soc: intel_adsp/ace: allows more spin relax loop per CPU #96989
+29
−1
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
1 task
|
Compliance check fails due to have spaces on second line of a |
zephyrbot
added
area: Boards/SoCs
area: Tests
zephyrbot
requested review from
TaiJuWu,
andyross,
ceolin,
cfriedt,
kv2019i,
lgirdwood,
lyakh,
marcinszkudlinski,
nashif,
npitre,
peter-mitsis,
softwarecki and
tmleman
lyakh
reviewed
Oct 6, 2025
| register uint32_t remaining = CONFIG_SOC_SERIES_INTEL_ADSP_ACE_NUM_SPIN_RELAX_NOPS; | ||
|
|
||
| #if defined(CONFIG_SOC_SERIES_INTEL_ADSP_ACE_NUM_SPIN_RELAX_NOPS_ADD_CPU_ID) | ||
| remaining += arch_proc_id(); |
There was a problem hiding this comment.
without going too deeply into this - is this enough? The variable is decremented by 4, so maybe arch_proc_id() * 4?
There was a problem hiding this comment.
I thought about it, though I feel like always adding 20 NOPs on CPU #5 seems too much.
nashif
previously approved these changes
Oct 7, 2025
peter-mitsis
previously approved these changes
Oct 7, 2025
This allows adding the CPU ID to the number of NOPs in the custom arch_spin_relax(). With the same number of NOPs for all CPUs, it is possible to have them all doing RCW transactions at the same time over and over again if they enter and exit the spin relax loop at the same time. This behavior has been observed when doing lots of context switching, like in the SMP switching stress test. So adds a new kconfig to fine tune the relax loop behavior if needed. The new kconfig allows adding the CPU ID to the number of NOPs which will add some minimal offsetting to workaround the above mentioned situation. Signed-off-by: Daniel Leung <[email protected]>
It has been observed that, during the switching stress test, the context switching becomes very slow due to enough CPUs doing RCW transaction on hardware bus (e.g. spin locks). Since the number of NOPs are the same for all CPUs, they are simply entering and exiting the relax loop at the same time, and hitting the bus with RCW transactions at the same time. Not exactly a deadlock but it slows down the execution enough to result in the test timing out. The SoC layer has added an option to offset the number of NOPs per CPU by adding the CPU to the number of NOPs. So enabling it for the Intel ADSP boards to workaround the above mentioned issue. I haven't encountered another slowdown after turning on this option. So hopefully this lowers the probability of that happening such that a simple retry can pass the test. Signed-off-by: Daniel Leung <[email protected]>
dcpleung
force-pushed
the
intel_adsp/more_relax_nops
branch
from
4bafba1 to
bcc3c19
Compare
|
Just to make checkpatch happy. |
|
peter-mitsis
approved these changes
Oct 9, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
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.
This allows adding the CPU ID to the number of NOPs in the custom arch_spin_relax(). With the same number of NOPs for all CPUs, it is possible to have them all doing RCW transactions at the same time over and over again if they enter and exit the spin relax loop at the same time. This behavior has been observed when doing lots of context switching, like in the SMP switching stress test. So adds a new kconfig to fine tune the relax loop behavior if needed. The new kconfig allows adding the CPU ID to the number of NOPs which will add some minimal offsetting to workaround the above mentioned situation.