net: fec: improve UDP performance

UDP performance is extremely poor.  iperf reports UDP receive speeds in
the range of 50Mbps with a high packet loss.  The interface typically
reports a few thousand overrun errors per iperf run.  This is far from
satisfactory.

Adjust the receive FIFO thresholds to reduce the number of errors.

Firstly, we decrease RAFL (receive almost full threshold) to the minimum
value of 4, which gives us the maximum available space in the FIFO prior
to indicating a FIFO overrun condition.

Secondly, we adjust the RSEM value to send a XOFF pause frame early
enough that an in-progress transmission can complete without overflowing
the FIFO.

Document these registers and the changes being made in the driver.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

1 files changed, 25 insertions, 3 deletions

diff --git a/drivers/net/ethernet/freescale/fec_main.c b/drivers/net/ethernet/freescale/fec_main.c
index b79d78fd6e7a..c30270168bec 100644
--- a/drivers/net/ethernet/freescale/fec_main.c
+++ b/drivers/net/ethernet/freescale/fec_main.c
@@ -72,12 +72,34 @@ static void set_multicast_list(struct net_device *ndev);
 
 #define DRIVER_NAME	"fec"
 
-/* Pause frame feild and FIFO threshold */
+/* Pause frame field and FIFO threshold */
 #define FEC_ENET_FCE	(1 << 5)
-#define FEC_ENET_RSEM_V	0x84
+/*
+ * RSEM: the FIFO threshold in 64-bit words (from zero bytes) at which
+ *  XOFF will be sent.  Minimum 0, maximum possible 255.
+ * RAFL: the number of remaining 64-bit words in the FIFO at which the
+ *  frame is marked with overrun.  Minimum 4, maximum possible 255.
+ *
+ * When XOFF is sent, the link partner should continue sending any
+ * packet currently in progress, and then pause transmission for the
+ * specified duration.  It should not terminate transmission of the
+ * packet.
+ *
+ * The FIFO can store the full frame and the FCS value, which on standard
+ * Ethernet is 1518 bytes, or 190 FIFO entries.  However, we do not know
+ * the size of the receive FIFO, so we are unable to properly calculate
+ * the thresholds.
+ *
+ * Practical tests with 1514 byte ethernet packets show that with
+ * RSEM = 132, RAFL = 8, we see a large quantity of overflowed UDP packets.
+ * Reducing RAFL to 4 has almost no effect.  Reducing RSEM to 78 gives us
+ * a small number of overflow errors, and at 77 it gives only a single
+ * error per UDP iperf run.
+ */
+#define FEC_ENET_RSEM_V	77
 #define FEC_ENET_RSFL_V	16
 #define FEC_ENET_RAEM_V	0x8
-#define FEC_ENET_RAFL_V	0x8
+#define FEC_ENET_RAFL_V	4
 #define FEC_ENET_OPD_V	0xFFF0
 
 /* Controller is ENET-MAC */