RE: [PATCH 06/14] net: axienet: Check for DMA mapping errors

From: Radhey Shyam Pandey
Date: Mon Jan 13 2020 - 00:54:48 EST


> -----Original Message-----
> From: Andre Przywara <andre.przywara@xxxxxxx>
> Sent: Friday, January 10, 2020 5:24 PM
> To: David S . Miller <davem@xxxxxxxxxxxxx>; Radhey Shyam Pandey
> <radheys@xxxxxxxxxx>
> Cc: Michal Simek <michals@xxxxxxxxxx>; Robert Hancock
> <hancock@xxxxxxxxxxxxx>; netdev@xxxxxxxxxxxxxxx; linux-arm-
> kernel@xxxxxxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
> Subject: [PATCH 06/14] net: axienet: Check for DMA mapping errors
>
> Especially with the default 32-bit DMA mask, DMA buffers are a limited
> resource, so their allocation can fail.
> So as the DMA API documentation requires, add error checking code after
> dma_map_single() calls to catch the case where we run out of "low" memory.
>
> Signed-off-by: Andre Przywara <andre.przywara@xxxxxxx>
> ---
> .../net/ethernet/xilinx/xilinx_axienet_main.c | 22 ++++++++++++++++++-
> 1 file changed, 21 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
> b/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
> index 82abe2b0f16a..8d2b67cbecf9 100644
> --- a/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
> +++ b/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
> @@ -248,6 +248,11 @@ static int axienet_dma_bd_init(struct net_device
> *ndev)
> skb->data,
> lp->max_frm_size,
> DMA_FROM_DEVICE);
> + if (dma_mapping_error(ndev->dev.parent, lp->rx_bd_v[i].phys))

Prefer using unlikely compiler hint for dma_mapping_error.
Also, we need to add error print to report this condition to the user,
in case it isn't there in dma_mapping_error implementation.

> {
> + dev_kfree_skb(skb);

free of skb is already handled in _release. We can reuse that?
> + goto out;
> + }
> +
> lp->rx_bd_v[i].cntrl = lp->max_frm_size;
> }
>
> @@ -668,6 +673,7 @@ axienet_start_xmit(struct sk_buff *skb, struct
> net_device *ndev)
> dma_addr_t tail_p;
> struct axienet_local *lp = netdev_priv(ndev);
> struct axidma_bd *cur_p;
> + u32 orig_tail_ptr = lp->tx_bd_tail;
>
> num_frag = skb_shinfo(skb)->nr_frags;
> cur_p = &lp->tx_bd_v[lp->tx_bd_tail];
> @@ -703,9 +709,11 @@ axienet_start_xmit(struct sk_buff *skb, struct
> net_device *ndev)
> cur_p->app0 |= 2; /* Tx Full Checksum Offload Enabled */
> }
>
> - cur_p->cntrl = skb_headlen(skb) | XAXIDMA_BD_CTRL_TXSOF_MASK;
> cur_p->phys = dma_map_single(ndev->dev.parent, skb->data,
> skb_headlen(skb), DMA_TO_DEVICE);
> + if (dma_mapping_error(ndev->dev.parent, cur_p->phys))
> + return NETDEV_TX_BUSY;

This is not ideally tx busy and related to available mem mapping in the system.
I just looked at other eth drivers and it seems they return TX_OK with drop
count stats incremented.

> + cur_p->cntrl = skb_headlen(skb) | XAXIDMA_BD_CTRL_TXSOF_MASK;
>
> for (ii = 0; ii < num_frag; ii++) {
> if (++lp->tx_bd_tail >= lp->tx_bd_num) @@ -716,6 +724,13 @@
> axienet_start_xmit(struct sk_buff *skb, struct net_device *ndev)
> skb_frag_address(frag),
> skb_frag_size(frag),
> DMA_TO_DEVICE);
> + if (dma_mapping_error(ndev->dev.parent, cur_p->phys)) {
> + axienet_free_tx_chain(ndev, orig_tail_ptr, ii + 1,
> + NULL);
> + lp->tx_bd_tail = orig_tail_ptr;
> +
> + return NETDEV_TX_BUSY;
> + }
> cur_p->cntrl = skb_frag_size(frag);
> }
>
> @@ -796,6 +811,11 @@ static void axienet_recv(struct net_device *ndev)
> cur_p->phys = dma_map_single(ndev->dev.parent, new_skb-
> >data,
> lp->max_frm_size,
> DMA_FROM_DEVICE);
> + if (dma_mapping_error(ndev->dev.parent, cur_p->phys)) {
> + dev_kfree_skb(new_skb);
> + return;
> + }
> +
> cur_p->cntrl = lp->max_frm_size;
> cur_p->status = 0;
> cur_p->skb = new_skb;
> --
> 2.17.1