On Mon, Jan 21, 2019 at 11:40:45PM +0530, Amit Kucheria wrote:
> On Tue, Jan 15, 2019 at 3:31 AM Matthias Kaehlcke <mka@xxxxxxxxxxxx> wrote:
> >
> > On Mon, Jan 14, 2019 at 03:51:10PM +0530, Amit Kucheria wrote:
> > > Since all cpus in the big and little clusters, respectively, are in the
> > > same frequency domain, use all of them for mitigation in the
> > > cooling-map. We end up with two cooling devices - one each for the big
> > > and little clusters.
> > >
> > > At the lower trip points we restrict ourselves to throttling only a few
> > > OPPs. At higher trip temperatures, allow ourselves to be throttled to
> > > any extent.
> > >
> > > Signed-off-by: Amit Kucheria <amit.kucheria@xxxxxxxxxx>
> > > ---
> > > arch/arm64/boot/dts/qcom/sdm845.dtsi | 177 ++++++++++++++++++++++++---
> > > 1 file changed, 161 insertions(+), 16 deletions(-)
> > >
> > > diff --git a/arch/arm64/boot/dts/qcom/sdm845.dtsi b/arch/arm64/boot/dts/qcom/sdm845.dtsi
> > > index fb7da678b116..7973e88bdf94 100644
> > > --- a/arch/arm64/boot/dts/qcom/sdm845.dtsi
> > > +++ b/arch/arm64/boot/dts/qcom/sdm845.dtsi
> > >
> > > ...
> > >
> > > @@ -1719,18 +1728,35 @@
> > > thermal-sensors = <&tsens0 1>;
> > >
> > > trips {
> > > - cpu_alert0: trip0 {
> > > + cpu0_alert0: trip-point@0 {
> >
> > Thanks for adapting the trip point names and labels in anticipation of
> > further additions!
> >
> > Seems you aren't overly convinced about the 'target/threshold'
> > terminology used by some other arm64 platforms ;-)
>
> target and threshold have an air of finality to them and doesn't lend
> itself to having a few trip points on the way to the critical trip,
> IMO.
>
> Let me know if you feel otherwise.
I can see your point, and it's also true that target/threshold seem to
imply the use of power_allocator, which may not always be the case.
> > > temperature = <95000>;
> > > hysteresis = <2000>;
> > > type = "passive";
> > > };
> >
> > I realized that we still have the potential problem of a name change
> > in the trip point node name if a 'threshold' node for IPA is added,
> > since this node will have a lower temperature than 95°. If this is
> > something to be concerned about it might be worth to add that extra
> > trip point already to avoid headaches or funky trip point enumeration,
> > even if we know that the value might not be the final one.
>
> I will squash both the DT changes in to a single change introducing 2
> passive trips and 1 critical trip to avoid the churn.
Sounds good, thanks!
> See if you like it better.
I didn't really dislike it, was just wondering if renaming nodes could
break existing users. I imagine it's not a huge problem after all,
since users with an older kernel version won't see the DT change and
probably should use the phandle anyway.
> > (I'm aware that we are also changing the node names and labels right
> > now, it seems less problematic at this point since the SDM845 thermal
> > zones are a fairly recent addition)
> >
> > > - cpu_crit0: trip1 {
> > > + cpu0_crit: cpu_crit@0 {
> >
> > nit: does the @0 add any value here? IIUC there can be only one
> > critical trip point, hence there will never be a cpu_crit@1 or
> > higher.
>
> Agreed. Will remove.
>
> > > temperature = <110000>;
> > > hysteresis = <1000>;
> > > type = "critical";
> > > };
> > > };
> > > +
> > > + cooling-maps {
> > > + map0 {
> > > + trip = <&cpu0_alert0>;
> > > + cooling-device = <&CPU0 THERMAL_NO_LIMIT 4>,
> > > + <&CPU1 THERMAL_NO_LIMIT 4>,
> > > + <&CPU2 THERMAL_NO_LIMIT 4>,
> > > + <&CPU3 THERMAL_NO_LIMIT 4>;
> > > + };
> >
> > Out of curiosity: how did you determing the max cooling state of 4?
>
> Just some basic testing by pinning a dhrystone benchmark to each of
> the cores along with some stress-ng threads. Lopping off the top 4
> OPPs seemed to mitigate anything I could throw at the board.
Thanks for sharing your approach!
> I'm unable to do the "device in a closed car on a hot summer day" type
> of tests on the dev board. Nevertheless, I've changed the patch now to
> only remove the boost frequency at 75 degrees and then full throttling
> at 95 degrees.
>
> I'd appreciate more "real world" testing to validate these.
Sure, we'll run some tests with the new configuration on our site.
Thanks
Matthias
