On Thu, Jun 12, 2025 at 06:28:47AM -0600, Jens Axboe wrote: > But ideally we'd have that, and just a plain doubly linked list on the > queue/dispatch side. Which makes the list handling there much easier to > follow, as per your patch. Quick hack from the weekend. This also never deletes the requests from the submission list for the queue_rqs case, so depending on the workload it should touch either the same amount of less cache lines as the existing version. Only very lightly tested, and ublk is broken and doesn't even compile as it's running out space in the io_uring pdu. I'll need help from someone who knows ublk for that. ---
