https://go.dev/427615 broke the linux-386-longtest builder as TestScript/cover_atomic_pkgall started to fail. My hypothesis is that we're just getting unlucky. The failing function, runtime/internal/sys.Len64, can't be intrinsified on 386, and is not nosplit so we might get a stack growth. However, I think this was really always the case, and we were just getting lucky that morestack just wasn't ever actually called. Something about this change adjusted stack use and now it is.
In either case, it's clear that the fix is to make Len64 nosplit since its use is in a nosplit function ((*timeHistogram).record) which was modified by the aforementioned CL.
The text was updated successfully, but these errors were encountered: