[U-Boot-Users] [PATCH] net: smc911x: sh: Add support RSK7203 board to smc911x

[U-Boot-Users] [PATCH] net: smc911x: sh: Add support RSK7203 board to smc911x

[Nobuhiro Iwamatsu]

The RSK board has the SMSC9118 wired up 'incorrectly'.
This problem can support in the software side.
I put a code to evade it.

Signed-off-by: Nobuhiro Iwamatsu <iwamatsu.nobuhiro@renesas.com>
---
 drivers/net/smc911x.c |   37 +++++++++++++++++++++++++++++++++++--
 1 files changed, 35 insertions(+), 2 deletions(-)

diff --git a/drivers/net/smc911x.c b/drivers/net/smc911x.c
index 1484b0b..ec731ea 100644
--- a/drivers/net/smc911x.c
+++ b/drivers/net/smc911x.c
@@ -53,6 +53,39 @@ static inline void reg_write(u32 addr, u32 val)
 	*(volatile u16*)addr = (u16)val;
 	*(volatile u16*)(addr + 2) = (u16)(val >> 16);
 }
+
+#if 0 //defined(CONFIG_RSK7203)
+/*
+ * The RSK board has the SMSC9118 wired up 'incorrectly'.
+ * Byte-swapping is necessary, and so poor performance is inevitable.
+ * It seems that this is a problem with the SMSC chip, which has an
+ * endian facility, but it only works at the word level so...
+ */
+/* 2301_6745 --> 0123_4567 */
+#define PKT_SWAP(x) ((vu_long)((((vu_long)x << 8) & 0x0000FF00)|\
+		(((vu_long)x >> 8) & 0x000000FF)|\
+		(((vu_long)x << 8) & 0xFF000000)|\
+		(((vu_long)x >> 8) & 0x00FF0000)))
+
+static inline u32 pkt_data_pull(u32 addr)
+{
+	volatile u16 *addr_16 = (u16 *)addr;
+	u32 d0 = (((*addr_16) << 16) & 0xFFFF0000);
+	u32 d1 = ((*(addr_16 + 1) & 0x0000FFFF));
+	return PKT_SWAP((d0|d1));
+}
+
+static inline void pkt_data_push(u32 addr, u32 val)
+{
+	*(volatile u16 *)(addr + 2) = PKT_SWAP((u16)val);
+	*(volatile u16 *)(addr) = PKT_SWAP((u16)(val >> 16));
+}
+
+#else
+# define pkt_data_pull(x) reg_read(x)
+# define pkt_data_push(x, l) reg_write(x, l)
+#endif /* CONFIG_RSK7203*/
+
 #else
 #error "SMC911X: undefined bus width"
 #endif /* CONFIG_DRIVER_SMC911X_16_BIT */
@@ -641,7 +674,7 @@ int eth_send(volatile void *packet, int length)
 	tmplen = (length + 3) / 4;

 	while (tmplen--)
-		reg_write(TX_DATA_FIFO, *data++);
+		pkt_data_push(TX_DATA_FIFO, *data++);

 	/* wait for transmission */
 	while (!((reg_read(TX_FIFO_INF) & TX_FIFO_INF_TSUSED) >> 16));
@@ -684,7 +717,7 @@ int eth_rx(void)

 		tmplen = (pktlen + 2+ 3) / 4;
 		while (tmplen--)
-			*data++ = reg_read(RX_DATA_FIFO);
+			*data++ = pkt_data_pull(RX_DATA_FIFO);

 		if (status & RX_STS_ES)
 			printf(DRIVERNAME
-- 
1.5.5.1

[U-Boot-Users] [PATCH] net: smc911x: sh: Add support RSK7203 board to smc911x

[Ben Warren]

Hi Nobuhiro,

Nobuhiro Iwamatsu wrote:
> The RSK board has the SMSC9118 wired up 'incorrectly'.
> This problem can support in the software side.
> I put a code to evade it.
>
>   
Putting a board fix in a common driver is not cool. Please provide more
generic word-swapping versions of reg_read and reg_write, or something
else like that.

regards,
Ben

[U-Boot-Users] [PATCH] net: smc911x: sh: Add support RSK7203 board to smc911x

[Nobuhiro Iwamatsu]

Hi, Ben.

Thank you for your comments.

2008/7/4 Ben Warren <biggerbadderben@gmail.com>:
> Hi Nobuhiro,
>
> Nobuhiro Iwamatsu wrote:
>> The RSK board has the SMSC9118 wired up 'incorrectly'.
>> This problem can support in the software side.
>> I put a code to evade it.
>>
>>
> Putting a board fix in a common driver is not cool.
Yes , I know.
However, this problem occurs in read/write of FIFO.
Therefore, I made a pkt_data_pull/pkt_data_push function newly.
And I change it with a preprocessor.

> Please provide more
> generic word-swapping versions of reg_read and reg_write, or something
> else like that.
OK, I will change swabXX and resend.

Best regards,
 Nobuhiro