GitLab

The LLCP stack is still using globals without
holding the concurrency lock. When turning the
screen off we were previously disabling the
NFCEE routing before disconnecting the active
target. The routing code would grab the lock
and then touch the globals, which the LLCP
was using illegally, causing a crash when the
LLCP triggered callbacks are delivered.

This change reorders the shutdown at screen
off to disconnect the remote target before
issuing the change to the NFCEE routing in
an attempt to stop the ongoing LLCP
transactions, thereby avoiding the crash.
Since the callbacks are asynchronous the
race is still there, but it's hit more like
10% of the time instead of 100% of the time.

Bug: 5366024
Change-Id: I31230166c1c5bf116195db350ac6605feb90e87f