Re: [net-next,v20,3/7] net: mtip: Add buffers management functions to the L2 switch driver

Re: [net-next,v20,3/7] net: mtip: Add buffers management functions to the L2 switch driver

[Jakub Kicinski]

This is an AI-generated review of your patch. The human sending this
email has considered the AI review valid, or at least plausible.

For local reproduction steps see https://netdev-ai.bots.linux.dev/ai-local.html
---
net: mtip: Add buffers management functions to the L2 switch driver

This patch adds buffer management functions for the MTIP L2 switch
driver, implementing mtip_free_buffers(), mtip_create_page_pool(), and
mtip_alloc_buffers() for RX page pool and TX bounce buffer handling.

> diff --git a/drivers/net/ethernet/freescale/mtipsw/mtipl2sw.c b/drivers/net/ethernet/freescale/mtipsw/mtipl2sw.c
> index 03c5163d6508..9c21a3d29c1c 100644
> --- a/drivers/net/ethernet/freescale/mtipsw/mtipl2sw.c
> +++ b/drivers/net/ethernet/freescale/mtipsw/mtipl2sw.c
> @@ -888,11 +888,96 @@ static void mtip_get_drvinfo(struct net_device *dev,
>
>  static void mtip_free_buffers(struct net_device *dev)
>  {
> +	struct mtip_ndev_priv *priv = netdev_priv(dev);
> +	struct switch_enet_private *fep = priv->fep;
> +	int i;
> +
> +	for (i = 0; i < RX_RING_SIZE; i++) {
> +		page_pool_put_full_page(fep->page_pool,
> +					fep->page[i], false);

Can fep->page[i] be NULL here?  When mtip_alloc_buffers() fails partway
through the RX allocation loop, it calls mtip_free_buffers() via the
error path.  At that point, only some fep->page[] entries have been
allocated - the rest remain NULL from the kzalloc initialization.

The call chain page_pool_put_full_page()->page_pool_put_netmem()->
page_pool_unref_and_test()->page_pool_unref_netmem()->
netmem_get_pp_ref_count_ref() will attempt to access &NULL->pp_ref_count
when passed a NULL page, causing a NULL pointer dereference.

Should there be a NULL check before calling page_pool_put_full_page()?

> +		fep->page[i] = NULL;
> +	}
> +
> +	page_pool_destroy(fep->page_pool);
> +	fep->page_pool = NULL;
> +
> +	for (i = 0; i < TX_RING_SIZE; i++)
> +		kfree(fep->tx_bounce[i]);
>  }

[ ... ]

> +static int mtip_alloc_buffers(struct net_device *dev)
> +{
> +	struct mtip_ndev_priv *priv = netdev_priv(dev);
> +	struct switch_enet_private *fep = priv->fep;
> +	struct cbd_t *bdp;
> +	struct page *page;
> +	int i, ret;
> +
> +	ret = mtip_create_page_pool(fep, RX_RING_SIZE);
> +	if (ret < 0) {
> +		dev_err(&fep->pdev->dev, "Failed to create page pool\n");
> +		return ret;
> +	}
> +
> +	bdp = fep->rx_bd_base;
> +	for (i = 0; i < RX_RING_SIZE; i++) {
> +		page = page_pool_dev_alloc_pages(fep->page_pool);
> +		if (!page) {
> +			dev_err(&fep->pdev->dev,
> +				"Failed to allocate page for rx buffer\n");
> +			goto err;
> +		}

[ ... ]

> + err:
> +	mtip_free_buffers(dev);

When this error path is taken, fep->page[] may contain NULL entries for
pages that were never allocated.

> +	return -ENOMEM;
>  }

Re: [net-next,v20,3/7] net: mtip: Add buffers management functions to the L2 switch driver

[Łukasz Majewski]

Hi Jakub,

> This is an AI-generated review of your patch. The human sending this
> email has considered the AI review valid, or at least plausible.
> 
> For local reproduction steps see
> https://netdev-ai.bots.linux.dev/ai-local.html ---
> net: mtip: Add buffers management functions to the L2 switch driver
> 
> This patch adds buffer management functions for the MTIP L2 switch
> driver, implementing mtip_free_buffers(), mtip_create_page_pool(), and
> mtip_alloc_buffers() for RX page pool and TX bounce buffer handling.
> 
> > diff --git a/drivers/net/ethernet/freescale/mtipsw/mtipl2sw.c
> > b/drivers/net/ethernet/freescale/mtipsw/mtipl2sw.c index
> > 03c5163d6508..9c21a3d29c1c 100644 ---
> > a/drivers/net/ethernet/freescale/mtipsw/mtipl2sw.c +++
> > b/drivers/net/ethernet/freescale/mtipsw/mtipl2sw.c @@ -888,11
> > +888,96 @@ static void mtip_get_drvinfo(struct net_device *dev,
> >
> >  static void mtip_free_buffers(struct net_device *dev)
> >  {
> > +	struct mtip_ndev_priv *priv = netdev_priv(dev);
> > +	struct switch_enet_private *fep = priv->fep;
> > +	int i;
> > +
> > +	for (i = 0; i < RX_RING_SIZE; i++) {
> > +		page_pool_put_full_page(fep->page_pool,
> > +					fep->page[i], false);  
> 
> Can fep->page[i] be NULL here?  When mtip_alloc_buffers() fails
> partway through the RX allocation loop, it calls mtip_free_buffers()
> via the error path.  At that point, only some fep->page[] entries
> have been allocated - the rest remain NULL from the kzalloc
> initialization.
> 
> The call chain page_pool_put_full_page()->page_pool_put_netmem()->
> page_pool_unref_and_test()->page_pool_unref_netmem()->
> netmem_get_pp_ref_count_ref() will attempt to access
> &NULL->pp_ref_count when passed a NULL page, causing a NULL pointer
> dereference.
> 
> Should there be a NULL check before calling page_pool_put_full_page()?
> 

Yes, such check shall be added.

> > +		fep->page[i] = NULL;
> > +	}
> > +
> > +	page_pool_destroy(fep->page_pool);
> > +	fep->page_pool = NULL;
> > +
> > +	for (i = 0; i < TX_RING_SIZE; i++)
> > +		kfree(fep->tx_bounce[i]);
> >  }  
> 
> [ ... ]
> 
> > +static int mtip_alloc_buffers(struct net_device *dev)
> > +{
> > +	struct mtip_ndev_priv *priv = netdev_priv(dev);
> > +	struct switch_enet_private *fep = priv->fep;
> > +	struct cbd_t *bdp;
> > +	struct page *page;
> > +	int i, ret;
> > +
> > +	ret = mtip_create_page_pool(fep, RX_RING_SIZE);
> > +	if (ret < 0) {
> > +		dev_err(&fep->pdev->dev, "Failed to create page
> > pool\n");
> > +		return ret;
> > +	}
> > +
> > +	bdp = fep->rx_bd_base;
> > +	for (i = 0; i < RX_RING_SIZE; i++) {
> > +		page = page_pool_dev_alloc_pages(fep->page_pool);
> > +		if (!page) {
> > +			dev_err(&fep->pdev->dev,
> > +				"Failed to allocate page for rx
> > buffer\n");
> > +			goto err;
> > +		}  
> 
> [ ... ]
> 
> > + err:
> > +	mtip_free_buffers(dev);  
> 
> When this error path is taken, fep->page[] may contain NULL entries
> for pages that were never allocated.
> 
> > +	return -ENOMEM;
> >  }  

I will add proper fix for v21.

-- 
Best regards,

Łukasz Majewski