Re: [PATCH net-next 3/8] cadence: macb: Add page pool support handle multi-descriptor frame rx

["Théo Lebrun" <theo.lebrun@bootlin.com>]

Hi Paolo,

On Thu Jan 15, 2026 at 11:25 PM CET, Paolo Valerio wrote:
> Use the page pool allocator for the data buffers and enable skb recycling
> support, instead of relying on netdev_alloc_skb allocating the entire skb
> during the refill.
>
> The patch also add support for receiving network frames that span multiple
> DMA descriptors in the Cadence MACB/GEM Ethernet driver.
>
> The patch removes the requirement that limited frame reception to
> a single descriptor (RX_SOF && RX_EOF), also avoiding potential
> contiguous multi-page allocation for large frames.

[...]

> --- a/drivers/net/ethernet/cadence/macb.h
> +++ b/drivers/net/ethernet/cadence/macb.h
> @@ -14,6 +14,7 @@
>  #include <linux/interrupt.h>
>  #include <linux/phy/phy.h>
>  #include <linux/workqueue.h>
> +#include <net/page_pool/helpers.h>
>  
>  #define MACB_GREGS_NBR 16
>  #define MACB_GREGS_VERSION 2
> @@ -1266,6 +1267,8 @@ struct macb_queue {
>  	void			*rx_buffers;
>  	struct napi_struct	napi_rx;
>  	struct queue_stats stats;
> +	struct page_pool	*page_pool;
> +	struct sk_buff		*skb;

There might be a cleanup issue with the GRO queue->skb field.
It gets cleaned up by gem_rx() in
 - various error codepaths with the free_frags label, or,
 - the happy path where we hand off the skb with napi_gro_receive().

If we get closed in the middle of a transfer we have queue->skb that is
valid. It sounds like an issue if we shutdown (and leak the SKB) or if
we reopen (and have the old SKB laying around). I have not encountered
this on HW, but I haven't tried.

Do you agree? A job for gem_free_rx_buffers()?

Thanks,

--
Théo Lebrun, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com