Re: [PATCH v3 4/8] RDMA/bng_re: Allocate required memory resources for Firmware channel

[Simon Horman <horms@kernel.org>]

On Mon, Nov 17, 2025 at 05:11:22PM +0000, Siva Reddy Kallam wrote:

...

> +static void bng_re_dev_uninit(struct bng_re_dev *rdev)
> +{
> +	bng_re_free_rcfw_channel(&rdev->rcfw);
> +	bng_re_destroy_chip_ctx(rdev);
> +	if (test_and_clear_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags))
> +		bnge_unregister_dev(rdev->aux_dev);
> +}
> +
>  static int bng_re_dev_init(struct bng_re_dev *rdev)
>  {
>  	int rc;
> @@ -170,14 +184,18 @@ static int bng_re_dev_init(struct bng_re_dev *rdev)
>  
>  	bng_re_query_hwrm_version(rdev);
>  
> +	rc = bng_re_alloc_fw_channel(&rdev->bng_res, &rdev->rcfw);
> +	if (rc) {
> +		ibdev_err(&rdev->ibdev,
> +			  "Failed to allocate RCFW Channel: %#x\n", rc);
> +		goto fail;
> +	}
> +
>  	return 0;
> -}
>  
> -static void bng_re_dev_uninit(struct bng_re_dev *rdev)
> -{
> -	bng_re_destroy_chip_ctx(rdev);
> -	if (test_and_clear_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags))
> -		bnge_unregister_dev(rdev->aux_dev);
> +fail:
> +	bng_re_dev_uninit(rdev);
> +	return rc;

Hi Siva,

IMHO, I think that it would best to handle unwind using a ladder
of goto statements, that reverse the order of the incremental
initialisation performed by this function.

As is, this may not have much effect, other than seeming to duplicate
bng_re_dev_uninit(). But I think that as bng_re_dev_init() grows,
as it does in this patch-set, this will lead to clearer error handling
(and ideally a lower chance of bugs later).

I would also suggest that it would be best to name the label
after what tit does, rather than somewhat general name 'fail'.

>  }
>  
>  static int bng_re_add_device(struct auxiliary_device *adev)

...