On Wed 13-03-19 19:21:37, Christopher Lameter wrote: > On Wed, 13 Mar 2019, Christoph Hellwig wrote: > > > On Wed, Mar 13, 2019 at 09:11:13AM +1100, Dave Chinner wrote: > > > On Tue, Mar 12, 2019 at 03:39:33AM -0700, Ira Weiny wrote: > > > > IMHO I don't think that the copy_file_range() is going to carry us through the > > > > next wave of user performance requirements. RDMA, while the first, is not the > > > > only technology which is looking to have direct access to files. XDP is > > > > another.[1] > > > > > > Sure, all I doing here was demonstrating that people have been > > > trying to get local direct access to file mappings to DMA directly > > > into them for a long time. Direct Io games like these are now > > > largely unnecessary because we now have much better APIs to do > > > zero-copy data transfer between files (which can do hardware offload > > > if it is available!). > > > > And that is just the file to file case. There are tons of other > > users of get_user_pages, including various drivers that do large > > amounts of I/O like video capture. For them it makes tons of sense > > to transfer directly to/from a mmap()ed file. > > That is very similar to the RDMA case and DAX etc. We need to have a way > to tell a filesystem that this is going to happen and that things need to > be setup for this to work properly. The way to tell filesystem what's happening is exactly what we are working on with these patches... > But if that has not been done then I think its proper to fail a long term > pin operation on page cache pages. Meaning the regular filesystems > maintain control of whats happening with their pages. And as I mentioned in my other email, we cannot just fail the pin for pagecache pages as that would regress existing applications. Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
