Skip to contentSkip to conversation

cpufreq/schedutil: When updating limitations, frequency modulation interval not become invalid.

PATCH In review Lizhe 2 participants 3 messages Feb 4, 2024 – Feb 5, 2024 1 message not in the archive CPU frequency scaling framework

lore
Share & tools
lore

Lizhe

author
#0 lore raw

PATCH 1/1 cpufreq/schedutil: When updating limitations, frequency modulation interval not become invalid.

If the current frequency scaling policy is schedutil.
echo schedutil > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor This would result in an invalid frequency modulation interval. In sugov_limit(), sg_policy->limits_changed is set to true.

---
 drivers/cpufreq/cpufreq.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/cpufreq/cpufreq.c +1 -1
index 44db4f59c4cc..a0af38fcb7e2 100644
@@ -2631,7 +2631,7 @@ static int cpufreq_set_policy(struct cpufreq_policy *policy,
 
 	if (new_gov == policy->governor) {
 		pr_debug("governor limits update\n");
-		cpufreq_governor_limits(policy);
+		cpufreq_policy_apply_limits(policy);
 		return 0;
 	}
 
Signature · 1 line
2.25.1
maintainer
#1 lore raw
Quoted 4 lines · Lizhe

If the current frequency scaling policy is schedutil.
echo schedutil > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor This would result in an invalid frequency modulation interval. In sugov_limit(), sg_policy->limits_changed is set to true.

That will only make us do an extra freq change. What's the problem with that ?

Quoted 17 lines
---
 drivers/cpufreq/cpufreq.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/cpufreq/cpufreq.c +1 -1
index 44db4f59c4cc..a0af38fcb7e2 100644
@@ -2631,7 +2631,7 @@ static int cpufreq_set_policy(struct cpufreq_policy *policy,
 
 	if (new_gov == policy->governor) {
 		pr_debug("governor limits update\n");
-		cpufreq_governor_limits(policy);
+		cpufreq_policy_apply_limits(policy);
 		return 0;
 	}
Signature · 1 line
viresh

A message that is not in the archive (sent off-list or not yet received): [email protected]

maintainer
#3 lore raw
Quoted 18 lines · lizhe

HI:
I think not.
It will cause the following changes:
limits_changed in sugov_limit() is set to true. This will cause the modulation interval to be invalid. The system modulation will ignore the modulation interval. A modulation request will be initiated immediately.

   As shown below : 
   sugov_should_update_freq(...)
  {
     if (unlikely(sg_policy->limits_changed)) {
          ....
          return true;
     }
  }
Quoted 2 lines

Repeatedly setting the schedutil modulation policy causes the modulation interval to be invalid. Why is this necessary ?

This is the penalty being paid to keep the interface simple and consistent. Playing with sysfs can do such minor things.

Signature · 1 line
viresh