On Tue, 26 Nov 2013, Rafael J. Wysocki wrote: > On Tuesday, November 26, 2013 10:48:56 AM Ulf Hansson wrote: > > On 26 November 2013 00:10, Rafael J. Wysocki <rjw@xxxxxxxxxxxxx> wrote: > > > On Monday, November 25, 2013 01:40:21 PM Ulf Hansson wrote: > > >> To put devices into low power state during sleep, it sometimes makes > > >> sense at subsystem-level to re-use device's runtime PM callbacks. Ulf, please be more careful about the distinction between "device" and "driver". Devices don't have callbacks; drivers do. You made this mistake a few times in the patch description. > > >> The PM core will at device_suspend_late disable runtime PM, after that > > >> we can safely operate on these callbacks. At suspend_late the device > > >> will be put into low power state by invoking the device's > > >> runtime_suspend callback, unless the runtime status is already > > >> suspended. At resume_early the state is restored by invoking the > > >> device's runtime_resume callback. Soon after the PM core will re-enable > > >> runtime PM before returning from device_resume_early. > > >> > > >> The new pm_generic functions are named pm_generic_suspend_late_runtime > > >> and pm_generic_resume_early_runtime, they are supposed to be used in > > >> pairs. > > > > > > What happens if the subsystem uses the new functions as its late suspend/ > > > early resume callbacks, but some of its drivers implement .suspend_late() > > > and .resume_early()? The same thing that happens whenever any PM-related callback is defined by both the subsystem and the driver: The PM core uses the subsystem's callback. (Or the pm_domain's callback, or the type's callback, or the class's callback.) Besides, Ulf specifically assumed at the top of this message that re-using the runtime-suspend callback makes sense at the subsystem level. This must mean it is appropriate for all drivers in the subsystem. So why would a driver want to override the subsystem's behavior? > > Good point. Normally, I think the decision of using these callbacks > > should be taken at the lowest level, in other words in the driver. > > Otherwise the lowest layer of .suspend_late will be by-passed. > > I'm not sure what's the point to add them, then. If the driver has to decide > anyway, it may simply populate its .suspend_late and .resume_early pointers > I think. Another possibility is to have the pm_generic_suspend_late_runtime routine check to see if the driver has defined a suspend_late callback. If the driver has its own callback, we could invoke it instead of invoking the runtime_suspend callback. However, I'm not sure that doing this would be worthwhile. And it would hang in an infinite loop if the driver set its suspend_late pointer to pm_generic_suspend_late_runtime! Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
