On Tue, Sep 06, 2022 at 12:03:29PM +0530, Kanchan Joshi wrote:
>On Tue, Sep 06, 2022 at 08:25:22AM +0200, Christoph Hellwig wrote:
>>On Mon, Sep 05, 2022 at 07:18:32PM +0530, Kanchan Joshi wrote:
>>>+static struct bio *bio_map_get(struct request *rq, unsigned int nr_vecs,
>>> 		gfp_t gfp_mask)
>>> {
>>>@@ -259,13 +252,31 @@ static int bio_map_user_iov(struct request *rq, struct iov_iter *iter,
>>> 		bio = bio_alloc_bioset(NULL, nr_vecs, opf, gfp_mask,
>>> 					&fs_bio_set);
>>> 		if (!bio)
>>>-			return -ENOMEM;
>>>+			return NULL;
>>
>>This context looks weird?  That bio_alloc_bioset should not be there,
>>as biosets are only used for file system I/O, which this is not.
>
>if you think it's a deal-breaker, maybe I can add a new bioset in nvme and
>pass that as argument to this helper. Would you prefer that over the
>current approach.

seems I responded without looking carefully. The bioset addition is not
part of this series. It got added earlier [1] [2], as part of optimizing
polled-io on file/passthru path.

[1] https://lore.kernel.org/linux-block/20220806152004.382170-3-axboe@kernel.dk/
[2] https://lore.kernel.org/linux-block/f2863702-e54c-cd74-efcf-8cb238be1a7c@kernel.dk/