e1000e : Correct Rx Threshold granularity

In e1000e driver, Rx descriptor queue is used such that hardware can add only
one descriptor at a time. So the WTHRESH granularity in RXDCTL should be set
to single descriptor. This would ensure that every time controller fills a Rx
descriptor, it is flushed to host memory. Earlier this granularity was in
cache line units i.e 2 descriptors. This leads to controller always waiting
for 2 descriptors before flushing them out. But since not more than one Rx BD
is actually available , the accumulation condition never gets hit.

Signed-off-by: Ruchika Gupta <ruchika.gupta@freescale.com>
Signed-off-by: Vakul Garg <vakul@freescale.com>
Acked-by: Roy Zang <tie-fei.zang@freescale.com>

1 files changed, 10 insertions, 0 deletions

diff --git a/drivers/net/e1000.c b/drivers/net/e1000.c
index 2d4da4b386..8ba98b27d5 100644
--- a/drivers/net/e1000.c
+++ b/drivers/net/e1000.c
@@ -1688,6 +1688,16 @@ e1000_init_hw(struct eth_device *nic)
 		E1000_WRITE_REG(hw, TXDCTL, ctrl);
 	}
 
+	/* Set the receive descriptor write back policy */
+
+	if (hw->mac_type >= e1000_82571) {
+		ctrl = E1000_READ_REG(hw, RXDCTL);
+		ctrl =
+		    (ctrl & ~E1000_RXDCTL_WTHRESH) |
+		    E1000_RXDCTL_FULL_RX_DESC_WB;
+		E1000_WRITE_REG(hw, RXDCTL, ctrl);
+	}
+
 	switch (hw->mac_type) {
 	default:
 		break;