Re: [PATCH v2 0/7] pmdomain/cpuidle-psci: Support s2idle/s2ram on PREEMPT_RT

From: Ulf Hansson
Date: Tue Jul 09 2024 - 11:32:03 EST


On Mon, 8 Jul 2024 at 15:53, Sebastian Andrzej Siewior
<bigeasy@xxxxxxxxxxxxx> wrote:
>
> On 2024-05-27 16:25:50 [+0200], Ulf Hansson wrote:
> > Updates in v2:
> > - Rebased and fixed a small issue in genpd, see patch3.
> > - Re-tested on v6.9-rt5 (PREEMPT_RT enabled)
> > - Re-tested on v6.10-rc1 (for regressions, PREEMPT_RT disabled)
> >
> > The hierarchical PM domain topology and the corresponding domain-idle-states
> > are currently disabled on a PREEMPT_RT based configuration. The main reason is
> > because spinlocks are turned into sleepable locks on PREEMPT_RT, which means
> > genpd and runtime PM can't be use in the atomic idle-path when
> > selecting/entering an idle-state.
> >
> > For s2idle/s2ram this is an unnecessary limitation that this series intends to
> > address. Note that, the support for cpuhotplug is left to future improvements.
> > More information about this are available in the commit messages.
>
> I looked at it and it seems limited to pmdomain/core.c, also I don't
> know if there is a ->set_performance_state callback set since the one I
> checked have mutex_t locking ;)
> So if this is needed, then be it. s2ram wouldn't be used in "production"
> but in "safe state" so I wouldn't worry too much about latency spikes.
> Not sure what it means for the other modes.
> I am not to worried for now, please don't let spread more than needed ;)

Thanks for taking a look and for providing your thoughts. Can I
consider that as an "ack" for the whole series?

Before I decide to apply this I am awaiting some additional
confirmation from Qcom guys. It's getting late for v6.11, so I may
need to make another re-spin, but let's see.

Kind regards
Uffe