On 6/13/24 2:07 AM, Christoph Hellwig wrote: > On Wed, Jun 12, 2024 at 02:53:27PM -0600, Jens Axboe wrote: >>> So we need to ensure REQ_POLLED doesn't even get set for any >>> other I/O. >> >> We happily allow polled IO for async polled IO, even if the destination >> queue isn't polled (or it doesn't exist). This is different than the old >> sync polled support. > > Yes, and for that to work we can't start returning -EOPNOTSUPP as in > this patch, as BLK_QC_T_NONE an be cleared for all kinds of reasons. > > So if we want some kind of error handling that people don't even > bother to poll for devices where it is not supported we need that > check much earlier (probably in io_uring). There's just no way we can do that, who knows if you'll run into a polled queue or not further down the stack. IMHO there's nothing wrong with the current code. If you do stupid things (do polled IO without having polled queues), then you get to collect stupid prizes (potentially excessive CPU usage). -- Jens Axboe
