Re: [PATCH v4 0/4] Enable ICE clock scaling

[Abhinaba Rakshit]

On Fri, Feb 06, 2026 at 01:14:28PM +0100, Konrad Dybcio wrote:
> On 2/6/26 1:03 PM, Abhinaba Rakshit wrote:
> > On Mon, Feb 02, 2026 at 04:01:38PM +0100, Konrad Dybcio wrote:
> >> On 2/2/26 7:36 AM, Abhinaba Rakshit wrote:
> >>> On Thu, Jan 29, 2026 at 01:17:51PM +0100, Konrad Dybcio wrote:
> >>>> On 1/28/26 9:46 AM, Abhinaba Rakshit wrote:
> >>>>> Introduce support for dynamic clock scaling of the ICE (Inline Crypto Engine)
> >>>>> using the OPP framework. During ICE device probe, the driver now attempts to
> >>>>> parse an optional OPP table from the ICE-specific device tree node to
> >>>>> determine minimum and maximum supported frequencies for DVFS-aware operations.
> >>>>> API qcom_ice_scale_clk is exposed by ICE driver and is invoked by UFS host
> >>>>> controller driver in response to clock scaling requests, ensuring coordination
> >>>>> between ICE and host controller.
> >>>>>
> >>>>> For MMC controllers that do not support clock scaling, the ICE clock frequency
> >>>>> is kept aligned with the MMC controller’s clock rate (TURBO) to ensure
> >>>>> consistent operation.
> >>>>
> >>>> You skipped that bit, so I had to do a little digging..
> >>>>
> >>>> This paragraph sounds scary on the surface, as leaving a TURBO vote hanging
> >>>> would absolutely wreck the power/thermal profile of a running device,
> >>>> however sdhci-msm's autosuspend functions quiesce the ICE by calling
> >>>> qcom_ice_suspend()
> >>>>
> >>>> I think you're missing a dev_pm_opp_set(dev, NULL) or so in that function
> >>>> and a mirrored restore in _resume
> >>>
> >>> Thanks for pointing this out, its an important piece which is missed.
> >>> We can use dev_pm_opp_set_rate(dev, 0/min_freq) in _suspend and restore the
> >>
> >> FWIW
> >>
> >> dev_pm_opp_set_rate(0) will drop the rpmh vote altogether and NOT
> >> disable the clock or change its rate
> >>
> >> dev_pm_opp_set_rate(min_freq) will *lower* the rpmh vote and DO
> >> set_rate (the clock is also left on)
> >>
> >> Konrad
> >>
> > 
> > Thanks for the info.
> > I guess, dev_pm_opp_set_rate(dev, 0) seems more ideal as this is
> > API is for full quiesce mode and the clocks are anyway gated in
> > the suspend call (clk_disable_unprepare).
> 
> Yeah, please make sure to call dev_pm_opp_set_rate(0) *after* you
> disable the clock though, to make sure we don't brownout

Sure, that makes sense.

Abhinaba Rakshit