On Tuesday, May 13, 2014 11:25:07 AM Alan Stern wrote: > Crossing emails again... > > On Tue, 13 May 2014, Rafael J. Wysocki wrote: > > > > There's nothing to prevent a runtime-suspended device from being > > > resumed in between the ->prepare() and ->suspend() callbacks. > > > > I'm moving the barrier from __device_suspend() to device_prepare(), so there > > shouldn't be surprise resumes in that time frame. > > A wakeup request from the hardware can cause a runtime resume, even > if most threads are in the freezer: > > Not all kernel threads get frozen. One of the unfrozen threads > could respond to the wakeup request by calling > pm_runtime_resume(). > > Some runtime PM callbacks are marked as IRQ-safe and can run > directly within an interrupt handler. > > > > Therefore it makes little sense to check the device's runtime status in > > > device_prepare(). The check should be done in __device_suspend(). > > > > If we do the barrier in device_prepare(), then I'm not sure what mechanism > > would cause the device to resume. > > See above. A wakeup request can arrive after the barrier has finished. > > > If there is one, the whole approach is in danger, because ->prepare() has to > > check if devices are runtime-suspended and has to be sure that their status > > won't change after it has returned 1. > > ->prepare() cannot guarantee in all cases that a device will remain in > runtime suspend. Fortunately, it doesn't need to. In fact (as I > mentioned sometime before), it doesn't even need to check whether the > device currently is runtime suspended -- it suffices to know that _if_ > the device is runtime suspended _then_ it has the proper settings for > system suspend. > > Regardless, status changes cannot cause a problem. If the device does > get runtime-resumed after ->prepare(), it will remain that way when > __device_suspend() runs. The device can't be runtime-suspended again, > because device_prepare() does pm_get_noresume(). > > Therefore, if the device is still runtime-suspended when > __device_suspend() runs, we can be sure that its status and state are > still the same as when ->prepare() ran. But if the device is runtime-suspended, we cannot know if it's going to resume a while later. That's the problem. Rafael -- 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
