linux-kernel - Re: [PATCH v1 0/5] cpuidle: governors: teo: Wakeup events classification change and some refinements

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <0532f98b-1f33-418b-ae94-d9bb57fb259a@arm.com>
Date: Fri, 16 Jan 2026 11:52:49 +0000
From: Christian Loehle <christian.loehle@....com>
To: "Rafael J. Wysocki" <rafael@...nel.org>,
 Linux PM <linux-pm@...r.kernel.org>
Cc: LKML <linux-kernel@...r.kernel.org>, Doug Smythies <dsmythies@...us.net>
Subject: Re: [PATCH v1 0/5] cpuidle: governors: teo: Wakeup events
 classification change and some refinements

On 1/14/26 19:42, Rafael J. Wysocki wrote:
> Hi All,
> 
> This material has been in my local queue for almost a full development cycle,
> so time to post it.
> 
> The motivation for the changes in this series is mostly theoretical, but I do
> see some idle power improvements from patch [4/5], for example, but nothing
> specifically worth reporting.
> 
> The first patch simply prevents idle states with zero-size bins from being
> selected sometimes when teo_select() runs with stopped tick.
> 
> Patch [2/5] avoids counting tick wakeups as intercepts unless there are
> sufficiently many intercepts within the tick period range to assume that
> the tick wakeup may have clobbered a genuine intercept.
> 
> Patch [3/5] simply updates a coefficient in one of the inequalities to be
> somewhat easier to interpret (this should be a cosmetic change).
> 
> Patch [4/5] changes the criteria used for classifying wakeup events as hits
> or intercepts to (hopefully) make the classification work better for large
> state bins.
> 
> Patch [5/5] refines the idle state lookup based on intercepts to first
> consider the state with the maximum intercepts metric, so that state is
> always taken into consideration.
> 
> Please see the individual patch changelogs for details.
> 
> Thanks!
> 
> 
> 

Hi Rafael,
I'll do the in-depth review, but have run some tests already.
They are attached, platform is the usual rk3399.
"teo" is mainline, "teo-$i" is with patches 1..$i applied.

There's a regression on teo-4 visible on the intercept heavy IO workloads,
for idle misses that isn't strong enough to reflect in score changes except
for the very slow mtdblock device.
interestingly though there also seems to be a regression in
mapper/dm-slow (dm device with 51ms delay on each IO), which is not
intercept heavy.
Looking at the state residencies it overuses the deepest state2 in
where state1 was preferred for the other teo variants.
I've attached that too for reference.
I'm assuming that is because of the new intercept-logic-exclusion clause.
teo-5 seems to be slightly better than teo-4 here, but still a regression
from the others.

Regards,
Christian
View attachment "teo-6.19-patches-rafael-wakeup-classification-dm-slow-residencies.txt" of type "text/plain" (9781 bytes)

View attachment "teo-6.19-patches-rafael-wakeup-classification-results.txt" of type "text/plain" (44962 bytes)