On Tue, Mar 18, 2025 at 10:49:25AM -0700, Bart Van Assche wrote: > On 3/17/25 6:06 PM, Kent Overstreet wrote: > > What bcachefs is doing is entirely in line with the behaviour the > > standard states. > > I do not agree with the above. > > Kent, do you plan to attend the LSF/MM/BPF summit next week? I'm > wondering whether allowing REQ_FUA|REQ_READ would be a good topic > for that summit. No, I won't be there this year. And I don't think it'd be the right forum for arguing over the meaning of an obscure line in the NVME spec, anyways :) It's certainly not in dispute that read fua is a documented, legitimate command, so there's no reason for the block layer to be rejecting it. Whether it has exactly the behaviour we want isn't a critical issue that has to be determined right now. The starting point for that will be to test device behaviour (with some simple performance tests, like I mentioned), and anyways it's outside the scope of the block layer.
