On Wed, Sep 13, 2023 at 01:28:10PM -0700, Keith Busch wrote: > From: Keith Busch <kbusch@xxxxxxxxxx> > > Some devices are reporting Controller Ready Modes Supported, but return > 0 for CRTO. These devices require a much higher time to ready than that, > so they are failing to initialize after the driver started preferring > that value over CAP.TO. > > The spec requires CAP.TO match the appropritate CRTO value, or be set to > 0xff if CRTO is larger than that. This means that CAP.TO can be used to > validate if CRTO is reliable, and provides an appropriate fallback for > setting the timeout value if not. Use whichever is larger. I need to send a pull request out today since we're quite a bit behind as it is. I've applied this for nvme-6.6 now since it fixes a regression that apparently quite a few people are encountering. If there are any objections, please let me know by EOD and I'll remove it from the queue.
