[PATCH] nvme: enable generic interface (/dev/ngX) for unknown command sets

[Christoph Hellwig]

On Fri, Mar 25, 2022 at 04:44:44PM +0530, Kanchan Joshi wrote:
> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
> index 677fa4bf76d3..3e5d9a3f4167 100644
> --- a/drivers/nvme/host/core.c
> +++ b/drivers/nvme/host/core.c
> @@ -1875,7 +1875,7 @@ static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
>  static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
>  {
>  	unsigned lbaf = nvme_lbaf_index(id->flbas);
> -	int ret;
> +	int ret = 0;
>  
>  	blk_mq_freeze_queue(ns->disk->queue);
>  	ns->lba_shift = id->lbaf[lbaf].ds;

nvme_update_ns_info can't really work for arbitrary command sets as
the concept of LBA sizes and format doesn't apply there.  So we'll
need a different stubbed out version that doesn't deal with the
block interface at all.

> @@ -4098,6 +4100,14 @@ static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
>  	default:
>  		dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
>  			ids.csi, nsid);
> +		if (!nvme_multi_css(ctrl)) {
> +			dev_warn(ctrl->device,
> +				"command set not reported for nsid: %d\n",
> +				nsid);
> +			break;
> +		}

Given that the NVM command set has a CSI of 0 getting random garbage
here sounds unlikely.  That being said if we want to support passthrough
for unknown command sets the warning above should probably go
away above as well.

> +		/* required to enable char-interface for unknown command sets*/
> +		nvme_alloc_ns(ctrl, nsid, &ids);

And this can't really work as nvme_alloc_ns looks at the Identify
Namespace structure for the NVM command set which can't really
work for arbitrary command sets.  We'll need a nvme_alloc_ns_generic
instead that looks at the Command Set Independent Identify Namespace
structure for a few things like namespace sharing capability and
does just the basic work to get the passthrough interface up.