On 7/1/19 9:20 PM, Chaitanya Kulkarni wrote: > Since discard requests are not as common as read and write requests > mark REQ_OP_DISCARD condition unlikely in the null_handle_rq() and > null_handle_bio(). We should let normal branch prediction handle this. What if you are running a pure discard workload? In general I'm not a huge fan of likely/unlikely annotations, they only tend to make sense when it's an unlikely() for an error case, not for something that could potentially be quite the opposite of an unlikely case. -- Jens Axboe
