On 1/11/21 21:37, Damien Le Moal wrote:

quoted

diff --git a/drivers/nvme/target/nvmet.h b/drivers/nvme/target/nvmet.h
index 7361665585a2..3fc84f79cce1 100644
--- a/drivers/nvme/target/nvmet.h
+++ b/drivers/nvme/target/nvmet.h

@@ -652,4 +652,20 @@ nvmet_bdev_execute_zone_append(struct nvmet_req *req)
 }
 #endif /* CONFIG_BLK_DEV_ZONED */
 
+static inline struct bio *nvmet_req_bio_get(struct nvmet_req *req,
+					    bio_end_io_t *bi_end_io)
+{
+	struct bio *bio;
+
+	if (req->transfer_len <= NVMET_MAX_INLINE_DATA_LEN) {
+		bio = &req->b.inline_bio;
+		bio_init(bio, req->inline_bvec, ARRAY_SIZE(req->inline_bvec));
+		return bio;
+	}
+
+	bio = bio_alloc(GFP_KERNEL, req->sg_cnt);

I have a doubt about the use of GFP_KERNEL here... Shouldn't these be GFP_NOIO ?
The code was like this so it is may be OK, but without GFP_NOIO, is forward
progress guaranteed ? No recursion possible ?

I've kept the original behavior, let check if this needs to be
GFP_KERNEL or not,
if so I'll send a separate patch with a proper message.

Thanks for pointing this out.