[PATCH net] net: ag71xx: Add missing check after DMA map

[PATCH net] net: ag71xx: Add missing check after DMA map

[Thomas Fourier]

The DMA map functions can fail and should be tested for errors.

Fixes: d51b6ce441d3 ("net: ethernet: add ag71xx driver")
Signed-off-by: Thomas Fourier <fourier.thomas@gmail.com>
---
 drivers/net/ethernet/atheros/ag71xx.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/atheros/ag71xx.c b/drivers/net/ethernet/atheros/ag71xx.c
index d8e6f23e1432..0e68ab225e0a 100644
--- a/drivers/net/ethernet/atheros/ag71xx.c
+++ b/drivers/net/ethernet/atheros/ag71xx.c
@@ -1198,7 +1198,8 @@ static int ag71xx_buffer_size(struct ag71xx *ag)
 
 static bool ag71xx_fill_rx_buf(struct ag71xx *ag, struct ag71xx_buf *buf,
 			       int offset,
-			       void *(*alloc)(unsigned int size))
+			       void *(*alloc)(unsigned int size),
+			       void (*free)(void *))
 {
 	struct ag71xx_ring *ring = &ag->rx_ring;
 	struct ag71xx_desc *desc;
@@ -1213,6 +1214,11 @@ static bool ag71xx_fill_rx_buf(struct ag71xx *ag, struct ag71xx_buf *buf,
 	buf->rx.rx_buf = data;
 	buf->rx.dma_addr = dma_map_single(&ag->pdev->dev, data, ag->rx_buf_size,
 					  DMA_FROM_DEVICE);
+	if (dma_mapping_error(&ag->pdev->dev, buf->rx.dma_addr)) {
+		free(data);
+		buf->rx.rx_buf = NULL;
+		return false;
+	}
 	desc->data = (u32)buf->rx.dma_addr + offset;
 	return true;
 }
@@ -1241,7 +1247,7 @@ static int ag71xx_ring_rx_init(struct ag71xx *ag)
 		struct ag71xx_desc *desc = ag71xx_ring_desc(ring, i);
 
 		if (!ag71xx_fill_rx_buf(ag, &ring->buf[i], ag->rx_buf_offset,
-					netdev_alloc_frag)) {
+					netdev_alloc_frag, skb_free_frag)) {
 			ret = -ENOMEM;
 			break;
 		}
@@ -1275,7 +1281,7 @@ static int ag71xx_ring_rx_refill(struct ag71xx *ag)
 
 		if (!ring->buf[i].rx.rx_buf &&
 		    !ag71xx_fill_rx_buf(ag, &ring->buf[i], offset,
-					napi_alloc_frag))
+					napi_alloc_frag, skb_free_frag))
 			break;
 
 		desc->ctrl = DESC_EMPTY;
@@ -1511,6 +1517,10 @@ static netdev_tx_t ag71xx_hard_start_xmit(struct sk_buff *skb,
 
 	dma_addr = dma_map_single(&ag->pdev->dev, skb->data, skb->len,
 				  DMA_TO_DEVICE);
+	if (dma_mapping_error(&ag->pdev->dev, dma_addr)) {
+		netif_dbg(ag, tx_err, ndev, "DMA mapping error\n");
+		goto err_drop;
+	}
 
 	i = ring->curr & ring_mask;
 	desc = ag71xx_ring_desc(ring, i);
-- 
2.43.0

Re: [PATCH net] net: ag71xx: Add missing check after DMA map

[Simon Horman]

On Wed, Jul 02, 2025 at 09:50:46AM +0200, Thomas Fourier wrote:
> The DMA map functions can fail and should be tested for errors.
> 
> Fixes: d51b6ce441d3 ("net: ethernet: add ag71xx driver")
> Signed-off-by: Thomas Fourier <fourier.thomas@gmail.com>
> ---
>  drivers/net/ethernet/atheros/ag71xx.c | 16 +++++++++++++---
>  1 file changed, 13 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/net/ethernet/atheros/ag71xx.c b/drivers/net/ethernet/atheros/ag71xx.c
> index d8e6f23e1432..0e68ab225e0a 100644
> --- a/drivers/net/ethernet/atheros/ag71xx.c
> +++ b/drivers/net/ethernet/atheros/ag71xx.c
> @@ -1198,7 +1198,8 @@ static int ag71xx_buffer_size(struct ag71xx *ag)
>  
>  static bool ag71xx_fill_rx_buf(struct ag71xx *ag, struct ag71xx_buf *buf,
>  			       int offset,
> -			       void *(*alloc)(unsigned int size))
> +			       void *(*alloc)(unsigned int size),
> +			       void (*free)(void *))

FWIIW, I'm of two minds about this free parameter.
On the one hand I like the symmetry with free.
But on the other the two callers both pass
skb_free_frag as this argument.

But I lean towards what you have being a good choice.

Reviewed-by: Simon Horman <horms@kernel.org>

Re: [PATCH net] net: ag71xx: Add missing check after DMA map

[Paolo Abeni]

On 7/2/25 9:50 AM, Thomas Fourier wrote:
@@ -1275,7 +1281,7 @@ static int ag71xx_ring_rx_refill(struct ag71xx *ag)
>  
>  		if (!ring->buf[i].rx.rx_buf &&
>  		    !ag71xx_fill_rx_buf(ag, &ring->buf[i], offset,
> -					napi_alloc_frag))
> +					napi_alloc_frag, skb_free_frag))

I guess that if we merge this patch in the current form, we are going to
receive soon some automated-tool-generated patch removing the 'free'
argument.

It's probably better to avoid such argument from the start.

Thanks,

Paolo