On Mon, 2018-11-26 at 18:29 +0000, Joseph Myers wrote: > On Mon, 26 Nov 2018, Mark Wielaard wrote: > > > Define a new target macro TARGET_HAS_DEFAULT_NOEXEC_STACK for those > > targets > > that have a non-executable default stack based on when they call > > file_end_indicate_exec_stack. > > Some targets (e.g. ia64) may default to no-exec stacks without > needing > that hook (e.g. if they use function descriptors, so trampolines have > no > need for an executable stack).
Yes, but for such targets it doesn't really matter whether or not they define this new target macro. The -Wtrampolines warning only warns when creating an executable runtime trampoline on the stack. It won't for non-executable trampolines. So yes, then this target macro doesn't have to be defined even if the stack is non-executable by default. Defining the TARGET_HAS_DEFAULT_NOEXEC_STACK macro for such a target will then make it so that -Wall enables -Wtrampolines by default, but -Wtrampolines still won't warn because the trampolines aren't executable code on the stack. But I believe that is the correct behavior. The warning should only happen if gcc would otherwise silently make the stack executable for a target that has a default non- executable stack (and it won't for target that uses function descriptors). I thought the documentation updates for the warning made that more clear, but maybe they can be improved even more? A case could be made that taking a nested function pointer that escapes the lexical scope from which it takes some context variable always is a dangerous thing to do and that it should even warn if supporting that doesn't involve placing an executable runtime trampoline on the stack. Which is actually the case I made in the original bug. But I don't believe I got consensus for that. Thanks, Mark
