02.08.2019 16:38, Peter De Schrijver пишет: > On Tue, Jul 30, 2019 at 08:40:19PM +0300, Dmitry Osipenko wrote: >> The PCLK clock is running off SCLK, which is a critical clock that is >> very unlikely to randomly change its rate. It is possible to get a >> lockup if kernel decides to enter LP2 cpuidle from a clk-notifier, which >> happens occasionally in a case of Tegra30 EMC driver that waits for the >> clk-change event in the clk-notify handler, because CCF's 'prepare' mutex >> in kept locked and thus clk_get_rate() wants to sleep with interrupts >> being disabled. >> > > I don't think this is the right solution. Eventually we will want to > scale sclk and pclk because the clock tree power of those is not > insignificant. Maybe register a notifier which updates the PMC timer > values when pclk changes? I also had a thought about the notifier for pclk, but wasn't sure if it's really worthwhile right now since there is no real use-case and it's not obvious when such case will materialize. So, I'd say that solution is correct but incomplete. I'll make a v4 with the notifier, since you're asking about it. Thanks for the review!
