On Tue, Dec 05, 2017 at 01:13:43AM +0000, Bart Van Assche wrote:

On Tue, 2017-12-05 at 09:04 +0800, Ming Lei wrote:

quoted

Then no reason to revert commit(0df21c86bdbf scsi: implement .get_budget an
.put_budget for blk-mq) for one issue which may never happen in reality since
this reproducer need out-of-tree patch.

Sorry but I disagree completely. You seem to overlook that there may be other
circumstances that trigger the same lockup, e.g. a SCSI queue full condition.

If the scsi_dev_queue_ready() returns false, .get_budget() catches that
and never add request to hctx->dispatch. And scsi_host_queue_ready()
always returns true, since we respect per-host queue depth by
blk_mq_get_driver_tag() before calling .queue_rq().

Or if I miss other cases, please point it out.

-- 
Ming