[PATCH] e1000: Reset rx ring index on receive overrun

[Samuel Thibault]

At high traffic rate, the rx ring may get completely filled before we
manage to consume it.  After it is filled, the kernel and device indexes
are not synchronized any more, so we have to reset them, otherwise the
kernel will be stuck waiting for the wrong slot to be filled.

Signed-off-by: Samuel Thibault <samuel.thibault@xxxxxxxxxxxx>

---
This is just a patch suggestion, I'm not an expert in network drivers, I
leave to actual driver authors to bake a better version.

diff --git a/drivers/net/ethernet/intel/e1000/e1000_main.c b/drivers/net/ethernet/intel/e1000/e1000_main.c
index 37caa88..77c8dbc 100644
--- a/drivers/net/ethernet/intel/e1000/e1000_main.c
+++ b/drivers/net/ethernet/intel/e1000/e1000_main.c
@@ -3759,6 +3759,21 @@ static irqreturn_t e1000_intr(int irq, void *data)
 	if (unlikely(test_bit(__E1000_DOWN, &adapter->flags)))
 		return IRQ_HANDLED;
 
+	if (unlikely(icr & E1000_ICR_RXO)) {
+		/* Receive Overrun */
+		u32 rctl;
+		int i;
+		rctl = er32(RCTL);
+		ew32(RCTL, rctl & ~E1000_RCTL_EN);
+		for (i = 0; i < adapter->num_rx_queues; i++) {
+			memset(adapter->rx_ring[i].desc, 0, adapter->rx_ring[i].size);
+			adapter->rx_ring[i].next_to_clean = 0;
+		}
+		ew32(RDH, 0);
+		ew32(RCTL, rctl);
+		adapter->netdev->stats.rx_fifo_errors++;
+	}
+
 	if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
 		hw->get_link_status = 1;
 		/* guard against interrupt when we're going down */
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/