Skip to contentSkip to conversation

Fix a regression in 6.13 kernel

PATCH Applied Maurizio Lombardi 3 participants 4 messages Mar 31, 2025 – Apr 3, 2025 NVMe

lore
Share & tools
lore
#0 lore raw

cover Fix a regression in 6.13 kernel

Some time ago Mikhail Gavrilov reported a regression against kernel 6.13 introduced by commit beadf0088501d9dcf2 ("nvme-pci: reverse request order in nvme_queue_rqs")

https://lore.kernel.org/linux-nvme/CABXGCsMcxu3pCF8jYPeqF_jN34saBwc8Fci+c-Dg2Lh7rqvuFQ@mail.gmail.com/

I was not able to reproduce the bug, but Mikhail reported that this patch fixed it for him:
https://lore.kernel.org/linux-nvme/CABXGCsMd_xv8jPDF_sFYhwd8GtANZ23nbaSJuCxQRO7cjPtgWg@mail.gmail.com/

I think the problem here is that nvme_write_sq_db() ends up being called even when the rqlist is empty and this couldn't happen before the changes introduced by beadf0088501d9dcf2

Maurizio Lombardi (1):
nvme-pci: do not call nvme_write_sq_db() if the commands list is empty

 drivers/nvme/host/pci.c | 3 +++
 1 file changed, 3 insertions(+)
Signature · 1 line
2.43.5
#1 lore raw

PATCH 1/1 nvme-pci: do not call nvme_write_sq_db() if the commands list is empty

nvme_submit_cmds() should check the rqlist
before calling nvme_write_sq_db();
if the list is empty, it must return immediately.

Fixes beadf0088501 ("nvme-pci: reverse request order in nvme_queue_rqs")
---
 drivers/nvme/host/pci.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/nvme/host/pci.c +3 -0
index 2883d17ee1eb..b178d52eac1b 100644
@@ -986,6 +986,9 @@ static void nvme_submit_cmds(struct nvme_queue *nvmeq, struct rq_list *rqlist)
 {
 	struct request *req;
 
+	if (rq_list_empty(rqlist))
+		return;
+
 	spin_lock(&nvmeq->sq_lock);
 	while ((req = rq_list_pop(rqlist))) {
 		struct nvme_iod *iod = blk_mq_rq_to_pdu(req);
Signature · 1 line
2.43.5
maintainer Applied
#2 lore raw

Re: [PATCH 1/1] nvme-pci: do not call nvme_write_sq_db() if the commands list is empty

Quoted 3 lines · Maurizio Lombardi

nvme_submit_cmds() should check the rqlist
before calling nvme_write_sq_db();
if the list is empty, it must return immediately.

I believe that could cause the driver to write the same value into a sq doorbell twice, which is a spec violation.

Thanks, applied.

maintainer Reviewed-by on v1 1/1
#3 lore raw

Re: [PATCH 1/1] nvme-pci: do not call nvme_write_sq_db() if the commands list is empty

In reply to Maurizio Lombardi
Quoted 3 lines · Maurizio Lombardi

nvme_submit_cmds() should check the rqlist
before calling nvme_write_sq_db();
if the list is empty, it must return immediately.

It would be useful for future readers to state why it is important.

Otherwise looks good: