Thanks, yeah I think I’ve settled on the same conclusion. Everything else about the board seems totally stable, so it seems highly unlikely that it only manifests in clang.
The only reason I’d think that clang was a canary for a deeper problem is that AFAIK it does a bit of multithreading in an early phase of the compiler, and if there were some sort of TLB sync issue that may be a fairly good canary for it. Large process, probably faulting in lots of pages with a bunch of threads running simultaneously would be a good test. Not a lot of apps behave that way.