RE: [PATCH] net/r8169: Update the new parser for the new firmware

-----Original Message-----
From: Francois Romieu [mailto:romieu@fr.zoreil.com] 
Sent: Monday, June 13, 2011 10:33 PM
To: Hayeswang
Cc: netdev@vger.kernel.org; linux-kernel@vger.kernel.org
Subject: Re: [PATCH] net/r8169: Update the new parser for the 
new firmware

Hayes Wang [off-list ref] :

quoted

Update the parser for the new firmware which is embedded 

some information.

quoted

The paser cannot be used for the old firmware. It is only 

for the new type one.

quoted

Signed-off-by: Hayes Wang <redacted>
---
 drivers/net/r8169.c |   59 

++++++++++++++++++++++++++++----------------------

quoted

 1 files changed, 33 insertions(+), 26 deletions(-)

diff --git a/drivers/net/r8169.c b/drivers/net/r8169.c index 

ef1ce2e..87b684f 100644

--- a/drivers/net/r8169.c
+++ b/drivers/net/r8169.c

@@ -36,11 +36,11 @@
 #define MODULENAME "r8169"
 #define PFX MODULENAME ": "
 
-#define FIRMWARE_8168D_1	"rtl_nic/rtl8168d-1.fw"
-#define FIRMWARE_8168D_2	"rtl_nic/rtl8168d-2.fw"
-#define FIRMWARE_8168E_1	"rtl_nic/rtl8168e-1.fw"
-#define FIRMWARE_8168E_2	"rtl_nic/rtl8168e-2.fw"
-#define FIRMWARE_8105E_1	"rtl_nic/rtl8105e-1.fw"
+#define FIRMWARE_8168D_1	"rtl_nic/rtl8168d-3.fw"
+#define FIRMWARE_8168D_2	"rtl_nic/rtl8168d-4.fw"
+#define FIRMWARE_8168E_1	"rtl_nic/rtl8168e-3.fw"
+#define FIRMWARE_8168E_2	"rtl_nic/rtl8168e-4.fw"
+#define FIRMWARE_8105E_1	"rtl_nic/rtl8105e-2.fw"

Why ?

I want to keep the old firmware used by the old paser. The old paser cannot use
the new firmware and the new paser cannot use the old firmware, so I separate
them.

I see it more as a calling convention to communicate with 
userspace than as a place to funnel a partial version information.

quoted

 #ifdef RTL8169_DEBUG
 #define assert(expr) \

@@ -594,6 +594,13 @@ struct ring_info {
 	u8		__pad[sizeof(void *) - sizeof(u32)];
 };
 
+struct fw_info {
+	char	version[32];
+	u32	fw_start;
+	u32	fw_len;
+	u8	chksum;

The chksum field is never used.

This field in the binary file makes sure the checksum is zero. Driver don't need
to access this address. But it needs in the binary file for checking.

[...]

quoted

@@ -1743,16 +1750,30 @@ static void rtl_writephy_batch(struct 

rtl8169_private *tp,  static void  rtl_phy_write_fw(struct 
rtl8169_private *tp, const struct firmware *fw)  {
-	__le32 *phytable = (__le32 *)fw->data;
+	__le32 *phytable;
 	struct net_device *dev = tp->dev;
-	size_t index, fw_size = fw->size / sizeof(*phytable);
+	struct fw_info *f_info;
+	size_t index, fw_size;

s/index/i/

Yes, I would replace it.

quoted

 	u32 predata, count;
+	u8 checksum;
+
+	f_info = (struct fw_info *)fw->data;

It will not work very well on big-endian computers. start and 
len should both use le32_to_cpu.

I would fix these. Thanks.

quoted

 
-	if (fw->size % sizeof(*phytable)) {
-		netif_err(tp, probe, dev, "odd sized firmware 

%zd\n", fw->size);

quoted

+	if (checksum != 0) {
+		netif_err(tp, probe, dev, "none zero 

checksum(%u)\n", checksum);

quoted

 		return;
 	}

Nit: "checksum" is not used beyond this point. It should be 
possible to put a few things in distinct functions to save 
some local variables.

I would separate this into another function.

quoted

+	netif_info(tp, probe, dev, "firmware: %s\n", f_info->version);
+
+	phytable = (__le32 *)(fw->data + f_info->fw_start);
+	fw_size = f_info->fw_len;
+
 	for (index = 0; index < fw_size; index++) {

It's a bit paranoid but I would feel safer if f_info->fw_len 
was compared to fw->size beforehand.

You are right. I would do it.

[...]

quoted

@@ -1892,14 +1913,6 @@ static void 

rtl_apply_firmware(struct rtl8169_private *tp)

quoted

 		rtl_phy_write_fw(tp, fw);
 }
 
-static void rtl_apply_firmware_cond(struct rtl8169_private *tp, u8 
reg, u16 val) -{
-	if (rtl_readphy(tp, reg) != val)
-		netif_warn(tp, hw, tp->dev, "chipset not ready 

for firmware\n");

quoted

-	else
-		rtl_apply_firmware(tp);
-}
-
 static void rtl8169s_hw_phy_config(struct rtl8169_private *tp)  {
 	static const struct phy_reg phy_reg_init[] = { @@ 

-2334,10 +2347,7 

quoted

@@ static void rtl8168d_1_hw_phy_config(struct rtl8169_private *tp)
 	rtl_w1w0_phy(tp, 0x02, 0x0100, 0x0600);
 	rtl_w1w0_phy(tp, 0x03, 0x0000, 0xe000);
 
-	rtl_writephy(tp, 0x1f, 0x0005);
-	rtl_writephy(tp, 0x05, 0x001b);
-
-	rtl_apply_firmware_cond(tp, MII_EXPANSION, 0xbf00);
+	rtl_apply_firmware(tp);
 
 	rtl_writephy(tp, 0x1f, 0x0000);
 }

@@ -2437,10 +2447,7 @@ static void 

rtl8168d_2_hw_phy_config(struct rtl8169_private *tp)

quoted

 	rtl_writephy(tp, 0x1f, 0x0002);
 	rtl_patchphy(tp, 0x0f, 0x0017);
 
-	rtl_writephy(tp, 0x1f, 0x0005);
-	rtl_writephy(tp, 0x05, 0x001b);
-
-	rtl_apply_firmware_cond(tp, MII_EXPANSION, 0xb300);
+	rtl_apply_firmware(tp);

Actually both the format and the content of the firmware are 
changed by the patch.

Imho the new firmware format could include some specific 
string so that the driver can identify the new firmware 
format and fallback to the old format otherwise. Any fixed 
magic prefix would be enough as the new format mandates the 
version information.

This way Linus's test machine won't risk of breaking 
(again...) if he builds a new kernel without updating the 
firmware at the same time.

I plan to let the old paser loads the old firmware and the new paser loads the
new firmware. If you build the new kernel without updating the firmware, you
just load no firmware because the new firmware doesn't exist. However, it is
more dangerous for the old paser to load the new firmware. That is why I create
the new ones, not replace the existing files.

 
Best Regards,
Hayes