On Wed, 25 Apr 2012 22:26:05 +0530 anish kumar <anish198519851985@xxxxxxxxx> wrote: > On Wed, 2012-04-25 at 15:26 +1000, NeilBrown wrote: > > On Tue, 24 Apr 2012 22:09:19 -0700 Dmitry Torokhov > > <dmitry.torokhov@xxxxxxxxx> wrote: > > > > > Hi Neil, > > > > > > On Wed, Apr 25, 2012 at 12:21:39PM +1000, NeilBrown wrote: > > > > > > > > If we press and release the power button before the press interrupt is > > > > handled - as can happen on resume - we lose the press event so the > > > > release event is ignored and we don't know what happened to cause the > > > > wakeup. > I didn't understand this.If power button is waking up the device then > obviously power button interrupt handler is called right?If yes then how > can we lose the press event?Is user space not ready to take the event? > May be I didn't understand it properly.Can you kindly explain? Yes, the interrupt handler is running. However the way the driver is currently written, an interrupt is not enough to generate an "button press" event. What happens is when the kernel-thread half of the interrupt handler finally runs, it samples the state of the button and generates an event based on that level. So if the button has been released again by the time the ISR runs, then only a "button release" event is generated. When this gets to the input layer, the input later *knows* that the button is currently released so it suppresses the new "button release" event. A "sync" event does still get through to the App, but they can come at all sorts of time even when nothing is happening to the button, so they are best ignored (when by themselves). What the driver needs to do is acknowledge that just getting an interrupt means that something changed, and to ensure that a change gets passed up to the input layer. Is that clearer? Thanks, NeilBrown > > > > > > What kind of latency do you observe? > > > > When I have debugging enabled, hundreds of milliseconds. > > > > When I don't have debugging enabled ... it doesn't tell me, but I'm fairly > > sure it is several tens of milliseconds and the button press can be quicker > > than that. > > > > If it will help I can try to instrument the driver and get some timings. > > > > > > > > > > > > > So make sure that each interrupt handled does generate an event. > > > > Because twl4030 queues interrupt events we will see two interrupts > > > > for a press-release even if we handle the first one later. This means > > > > that such a sequence will be reported as two button presses. This > > > > is unfortunate but is better than no button presses. > > > > Possibly we could set the PENDDIS_MASK to disable queuing of > > > > interrupts, but that might adversely affect other interrupt sources. > > > > > > > > > > It looks like we'd have to modify every driver to ensure consistent > > > behavior as we do not have any guarantees on how long resume takes. > > > Maybe this is something that input core needs to implement? > > > > Well if every driver is buggy.... > > > > I don't see how this could be implemented in the input core. And even if it > > was, you'd probably need to change each driver to interact with this new > > functionality which would be much the same work as changing them to work with > > the current functionality.... > > But maybe I have no imagination - if you can suggest a way that the input core > > could support this without changing the drivers, I'm happy to try it out. > > > > Thanks, > > NeilBrown >
Attachment:
signature.asc
Description: PGP signature
