On 6/12/25 6:23 AM, Christoph Hellwig wrote: > On Thu, Jun 12, 2025 at 06:21:14AM -0600, Jens Axboe wrote: >> It's certainly going to make the cached handling more expensive, as the >> doubly linked behavior there is just pointless. Generally LIFO behavior >> there is preferable. I'd strongly suggest we use the doubly linked side >> for dispatch, and retain singly linked for cached + completion. If not >> I'm 100% sure we're going to be revisiting this again down the line, and >> redo those parts yet again. > > Yeah. For cached requests and completions it might even make sense > to have a simple fixed size array FIFO buffer.. I did ponder that in the past too, as that's clearly better. Experimentally we need ~32 slots in there though, which is 256b of storage. Pretty sure I have patches laying around somewhere that did that, but didn't like the plug and batch size growth on the stack. Maybe overthinking that part... 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. -- Jens Axboe
