On 5/28/24 8:34 AM, Jens Axboe wrote: > You could just make it io_kiocb based, but I did not want to get into > foreign requests on remote rings. What would you envision with that > approach, using our normal ring task_work for this instead? That would > be an approach, obviously this one took a different path from the > previous task_work driven approach. Foreign io_kiocb itself isn't even enough, you'd need an overflow cqe allocated upfront too for posting directly without waiting for the reply. Basically it needs to be confident it'll post successfully on the remote ring, otherwise we cannot complete the source msg ring sqe inline. -- Jens Axboe
