[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20250617145944.10d444f4@kernel.org>
Date: Tue, 17 Jun 2025 14:59:44 -0700
From: Jakub Kicinski <kuba@...nel.org>
To: Thomas Fourier <fourier.thomas@...il.com>
Cc: Chris Snook <chris.snook@...il.com>, Andrew Lunn
<andrew+netdev@...n.ch>, "David S. Miller" <davem@...emloft.net>, Eric
Dumazet <edumazet@...gle.com>, Paolo Abeni <pabeni@...hat.com>, Ingo Molnar
<mingo@...nel.org>, Thomas Gleixner <tglx@...utronix.de>, Jeff Garzik
<jeff@...zik.org>, netdev@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH net] ethernet: alt1: Fix missing DMA mapping tests
On Mon, 16 Jun 2025 15:59:55 +0200 Thomas Fourier wrote:
> According to Shuah Khan[1], all `dma_map()` functions should be tested
> before using the pointer. This patch checks for errors after all
> `dma_map()` calls.
The link and reference to Shuah's presentation is not necessary.
Just say that the DMA functions can fail so we need the error check.
> In `atl1_alloc_rx_buffers()`, when the dma mapping fails,
> the buffer is deallocated ans marked as such.
>
> In `atl1_tx_map()`, the previously dma_mapped buffers are de-mapped and an
> error is returned.
>
> [1] https://events.static.linuxfound.org/sites/events/files/slides/Shuah_Khan_dma_map_error.pdf
>
> Fixes: f3cc28c79760 ("Add Attansic L1 ethernet driver.")
> Signed-off-by: Thomas Fourier <fourier.thomas@...il.com>
Please don't send the new versions in reply to old ones.
For those of us who use their inbox as a FIFO of pending submissions
it messes up ordering.
> diff --git a/drivers/net/ethernet/atheros/atlx/atl1.c b/drivers/net/ethernet/atheros/atlx/atl1.c
> index cfdb546a09e7..dd0e01d3a023 100644
> --- a/drivers/net/ethernet/atheros/atlx/atl1.c
> +++ b/drivers/net/ethernet/atheros/atlx/atl1.c
> @@ -1869,6 +1869,14 @@ static u16 atl1_alloc_rx_buffers(struct atl1_adapter *adapter)
> buffer_info->dma = dma_map_page(&pdev->dev, page, offset,
> adapter->rx_buffer_len,
> DMA_FROM_DEVICE);
> + if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
> + buffer_info->alloced = 0;
> + buffer_info->skb = NULL;
> + buffer_info->dma = 0;
maybe you could move the map call a little further up, before all the
buffer_info fields are populated? So we dont have to clean them up?
> + kfree_skb(skb);
> + adapter->soft_stats.rx_dropped++;
> + break;
> + }
> rfd_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
> rfd_desc->buf_len = cpu_to_le16(adapter->rx_buffer_len);
> rfd_desc->coalese = 0;
> @@ -2183,8 +2191,8 @@ static int atl1_tx_csum(struct atl1_adapter *adapter, struct sk_buff *skb,
> return 0;
> }
>
> -static void atl1_tx_map(struct atl1_adapter *adapter, struct sk_buff *skb,
> - struct tx_packet_desc *ptpd)
> +static int atl1_tx_map(struct atl1_adapter *adapter, struct sk_buff *skb,
> + struct tx_packet_desc *ptpd)
> {
> struct atl1_tpd_ring *tpd_ring = &adapter->tpd_ring;
> struct atl1_buffer *buffer_info;
> @@ -2194,6 +2202,7 @@ static void atl1_tx_map(struct atl1_adapter *adapter, struct sk_buff *skb,
> unsigned int nr_frags;
> unsigned int f;
> int retval;
> + u16 first_mapped;
> u16 next_to_use;
> u16 data_len;
> u8 hdr_len;
> @@ -2201,6 +2210,7 @@ static void atl1_tx_map(struct atl1_adapter *adapter, struct sk_buff *skb,
> buf_len -= skb->data_len;
> nr_frags = skb_shinfo(skb)->nr_frags;
> next_to_use = atomic_read(&tpd_ring->next_to_use);
> + first_mapped = next_to_use;
> buffer_info = &tpd_ring->buffer_info[next_to_use];
> BUG_ON(buffer_info->skb);
> /* put skb in last TPD */
> @@ -2216,6 +2226,8 @@ static void atl1_tx_map(struct atl1_adapter *adapter, struct sk_buff *skb,
> buffer_info->dma = dma_map_page(&adapter->pdev->dev, page,
> offset, hdr_len,
> DMA_TO_DEVICE);
> + if (dma_mapping_error(&adapter->pdev->dev, buffer_info->dma))
> + goto dma_err;
>
> if (++next_to_use == tpd_ring->count)
> next_to_use = 0;
> @@ -2242,6 +2254,9 @@ static void atl1_tx_map(struct atl1_adapter *adapter, struct sk_buff *skb,
> page, offset,
> buffer_info->length,
> DMA_TO_DEVICE);
> + if (dma_mapping_error(&adapter->pdev->dev,
> + buffer_info->dma))
> + goto dma_err;
> if (++next_to_use == tpd_ring->count)
> next_to_use = 0;
> }
> @@ -2254,6 +2269,8 @@ static void atl1_tx_map(struct atl1_adapter *adapter, struct sk_buff *skb,
> buffer_info->dma = dma_map_page(&adapter->pdev->dev, page,
> offset, buf_len,
> DMA_TO_DEVICE);
> + if (dma_mapping_error(&adapter->pdev->dev, buffer_info->dma))
> + goto dma_err;
> if (++next_to_use == tpd_ring->count)
> next_to_use = 0;
> }
> @@ -2277,6 +2294,9 @@ static void atl1_tx_map(struct atl1_adapter *adapter, struct sk_buff *skb,
> buffer_info->dma = skb_frag_dma_map(&adapter->pdev->dev,
> frag, i * ATL1_MAX_TX_BUF_LEN,
> buffer_info->length, DMA_TO_DEVICE);
> + if (dma_mapping_error(&adapter->pdev->dev,
> + buffer_info->dma))
> + goto dma_err;
>
> if (++next_to_use == tpd_ring->count)
> next_to_use = 0;
> @@ -2285,6 +2305,22 @@ static void atl1_tx_map(struct atl1_adapter *adapter, struct sk_buff *skb,
>
> /* last tpd's buffer-info */
> buffer_info->skb = skb;
> +
> + return 0;
> +
> + dma_err:
> + while (first_mapped != next_to_use) {
> + buffer_info = &tpd_ring->buffer_info[first_mapped];
> + dma_unmap_page(&adapter->pdev->dev,
> + buffer_info->dma,
> + buffer_info->length,
> + DMA_TO_DEVICE);
> + buffer_info->dma = 0;
> +
> + if (++first_mapped == tpd_ring->count)
> + first_mapped = 0;
> + }
> + return -ENOMEM;
> }
>
> static void atl1_tx_queue(struct atl1_adapter *adapter, u16 count,
> @@ -2419,7 +2455,8 @@ static netdev_tx_t atl1_xmit_frame(struct sk_buff *skb,
> }
> }
>
> - atl1_tx_map(adapter, skb, ptpd);
> + if (atl1_tx_map(adapter, skb, ptpd))
> + return NETDEV_TX_BUSY;
You should drop the packet. Occasional packet loss is better than
having a packet that we can't map for some platform reasons blocking
the queue
> atl1_tx_queue(adapter, count, ptpd);
> atl1_update_mailbox(adapter);
> return NETDEV_TX_OK;
Powered by blists - more mailing lists