Re: [REGRESSION] cpuidle: psci: boot stall/hang on i.MX8MP
From: Dario Binacchi
Date: Wed Nov 11 2026 - 02:21:04 EST
Hi Thorsten,
On Tue, Nov 10, 2026 at 10:12 AM Thorsten Leemhuis
<regressions@xxxxxxxxxxxxx> wrote:
>
> On 10/30/26 18:19, Dario Binacchi wrote:
> > Hi Ulf,
> >
> > on an i.MX8M Plus Dual board (Engicam i.Core MX8M Plus), mainline
> > c339f4010dea sometimes stalls for ~0.9 s at boot, or hangs until the
> > watchdog resets the board, right after:
> >
> > [ 0.806123] ledtrig-cpu: registered to indicate activity on CPUs
> >
> > On a good boot the next message comes ~11 ms later.
> >
> > About 1 boot out of 3 stalls or hangs. With cpuidle.off=1, or
> > with commit 65705b18162b ("cpuidle: psci: Move initialization a bit
> > earlier in the boot sequence") reverted, 10 out of 10 boots are fine.
>
> Hmmm, nothing happened here. Might have fallen through the cracks, so
> maybe this mail will get thing rolling. If not, I'd suggest you submit a
> revert.
>
> Or was there progress or even a fix and I just missed it?
Sorry for not following up. I can't reproduce it anymore with
current mainline (b3dfbde62cf1, v7.4-rc2+) on my current setup, and
I'm not sure what changed. I'll keep an eye on it in my next tests
and report back if it shows up again.
#regzbot inconclusive: can't reproduce anymore
Grazie e ciao,
Dario
>
> Ciao, Thorsten
>
> > I haven't dug into it much. The stall seems to be in
> > pl330_driver_init(), so my guess is that it is related to the AMBA bus
> > reading the ID registers of the CoreSight ETMs while a CPU is already
> > in cpu_pd_wait, but I may well be wrong.
> >
> > Any idea?
> >
> > #regzbot introduced: 65705b18162b
> >
> > Thanks,
> > Dario
>
--
Dario Binacchi
Senior Embedded Software Engineer
M. +39 328 0625246
dario.binacchi@xxxxxxxxxxxxxxxxxxxx
―――――――――――――――
Amarula Solutions SRL
Via Felice Cavallotti 25D, 41012 Carpi, MO, IT
info@xxxxxxxxxxxxxxxxxxxx
www.amarulasolutions.com