On 22/10/2019 17:26, Edward Cree wrote:

quoted

@@ -764,6 +872,16 @@ void efx_init_rx_queue(struct efx_rx_queue *rx_queue)
 	rx_queue->fast_fill_trigger = trigger;
 	rx_queue->refill_enabled = true;
 
+	/* Initialise XDP queue information */
+	rc = xdp_rxq_info_reg(&rx_queue->xdp_rxq_info, efx->net_dev,
+			      rx_queue->core_index);
+
+	if (rc) {
+		netif_err(efx, rx_err, efx->net_dev,
+			  "Failure to initialise XDP queue information rc=%d\n",
+			  rc);
+	}

What happens if we try to use XDP after this has failed?
Should we set some kind of "XDP broken" flag to prevent that?

-Ed

Agreed. I've added a flag to track which rx queues fail, and then a per-nic flag
to indicate that at least one rx queue has failed, and hence that we shouldn't 
allow XDP programs to bind.