On Tue, 23 Jun 2009, Rafael J. Wysocki wrote: > In short, I think suspending (or queuing a suspend request) should fail if the > usage counter is nonzero, but the resuming (or queuing up a resume request) > should be possible regardless of its value. The reason is that multiple > threads may in theory attempt to resume the device at the same time. Agreed. Suspends and resumes aren't symmetrical -- a single resume request must outweigh numerous suspend requests. > However, I'm not sure if the core should manipulate the usage counter by > itself, because it's sort of problematic (there's no good approach to decide > when to decrement the counter). Yes. The idea behind my previous message was that it's not really so easy for the core to decide when to _increment_ the counter either. > So, I'd let the callers use pm_runtime_get() to increment the counter > and pm_runtime_put() to decrement it, possibly queuing up an idle notification > if the counter happens to reach 0. Also, I'm not sure if unbalanced > pm_runtime_put() should be regarded as a bug. It should be. Once the counter is messed up, runtime PM wouldn't be able to work properly. But maybe you should add a pm_set_counter call so that drivers can recover from imbalances. One question still remains: If the counter is 0 at the end of a successful pm_runtime_resume, should the core then call pm_notify_idle? Or should we make the driver responsible for that too? > At the same time, I'd like the core to use runtime_status and the other > fields in dev_pm_info, except for the usage counter, to ensure that all > operations are only carried out when it makes sense. Yes. In fact, I'd say that when the counter is positive it doesn't make sense to allow a runtime suspend -- so you don't need that exception in your statement above. :-) Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
