On Fri, May 12, 2023 at 01:20:15PM -0700, Christoph Hellwig wrote: > On Fri, May 12, 2023 at 09:08:54AM -0600, Jens Axboe wrote: > > On 5/12/23 9:03?AM, Ming Lei wrote: > > > Passthrough(pt) request shouldn't be queued to scheduler, especially some > > > schedulers(such as bfq) supposes that req->bio is always available and > > > blk-cgroup can be retrieved via bio. > > > > > > Sometimes pt request could be part of error handling, so it is better to always > > > queue it into hctx->dispatch directly. > > > > > > Fix this issue by queuing pt request from plug list to hctx->dispatch > > > directly. > > > > Why not just add the check to the BFQ insertion? That would be a lot > > more trivial and would not be poluting the core with this stuff. > > Because we really need to keep the passthrough code separate. The > fact that a passthrough request can leak into common code in various > places is really a bit of a problem. We have most of these nicely > separate with two exceptions: > > - the plug list > - the requeue list > > The higher level and the more obvious we special case the passthrough > request there, the better for debuggability and maintainability. Agree, and there could be more things involved: 1) pt request shares same tags, so scheduler tags is used for allocating pt request 2) RQF_ELV is set for pt request, and e->type->ops.requeue_request()/ e->type->ops.completed_request() still might be called for pt request. Probably long-term one dedicated request queue(BLK_MQ_F_TAG_QUEUE_SHARED/ BLK_MQ_F_NO_SCHED) could be used for handling passthrough only if the device exposes pt interface. Then I guess core code may be cleaned a bit and becomes easier to improve both two paths, but plug handling still have to cover both. thanks, Ming
