07:15 <wens> wondering how people deal with intermediate stable clocks in the context of cpufreq/devfreq? i.e. when the rate of the upstream PLL is reconfigured, the consumer is switched over to a stable clock. The gotcha is that stable clock is faster than the lowest OPP, and so the device might need a voltage boost.

07:16 <wens> one not so great option I have is to switch to the slowest stable clock, the main crystal, and not need to deal with voltage boost.

07:18 <wens> For more context: https://lore.kernel.org/linux-mediatek/CAGXv+5EfoOx8LHwHeL+wva_M0KX4S3qLNsBgk_003hvXxYPRVQ@mail.gmail.com/

07:18 <wens> This is for a cache coherency interconnect devfreq driver.

14:47 <CounterPillow> robher: in https://lore.kernel.org/linux-media/20210324151715.GA3070006@robh.at.kernel.org/ you mention that bindings with only one interrupt shouldn't use interrupt-names

14:47 <CounterPillow> the rockchip-vpu.yaml binding requires interrupt-names in a case where only the decoder is present, and only has a decoder "vdpu" interrupt. I followed this example by extending this to require that interrupt-names be present in the encoder-only configuration as well, with just the "vepu" interrupt. Was this the wrong choice?

14:51 <robher> CounterPillow: There's no shortage of bad examples... If the vpu can have both encoder and decoder and has cases of multiple interrupts and single interrupt, then using interrupt-names is perfectly justified.

14:51 <robher> BTW, #devicetree is the appropriate channel for DT issues.

14:51 <CounterPillow> Oh, nice