Handling fault in Umbra

[toshipiazza]

32-bit users of Umbra require custom fault handling logic when writing to a read-only page shared shadow memory page. Currently, we write the last app address to a spill slot, which was already going to be a temporary hacky solution.

Here are some issues with the current setup:

1. users of the drtaint library cannot use this particular spill slot
2. it is slow to pessimistically save to this spill slot, despite the possibility that no fault ever occurs

drmemory's fastpath for handling faults in the inlined code looks like this