Re: [PATCH v3] can: mcp251xfd: fix infinite loop when xmit fails

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 11.03.2024 12:11:43, Vitor Soares wrote:
> From: Vitor Soares <vitor.soares@xxxxxxxxxxx>
> 
> When the mcp251xfd_start_xmit() function fails, the driver stops
> processing messages, and the interrupt routine does not return,
> running indefinitely even after killing the running application.
> 
> Error messages:
> [  441.298819] mcp251xfd spi2.0 can0: ERROR in mcp251xfd_start_xmit: -16
> [  441.306498] mcp251xfd spi2.0 can0: Transmit Event FIFO buffer not empty. (seq=0x000017c7, tef_tail=0x000017cf, tef_head=0x000017d0, tx_head=0x000017d3).
> ... and repeat forever.
> 
> The issue can be triggered when multiple devices share the same
> SPI interface. And there is concurrent access to the bus.
> 
> The problem occurs because tx_ring->head increments even if
> mcp251xfd_start_xmit() fails. Consequently, the driver skips one
> TX package while still expecting a response in
> mcp251xfd_handle_tefif_one().
> 
> This patch resolves the issue by decreasing tx_ring->head if
> mcp251xfd_start_xmit() fails. With the fix, if we trigger the issue and
> the err = -EBUSY, the driver returns NETDEV_TX_BUSY. The network stack
> retries to transmit the message.
> Otherwise, it prints an error and discards the message.
> 
> Fixes: 55e5b97f003e ("can: mcp25xxfd: add driver for Microchip MCP25xxFD SPI CAN")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Vitor Soares <vitor.soares@xxxxxxxxxxx>
> ---
> 
> V2->V3:
>   - Add tx_dropped stats.
>   - netdev_sent_queue() only if can_put_echo_skb() succeed.
> 
> V1->V2:
>   - Return NETDEV_TX_BUSY if mcp251xfd_tx_obj_write() == -EBUSY.
>   - Rework the commit message to address the change above.
>   - Change can_put_echo_skb() to be called after mcp251xfd_tx_obj_write() succeed.
>     Otherwise, we get Kernel NULL pointer dereference error.
> 
>  drivers/net/can/spi/mcp251xfd/mcp251xfd-tx.c | 34 ++++++++++++--------
>  1 file changed, 21 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/net/can/spi/mcp251xfd/mcp251xfd-tx.c b/drivers/net/can/spi/mcp251xfd/mcp251xfd-tx.c
> index 160528d3cc26..146c44e47c60 100644
> --- a/drivers/net/can/spi/mcp251xfd/mcp251xfd-tx.c
> +++ b/drivers/net/can/spi/mcp251xfd/mcp251xfd-tx.c
> @@ -166,6 +166,7 @@ netdev_tx_t mcp251xfd_start_xmit(struct sk_buff *skb,
>  				 struct net_device *ndev)
>  {
>  	struct mcp251xfd_priv *priv = netdev_priv(ndev);
> +	struct net_device_stats *stats = &ndev->stats;
>  	struct mcp251xfd_tx_ring *tx_ring = priv->tx;
>  	struct mcp251xfd_tx_obj *tx_obj;
>  	unsigned int frame_len;
> @@ -181,25 +182,32 @@ netdev_tx_t mcp251xfd_start_xmit(struct sk_buff *skb,
>  	tx_obj = mcp251xfd_get_tx_obj_next(tx_ring);
>  	mcp251xfd_tx_obj_from_skb(priv, tx_obj, skb, tx_ring->head);
>  
> -	/* Stop queue if we occupy the complete TX FIFO */
>  	tx_head = mcp251xfd_get_tx_head(tx_ring);
> -	tx_ring->head++;
> -	if (mcp251xfd_get_tx_free(tx_ring) == 0)
> -		netif_stop_queue(ndev);
> -
>  	frame_len = can_skb_get_frame_len(skb);
> -	err = can_put_echo_skb(skb, ndev, tx_head, frame_len);
> -	if (!err)
> -		netdev_sent_queue(priv->ndev, frame_len);
> +
> +	tx_ring->head++;
>  
>  	err = mcp251xfd_tx_obj_write(priv, tx_obj);
> -	if (err)
> -		goto out_err;
> +	if (err) {
> +		tx_ring->head--;
>  
> -	return NETDEV_TX_OK;
> +		if (err == -EBUSY)
> +			return NETDEV_TX_BUSY;
>  
> - out_err:
> -	netdev_err(priv->ndev, "ERROR in %s: %d\n", __func__, err);
> +		stats->tx_dropped++;
> +
> +		if (net_ratelimit())
> +			netdev_err(priv->ndev,
> +				   "ERROR in %s: %d\n", __func__, err);
> +	} else {
> +		err = can_put_echo_skb(skb, ndev, tx_head, frame_len);
> +		if (!err)
> +			netdev_sent_queue(priv->ndev, frame_len);
> +
> +		/* Stop queue if we occupy the complete TX FIFO */
> +		if (mcp251xfd_get_tx_free(tx_ring) == 0)
> +			netif_stop_queue(ndev);
> +	}

Once you have sent the TX objects to the chip, it can trigger a TX
complete IRQ. This is not very likely as the SPI sends asynchronously
and the xmit handler cannot sleep.

This means you have to call can_put_echo_skb() and stop the queue if
needed, _before_ you call mcp251xfd_tx_obj_write(). In case of an error,
you have to roll-back these.

regards,
Marc

-- 
Pengutronix e.K.                 | Marc Kleine-Budde          |
Embedded Linux                   | https://www.pengutronix.de |
Vertretung Nürnberg              | Phone: +49-5121-206917-129 |
Amtsgericht Hildesheim, HRA 2686 | Fax:   +49-5121-206917-9   |

Attachment: signature.asc
Description: PGP signature


[Index of Archives]     [Automotive Discussions]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Security]     [Bugtraq]     [Linux]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]     [CAN Bus]

  Powered by Linux