On Fri, Apr 07, 2023 at 12:16:33PM -0700, Keith Busch wrote: >From: Keith Busch > >Set the bio's bi_end_io to handle the cleanup so that uring_cmd doesn't >need this complex pdu->{bio,req} switchero and restore. > >Signed-off-by: Keith Busch >--- > drivers/nvme/host/ioctl.c | 26 +++++++++----------------- > 1 file changed, 9 insertions(+), 17 deletions(-) > >diff --git a/drivers/nvme/host/ioctl.c b/drivers/nvme/host/ioctl.c >index d24ea2e051564..278c57ee0db91 100644 >--- a/drivers/nvme/host/ioctl.c >+++ b/drivers/nvme/host/ioctl.c >@@ -159,6 +159,11 @@ static struct request *nvme_alloc_user_request(struct request_queue *q, > return req; > } > >+static void nvme_uring_bio_end_io(struct bio *bio) >+{ >+ blk_rq_unmap_user(bio); >+} >+ > static int nvme_map_user_request(struct request *req, u64 ubuffer, > unsigned bufflen, void __user *meta_buffer, unsigned meta_len, > u32 meta_seed, void **metap, struct io_uring_cmd *ioucmd, >@@ -204,6 +209,7 @@ static int nvme_map_user_request(struct request *req, u64 ubuffer, > *metap = meta; > } > >+ bio->bi_end_io = nvme_uring_bio_end_io; > return ret; > > out_unmap: >@@ -249,8 +255,6 @@ static int nvme_submit_user_cmd(struct request_queue *q, > if (meta) > ret = nvme_finish_user_metadata(req, meta_buffer, meta, > meta_len, ret); >- if (bio) >- blk_rq_unmap_user(bio); Is it safe to call blk_rq_unamp_user in irq context? Agree that current code does some complex stuff, but that's to ensure what the code-comment [1] says. Also for polled-io, new-code will hit this warn-on [2] on calling bio_put_percpu_cache. [1] 623 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while 624 * still in process context. 625 */ 626 int blk_rq_map_user_iov(struct request_queue *q, struct request *rq, 627 struct rq_map_data *map_data, [2] 773 if ((bio->bi_opf & REQ_POLLED) && !WARN_ON_ONCE(in_interrupt())) { 774 bio->bi_next = cache->free_list;