Skip to contentSkip to conversation

sched/core: Fix reset-on-fork from RT with uclamp

PATCH Superseded Quentin Perret 4 participants 11 messages Apr 14, 2020 – Apr 16, 2020 Scheduler

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sched/core: Fix reset-on-fork from RT with uclamp

uclamp_fork() resets the uclamp values to their default when the reset-on-fork flag is set. It also checks whether the task has a RT policy, and sets its uclamp.min to 1024 accordingly. However, during reset-on-fork, the task's policy is lowered to SCHED_NORMAL right after, hence leading to an erroneous uclamp.min setting for the new task if it was forked from RT.

Fix this by removing the unnecessary check on rt_policy() in uclamp_fork() as this doesn't make sense if the reset-on-fork flag is set.

---
 kernel/sched/core.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/sched/core.c +0 -4
index 3a61a3b8eaa9..9ea3e484eea2 100644
@@ -1234,10 +1234,6 @@ static void uclamp_fork(struct task_struct *p)
 	for_each_clamp_id(clamp_id) {
 		unsigned int clamp_value = uclamp_none(clamp_id);
 
-		/* By default, RT tasks always get 100% boost */
-		if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN))
-			clamp_value = uclamp_none(UCLAMP_MAX);
-
 		uclamp_se_set(&p->uclamp_req[clamp_id], clamp_value, false);
 	}
 }
Signature · 1 line
2.26.0.110.g2183baf09c-goog
#1 lore raw
Quoted 22 lines · Quentin Perret

uclamp_fork() resets the uclamp values to their default when the reset-on-fork flag is set. It also checks whether the task has a RT policy, and sets its uclamp.min to 1024 accordingly. However, during reset-on-fork, the task's policy is lowered to SCHED_NORMAL right after, hence leading to an erroneous uclamp.min setting for the new task if it was forked from RT.

Fix this by removing the unnecessary check on rt_policy() in uclamp_fork() as this doesn't make sense if the reset-on-fork flag is set.

---
 kernel/sched/core.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/sched/core.c +0 -0
index 3a61a3b8eaa9..9ea3e484eea2 100644

kernel/sched/core.c static void uclamp_fork(struct task_struct *p) · line 1240

Quoted 7 lines
kernel/sched/core.c +0 -4
@@ -1234,10 +1234,6 @@ static void uclamp_fork(struct task_struct *p)
 	for_each_clamp_id(clamp_id) {
 		unsigned int clamp_value = uclamp_none(clamp_id);
 
-		/* By default, RT tasks always get 100% boost */
-		if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN))
-			clamp_value = uclamp_none(UCLAMP_MAX);
-

Shouldn't this be conditional on p->sched_reset_on_fork instead of deleting the code?

thanks,

- Joel

Quoted 6 lines
kernel/sched/core.c +0 -0
 		uclamp_se_set(&p->uclamp_req[clamp_id], clamp_value, false);
 	}
 }
Signature · 1 line
2.26.0.110.g2183baf09c-goog
#2 lore raw
Quoted 2 lines · Joel Fernandes

Shouldn't this be conditional on p->sched_reset_on_fork instead of deleting the code?

Right, it's not obvious from the diff, but this code _is_ conditional on p->sched_reset_on_fork already. This is what the whole function looks like with my patch applied:

---8<---
static void uclamp_fork(struct task_struct *p)

{
        enum uclamp_id clamp_id;

        for_each_clamp_id(clamp_id)
                p->uclamp[clamp_id].active = false;

        if (likely(!p->sched_reset_on_fork))
                return;

        for_each_clamp_id(clamp_id) {
                unsigned int clamp_value = uclamp_none(clamp_id);

                uclamp_se_set(&p->uclamp_req[clamp_id], clamp_value, false);
        }
}

--->8---

Thanks,
Quentin

#3 lore raw
Quoted 10 lines · Quentin Perret
Quoted 2 lines · Joel Fernandes

Shouldn't this be conditional on p->sched_reset_on_fork instead of deleting the code?

Right, it's not obvious from the diff, but this code _is_ conditional on p->sched_reset_on_fork already. This is what the whole function looks like with my patch applied:

---8<---
static void uclamp_fork(struct task_struct *p)

Quoted 14 lines
{
        enum uclamp_id clamp_id;

        for_each_clamp_id(clamp_id)
                p->uclamp[clamp_id].active = false;

        if (likely(!p->sched_reset_on_fork))
                return;

        for_each_clamp_id(clamp_id) {
                unsigned int clamp_value = uclamp_none(clamp_id);

                uclamp_se_set(&p->uclamp_req[clamp_id], clamp_value, false);
        }

Oh ok, thanks for clarification.

- Joel

Quoted 5 lines

}
--->8---

Thanks,
Quentin

Reviewed-by on v1 1/1
#4 lore raw
In reply to Quentin Perret
Quoted 6 lines · Quentin Perret

uclamp_fork() resets the uclamp values to their default when the reset-on-fork flag is set. It also checks whether the task has a RT policy, and sets its uclamp.min to 1024 accordingly. However, during reset-on-fork, the task's policy is lowered to SCHED_NORMAL right after, hence leading to an erroneous uclamp.min setting for the new task if it was forked from RT.

Right, good catch. Thanks.

Quoted 6 lines

Fix this by removing the unnecessary check on rt_policy() in uclamp_fork() as this doesn't make sense if the reset-on-fork flag is set.

Fixes 1a00d999971c ("sched/uclamp: Set default clamps for RT tasks")
Quoted 22 lines
---
 kernel/sched/core.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/sched/core.c +0 -4
index 3a61a3b8eaa9..9ea3e484eea2 100644
@@ -1234,10 +1234,6 @@ static void uclamp_fork(struct task_struct *p)
 	for_each_clamp_id(clamp_id) {
 		unsigned int clamp_value = uclamp_none(clamp_id);
 
-		/* By default, RT tasks always get 100% boost */
-		if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN))
-			clamp_value = uclamp_none(UCLAMP_MAX);
-
 		uclamp_se_set(&p->uclamp_req[clamp_id], clamp_value, false);
 	}
 }
Signature · 1 line
2.26.0.110.g2183baf09c-goog
Signature · 3 lines
#include <best/regards.h>

Patrick Bellasi
#6 lore raw
In reply to Quentin Perret

Hi,

Quoted 23 lines · Quentin Perret

uclamp_fork() resets the uclamp values to their default when the reset-on-fork flag is set. It also checks whether the task has a RT policy, and sets its uclamp.min to 1024 accordingly. However, during reset-on-fork, the task's policy is lowered to SCHED_NORMAL right after, hence leading to an erroneous uclamp.min setting for the new task if it was forked from RT.

Fix this by removing the unnecessary check on rt_policy() in uclamp_fork() as this doesn't make sense if the reset-on-fork flag is set.

---
 kernel/sched/core.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/sched/core.c +0 -0
index 3a61a3b8eaa9..9ea3e484eea2 100644

kernel/sched/core.c static void uclamp_fork(struct task_struct *p) · line 1237

Quoted 8 lines
kernel/sched/core.c +0 -4
@@ -1234,10 +1234,6 @@ static void uclamp_fork(struct task_struct *p)
        for_each_clamp_id(clamp_id) {
                unsigned int clamp_value = uclamp_none(clamp_id);
 
-               /* By default, RT tasks always get 100% boost */
-               if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN))
-                       clamp_value = uclamp_none(UCLAMP_MAX);
-
                uclamp_se_set(&p->uclamp_req[clamp_id], clamp_value, false);

The local variable "clamp_value" doesn't have a lot of value anymore, does it? (Pun intended). Remove it?

-Doug

#7 lore raw

Hi Doug,

Quoted 29 lines · Doug Anderson

Hi,

Quoted 26 lines · Quentin Perret

uclamp_fork() resets the uclamp values to their default when the reset-on-fork flag is set. It also checks whether the task has a RT policy, and sets its uclamp.min to 1024 accordingly. However, during reset-on-fork, the task's policy is lowered to SCHED_NORMAL right after, hence leading to an erroneous uclamp.min setting for the new task if it was forked from RT.

Fix this by removing the unnecessary check on rt_policy() in uclamp_fork() as this doesn't make sense if the reset-on-fork flag is set.

---
 kernel/sched/core.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/sched/core.c +0 -0
index 3a61a3b8eaa9..9ea3e484eea2 100644
@@ -1234,10 +1234,6 @@ static void uclamp_fork(struct task_struct *p)
        for_each_clamp_id(clamp_id) {
                unsigned int clamp_value = uclamp_none(clamp_id);
 
Quoted 8 lines

kernel/sched/core.c static void uclamp_fork(struct task_struct *p) · line 1237

Quoted 5 lines
kernel/sched/core.c +0 -4
-               /* By default, RT tasks always get 100% boost */
-               if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN))
-                       clamp_value = uclamp_none(UCLAMP_MAX);
-
                uclamp_se_set(&p->uclamp_req[clamp_id], clamp_value, false);

The local variable "clamp_value" doesn't have a lot of value anymore, does it? (Pun intended).

:)

Quoted 1 line

Remove it?

Right, but I figured the generated code should be similar, and 'uclamp_se_set(&p->uclamp_req[clamp_id], uclamp_none(clamp_id), false);' doesn't fit in 80 cols at this identation level, so I kept the local var. No strong opinion, though.

Thanks,
Quentin

#8 lore raw

Hi,

Quoted 42 lines · Quentin Perret

Hi Doug,

Quoted 29 lines · Doug Anderson

Hi,

Quoted 26 lines · Quentin Perret

uclamp_fork() resets the uclamp values to their default when the reset-on-fork flag is set. It also checks whether the task has a RT policy, and sets its uclamp.min to 1024 accordingly. However, during reset-on-fork, the task's policy is lowered to SCHED_NORMAL right after, hence leading to an erroneous uclamp.min setting for the new task if it was forked from RT.

Fix this by removing the unnecessary check on rt_policy() in uclamp_fork() as this doesn't make sense if the reset-on-fork flag is set.

---
 kernel/sched/core.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/sched/core.c +0 -0
index 3a61a3b8eaa9..9ea3e484eea2 100644
@@ -1234,10 +1234,6 @@ static void uclamp_fork(struct task_struct *p)
        for_each_clamp_id(clamp_id) {
                unsigned int clamp_value = uclamp_none(clamp_id);
 
Quoted 8 lines

kernel/sched/core.c static void uclamp_fork(struct task_struct *p) · line 1237

Quoted 5 lines
kernel/sched/core.c +0 -4
-               /* By default, RT tasks always get 100% boost */
-               if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN))
-                       clamp_value = uclamp_none(UCLAMP_MAX);
-
                uclamp_se_set(&p->uclamp_req[clamp_id], clamp_value, false);

The local variable "clamp_value" doesn't have a lot of value anymore, does it? (Pun intended).

Quoted 8 lines

:)

Quoted 1 line

Remove it?

Right, but I figured the generated code should be similar, and 'uclamp_se_set(&p->uclamp_req[clamp_id], uclamp_none(clamp_id), false);' doesn't fit in 80 cols at this identation level, so I kept the local var. No strong opinion, though.

OK. It's definitely a bikeshed color issue and since you'll spend more time in this bikeshed than I will I'll leave it to you to pick the color.

I'm not at all an expert on this code but it sure looks sane to me. If you think my review tag is worth anything in this context feel free to add it, but since you already have Patrick's mine probably adds very little value.

-Doug

#9 lore raw
Quoted 8 lines · Doug Anderson

Hi,

Quoted 5 lines · Quentin Perret

Hi Doug,

Quoted 2 lines · Doug Anderson

Hi,
On Tue, Apr 14, 2020 at 9:13 AM Quentin Perret <[email protected]> wrote:

[...]

Quoted 26 lines
Quoted 17 lines
Quoted 8 lines
Quoted 8 lines
kernel/sched/core.c +0 -0
index 3a61a3b8eaa9..9ea3e484eea2 100644
@@ -1234,10 +1234,6 @@ static void uclamp_fork(struct task_struct *p)
        for_each_clamp_id(clamp_id) {
                unsigned int clamp_value = uclamp_none(clamp_id);
 
Quoted 8 lines

kernel/sched/core.c static void uclamp_fork(struct task_struct *p) · line 1237

Quoted 5 lines
kernel/sched/core.c +0 -4
-               /* By default, RT tasks always get 100% boost */
-               if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN))
-                       clamp_value = uclamp_none(UCLAMP_MAX);
-
                uclamp_se_set(&p->uclamp_req[clamp_id], clamp_value, false);

The local variable "clamp_value" doesn't have a lot of value anymore, does it? (Pun intended).

Quoted 8 lines

:)

Quoted 1 line

Remove it?

Right, but I figured the generated code should be similar, and 'uclamp_se_set(&p->uclamp_req[clamp_id], uclamp_none(clamp_id), false);' doesn't fit in 80 cols at this identation level, so I kept the local var. No strong opinion, though.

Quoted 8 lines

OK. It's definitely a bikeshed color issue and since you'll spend more time in this bikeshed than I will I'll leave it to you to pick the color.

I'm not at all an expert on this code but it sure looks sane to me. If you think my review tag is worth anything in this context feel free to add it, but since you already have Patrick's mine probably adds very little value.

Honestly I was almost there to ask for the same cleanup. :)

If you get rid of the useless variable we can go back to the exact same code modified by the commit we are fixing, i.e. 1a00d999971c.

So, not strong opinions here too, but if it's not a big problem I would post a cleaned up v2.

Best,
Patrick

Signature · 3 lines
#include <best/regards.h>

Patrick Bellasi
#10 lore raw
Quoted 7 lines · Patrick Bellasi

Honestly I was almost there to ask for the same cleanup. :)

If you get rid of the useless variable we can go back to the exact same code modified by the commit we are fixing, i.e. 1a00d999971c.

So, not strong opinions here too, but if it's not a big problem I would post a cleaned up v2.

Alright, that works for me. I'll send a v2 shortly.

Thanks!
Quentin