Re: [PATCH v25 0/7] soc: mediatek: SVS: introduce MTK SVS

[Kevin Hilman <khilman@kernel.org>]

Chen-Yu Tsai <wenst@chromium.org> writes:

> n Wed, May 18, 2022 at 8:03 AM Kevin Hilman <khilman@kernel.org> wrote:
>>
>> Kevin Hilman <khilman@kernel.org> writes:
>>
>> > Chen-Yu Tsai <wenst@chromium.org> writes:
>> >
>> >> On Mon, May 16, 2022 at 8:43 AM Roger Lu <roger.lu@mediatek.com> wrote:
>> >>>
>> >>> The Smart Voltage Scaling(SVS) engine is a piece of hardware
>> >>> which calculates suitable SVS bank voltages to OPP voltage table.
>> >>> Then, DVFS driver could apply those SVS bank voltages to PMIC/Buck
>> >>> when receiving OPP_EVENT_ADJUST_VOLTAGE.
>> >>>
>> >>> 1. SVS driver uses OPP adjust event in [1] to update OPP table voltage part.
>> >>> 2. SVS driver gets thermal/GPU device by node [2][3] and CPU device by get_cpu_device().
>> >>> After retrieving subsys device, SVS driver calls device_link_add() to make sure probe/suspend callback priority.
>> >>>
>> >>> [1] https://git.kernel.org/pub/scm/linux/kernel/git/vireshk/pm.git/commit/?h=opp/linux-next&id=25cb20a212a1f989385dfe23230817e69c62bee5
>> >>> [2] https://git.kernel.org/pub/scm/linux/kernel/git/vireshk/pm.git/commit/?h=opp/linux-next&id=b325ce39785b1408040d90365a6ab1aa36e94f87
>> >>> [3] https://git.kernel.org/pub/scm/linux/kernel/git/matthias.bgg/linux.git/commit/?h=v5.16-next/dts64&id=a8168cebf1bca1b5269e8a7eb2626fb76814d6e2
>> >>>
>> >>> Change since v24:
>> >>> - Rebase to Linux 5.18-rc6
>> >>> - Show specific fail log in svs_platform_probe() to help catch which step fails quickly
>> >>> - Remove struct svs_bank member "pd_dev" because all subsys device's power domain has been merged into one node like above [3]
>> >>>
>> >>> Test in below environment:
>> >>> SW: Integration Tree [4] + Thermal patch [5] + SVS v25 (this patchset)
>> >>> HW: mt8183-Krane
>> >>>
>> >>> [4] https://github.com/wens/linux/commits/mt8183-cpufreq-cci-svs-test
>> >>
>> >> I've updated my branch to include all the latest versions of the relevant
>> >> patch series:
>> >>
>> >> - anx7625 DPI bus type series v2 (so the display works)
>> >> - MT8183 thermal series v9 (this seems to have been overlooked by the
>> >> maintainer)
>> >> - MTK SVS driver series v25
>> >> - devfreq: cpu based scaling support to passive governor series v5
>> >> - MTK CCI devfreq series v4
>> >> - MT8183 cpufreq series v7
>> >> - Additional WIP patches for panfrost MTK devfreq
>> >
>> > Thanks for preparing an integration branch Chen-Yu.
>> >
>> > I'm testing this on mt8183-pumpkin with one patch to add the CCI
>> > regulator[1], and the defconfig you posted in a previous rev of this
>> > series, but the CCI driver still causes a fault on boot[2] on my
>> > platform.
>> >
>> > I mentioned in earlier reviews that I think there's potentially a race
>> > between CCI and SVS loading since they are co-dependent.  My hunch is
>> > that this is still not being handled properly.
>>
>> Ah, actually it's crashing when I try to boot the platform with
>> `maxcpus=4` on the cmdline (which I have to do because mt8183-pumpkin is
>> unstable upstream with the 2nd cluster enabled.)
>>
>> The CCI driver should be a bit more robust about detecting
>> available/online CPUs
>
> This all seems to be handled in the devfreq passive governor.

Well, that's the initial crash.  But the SVS driver will also go through
its svs_mt8183_banks[] array (including both big & little clusters) and
try to init SVS, so presumably that will have some problems also if only
one cluster is enabled.

> And presumably we'd like to have CCI devfreq running even if just one
> core was booted.

Yes, I assume so also.

> Added Chanwoo for more ideas.

OK, thanks.

Kevin

_______________________________________________
Linux-mediatek mailing list
Linux-mediatek@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-mediatek