-
> > 1 file changed, 5 insertions(+), 1 deletion(-)
> >
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index 290f9e3..079760b 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -794,7 +794,11 @@ void post_init_entity_util_avg(struct task_struct *p)
> >
> > if (cap > 0) {
> > if (cfs_rq->avg.util_avg != 0) {
>
> We should now use cpu_util() instead of cfs_rq->avg.util_avg which
> takes into account other classes
>
> > - sa->util_avg = cfs_rq->avg.util_avg * se->load.weight;
> > + if (p->sched_class == &fair_sched_class)
> > + sa->util_avg = cfs_rq->avg.util_avg * se_weight(se);
> > + else
> > + sa->util_avg = cfs_rq->avg.util_avg * se->load.weight;
>
> Why this else keeps using se->load.weight ?
>
> Either we uses sa->util_avg = cfs_rq->avg.util_avg * se_weight(se);
> for all classes
>
> Or we want a different init value for other classes. But in this case
> se->load.weight is meaningless and we should simply set them to 0
> although we could probably compute a value based on bandwidth for
> deadline class.
>
> ---
> kernel/sched/fair.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 290f9e3..c6186cc 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -794,7 +794,7 @@ void post_init_entity_util_avg(struct task_struct *p)
>
> if (cap > 0) {
> if (cfs_rq->avg.util_avg != 0) {
> - sa->util_avg = cfs_rq->avg.util_avg * se->load.weight;
> + sa->util_avg = cfs_rq->avg.util_avg * se_weight(se);
Please refer to this MessageID: 20161208012722.GA4128@geo in lkml web site
if you want. Just a notice and no matter here. My head do not work now.
I can't remember more things that time..
> sa->util_avg /= (cfs_rq->avg.load_avg + 1);
>
> if (sa->util_avg > cap)
> --
> 1.9.1
>