[PATCH v2 0/2] cpufreq: mediatek: Suspend at the opp-suspend OPP, and mark one on MT8173

[Viresh Kumar]

On 09-10-26, 14:44, Ryan Brue wrote:
> mediatek-cpufreq has no .suspend callback, so each cluster enters system
> sleep at whatever OPP the governor last chose. On the Amazon Fire HD 10
> (2017), an MT8173 tablet that is not upstream yet, entering
> suspend-to-RAM with both clusters at their highest OPP (2106 and
> 1703 MHz) measured 63.5 and 63.6 mA, against 52.0 and 52.0 mA with this
> series. Those are 15-minute windows of the PMIC's coulomb counter,
> alternating with and without the change; one count is about 2.9 mA.
> Earlier, on the board's previous firmware path, s2idle drew 210-232 mA
> against 158 mA.
> 
> Patch 1 takes the suspend frequency from the OPP table's opp-suspend
> entry, as cpufreq-dt does, rather than always using the lowest OPP. That
> keeps it opt-in per SoC: this driver also runs mt2701, mt7622, mt8183,
> mt8186 and others I cannot test, and no MediaTek table has opp-suspend
> today, so they are unchanged. Patch 2 marks the lowest OPP of both MT8173
> clusters.
> 
> Patch 2 also reaches elm, hana and the EVB, which I cannot test either.
> What they get is a switch to 507 MHz in cpufreq_suspend(), which runs
> before any device is suspended, so the regulators it needs should still
> be up. cpufreq_suspend() also runs from device_shutdown(), so reboot and
> power-off switch first as well; on the tablet, reboot -f still resets in
> 1.4 s.
> 
> Patch 2 does nothing without patch 1 and is harmless on its own, so the
> two can go through the cpufreq and MediaTek trees independently.

Applied. Thanks.

-- 
viresh