Skip to contentSkip to conversation

sched/fair: Fix frequency selection for non-invariant case

PATCH In review Wentao Guan 2 participants 6 messages Aug 15, 2025 Stable branch

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sched/fair: Fix frequency selection for non-invariant case

Hi,

    Please apply the commit e37617c8e53a1f7fcba6d5e1041f4fd8a2425c27,

it fix the REGRESSION by ada8d7fa0ad49a2a078f97f7f6e02d24d3c357a3 ("sched/cpufreq: Rework schedutil governor performance estimation") which introduced in v6.6.89.

Best Regards
Wentao Guan

maintainer
#1 lore raw
Quoted 5 lines · Wentao Guan

Hi,

    Please apply the commit e37617c8e53a1f7fcba6d5e1041f4fd8a2425c27,

it fix the REGRESSION by ada8d7fa0ad49a2a078f97f7f6e02d24d3c357a3 ("sched/cpufreq: Rework schedutil governor performance estimation") which introduced in v6.6.89.

Please provide a working, and tested, backport of that commit and we will be glad to queue it up.

thanks,

greg k-h

author Tested-by Wyes Karny
#2 lore raw

From ddd3f1ed6e88d88502d2b42e159ac976b4194b78 Mon Sep 17 00:00:00 2001
From: Vincent Guittot <[email protected]>
Date: Sun, 14 Jan 2024 19:36:00 +0100
Subject: [PATCH] sched/fair: Fix frequency selection for non-invariant case

Linus reported a ~50% performance regression on single-threaded workloads on his AMD Ryzen system, and bisected it to:

9c0b4bb7f630 ("sched/cpufreq: Rework schedutil governor performance estimation")

When frequency invariance is not enabled, get_capacity_ref_freq(policy) is supposed to return the current frequency and the performance margin applied by map_util_perf(), enabling the utilization to go above the maximum compute capacity and to select a higher frequency than the current one.

After the changes in 9c0b4bb7f630, the performance margin was applied earlier in the path to take into account utilization clampings and we couldn't get a utilization higher than the maximum compute capacity, and the CPU remained 'stuck' at lower frequencies.

To fix this, we must use a frequency above the current frequency to get a chance to select a higher OPP when the current one becomes fully used. Apply the same margin and return a frequency 25% higher than the current one in order to switch to the next OPP before we fully use the CPU at the current one.

[ mingo: Clarified the changelog. ]

Fixes 9c0b4bb7f630 ("sched/cpufreq: Rework schedutil governor performance estimation")
(cherry picked from commit e37617c8e53a1f7fcba6d5e1041f4fd8a2425c27)

---
I tested the patch two days ago, and the upstream commit can be directly apply.
---

---
 kernel/sched/cpufreq_schedutil.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
kernel/sched/cpufreq_schedutil.c +5 -1
index cfe7c625d2ad6..819ec1ccc08cf 100644
@@ -156,7 +156,11 @@ unsigned long get_capacity_ref_freq(struct cpufreq_policy *policy)
 	if (arch_scale_freq_invariant())
 		return policy->cpuinfo.max_freq;
 
-	return policy->cur;
+	/*
+	 * Apply a 25% margin so that we select a higher frequency than
+	 * the current one before the CPU is fully busy:
+	 */
+	return policy->cur + (policy->cur >> 2);
 }
 
 /**
maintainer
#3 lore raw
Quoted 52 lines · Wentao Guan

From ddd3f1ed6e88d88502d2b42e159ac976b4194b78 Mon Sep 17 00:00:00 2001
From: Vincent Guittot <[email protected]>
Date: Sun, 14 Jan 2024 19:36:00 +0100
Subject: [PATCH] sched/fair: Fix frequency selection for non-invariant case

Linus reported a ~50% performance regression on single-threaded workloads on his AMD Ryzen system, and bisected it to:

9c0b4bb7f630 ("sched/cpufreq: Rework schedutil governor performance estimation")

When frequency invariance is not enabled, get_capacity_ref_freq(policy) is supposed to return the current frequency and the performance margin applied by map_util_perf(), enabling the utilization to go above the maximum compute capacity and to select a higher frequency than the current one.

After the changes in 9c0b4bb7f630, the performance margin was applied earlier in the path to take into account utilization clampings and we couldn't get a utilization higher than the maximum compute capacity, and the CPU remained 'stuck' at lower frequencies.

To fix this, we must use a frequency above the current frequency to get a chance to select a higher OPP when the current one becomes fully used. Apply the same margin and return a frequency 25% higher than the current one in order to switch to the next OPP before we fully use the CPU at the current one.

[ mingo: Clarified the changelog. ]

Fixes 9c0b4bb7f630 ("sched/cpufreq: Rework schedutil governor performance estimation")
(cherry picked from commit e37617c8e53a1f7fcba6d5e1041f4fd8a2425c27)

---
I tested the patch two days ago, and the upstream commit can be directly apply.
---

---
 kernel/sched/cpufreq_schedutil.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
kernel/sched/cpufreq_schedutil.c +0 -0
index cfe7c625d2ad6..819ec1ccc08cf 100644
@@ -156,7 +156,11 @@ unsigned long get_capacity_ref_freq(struct cpufreq_policy *policy)
 	if (arch_scale_freq_invariant())
 		return policy->cpuinfo.max_freq;

kernel/sched/cpufreq_schedutil.c unsigned long get_capacity_ref_freq(struct cpufreq_policy *policy) · line 166

Quoted 10 lines
kernel/sched/cpufreq_schedutil.c +5 -1
 
-	return policy->cur;
+	/*
+	 * Apply a 25% margin so that we select a higher frequency than
+	 * the current one before the CPU is fully busy:
+	 */
+	return policy->cur + (policy->cur >> 2);
 }
 
 /**

Does not apply against the latest 6.6.y kernel release :(

thanks,

greg k-h

#4 lore raw

Dear Greg,

    Oh, I miss the thing which first apply b3edde44e5d4504c23a176819865cd603fd16d6c

("cpufreq/schedutil: Use a fixed reference frequency"), then apply e37617c8e53a1f7fcba6d5e1041f4fd8a2425c27
("sched/fair: Fix frequency selection for non-invariant case").

sorry...

Best Regards
Wentao Guan.

#5 lore raw

Dear Greg,

Oops, I missed the whole things, it will build failed, i will send the full patches series later.

Best Regards
Wentao Guan