nvmet: synchronize sqhd update

In testing target io in read write mix, we did indeed get into cases where
sqhd didn't update properly and slowly missed enough updates to shutdown
the queue.

Protect the updating sqhd by using cmpxchg, and for that turn the sqhd
field into a u32 so that cmpxchg works on it for all architectures.

Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

2 files changed, 13 insertions, 4 deletions

diff --git a/drivers/nvme/target/core.c b/drivers/nvme/target/core.c
index 1b208be..645ba7e 100644
--- a/drivers/nvme/target/core.c
+++ b/drivers/nvme/target/core.c
@@ -387,12 +387,21 @@ struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid)
 
 static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
 {
+	u32 old_sqhd, new_sqhd;
+	u16 sqhd;
+
 	if (status)
 		nvmet_set_status(req, status);
 
-	if (req->sq->size)
-		req->sq->sqhd = (req->sq->sqhd + 1) % req->sq->size;
-	req->rsp->sq_head = cpu_to_le16(req->sq->sqhd);
+	if (req->sq->size) {
+		do {
+			old_sqhd = req->sq->sqhd;
+			new_sqhd = (old_sqhd + 1) % req->sq->size;
+		} while (cmpxchg(&req->sq->sqhd, old_sqhd, new_sqhd) !=
+					old_sqhd);
+	}
+	sqhd = req->sq->sqhd & 0x0000FFFF;
+	req->rsp->sq_head = cpu_to_le16(sqhd);
 	req->rsp->sq_id = cpu_to_le16(req->sq->qid);
 	req->rsp->command_id = req->cmd->common.command_id;
 
diff --git a/drivers/nvme/target/nvmet.h b/drivers/nvme/target/nvmet.h
index 7b8e20a..87e429b 100644
--- a/drivers/nvme/target/nvmet.h
+++ b/drivers/nvme/target/nvmet.h
@@ -74,7 +74,7 @@ struct nvmet_sq {
 	struct percpu_ref	ref;
 	u16			qid;
 	u16			size;
-	u16			sqhd;
+	u32			sqhd;
 	struct completion	free_done;
 	struct completion	confirm_done;
 };