Hi Daniel Thank you for your the new information I have a question trouble to you We should choose which per-core thermal zone as the IPA's input reference temperature in the current kernel version? thank you. On 27/11/2020, Lukasz Luba <lukasz.luba@xxxxxxx> wrote: > > > On 11/27/20 1:26 PM, Daniel Lezcano wrote: >> >> Hi Lukasz, >> >> On 27/11/2020 10:27, Lukasz Luba wrote: >>> >>> >>> On 11/27/20 8:35 AM, gao.yunxiao6@xxxxxxxxx wrote: >>>> From: "jeson.gao" <jeson.gao@xxxxxxxxxx> >>>> >>>> virtual thermal node definition description in dts file >>>> >>>> Signed-off-by: jeson.gao <jeson.gao@xxxxxxxxxx> >>>> --- >> >> [ ... ] >> >>> It's coming back. There were attempts to solve this problem. >>> Javi tried to solved this using hierarchical thermal zones [1]. >>> It was even agreed (IIRC during LPC) but couldn't continue. Then Eduardo >>> was going to continue this (last message at [3]). Unfortunately, >>> development stopped. >>> >>> I also have out-of-tree similar implementation for my Odroid-xu4, >>> which does no have an 'SoC' sensor, but have CPU sensors and needs >>> some aggregation function to get temperature. >>> >>> I can pick up Javi's patches and continue 'hierarchical thermal zones' >>> approach. >>> >>> Javi, Daniel, Rui what do you think? >> >> I already worked on the hierarchical thermal zones and my opinion is >> that fits not really well. >> >> We want to define a new feature because the thermal framework is built >> on the 1:1 relationship between a governor and a thermal zone. >> >> Practically speaking, we want to mitigate two thermal zones from one >> governor, especially here the IPA governor. >> >> The DTPM framework is being implemented to solve that by providing an >> automatic power rebalancing between the power manageable capable devices. >> >> In our case, the IPA would stick on the 'sustainable-power' resulting on >> the aggregation of the two performance domains and set the power limit >> on the parent node. The automatic power rebalancing will ensure maximum >> throughput between the two performance domains instead of capping the >> whole. >> >> > > Make sense. Thank you for sharing valuable opinion. > > Regards, > Lukasz >
