On Tue, Jun 23, 2020 at 08:27:52PM -0700, Chaitanya Kulkarni wrote:
> There are many places where trace API accepts the struct request_queue*
> parameter which can be derived from other function parameters.
>
> This patch removes the struct request queue parameter from the
> blktrace framework and adjusts the tracepoints definition and usage
> along with the tracing API itself.
Good idea, and I had a half-ready patch for this already as well.
One issue, and two extra requests below:
> if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
> - trace_block_bio_complete(bio->bi_disk->queue, bio);
> + trace_block_bio_complete(bio);
This one can also be called for a different queue than
bio->bi_disk->queue, so for this one particular tracepoint we'll need
to keep the request_queue argument.
> diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
> index fdcc2c1dd178..a3cade16ef80 100644
> --- a/block/blk-mq-sched.c
> +++ b/block/blk-mq-sched.c
> @@ -409,7 +409,7 @@ EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge);
>
> void blk_mq_sched_request_inserted(struct request *rq)
> {
> - trace_block_rq_insert(rq->q, rq);
> + trace_block_rq_insert(rq);
> }
> EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted);
As a follow on patch we should also remove this function.
> }
>
> spin_lock(&ctx->lock);
> @@ -2111,7 +2111,7 @@ blk_qc_t blk_mq_make_request(struct request_queue *q, struct bio *bio)
> goto queue_exit;
> }
>
> - trace_block_getrq(q, bio, bio->bi_opf);
> + trace_block_getrq(bio, bio->bi_opf);
The second argument can be removed as well. Maybe as another patch.
