On Wed, Feb 28, 2024 at 12:34:41PM -0500, Kent Overstreet wrote: > But even better would be if we could get lockdep to support folio locks, > then we could define a lockdep comparison function. Of course, there's > no place to stick a lockdep map, but I think technically lockdep could > do everything it needs to do without state in the lock itself, if only > that code didn't make my eyes bleed whenever I have to dig into it... I don't think lockdep can support folio locks. They're taken in process context and (for ->read_folio) released in interrupt context. Also we can return to userspace while holding folio locks (I believe ...) DEPT claims to be able to handle them, but I haven't looked in detail.
