On Mon, Oct 02, 2017 at 07:33:29PM -0400, Nicolas Pitre wrote: > On Tue, 3 Oct 2017, Richard Weinberger wrote: > > > On Mon, Oct 2, 2017 at 12:29 AM, Nicolas Pitre <nicolas.pitre@xxxxxxxxxx> wrote: > > > On Sun, 1 Oct 2017, Christoph Hellwig wrote: > > > > > >> up_read(&mm->mmap_sem) in the fault path is a still a complete > > >> no-go, > > >> > > >> NAK > > > > > > Care to elaborate? > > > > > > What about mm/filemap.c:__lock_page_or_retry() then? > > > > As soon you up_read() in the page fault path other tasks will race > > with you before > > you're able to grab the write lock. > > But I _know_ that. > > Could you highlight an area in my code where this is not accounted for? Existing users of lock_page_or_retry return VM_FAULT_RETRY right after up()ing mmap_sem, and they must already have a reference to the page which is the only thing touched until then. Your patch instead goes for an exclusive mmap_sem if it can, and even if there is nothing that breaks with that scheme right now there s nothing documenting that this actually safe, and we are way down in the complex page fault path. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
