#5703 HIGH 8.2.0 (: Lid switch detection is unreliable.

[Zarro Boogs per Child]

#5703: Lid switch detection is unreliable.
-------------------------+--------------------------------------------------
   Reporter:  cjb        |       Owner:  dilinger            
       Type:  defect     |      Status:  new                 
   Priority:  high       |   Milestone:  8.2.0 (was Update.2)
  Component:  kernel     |     Version:                      
 Resolution:             |    Keywords:  power               
Next_action:  never set  |    Verified:  0                   
  Blockedby:             |    Blocking:  6590                
-------------------------+--------------------------------------------------

Comment(by wmb at firmworks.com):

 I don't think we can distinguish between 2 edges (close open) and 3 edges
 (close open close) without using pin readback.

 Here is an idea for a way to setup and use the GPIO circuit so as to avoid
 the inherent problems with the edge detection.  (The problem is that you
 can't read back the input pin state when the edge detectors are on, and
 when you turn them off, you have to wait 100 uS before you can see the pin
 state.  It's best not to have delays in an interrupt routine.)

 The idea depends on the facts that 0 to 1 events are detected even when
 both edge enables are off and that you can invert the GPIO pin at the
 source.

 The trick is to leave both edge enables off (the master event enable can
 stay on).
 When the GPIO pin is currently high, as determined by readback, set the
 "input invert" bit to make its internal version "go low" with respect to
 the event detector. That high-to-low internal transition does not cause an
 event.

 When the external signal goes low, the inverted internal form goes high,
 triggering an event. The interrupt handler uses the readback to verify
 that it is indeed still in the expected new state (internal version high),
 and if so, clears the inverter so the next event will occur when the
 external signal goes high again.  If the readback show that the pin is
 still in the old state, the event can be ignored as a glitch (i.e. a brief
 transition that was subsequently "undone").

 The posedge and negedge detectors never come into play. Only the implicit
 positive edge detector represented by the PME EdgeLatch is used.  There is
 no need for any delays.  There are no worries about missed events because
 the action to take depends on the readback state, not on the assumption
 that there was exactly one edge.

-- 
Ticket URL: <http://dev.laptop.org/ticket/5703#comment:16>
One Laptop Per Child <http://laptop.org/>
OLPC bug tracking system