Re: [PATCH 09/17] nvmet: prepare metadata request

[Christoph Hellwig <hch@lst.de>]

On Fri, Mar 27, 2020 at 08:15:37PM +0300, Max Gurtovoy wrote:
> From: Israel Rukshin <israelr@mellanox.com>
> 
> Allocate the metadata SGL buffers and set metadata fields for the
> request. This is a preparation for adding metadata support over the
> fabrics.

I don't think having this as a separate prep patch is a good idea, it
should go with the code making use of it.

> -						       req->transfer_len);
> -			if (req->sg) {
> -				req->p2p_dev = p2p_dev;
> -				return 0;
> +						       data_len);
> +			if (!req->sg)
> +				goto fallback;
> +
> +			if (req->md_len) {
> +				req->md_sg =
> +					pci_p2pmem_alloc_sgl(p2p_dev,
> +							     &req->md_sg_cnt,
> +							     req->md_len);
> +				if (!req->md_sg) {
> +					pci_p2pmem_free_sgl(req->p2p_dev,
> +							    req->sg);
> +					goto fallback;
> +				}
>  			}
> +			req->p2p_dev = p2p_dev;
> +			return 0;
>  		}
>  
>  		/*
> @@ -984,23 +1001,40 @@ int nvmet_req_alloc_sgl(struct nvmet_req *req)
>  		 */
>  	}
>  
> -	req->sg = sgl_alloc(req->transfer_len, GFP_KERNEL, &req->sg_cnt);
> +fallback:
> +	req->sg = sgl_alloc(data_len, GFP_KERNEL, &req->sg_cnt);
>  	if (unlikely(!req->sg))
>  		return -ENOMEM;
>  
> +	if (req->md_len) {
> +		req->md_sg = sgl_alloc(req->md_len, GFP_KERNEL,
> +					 &req->md_sg_cnt);
> +		if (unlikely(!req->md_sg)) {
> +			sgl_free(req->sg);
> +			return -ENOMEM;
> +		}
> +	}
> +
>  	return 0;
>  }
>  EXPORT_SYMBOL_GPL(nvmet_req_alloc_sgl);

I suspect this might be a little cleaner with a nvmet_alloc_sgl
helper that returns a scatterlist instead of duplicating all the
code for data vs metadata.

_______________________________________________
linux-nvme mailing list
linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme