On Tue, Oct 08, 2024 at 02:25:35PM +0200, Christoph Hellwig wrote: > On Mon, Oct 07, 2024 at 12:10:11PM +0200, Javier González wrote: > > In summary, what we are asking for is to take the patches that cover the > > current use-case, and work together on what might be needed for better > > FS support. > > And I really do not think it is a good idea. For one it actually > works against the stated intent of the FDP spec. Second extending > the hints to per per-I/O in the io_uring patch is actively breaking > the nice per-file I/O hint abstraction we have right now, and is > really unsuitable when actually used on a file and not just a block > device. And if you are only on a block device I think passthrough > of some form is still the far better option, despite the problems > with it mentioned by Keith. Then let's just continue with patches 1 and 2. They introduce no new user or kernel APIs, and people have already reported improvements using it. Further, it is just a hint, it doesn't lock the kernel into anything that may hinder future inovations and enhancements. So let's unblock users and refocus *our* time to something more productive, please? And to be honest, the per-io hints for generic read/write use is only valuable for my users if metadata is also exposed to userspace. I know Javier's team is working on that in parallel, so per-io hints are a lower priority for me until that part settles.
