pm: Optimize power state selection with early exit when no state is available #88149
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
requested review from
bjarki-andreasen,
ceolin,
JordanYates,
nashif,
teburd and
tmleman
nordic-krch
force-pushed
the
pm_policy_opt
branch
2 times, most recently
from
44c07a4 to
0ad194c
Compare
When all states are locked or latency requirement cannot be met by any power state it is important to be able to quickly exit suspend procedure because that usually means that application requires high performance. Add function for detecting if any power state is available. Additionally, add function pm_policy_state_is_available for checking if given state is available which means that it is not locked and fulfills current latency requirement. Signed-off-by: Krzysztof Chruściński <[email protected]>
Update power state selection. Previously, it was iterating over states starting from the last one so the most common short sleep periods were taking the longest time to select. Order is now swapped so that short sleeps will get power state as quick as possible. Signed-off-by: Krzysztof Chruściński <[email protected]>
nordic-krch
force-pushed
the
pm_policy_opt
branch
from
0ad194c to
084b967
Compare
|
@bjarki-andreasen can you reapprove? I had done minor fixes. |
|
@ceolin @JordanYates can you take a look? |
bjarki-andreasen
approved these changes
Apr 10, 2025
There was a problem hiding this comment.
Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.
Comments suppressed due to low confidence (2)
subsys/pm/policy/policy_state_lock.c:76
- [nitpick] The error message 'Unbalanced state lock get/put' could include additional context (such as the affected state and substate) to aid debugging when an imbalance occurs.
__ASSERT(lock_cnt[i] > 0, "Unbalanced state lock get/put");
subsys/pm/policy/policy_default.c:38
- [nitpick] The early exit using break may result in returning a NULL state if no valid candidate has been selected before the break is hit; confirm that this behavior is intended for cases when the first state fails the min residency check.
if (ticks < min_residency_ticks) {
kartben
reviewed
Apr 18, 2025
Comment on lines
+152
to
+163
| bool pm_policy_state_is_available(enum pm_state state, uint8_t substate_id); | ||
|
|
||
| /** | ||
| * @brief Check if any power state can be used. | ||
| * | ||
| * Function allows to quickly check if any power state is available and exit | ||
| * suspend operation early. | ||
| * | ||
| * @retval true if any power state is active. | ||
| * @retval false if all power states are unavailable. | ||
| */ | ||
| bool pm_policy_state_any_active(void); |
There was a problem hiding this comment.
please add these to release notes
ceolin
approved these changes
Apr 20, 2025
tmleman
reviewed
Apr 24, 2025
Comment on lines
+151
to
+155
| if (!pm_policy_state_any_active()) { | ||
| /* Return early if all states are unavailable. */ | ||
| return false; | ||
| } | ||
|
|
There was a problem hiding this comment.
This if breaks the functionality provided by the pm_state_force function.
https://docs.zephyrproject.org/latest/doxygen/html/group__subsys__pm__sys.html#ga075be307983f4efdcc93252a31a4258a
/**
* @brief Force usage of given power state.
*
* This function overrides decision made by PM policy forcing
* usage of given power state upon next entry of the idle thread.
*
* @note This function can only run in thread context
*
* @param cpu CPU index.
* @param info Power state which should be used in the ongoing
* suspend operation.
*/
bool pm_state_force(uint8_t cpu, const struct pm_state_info *info);My interpretation is that this function should result in overriding restrictions such as pm policy lock or max latency.
There was a problem hiding this comment.
I've opened #89037 to fix that.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Issues found in power management:
Lack of early exit when there are no power states available
When low latency requirement is set (e.g. 0us) then
pm_suspenddoes not perform faster as it still iterates over all power modes only to find that there is no power mode that fulfills the requirement. When all power modes are lockedpm_suspendbehavior is the same as when latency requirement is set. In both cases user sets those to get higher performance so ideally,pm_suspendshould be limited.Added
pm_policy_state_any_activewhich is using bitmasks to determine if any state is available. It is used early in thepm_suspendand allows early exit. Addedpm_policy_state_is_availablewhich checks if state is locked or latency requirement is not met.Default algorithm for picking power state iterates always from the last power state
Algorithm for picking power state iterates always from the last power state (the longest sleep) and because of that
pm_suspendis the longest for the shortest (most common) sleep periods.Order is reverted.
Performance results
Timings were measured on nrf54h20 with 2 power states. Code includes other proposed improvements: #88144 and #88146. It measures the time between entering
pm_suspendand selecting the power state or early exit. Results show significant improvement for case where no state is available and 33% improvement for shortest sleep.Results show that high performance case is significantly improved.
Short sleep times is slightly improved.