Hi all,
I am developing PPC record/replay for
panda.
I am encountering an issue where the atomic compare-and-swap using lwarx/stwcx is a source of nondeterminism during replay. The second bne- sometimes takes different branches during record and replay, so I am investigating why this is the case and whether I need to write this to a record log.
Can you point out what helper functions lwarx uses to reserve memory locations, and what may be causing the reservation to be lost nondeterministically?
Thanks.