On 10/15/18 4:38 AM, Ming Lei wrote: >> @@ -320,7 +320,7 @@ static void vdc_end_one(struct vdc_port *port, struct vio_dring_state *dr, >> >> rqe->req = NULL; >> >> - __blk_end_request(req, (desc->status ? BLK_STS_IOERR : 0), desc->size); >> + blk_mq_end_request(req, desc->status ? BLK_STS_IOERR : 0); > > blk_mq_end_request() may trigger BUG in case of partial completion, > however looks it is fine for __blk_end_request(). True, I'll change this to the blk_update_request() + __blk_mq_end_request() combo instead. > Most of conversions have the above pattern, maybe it is easier to > introduce the following block API: > > struct reuqest_queue *blk_mq_alloc_and_init_sq(set, ops, queue_depth) Yeah I did consider that, I'll make that addition. -- Jens Axboe
