On Sat, Jul 31, 2021 at 10:41 AM Stephen Boyd <sboyd@xxxxxxxxxx> wrote: > Quoting Russell King (Oracle) (2021-07-28 13:40:34) > > > I adapted the Subject in the hope to catch Stephen's and Michael's > > > attention. My impression is that this thread isn't on their radar yet, > > > but the topic here seems important enough to get a matching Subject. > I still wonder if it would be better if we had some sort of DT binding > that said "turn this clk on when the driver probes this device and keep > it on until the driver is unbound". DT is not the only way the clocks are established in the kernel. > That would probably work well for > quite a few drivers that don't want to ever call clk_get() or > clk_prepare_enable() and could tie into the assigned-clock-rates > property in a way that let us keep the platform details out of the > drivers. > We could also go one step further and make a clk pm domain when > this new property is present and then have the clk be enabled based on > runtime PM of the device (and if runtime PM is disabled then just enable > it at driver probe time). This sounds like a good idea from a usage perspective. -- With Best Regards, Andy Shevchenko