On 08.02.2025 00:00:43, Vincent Mailhol wrote:

On 07/02/2025 at 21:15, Marc Kleine-Budde wrote:

quoted

On 07.02.2025 15:44:59, Ming Yu wrote:

(...)

quoted

quoted

+static netdev_tx_t nct6694_can_start_xmit(struct sk_buff *skb,
+					  struct net_device *ndev)
+{
+	struct nct6694_can_priv *priv = netdev_priv(ndev);
+
+	if (can_dev_dropped_skb(ndev, skb))
+		return NETDEV_TX_OK;
+
+	netif_stop_queue(ndev);
+	can_put_echo_skb(skb, ndev, 0, 0);
+	queue_work(priv->wq, &priv->tx_work);

What is the reason to use a work queue here? xmit() is not a hard IRQ.
Also, the other USB CAN devices just directly send the USB message in
their xmit() without the need to rely on such worker.

Sorry if this was discussed in the past, I can not remember if this
question has already been raised.

AFAICS xmit uses nct6694_write_msg(), which uses several usb_bulk_msg(),
that can only be used from task context.

You can build a chain of usb_fill_bulk_urb(), usb_submit_urb(),
callbacks instead, but that's hard to read compared to blocking
usb_bulk_msg()s.

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   |