[PATCH] nvme-multipath: allocate MQ disk

[PATCH] nvme-multipath: allocate MQ disk

[Vitaly Mayatskikh]

nvme_mpath_alloc_disk() allocates gendisk with blk_alloc_disk(), which
results in no MQ support even if the head device has it.

Signed-off-by: Vitaly Mayatskikh <vitaly@enfabrica.net>
---
 drivers/nvme/host/multipath.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/nvme/host/multipath.c b/drivers/nvme/host/multipath.c
index d3e2440d8abb..3f085f838910 100644
--- a/drivers/nvme/host/multipath.c
+++ b/drivers/nvme/host/multipath.c
@@ -490,7 +490,7 @@ int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl, struct nvme_ns_head *head)
 	    !nvme_is_unique_nsid(ctrl, head) || !multipath)
 		return 0;
 
-	head->disk = blk_alloc_disk(ctrl->numa_node);
+	head->disk = blk_mq_alloc_disk(ctrl->tagset, NULL);
 	if (!head->disk)
 		return -ENOMEM;
 	head->disk->fops = &nvme_ns_head_ops;
-- 
2.25.1

Re: [PATCH] nvme-multipath: allocate MQ disk

[Keith Busch]

On Thu, Jun 09, 2022 at 01:37:34PM -0400, Vitaly Mayatskikh wrote:
> @@ -490,7 +490,7 @@ int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl, struct nvme_ns_head *head)
>  	    !nvme_is_unique_nsid(ctrl, head) || !multipath)
>  		return 0;
>  
> -	head->disk = blk_alloc_disk(ctrl->numa_node);
> +	head->disk = blk_mq_alloc_disk(ctrl->tagset, NULL);

No, we don't use request based queues for stacking drivers. Especially not one
that re-uses the same tagset with the lower layer it stacks on.

Re: [PATCH] nvme-multipath: allocate MQ disk

[Vitaly Mayatskikh]

On Thu, Jun 9, 2022 at 1:49 PM Keith Busch <kbusch@kernel.org> wrote:

> No, we don't use request based queues for stacking drivers. Especially not one
> that re-uses the same tagset with the lower layer it stacks on.

Yeah, I screwed up my test bench and misdiagnosed the problem. Sorry about that.

Re: [PATCH] nvme-multipath: allocate MQ disk

[Christoph Hellwig]

On Thu, Jun 09, 2022 at 01:37:34PM -0400, Vitaly Mayatskikh wrote:
> nvme_mpath_alloc_disk() allocates gendisk with blk_alloc_disk(), which
> results in no MQ support even if the head device has it.

This can't work.  Can you explain hw you came up with this idea and
why you decided to send it out without any testing at all?

Re: [PATCH] nvme-multipath: allocate MQ disk

[Vitaly Mayatskikh]

On Fri, Jun 10, 2022 at 1:59 AM Christoph Hellwig <hch@lst.de> wrote:

> This can't work.  Can you explain hw you came up with this idea and
> why you decided to send it out without any testing at all?

That was a mistake in my setup that I misdiagnosed. Sorry.

The patch actually worked (as in "nothing crashed") in the simplest test.