On Wed, Sep 16, 2020 at 10:30:42PM +0100, Daniel Kiss wrote: > > Thanks for the summary -- yeah, that was my suspicion, that some > attribute was being lost somewhere. And I think if we generalize this, > and don't just try to attach "frame-pointer" attr to the function, we > probably also solve the BTI issue that Mark still pointed out with > these module_ctor/dtors. > > I was trying to see if there was a generic way to attach all the > common attributes to the function generated here: > > https://github.com/llvm/llvm-project/blob/master/llvm/lib/Transforms/Utils/ > ModuleUtils.cpp#L122 > -- but we probably can't attach all attributes, and need to remove a > bunch of them again like the sanitizers (or alternatively just select > the ones we need). But, I'm still digging for the function that > attaches all the common attributes… > > > We had the problem with not just the sanitisers. Same problem pops with > functions > that created elsewhere in clang (e.g _clang_call_terminate ) or llvm. > > In case of BTI the flag even controllable by function attributes which makes > it > more trickier so > the module flags found the only reliable way to pass this information down. > Scanning existing functions is fragile for data only compilation units for > example. > > Our solution, not generic enough but might help. > https://reviews.llvm.org/D85649
Thanks for the pointer -- I've subscribed to that now. Just to check my understanding, is the issue that generated functions don't implicitly get function attributes like "branch-target-enforcement", and so the BTI insertion pass skips those? I'm guessing that it's unlikely this'll be fixed for an LLVM 11 release? On the kernel side I guess we'll have to guard affected features as being incompatible with BTI until there's a viable fix on the compiler side. :/ Thanks, Mark.
