Ayn Odin 2 mini - No CPU clusters found

[dexx0008]

Ayn Odin 2 mini - No CPU cluster found

Is there a way to add support for Odin 2 mini and/or provide information to improve support for this device?

[AurelioB]

That's the first Ayn device I've seen that behaves differently. It's weird that it passed the first check (PServer being available) only to fail at detecting clusters. That may be due to PServer not forwarding stout to the caller. I don't own that device so I'm very limited on what I can do myself but if your device is rooted this is what anyone interested could do:

Check if you have a "run root script" or something similar in settings

Run the following scripts from a linux PC with adb and the device connected, and paste the output here:

• scripts/detect-soc.sh
• scripts/report-cpufreq.sh

[ElHereward]

@AurelioB Can you maybe save the printf output inside the script so you dont need the adb or root?

[ElHereward]

report-cpufreq.txt
detect-soc.txt
i used logcat from host - i hope the files are helpfull

[iloveswiming909]

@AurelioB I'm hitting the same "no CPU clusters found" symptom on my
Odin 2 Mini and I think I have data that confirms your stdout-forwarding
hypothesis. I have limited coding/debugging expierience so the follow is just a copy and paste from some tests i ran using Claude opus 4.7 and my (unrooted) odin 2 mini:

---

clustertune-detect-utf8.log

I built a debug version with Log.d lines added at each detection
decision point and adb logcat -s ClusterTune shows this on every
detection cycle:

D ClusterTune: Unprivileged listing of /sys/devices/system/cpu/cpufreq -> 3 entries: [/sys/devices/system/cpu/cpufreq/policy0, /sys/devices/system/cpu/cpufreq/policy7, /sys/devices/system/cpu/cpufreq/policy3]
D ClusterTune: policy0 reads: cpuIds=[0], scalingMax=null, cpuinfoMax=null, minFreq=0, available=0 steps, timeInStateMax=null
D ClusterTune: policy0 rejected: no usable max frequency from any source
D ClusterTune: policy3 reads: cpuIds=[3], scalingMax=null, cpuinfoMax=null, minFreq=0, available=0 steps, timeInStateMax=null
D ClusterTune: policy3 rejected: no usable max frequency from any source
D ClusterTune: policy7 reads: cpuIds=[7], scalingMax=null, cpuinfoMax=null, minFreq=0, available=0 steps, timeInStateMax=null
D ClusterTune: policy7 rejected: no usable max frequency from any source
D ClusterTune: detectPolicies finished: 0 policies parsed (ids=[])

So:
File.listFiles() on /sys/devices/system/cpu/cpufreq works fine on
the Mini and returns the expected 3-cluster Snapdragon 8 Gen 2
layout (policy0/3/7).

Every PServer-mediated read inside each policy directory comes back
null. affected_cpus falling back to [id] is consistent with that
read returning null too. scaling_max_freq and cpuinfo_max_freq
are not files that can legitimately be empty on a working SoC, so
PServer is silently dropping the response, not the kernel.

The pservice service registers as PServerBinder (RootExec finds it,
pServerAvailable is true) and the transactions don't throw — they
return success with an empty/null reply.

So the issue is in RootExec.executeAsRoot / how PServer on the Mini
expects the binder transaction to be shaped. Three things in there
look like potential mismatch points to me:
Transaction code 0. binder.transact(0, data, reply, 0) —
wondered if the Mini's pservice expects a different code. A wrong
code on a binder usually returns success with an empty reply,
which matches what we see.

The "1" flag in data.writeStringArray(arrayOf(cmd, "1")).
Looks like a "want stdout" toggle — if the Mini's pservice expects
a different value or doesn't recognise this signalling, the command
would still execute but stdout would be discarded.

reply.createByteArray() in decodeReply. If the Mini's
pservice writes the reply with writeString or any non-byte-array
format, createByteArray returns null and we get exactly this
symptom.

Without disassembling the Mini's /system/bin/pservice I can't tell
which of those three (or what combination) it actually wants. Do you
remember how you reverse-engineered the protocol on the Odin 3 — was
there an APK whose source you could refer to, or did you trace a live
process? If so, the same approach on the Mini would probably resolve
this quickly.

I'm happy to run further diagnostic builds on my device — e.g. a
version that probes alternative transaction codes, or one that dumps
the raw reply Parcel as hex for you to inspect. Just let me know what
would help.
For reference, my debug build with the logging is at https://github.com/iloveswiming909/cluster-tune, built off your v0.3.0 source.

[iloveswiming909]

@dexx0008 @ElHereward , I have made a workaround that uses run script to change clock speeds. Feel free to try it out and give feedback

[ElHereward]

@iloveswiming909 thx man - i will test it later and will report

[dexx0008]

same, I will test sometime today- appreciate the work!

[iloveswiming909]

Any luck @dexx0008 @ElHereward ?

[ElHereward]

@iloveswiming909 Yes, I tested it on my Odin2 Mini and it worked sometimes. Often, however, it didn't—but unfortunately, I can't tell you why. I suspect it has something to do with the CPU frequencies. It seems to me that it doesn't accept certain combinations of CPU clusters. So I tried using the same frequencies as on my Thor, but it wasn’t always successful. Maybe the power profile or fan profile plays a role here too. I haven’t been able to check that yet. Did you have any issues with that, or did it work for you right from the start? What frequencies are you using?
I was also thinking about whether we could just take the scripts and run them manually as root, and then just keep the scripts in Git?
Regardless of the testing, I wanted to thank you very much for your work and effort

[Zhonecage]

For what it's worth, there is an app called RootlessJamesDSP. It improves audio on AYN / Retroid devices using Snapdragon 8 Gen 2. It works on all AYN and Retroid devices with that chipset except the AYN Odin 2 Mini. Maybe the issue with Clustertune is related. There is definitely something strange going on with the Odin 2 Mini.

[dexx0008]

I had intermittent luck as well. One thing I did just run across this week is this post where someone has some scripts that apparently work for odin2mini that accomplish what clustertune does; and apparently works with the mini.

"On the AYN Discord, a fellow user tested it on their Odin 2 Mini and it worked perfectly. They even shared videos and reported good temperatures using the Balanced and Balanced Full GPU scripts." Translated from the comment.
threads:
https://www.reddit.com/r/OdinHandheld/comments/1ucnd3n/clustertune_on_odin_2_mini/
https://www.reddit.com/r/OdinHandheld/comments/1sylay3/comment/ojlsluc/

I haven't looked at these scripts yet, been a bit busy, but perhaps it sheds some more light to the odin 2 mini's differences and could be implemented here.