Skip to contentSkip to conversation

nvme-pci: check that sqid match in nvme cqe

PATCH New Max Gurtovoy 2 participants 5 messages Jan 1, 2019 – Jan 9, 2019 NVMe

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 nvme-pci: check that sqid match in nvme cqe

This patch protects against faulty drives (e.g: a drive might send the completion on a wrong msix).

Signed-off-by Max Gurtovoy <maxg at mellanox.com>
---
 drivers/nvme/host/pci.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/nvme/host/pci.c +7 -0
index 5a0bf6a..ab7ff34 100644
@@ -983,6 +983,13 @@ static inline void nvme_handle_cqe(struct nvme_queue *nvmeq, u16 idx)
 	volatile struct nvme_completion *cqe = &nvmeq->cqes[idx];
 	struct request *req;
 
+	if (unlikely(le16_to_cpu(cqe->sq_id) != nvmeq->qid)) {
+		dev_warn(nvmeq->dev->ctrl.device,
+			 "got completion on sqid %d instead of sqid %d\n",
+			 nvmeq->qid, le16_to_cpu(cqe->sq_id));
+		return;
+	}
+
 	if (unlikely(cqe->command_id >= nvmeq->q_depth)) {
 		dev_warn(nvmeq->dev->ctrl.device,
 			"invalid id %d completed on queue %d\n",
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH 1/1 nvme-pci: check that sqid match in nvme cqe

Quoted 2 lines · Max Gurtovoy

This patch protects against faulty drives (e.g: a drive might send the completion on a wrong msix).

Where/how do you see this issue?

#2 lore raw

PATCH 1/1 nvme-pci: check that sqid match in nvme cqe

Quoted 4 lines · Christoph Hellwig
Quoted 2 lines · Max Gurtovoy

This patch protects against faulty drives (e.g: a drive might send the completion on a wrong msix).

Where/how do you see this issue?

maybe it wasn't a good example but let's say I can change 1 line in QEMU code (@hw/block/nvme.c) to trigger this issue.

if I'll change

req->cqe.sq_id = cpu_to_le16(sq->sqid);

to be

req->cqe.sq_id = 0;

we'll always get wrong sq_id except of admin queue commands.

This patch tries to protect against HW/FW bugs and we need to decide if it's in of our driver interest to do so.

if not, let's abandon this commit.

#3 lore raw

PATCH 1/1 nvme-pci: check that sqid match in nvme cqe

Quoted 9 lines · Max Gurtovoy

maybe it wasn't a good example but let's say I can change 1 line in QEMU code (@hw/block/nvme.c) to trigger this issue.

if I'll change

req->cqe.sq_id = cpu_to_le16(sq->sqid);

to be

Quoted 8 lines

req->cqe.sq_id = 0;

we'll always get wrong sq_id except of admin queue commands.

This patch tries to protect against HW/FW bugs and we need to decide if it's in of our driver interest to do so.

if not, let's abandon this commit.

Yes, but hardware that gets this wrong will just blow up and timeout anyway. I'm not sure this case is worth adding a special detetion in the slow path.

#4 lore raw

PATCH 1/1 nvme-pci: check that sqid match in nvme cqe

Quoted 3 lines · Christoph Hellwig

Yes, but hardware that gets this wrong will just blow up and timeout anyway. I'm not sure this case is worth adding a special detetion in the slow path.

s/slow/fast/