On 3/24/21 4:02 PM, Andrew Morton wrote: > On Wed, 24 Mar 2021 13:48:08 -0600 Jens Axboe <axboe@xxxxxxxxx> wrote: > >> On Tue, Mar 2, 2021 at 4:43 PM <akpm@xxxxxxxxxxxxxxxxxxxx> wrote: >>> >>> >>> The patch titled >>> Subject: mm: provide filemap_range_needs_writeback() helper >>> has been added to the -mm tree. Its filename is >>> mm-provide-filemap_range_needs_writeback-helper.patch >> >> Are you still planning on flushing this out for 5.12?? > > Oh. No, I wasn't planning on that. I saw nothing which made me think > it was that urgent. Ehm ok, I think that was the plan though when we originally talked about this series. At least that was what was in my original emails on this topic :-) > What is the justification? How much performance benefit are we talking > here? Well it's pretty huge, since if we return "nope can't do this non-blocking" it'll mean that io_uring has to bump the operation to an async threads. So CPU cost goes way up, and latencies does too. Problem is, we're now at -rc4, and it was my understanding that we'd get this in for 5.12, but obviously way sooner than this. But I kind of lost track when it went into your tree, until I started thinking about it earlier today. While it's not the end of the world to wait for 5.13, though the current situation does suck a lot for certain workloads. -- Jens Axboe
