Collin Funk wrote: > Thanks for checking. I pushed the attached patch with those changes.
Thanks. > I assumed uintptr_t was correct for performing arithmetic on any > pointer. I know CHERI has some limitations, but I am not familiar with > them. Perhaps I will get around to reading the manual at some point [1]. Long story short: Modern hardware extensions add extra information into a pointer, such as bounds or a cryptographically secure signature. When you add an offset to a pointer, these extra fields remain unchanged. When you convert a pointer to an uintptr_t and back, these fields are revalidated. You can assume that the CPU contains some caches for optimizing this revalidation away in most cases (otherwise the processing would be too slow). But when converting an uintptr_t to a pointer, some of these optimizations don't apply. Bruno
