On 09-01-20, 10:44, Saravana Kannan wrote: > Agree, the OPP framework itself shouldn't be responsible. And I'm not > doing anything here? Just giving those devices a way to look up what > their suspend bandwidth is? So they can vote for it when they suspend? I think this will originate by itself from the device in case of interconnects as well and you don't need to separately have that for interconnects. For example, the device (lets say GPU) will have one of its OPP (and frequency, maybe the lowest one) marked as suspend-OPP. Then the driver which is doing the co-relation normally between GPU/DDR/Cache OPPs should be able to do this conversion as well without any extra help from the interconnect table. If the minimum freq of the device correspond to the minimum freq of the DDR/Cache during normal operation, that should still work during suspend times, isn't it ? > Ok, but you want this done only for "exact" or for all the other > helpers too? All helpers that you need for PM domains and interconnects. > Also, this means that I'll have to implement a > _opp_compare_key2() or whatever because the generic one that > automatically picks the key is still needed for the generic code. Is > that fine by you? I am not concerned about the number of helpers but their optimization. I will leave it for you to do that and review it when I see how you have done it :) -- viresh
