Re: [PATCH v1 1/5] cpuidle: governors: teo: Avoid selecting states with zero-size bins

[Christian Loehle <christian.loehle@arm.com>]

On 1/23/26 20:46, Rafael J. Wysocki wrote:
> On Wed, Jan 21, 2026 at 2:10 PM Christian Loehle
> <christian.loehle@arm.com> wrote:
>>
>> On 1/14/26 19:44, Rafael J. Wysocki wrote:
>>> From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
>>>
>>> If the last two enabled idle states have the same target residency which
>>> is at least equal to TICK_NSET, teo may select the next-to-last one even
>>
>> s/TICK_NSET/TICK_NSEC
> 
> Yup, thanks!
> 
>>> though the size of that state's bin is 0, which is confusing.
>>>
>>> Prevent that from happening by adding a target residency check to the
>>> relevant code path.
>>>
>>> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
>>> ---
>>>  drivers/cpuidle/governors/teo.c |   10 ++++++++++
>>>  1 file changed, 10 insertions(+)
>>>
>>> --- a/drivers/cpuidle/governors/teo.c
>>> +++ b/drivers/cpuidle/governors/teo.c
>>> @@ -388,6 +388,15 @@ static int teo_select(struct cpuidle_dri
>>>                       while (min_idx < idx &&
>>>                              drv->states[min_idx].target_residency_ns < TICK_NSEC)
>>>                               min_idx++;
>>> +
>>> +                     /*
>>> +                      * Avoid selecting a state with a lower index, but with
>>> +                      * the same target residency as the current candidate
>>> +                      * one.
>>> +                      */
>>> +                     if (drv->states[min_idx].target_residency_ns ==
>>> +                                     drv->states[idx].target_residency_ns)
>>
>> We need to check that min_idx isn't disabled though, otherwise we now skip a
>> potential (enabled) idx==1 if min_idx==2 and min_idx is disabled.
> 
> Not really because idx is the current candidate state and it is
> enabled.  We'll use idx if this check is true, not min_idx.
> 

Are you sure?
I meant initially:

		for (i = intercept_max_idx; i >= min_idx; i--) {
			intercept_sum += cpu_data->state_bins[i].intercepts;

			if (dev->states_usage[i].disable)
				continue;

			idx = i;
			if (2 * intercept_sum > idx_intercept_sum)
				break;
		}
might skip an idx==3 if it enters with min_idx==2 (sorry, messed up the +-1 in the initial mail)
even though idx==3 might have the same residency as idx==2.
So if idx==2 is disabled we could've selected idx==3, but now won't and will go for idx==1 or
whatever is the next shallower enabled state.

Additionally an issue with this and patch 5/5:

		if (min_idx >= intercept_max_idx) {
			idx = min_idx;
			goto constraint; // CL: this will just select min_idx
		}

will use min_idx even though it might be disabled and also the scenario

I think we should just add something like

------8<-------

cpuidle: teo: Fix intercept-logic selecting disabled

Prevent min_idx to be set to a disabled state, which could lead to
both a disabled state being returned by teo, but also an equally good
state being skipped because it has a higher index than a disabled state.

Signed-off-by: Christian Loehle <christian.loehle@arm.com>
---
 drivers/cpuidle/governors/teo.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/drivers/cpuidle/governors/teo.c b/drivers/cpuidle/governors/teo.c
index 4cf6302f99ad..94c5ef5df467 100644
--- a/drivers/cpuidle/governors/teo.c
+++ b/drivers/cpuidle/governors/teo.c
@@ -420,9 +420,11 @@ static int teo_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
 			 * candidate one whose target residency is at least
 			 * equal to the tick period length.
 			 */
-			while (min_idx < idx &&
-			       drv->states[min_idx].target_residency_ns < TICK_NSEC)
-				min_idx++;
+			while (i < idx && drv->states[i].target_residency_ns < TICK_NSEC) {
+				i++;
+				if (!dev->states_usage[i].disable)
+					min_idx = i;
+			}
 
 			/*
 			 * Avoid selecting a state with a lower index, but with
-- 
2.34.1