On 2024/02/25 13:01, Stuart Henderson wrote:
> On 2024/02/25 12:46, Stuart Henderson wrote:
> > I will look at rspamd and luajit later today.
> 
> > > lang/luajit.log
> > > ld.lld: warning: lj_vm_unwind_c_eh: missing endbr64
> > > ld.lld: warning: lj_vm_unwind_rethrow: missing endbr64
> > > ld.lld: warning: lj_vm_unwind_ff_eh: missing endbr64
> > > ld.lld: warning: lj_cont_cat: missing endbr64
> > > ld.lld: warning: lj_cont_ra: missing endbr64
> > > ld.lld: warning: lj_cont_nop: missing endbr64
> > > ld.lld: warning: lj_cont_condt: missing endbr64
> > > etc
> 
> ah, actually I think I'll need help with luajit, these look like they
> may be encoding luajit bytecode so I'm not sure how to deal with them
> or whether they're actually needed at all.
> 
> The trivial "make test" example which just prints a string works on an
> IBT machine but there are missing endbr64 in various fuctions with an
> sse/x87 name - if there are any lua-knowledgeable people reading that
> can show me something that would be likely to use these that I can run
> to test, that would be handy.
> 
> (No need to worry about arm64 for lang/luajit, it uses the 2.0 branch
> which doesn't support arm64 anyway).
> 

semarie points out that luajit already has USE_NOBTCFI (I missed it because
it was low down in the Makefile - personally I try to put things like
USE_WXNEEDED or USE_NOBTCFI right up top so they're easily visible), so
it uses -z nobtcfi.

Building without USE_NOBTCFI I confirm that luajit does crash on an IBT
machine.

Of the ports using luajit's library, most of them also have USE_NOBTCFI.
Only games/luasteam doesn't. (graphics/gimp/snapshot only uses the luajit
binary so doesn't need USE_NOBTCFI itself for any luajit-related reason).

So games/luasteam needs USE_NOBTCFI and a bump.