> From: dmitry.torokhov@xxxxxxxxx <dmitry.torokhov@xxxxxxxxx> > Sent: Thursday, October 3, 2019 10:46 AM > ... > Right. I think in practice the current suspend implementation can work > as freeze() for the HV keyboard, because in suspend you shut off vmbus > channel, so there should not be wakeup signals anymore. What you do not > want is to have the current resume to be used in place of thaw(), as > there you re-enable the vmbus channel and resume sending wakeup requests > as you are writing out the hibernation image to storage. > > I think if vmbus allowed HV keyboard driver to supply empty thaw() and > poweroff() implementations, while using suspend() as freeze() and > resume() as restore(), it would solve the issue for you. > > Dmitry BTW, IMO thaw() should not be empty, because drivers/base/power/main.c: pm_op() uses ops->thaw for both PM_EVENT_THAW and PM_EVENT_RECOVER. PMSG_RECOVER is the same as PM_EVENT_RECOVER. If some step in hibernation_snapshot() -> create_image() fails, we call dpm_resume_start(PMSG_RECOVER) at the end of create_image(), and call dpm_resume() in hibernation_snapshot(). An empty thaw() can not bring the device back to normal. Thanks, -- Dexuan
