On Sat, Mar 15, 2025 at 10:47:09AM -0600, Jens Axboe wrote: > On 3/11/25 2:15 PM, Kent Overstreet wrote: > > REQ_FUA|REQ_READ means "do a read that bypasses the controller cache", > > the same as writes. > > > > This is useful for when the filesystem gets a checksum error, it's > > possible that a bit was flipped in the controller cache, and when we > > retry we want to retry the entire IO, not just from cache. > > > > The nvme driver already passes through REQ_FUA for reads, not just > > writes, so disabling the warning is sufficient to start using it, and > > bcachefs is implementing additional retries for checksum errors so can > > immediately use it. > > This one got effectively nak'ed by various folks here: > > > Link: > > https://lore.kernel.org/linux-block/[email protected]/ > > yet it's part of this series and in linux-next? Hmm?
As I explained in that thread, they were only thinking about the caching of writes. That's not what we're concerned about; when we retry a read due to a checksum error we do not want the previous _read_ cached.
