Re: [PATCH net-next 2/5] net: dpaa2-mac: retrieve MAC statistics in one firmware command

[Simon Horman <horms@kernel.org>]

On Wed, Feb 25, 2026 at 05:06:45PM +0200, Ioana Ciornei wrote:

...

> diff --git a/drivers/net/ethernet/freescale/dpaa2/dpaa2-mac.c b/drivers/net/ethernet/freescale/dpaa2/dpaa2-mac.c

...

> +static void dpaa2_mac_setup_stats(struct dpaa2_mac *mac, struct dpaa2_mac_stats *stats,
> +				  size_t num_stats, const struct dpmac_counter *counters)
> +{
> +	struct device *dev = mac->net_dev->dev.parent;
> +	u32 *cnt_idx;

Hi Ioana,

The type of cnt_idx is u32.

> +
> +	stats->idx_dma_mem = kcalloc(num_stats, sizeof(u32), GFP_KERNEL);
> +	if (!stats->idx_dma_mem)
> +		goto out;
> +
> +	stats->values_dma_mem = kcalloc(num_stats, sizeof(u64), GFP_KERNEL);
> +	if (!stats->values_dma_mem)
> +		goto err_alloc_values;
> +
> +	cnt_idx = stats->idx_dma_mem;

As is that of idx_dma_mem. So the types match here.

> +	for (size_t i = 0; i < num_stats; i++)
> +		*cnt_idx++ = cpu_to_le32((u32)(counters[i].id));

But here __le32 values are assigned to elements of cnt_idx.

I think that the type of both cnt_idx and stats->idx_dma_mem
should probably be __le32 * rather than u32 *.

Flagged by Sparse v0.6.5-rc1.


...

>  void dpaa2_mac_get_ethtool_stats(struct dpaa2_mac *mac, u64 *data)
>  {
> +	struct device *dev = mac->net_dev->dev.parent;
>  	struct fsl_mc_device *dpmac_dev = mac->mc_dev;
> +	u64 *cnt_values;
>  	int i, err;
>  	u64 value;

...

> +	cnt_values = mac->ethtool_stats.values_dma_mem;
> +	for (i = 0; i < DPAA2_MAC_NUM_ETHTOOL_STATS; i++)
> +		*(data + i) = le64_to_cpu(*cnt_values++);

Likewise, I think the type of both cnt_values and
mac->ethtool_stats.values_dma_mem should be __le64 8 rather than u64 *.

And there is a similar problem in patch 3/5 centering on the use of
le64_to_cpu() in dpaa2_mac_transfer_stats().

Also flagged by Sparse.

...