https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114137
--- Comment #6 from Alexandre Oliva <aoliva at gcc dot gnu.org> --- Thanks for the report. Something's very rotten here. cfrvisited is an automatic variable, why oh why would we have GOTOFF unspecs for it?!? I'd be interested in a preprocessed testcase that will trigger the problem. Failing that, I suppose I could try to drive a remote debug session if you're up for it. If it is indeed something related with GC as suggested (and it sounds plausible), the exact details of when it hits will depend on local hardware details and not necessarily carry over to other machines.