On Thu, Aug 28, 2025 at 04:31:56PM -0700, Linus Torvalds wrote: > Same largely goes for that > > > - struct mount **mnt_pprev_for_sb;/* except that LSB of pprev will be stolen */ > > + unsigned long mnt_pprev_for_sb; /* except that LSB of pprev is stolen */ > > change, but at least there it's now a 'unsigned long', so it will > *always* complain if a cast is missing in either direction. That's > better, but still horrendously ugly. > > If you want to use an opaque type, then please make it be truly > opaque. Not 'unsigned long'. And certainly not 'void *'. Make it be > something that is still type-safe - you can make up a pointer to > struct name that is never actually declared, so that it's basically a > unique type (or two separate types for mnt_pprev_for_sb and > > I'm not even clear on why you did this change, but if you want to have > specific types for some reason, make them *really* specific. Don't > make them 'void *', and 'unsigned long'. What I want to avoid is compiler seeing something like (unsigned long)READ_ONCE(m->mnt_pprev_for_sb) & 1 and going "that thing is a pointer to struct mount *, either the address is even or it's an undefined behaviour and I can do whatever I want anyway; optimize it to 0". unsigned long is a brute-force way to avoid that - it avoids UB (OK, avoids it as long as no struct mount instance has an odd address), so compiler can't start playing silly buggers. If you have a prettier approach, I'd like to hear it - I obviously do not enjoy the way this one looks.
