On Mon, Mar 17, 2025 at 09:06:13PM -0400, Kent Overstreet wrote: > What bcachefs is doing is entirely in line with the behaviour the > standard states. Setting the FUA bit on any READ command (if supported by the device) is entirely in line with the behaviour the standards. It's just not going to do what you hope. And while you claim that it helps you with data recovery, you've not shown either a practical example where it does, or a theory based on hardware architeture how it could. > It's an amusing state of affairs, but it'd be easily resolved with an > admin level NVME command to flip a state bit (like the read recovery > level we were also talking about), and anyways multihoming capable NVME > devices are an entirely different market from the conventional stuff. Please joing the NVMe technical working group and write a proposal. NVMe requires you to clearly state the benefits of the proposal, which means you have to actually clearly write that down first.
