> If the test is just as legitimate without PROT_EXEC, what’s the > justification for not just removing PROT_EXEC entirely rather than > making its behaviour depend on the sysctl, which could become confusing > (and complicates the test)? IMO either the test should be skipped for > !allow_wx or it should always just make a RW mapping; this choice is > rather odd.
The short answer is, we didn’t know whether PROT_EXEC was essential, so this seemed like the safest change to preserve the intent but also work correctly with W^X. It was also the easiest way to increase test coverage of all these cases with the least code, since some systems will run with W+X and others with W^X. Eric
