On Tue 27-03-18 16:31:23, Mateusz Guzik wrote: > On Tue, Mar 27, 2018 at 08:29:39AM +0200, Michal Hocko wrote: > > On Tue 27-03-18 02:20:39, Yang Shi wrote: > > [...] > > The patch looks reasonable to me. Maybe it would be better to be more > > explicit about the purpose of the patch. As others noticed, this alone > > wouldn't solve the mmap_sem contention issues. I _think_ that if you > > were more explicit about the mmap_sem abuse it would trigger less > > questions. > > > > >From what I gather even with other fixes the kernel will still end up > grabbing the semaphore. In this case I don't see what's the upside of > adding the spinlock for args. The downside is growth of mm_struct. Because accessing the specific address in the address space can be later changed to use a more fine-grained locking. There are people experimenting with range locking. These mmap_sem abusers, on the other hand, will require the full range lock without a good reason. So it is really worth it to remove them and replace by a more fine grained locking. If the mm_struct grow is a real concern (I haven't checked that) then we can use a set of hashed locks or something else. -- Michal Hocko SUSE Labs
