On Mon 06-12-21 20:01:34, David Hildenbrand wrote: [...] > Yes, that's what I refer to as fixing it in the caller -- similar to > [1]. Michals point is to not require such node_online() checks at all, > neither in the caller nor in the buddy. > > I see 2 options short-term > > 1) What we have in [1]. > 2) What I proposed in [2], fixing it for all such instances until we > have something better. > > Long term I tend to agree that what Michal proposes is better. > > Short term I tend to like [2], because it avoids having to mess with all > such instances to eventually get it right and the temporary overhead > until we have the code reworked should be really negligible ... I do dislike both but if I were to chose which to chose between the two then 2 is surely more targeted. We really do not want to spread this into bulk/pcp or whatever other allocator there is. The problem is that somebody might still try to access NODE_DATA (e.g. via a helper that hides that fact). Anyway, I am not sure whether authors of the patch can reproduce the problem and whether they can run a testing code on their machine. If yes it would be great to try with http://lkml.kernel.org/r/Ya89aqij6nMwJrIZ@xxxxxxxxxxxxxx that I have just sent. > [1] https://lkml.kernel.org/r/20211108202325.20304-1-amakhalov@xxxxxxxxxx > [2] > https://lkml.kernel.org/r/51c65635-1dae-6ba4-daf9-db9df0ec35d8@xxxxxxxxxx > > -- > Thanks, > > David / dhildenb -- Michal Hocko SUSE Labs
