On Tue, Aug 12, 2025 at 04:03:17PM -0400, Kent Overstreet wrote: > On Tue, Aug 12, 2025 at 01:35:56PM -0600, Keith Busch wrote: > > On Mon, Aug 11, 2025 at 10:26:03AM -0400, Kent Overstreet wrote: > > > On the other hand, for the only incidences I can remotely refer to in > > > the past year and a half, there has been: > > ... > > > > > - the block layer developer who went on a four email rant where he, > > > charitably, misread the spec or the patchset or both; all this over a > > > patch to simply bring a warning in line with the actual NVME and SCSI > > > specs. > > > > Are you talking about this thread? > > > > https://lore.kernel.org/linux-block/20250311201518.3573009-14-kent.overstreet@xxxxxxxxx/ > > > > I try to closely follow those lists, and that's the only thread I recall > > that even slightly rings a bell from your description, however it's not > > an accurate description (you were the one who misread the specs there; I > > tried to help bridge the gap). I recall the interaction was pretty tame > > though, so maybe you're talking about something else. Perhaps a link for > > context if I got it wrong? > > I've since seen a lot of actual test data from SCSI hard drives - fua > reads are definitely not cached, without exception across manufacturers. > > On NVME the situation is much murkier. Okay, I take it I got the right thread then. I just wanted to get the context. For the record, all the specs align with what read fua does (anyone interested can visit the linked thread, I don't want to hijack this one for it).
