Skip to contentSkip to conversation

bnx2x: annotate nvram dword reading/writing

PATCH In review Harvey Harrison 2 participants 4 messages Jan 21, 2009 – Jan 22, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 9/9 bnx2x: annotate nvram dword reading/writing

I believe this exposed an endian bug as the shifting of bytes from the data buffer was done in cpu-order, then masked into a be32 and the combined value was then converted to cpu-order, this does all the masking in be-byteorder and passes a cpu-ordered value to the write routine.

---
 drivers/net/bnx2x_main.c |   17 +++++++----------
 1 files changed, 7 insertions(+), 10 deletions(-)
drivers/net/bnx2x_main.c +7 -10
index 5e5e008..d0b0f6e 100644
@@ -8011,7 +8011,7 @@ static void bnx2x_disable_nvram_access(struct bnx2x *bp)
 			MCPR_NVM_ACCESS_ENABLE_WR_EN)));
 }
 
-static int bnx2x_nvram_read_dword(struct bnx2x *bp, u32 offset, u32 *ret_val,
+static int bnx2x_nvram_read_dword(struct bnx2x *bp, u32 offset, __be32 *ret_val,
 				  u32 cmd_flags)
 {
 	int count, i, rc;
@@ -8047,8 +8047,7 @@ static int bnx2x_nvram_read_dword(struct bnx2x *bp, u32 offset, u32 *ret_val,
 			/* we read nvram data in cpu order
 			 * but ethtool sees it as an array of bytes
 			 * converting to big-endian will do the work */
-			val = cpu_to_be32(val);
-			*ret_val = val;
+			*ret_val = cpu_to_be32(val);
 			rc = 0;
 			break;
 		}
@@ -8062,7 +8061,7 @@ static int bnx2x_nvram_read(struct bnx2x *bp, u32 offset, u8 *ret_buf,
 {
 	int rc;
 	u32 cmd_flags;
-	u32 val;
+	__be32 val;
 
 	if ((offset & 0x03) || (buf_size & 0x03) || (buf_size == 0)) {
 		DP(BNX2X_MSG_NVM,
@@ -8181,7 +8180,7 @@ static int bnx2x_nvram_write1(struct bnx2x *bp, u32 offset, u8 *data_buf,
 	int rc;
 	u32 cmd_flags;
 	u32 align_offset;
-	u32 val;
+	__be32 val;
 
 	if (offset + buf_size > bp->common.flash_size) {
 		DP(BNX2X_MSG_NVM, "Invalid parameter: offset (0x%x) +"
@@ -8203,14 +8202,12 @@ static int bnx2x_nvram_write1(struct bnx2x *bp, u32 offset, u8 *data_buf,
 	rc = bnx2x_nvram_read_dword(bp, align_offset, &val, cmd_flags);
 
 	if (rc == 0) {
-		val &= ~(0xff << BYTE_OFFSET(offset));
-		val |= (*data_buf << BYTE_OFFSET(offset));
-
 		/* nvram data is returned as an array of bytes
 		 * convert it back to cpu order */
-		val = be32_to_cpu(val);
+		val &= ~cpu_to_be32(0xff << (8 * (offset & 0x03)));
+		val |= cpu_to_be32(*data_buf << (8 * (offset & 0x03)));
 
-		rc = bnx2x_nvram_write_dword(bp, align_offset, val,
+		rc = bnx2x_nvram_write_dword(bp, align_offset, be32_to_cpu(val),
 					     cmd_flags);
 	}
 
Signature · 1 line
1.6.1.249.g455e5
#1 lore raw
Quoted 7 lines · Harvey Harrison

I believe this exposed an endian bug as the shifting of bytes from the data buffer was done in cpu-order, then masked into a be32 and the combined value was then converted to cpu-order, this does all the masking in be-byteorder and passes a cpu-ordered value to the write routine.

This change is breaking the FW upgrade utility which uses this interface since it is changing the content. I will need to work with the engineer that owns this utility, but this change will probably stay out for a while

#2 lore raw
Quoted 7 lines · Eilon Greenstein
Quoted 6 lines · Harvey Harrison

I believe this exposed an endian bug as the shifting of bytes from the data buffer was done in cpu-order, then masked into a be32 and the combined value was then converted to cpu-order, this does all the masking in be-byteorder and passes a cpu-ordered value to the write routine.

Quoted 7 lines
Quoted 1 line

This change is breaking the FW upgrade utility which uses this interface since it is changing the content. I will need to work with the engineer that owns this utility, but this change will probably stay out for a while

No worries, can you explain briefly what the code is trying to accomplish? Maybe it was written assuming a le-machine, because this is going to work differently on a be-machine.

Harvey

#3 lore raw
Quoted 16 lines · Harvey Harrison
Quoted 7 lines · Eilon Greenstein
Quoted 6 lines · Harvey Harrison

I believe this exposed an endian bug as the shifting of bytes from the data buffer was done in cpu-order, then masked into a be32 and the combined value was then converted to cpu-order, this does all the masking in be-byteorder and passes a cpu-ordered value to the write routine.

Quoted 7 lines
Quoted 1 line

This change is breaking the FW upgrade utility which uses this interface since it is changing the content. I will need to work with the engineer that owns this utility, but this change will probably stay out for a while

Quoted 5 lines

No worries, can you explain briefly what the code is trying to accomplish? Maybe it was written assuming a le-machine, because this is going to work differently on a be-machine.

Harvey

The utility simply works with the current implementation (on both be and le) and this is why changing the logic breaks it. I will look into it to see if we can change the utility logic as well (maybe according to the driver version) - it is not that easy since this utility supports other drivers as well.

Harvey - thanks again for this patch, it is definitely a step in the right direction.

Eilon