Skip to contentSkip to conversation

: arm: Kirkwood: Various egiga fixes

PATCH New Simon Kagstrom 6 participants 22 messages Jul 8, 2009 – Jul 21, 2009 Network drivers

lore
Share & tools
lore
#0 lore raw

cover : arm: Kirkwood: Various egiga fixes

Hi again!

This is a refreshed version of the
http://lists.denx.de/pipermail/u-boot/2009-July/055370.html patch series. I've added one more patch to fix an issue with the MAC address under Linux. The patches are now:

1. (new) Set MAC address during egiga registration (so that Linux gets it even if the network interface has not been initialized under U-boot).

2. Correct the optimization bug in kgwbe_send. This patch now uses IO accessor functions instead of volatile (and also corrects recv).

3. Check the error summary bit during send (as before but based on the above patches)

4. Allow sending data which is not 8-byte aligned (as before but based on the above patches)

// Simon

#1 lore raw

PATCH 1/4 : arm: Kirkwood: Set MAC address during registration for kirkwood egiga

This patch sets the MAC address during registration in addition to during device init. Since U-boot might not access the ethernet device, Linux might end up with the MAC address unset.

---
 drivers/net/kirkwood_egiga.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)
drivers/net/kirkwood_egiga.c +2 -0
index 3c5db19..d760e1d 100644
@@ -653,6 +653,8 @@ int kirkwood_egiga_initialize(bd_t * bis)
 		dev->send = (void *)kwgbe_send;
 		dev->recv = (void *)kwgbe_recv;
 
+		port_uc_addr_set(dkwgbe->regs, dev->enetaddr);
+
 		eth_register(dev);
 
 #if defined(CONFIG_MII) || defined(CONFIG_CMD_MII)
Signature · 1 line
1.6.0.4
#2 lore raw

PATCH 1/4 : arm: Kirkwood: Set MAC address during registration for kirkwood egiga

Quoted 23 lines

-----Original Message-----
From: Simon Kagstrom [mailto:simon.kagstrom at netinsight.net]
Sent: Wednesday, July 08, 2009 4:32 PM
Cc: U-Boot ML; Prafulla Wadaskar
Subject: [PATCH 1/4]: arm: Kirkwood: Set MAC address during registration for kirkwood egiga

This patch sets the MAC address during registration in addition to during device init. Since U-boot might not access the ethernet device, Linux might end up with the MAC address unset.

---
 drivers/net/kirkwood_egiga.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)
(quoted diff) +0 -0

b/drivers/net/kirkwood_egiga.c index 3c5db19..d760e1d 100644

drivers/net/kirkwood_egiga.c int kirkwood_egiga_initialize(bd_t * bis)

Quoted 9 lines
drivers/net/kirkwood_egiga.c +2 -1
@@ -653,6 +653,8 @@ int kirkwood_egiga_initialize(bd_t * bis)
 		dev->send = (void *)kwgbe_send;
 		dev->recv = (void *)kwgbe_recv;
 
+		port_uc_addr_set(dkwgbe->regs, dev->enetaddr);
+
 		eth_register(dev);
 
 #if defined(CONFIG_MII) || defined(CONFIG_CMD_MII)
--

Ack. Good catch, I had removed it during code size reduction :-(

Regards..
Prafulla . .

Quoted 3 lines

1.6.0.4

#3 lore raw

PATCH 1/4 : arm: Kirkwood: Set MAC address during registration for kirkwood egiga

In reply to Simon Kagstrom
Quoted 4 lines · Simon Kagstrom

This patch sets the MAC address during registration in addition to during device init. Since U-boot might not access the ethernet device, Linux might end up with the MAC address unset.

I think this violates U-boot policy of only touching hardware if it's used, but Wolfgang can say for sure. In general, initialize() functions should just set up data structures and register the device. There's a long history of MAC-address issues with ARM chips and Linux that I've happily stayed clear of so can't claim to be an expert here.

regards,
Ben

#4 lore raw

PATCH 1/4 : arm: Kirkwood: Set MAC address during registration for kirkwood egiga

Quoted 10 lines · On Thu, 09 Jul 2009 22:44:10 -0700 Ben Warren <[email protected]> wrote:
Quoted 4 lines · Simon Kagstrom

This patch sets the MAC address during registration in addition to during device init. Since U-boot might not access the ethernet device, Linux might end up with the MAC address unset.

I think this violates U-boot policy of only touching hardware if it's used, but Wolfgang can say for sure. In general, initialize() functions should just set up data structures and register the device. There's a long history of MAC-address issues with ARM chips and Linux that I've happily stayed clear of so can't claim to be an expert here.

OK, I've tried looking around at how other boards do it, and at least the most similar (mv6436x_eth_initialize() for other Marvell boards) do it the same way as well as some others (ax88180.c).

Should Linux otherwise read the passed U-boot environment and set it up by itself?

// Simon

#5 lore raw

PATCH 1/4 : arm: Kirkwood: Set MAC address during registration for kirkwood egiga

Quoted 24 lines

-----Original Message-----
From: u-boot-bounces at lists.denx.de [mailto:u-boot-bounces at lists.denx.de] On Behalf Of Simon Kagstrom
Sent: Friday, July 10, 2009 12:55 PM
To: Ben Warren
Cc: U-Boot ML
Subject: Re: [U-Boot] [PATCH 1/4]: arm: Kirkwood: Set MAC address during registration for kirkwood egiga

Quoted 2 lines · On Thu, 09 Jul 2009 22:44:10 -0700 Ben Warren <[email protected]> wrote:
Quoted 1 line · Simon Kagstrom

This patch sets the MAC address during registration in

addition to

Quoted 4 lines
Quoted 3 lines

during device init. Since U-boot might not access the ethernet device, Linux might end up with the MAC address unset.

I think this violates U-boot policy of only touching

hardware if it's

Quoted 2 lines

used, but Wolfgang can say for sure. In general, initialize() functions should just set up data structures and register

Quoted 8 lines

the device.

Quoted 1 line

There's a long history of MAC-address issues with ARM chips

and Linux

Quoted 1 line

that I've happily stayed clear of so can't claim to be an

expert here.

OK, I've tried looking around at how other boards do it, and at least the most similar (mv6436x_eth_initialize() for other

This is a wrong reference, even this was my initial reference, but there are no updates to this code, we were planning to clean it but we do not have hardware to test it. Does any one have it?

Quoted 1 line

Marvell boards) do it the same way as well as some others (ax88180.c).

If posible can be improved.

Quoted 3 lines

Should Linux otherwise read the passed U-boot environment and set it up by itself?

I think this is better approach,
Regards..
Prafulla . .

Quoted 7 lines

// Simon
_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot

#6 lore raw

PATCH 1/4 : arm: Kirkwood: Set MAC address during registration for kirkwood egiga

In reply to Simon Kagstrom

Dear Simon Kagstrom,

Quoted 4 lines · In message <[email protected]> you wrote:

OK, I've tried looking around at how other boards do it, and at least the most similar (mv6436x_eth_initialize() for other Marvell boards) do it the same way as well as some others (ax88180.c).

Then those slipped through the review process. Patches to clean these up are welcome.

Quoted 2 lines

Should Linux otherwise read the passed U-boot environment and set it up by itself?

Unfortunately ARM Linux does not provide yet a standard way to pass the MAC address from the boot loader. In the long term this problem will be solved by using the device tree for ARM; in the meantime, some drivers use a "ethaddr=" command line parameter (but that will probably bring up discussions on the Linux front again).

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Few people do business well who do nothing else.
                                       -- Philip Earl of Chesterfield
#7 lore raw

PATCH 1/4 : arm: Kirkwood: Set MAC address during registration for kirkwood egiga

Quoted 8 lines · On Sat, 11 Jul 2009 11:27:57 +0200 Wolfgang Denk <[email protected]> wrote:
Quoted 2 lines

Should Linux otherwise read the passed U-boot environment and set it up by itself?

Unfortunately ARM Linux does not provide yet a standard way to pass the MAC address from the boot loader. In the long term this problem will be solved by using the device tree for ARM; in the meantime, some drivers use a "ethaddr=" command line parameter (but that will probably bring up discussions on the Linux front again).

OK, so until ARM has moved to the device tree approach (and from LKML I see there is some vocal opposition to this), we'd still need to resort to some driver-specific hack [with Linux support] for this?

Or is it OK to do some board-specific "hack" to initialize it like

board/atmel/at91rm9200ek/misc.c

does it?

Coming from a PowerPC-setting, I must say I miss the device trees though :-)

// Simon

#8 lore raw

PATCH 1/4 : arm: Kirkwood: Set MAC address during registration for kirkwood egiga

Dear Simon Kagstrom,

Quoted 3 lines · In message <[email protected]> you wrote:

OK, so until ARM has moved to the device tree approach (and from LKML I see there is some vocal opposition to this), we'd still need to resort to some driver-specific hack [with Linux support] for this?

Driver-specific hack or fix yes - but in Linux, not in U-Boot.

Quoted 5 lines

Or is it OK to do some board-specific "hack" to initialize it like

board/atmel/at91rm9200ek/misc.c

does it?

I this driver does something like this, then it should ne fixed.

Quoted 2 lines

Coming from a PowerPC-setting, I must say I miss the device trees though :-)

Me, too. Let's keep the fingers crossed that gcl makes fast progress.

Best regards,

Wolfgang Denk

Signature · 4 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
"I didn't know it was impossible when I did it."
#9 lore raw

PATCH 1/4 : arm: Kirkwood: Set MAC address during registration for kirkwood egiga

In reply to Ben Warren

Dear Ben Warren,

Quoted 10 lines · In message <[email protected]> you wrote:
Quoted 4 lines · Simon Kagstrom

This patch sets the MAC address during registration in addition to during device init. Since U-boot might not access the ethernet device, Linux might end up with the MAC address unset.

I think this violates U-boot policy of only touching hardware if it's used, but Wolfgang can say for sure. In general, initialize() functions should just set up data structures and register the device. There's a long history of MAC-address issues with ARM chips and Linux that I've happily stayed clear of so can't claim to be an expert here.

You are correct. Devices shall only be initialized when actually used in U-Boot (upon first use).

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
There is an order of things in this universe.
	-- Apollo, "Who Mourns for Adonais?" stardate 3468.1
#10 lore raw

PATCH 2/4 : arm: Kirkwood: Fix compiler optimization bug for kwgbe_send

In reply to Simon Kagstrom

kwgbe_send/recv both have loops waiting for the hardware to set a bit. GCC 4.3.3 cleverly optimizes the send case to ... a while(1); loop. This patch uses readl to force a read from device memory. Other volatile accesses have also been replaced with readl/writel where appropriate (as per suggestions on the U-boot mailing list).

---
 drivers/net/kirkwood_egiga.c |   31 +++++++++++++++++++------------
 1 files changed, 19 insertions(+), 12 deletions(-)
drivers/net/kirkwood_egiga.c +19 -12
index d760e1d..9cab7fd 100644
@@ -49,7 +49,7 @@ static int smi_reg_read(char *devname, u8 phy_adr, u8 reg_ofs, u16 * data)
 	struct kwgbe_device *dkwgbe = to_dkwgbe(dev);
 	struct kwgbe_registers *regs = dkwgbe->regs;
 	u32 smi_reg;
-	volatile u32 timeout;
+	u32 timeout;
 
 	/* Phyadr read request */
 	if (phy_adr == 0xEE && reg_ofs == 0xEE) {
@@ -124,7 +124,7 @@ static int smi_reg_write(char *devname, u8 phy_adr, u8 reg_ofs, u16 data)
 	struct kwgbe_device *dkwgbe = to_dkwgbe(dev);
 	struct kwgbe_registers *regs = dkwgbe->regs;
 	u32 smi_reg;
-	volatile u32 timeout;
+	u32 timeout;
 
 	/* Phyadr write request*/
 	if (phy_adr == 0xEE && reg_ofs == 0xEE) {
@@ -370,7 +370,7 @@ static void port_uc_addr_set(struct kwgbe_registers *regs, u8 * p_addr)
  */
 static void kwgbe_init_rx_desc_ring(struct kwgbe_device *dkwgbe)
 {
-	volatile struct kwgbe_rxdesc *p_rx_desc;
+	struct kwgbe_rxdesc *p_rx_desc;
 	int i;
 
 	/* initialize the Rx descriptors ring */
@@ -487,6 +487,7 @@ static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
 	struct kwgbe_device *dkwgbe = to_dkwgbe(dev);
 	struct kwgbe_registers *regs = dkwgbe->regs;
 	struct kwgbe_txdesc *p_txdesc = dkwgbe->p_txdesc;
+	u32 cmd_sts;
 
 	if ((u32) dataptr & 0x07) {
 		printf("Err..(%s) xmit dataptr not 64bit aligned\n",
@@ -507,21 +508,24 @@ static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
 	/*
 	 * wait for packet xmit completion
 	 */
-	while (p_txdesc->cmd_sts & KWGBE_BUFFER_OWNED_BY_DMA) {
+	cmd_sts = readl(&p_txdesc->cmd_sts);
+	while (cmd_sts & KWGBE_BUFFER_OWNED_BY_DMA) {
 		/* return fail if error is detected */
-		if (p_txdesc->cmd_sts & (KWGBE_UR_ERROR | KWGBE_RL_ERROR)) {
+		if (cmd_sts & (KWGBE_UR_ERROR | KWGBE_RL_ERROR)) {
 			printf("Err..(%s) in xmit packet\n", __FUNCTION__);
 			return -1;
 		}
+		cmd_sts = readl(&p_txdesc->cmd_sts);
 	};
 	return 0;
 }
 
 static int kwgbe_recv(struct eth_device *dev)
 {
-	volatile struct kwgbe_device *dkwgbe = to_dkwgbe(dev);
-	volatile struct kwgbe_rxdesc *p_rxdesc_curr = dkwgbe->p_rxdesc_curr;
-	volatile u32 timeout = 0;
+	struct kwgbe_device *dkwgbe = to_dkwgbe(dev);
+	struct kwgbe_rxdesc *p_rxdesc_curr = dkwgbe->p_rxdesc_curr;
+	u32 cmd_sts;
+	u32 timeout = 0;
 
 	/* wait untill rx packet available or timeout */
 	do {
@@ -531,7 +535,7 @@ static int kwgbe_recv(struct eth_device *dev)
 			debug("%s time out...\n", __FUNCTION__);
 			return -1;
 		}
-	} while (p_rxdesc_curr->cmd_sts & KWGBE_BUFFER_OWNED_BY_DMA);
+	} while (readl(&p_rxdesc_curr->cmd_sts) & KWGBE_BUFFER_OWNED_BY_DMA);
 
 	if (p_rxdesc_curr->byte_cnt != 0) {
 		debug("%s: Received %d byte Packet @ 0x%x (cmd_sts= %08x)\n",
@@ -545,14 +549,16 @@ static int kwgbe_recv(struct eth_device *dev)
 	 * OR the error summary bit is on,
 	 * the packets needs to be dropeed.
 	 */
-	if ((p_rxdesc_curr->cmd_sts &
+	cmd_sts = readl(&p_rxdesc_curr->cmd_sts);
+
+	if ((cmd_sts &
 		(KWGBE_RX_FIRST_DESC | KWGBE_RX_LAST_DESC))
 		!= (KWGBE_RX_FIRST_DESC | KWGBE_RX_LAST_DESC)) {
 
 		printf("Err..(%s) Dropping packet spread on"
 			" multiple descriptors\n", __FUNCTION__);
 
-	} else if (p_rxdesc_curr->cmd_sts & KWGBE_ERROR_SUMMARY) {
+	} else if (cmd_sts & KWGBE_ERROR_SUMMARY) {
 
 		printf("Err..(%s) Dropping packet with errors\n",
 			__FUNCTION__);
@@ -574,7 +580,8 @@ static int kwgbe_recv(struct eth_device *dev)
 	p_rxdesc_curr->buf_size = PKTSIZE_ALIGN;
 	p_rxdesc_curr->byte_cnt = 0;
 
-	dkwgbe->p_rxdesc_curr = p_rxdesc_curr->nxtdesc_p;
+	writel((unsigned)p_rxdesc_curr->nxtdesc_p, &dkwgbe->p_rxdesc_curr);
+
 	return 0;
 }
 
Signature · 1 line
1.6.0.4
Applied
#11 lore raw

PATCH 2/4 : arm: Kirkwood: Fix compiler optimization bug for kwgbe_send

Quoted 11 lines · Simon Kagstrom

kwgbe_send/recv both have loops waiting for the hardware to set a bit. GCC 4.3.3 cleverly optimizes the send case to ... a while(1); loop. This patch uses readl to force a read from device memory. Other volatile accesses have also been replaced with readl/writel where appropriate (as per suggestions on the U-boot mailing list).

---
 drivers/net/kirkwood_egiga.c |   31 +++++++++++++++++++------------
 1 files changed, 19 insertions(+), 12 deletions(-)

Applied to net repo.

thanks,

Applied
#12 lore raw

PATCH 2/4 : arm: Kirkwood: Fix compiler optimization bug for kwgbe_send

In reply to Simon Kagstrom
Quoted 8 lines · Simon Kagstrom

kwgbe_send/recv both have loops waiting for the hardware to set a bit. GCC 4.3.3 cleverly optimizes the send case to ... a while(1); loop. This patch uses readl to force a read from device memory. Other volatile accesses have also been replaced with readl/writel where appropriate (as per suggestions on the U-boot mailing list).

---

Applied to net repo.

thanks,
Ben

#13 lore raw

PATCH 3/4 : arm: Kirkwood: Check the error summary bit for error detection

In reply to Simon Kagstrom

The Marvell documentation for the 88f6281 states that the error coding is only valid if the error summary and last frame bits in the transmit descriptor status field are set. This patch adds checks for these for transmit (I would get transmit errors on bootp with the current check, which I believe are spurious).

---
 drivers/net/kirkwood_egiga.c |    4 +++-
 drivers/net/kirkwood_egiga.h |    1 +
 2 files changed, 4 insertions(+), 1 deletions(-)
drivers/net/kirkwood_egiga.c +3 -1
index 9cab7fd..537343f 100644
@@ -511,7 +511,9 @@ static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
 	cmd_sts = readl(&p_txdesc->cmd_sts);
 	while (cmd_sts & KWGBE_BUFFER_OWNED_BY_DMA) {
 		/* return fail if error is detected */
-		if (cmd_sts & (KWGBE_UR_ERROR | KWGBE_RL_ERROR)) {
+		if ((cmd_sts & (KWGBE_ERROR_SUMMARY | KWGBE_TX_LAST_FRAME)) ==
+				(KWGBE_ERROR_SUMMARY | KWGBE_TX_LAST_FRAME) &&
+				cmd_sts & (KWGBE_UR_ERROR | KWGBE_RL_ERROR)) {
 			printf("Err..(%s) in xmit packet\n", __FUNCTION__);
 			return -1;
 		}
drivers/net/kirkwood_egiga.h +1 -0
index 8b67c9c..9c893d1 100644
@@ -256,6 +256,7 @@
 #define KWGBE_UR_ERROR			(1 << 1)
 #define KWGBE_RL_ERROR			(1 << 2)
 #define KWGBE_LLC_SNAP_FORMAT		(1 << 9)
+#define KWGBE_TX_LAST_FRAME		(1 << 20)
 
 /* Rx descriptors status */
 #define KWGBE_CRC_ERROR			0
Signature · 1 line
1.6.0.4
Applied
#14 lore raw

PATCH 3/4 : arm: Kirkwood: Check the error summary bit for error detection

Quoted 8 lines · Simon Kagstrom

The Marvell documentation for the 88f6281 states that the error coding is only valid if the error summary and last frame bits in the transmit descriptor status field are set. This patch adds checks for these for transmit (I would get transmit errors on bootp with the current check, which I believe are spurious).

---

Applied to net repo.

thanks,
Ben

#15 lore raw

PATCH 4/4 : arm: Kirkwood: See to it that sent data is 8-byte aligned

In reply to Simon Kagstrom

U-boot might use non-8-byte-aligned addresses for sending data, which the kwgbe_send doesn't accept (bootp does this for me). This patch copies the data to be sent to a temporary buffer if it is non-aligned.

---
 drivers/net/kirkwood_egiga.c |   26 ++++++++++++++++++++------
 1 files changed, 20 insertions(+), 6 deletions(-)
drivers/net/kirkwood_egiga.c +20 -6
index 537343f..24269c1 100644
@@ -481,7 +481,7 @@ static int kwgbe_halt(struct eth_device *dev)
 	return 0;
 }
 
-static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
+static int kwgbe_send_aligned(struct eth_device *dev, volatile void *dataptr,
 		      int datasize)
 {
 	struct kwgbe_device *dkwgbe = to_dkwgbe(dev);
@@ -489,11 +489,6 @@ static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
 	struct kwgbe_txdesc *p_txdesc = dkwgbe->p_txdesc;
 	u32 cmd_sts;
 
-	if ((u32) dataptr & 0x07) {
-		printf("Err..(%s) xmit dataptr not 64bit aligned\n",
-			__FUNCTION__);
-		return -1;
-	}
 	p_txdesc->cmd_sts = KWGBE_ZERO_PADDING | KWGBE_GEN_CRC;
 	p_txdesc->cmd_sts |= KWGBE_TX_FIRST_DESC | KWGBE_TX_LAST_DESC;
 	p_txdesc->cmd_sts |= KWGBE_BUFFER_OWNED_BY_DMA;
@@ -522,6 +517,25 @@ static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
 	return 0;
 }
 
+static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
+		      int datasize)
+{
+	static u8 __attribute__((aligned(8))) aligned_buf[9000];
+	void *p = (void *)dataptr;
+
+	if ((u32) dataptr & 0x07) {
+		if (datasize > sizeof(aligned_buf)) {
+			printf("Err..(%s) Non-aligned data too large (%d)\n",
+				__FUNCTION__, datasize);
+			return -1;
+		}
+		memcpy(aligned_buf, p, datasize);
+		p = aligned_buf;
+	}
+
+	return kwgbe_send_aligned(dev, p, datasize);
+}
+
 static int kwgbe_recv(struct eth_device *dev)
 {
 	struct kwgbe_device *dkwgbe = to_dkwgbe(dev);
Signature · 1 line
1.6.0.4
#16 lore raw

PATCH 4/4 : arm: Kirkwood: See to it that sent data is 8-byte aligned

Quoted 9 lines

-----Original Message-----
From: Simon Kagstrom [mailto:simon.kagstrom at netinsight.net]
Sent: Wednesday, July 08, 2009 4:36 PM
To: U-Boot ML; Prafulla Wadaskar
Subject: [PATCH 4/4]: arm: Kirkwood: See to it that sent data is 8-byte aligned

Quoted 4 lines

U-boot might use non-8-byte-aligned addresses for sending data, which the kwgbe_send doesn't accept (bootp does this for me). This patch copies the data to be sent to a temporary buffer if it is non-aligned.

Kirkwood egiga driver may not be the only client for this kind of requirements. Any Ethernet controller using DMA will require aligned data pointer. I would suggest either upper layers should pass the aligned buffer pointers
Or
this patch should go in net/ethic, so that other drivers can benefit

Regards..
Prafulla . .

#17 lore raw

PATCH 4/4 : arm: Kirkwood: See to it that sent data is 8-byte aligned

Quoted 8 lines · On Wed, 8 Jul 2009 05:17:08 -0700 Prafulla Wadaskar <[email protected]>…
Quoted 4 lines

U-boot might use non-8-byte-aligned addresses for sending data, which the kwgbe_send doesn't accept (bootp does this for me). This patch copies the data to be sent to a temporary buffer if it is non-aligned.

Kirkwood egiga driver may not be the only client for this kind of requirements. Any Ethernet controller using DMA will require aligned data pointer. I would suggest either upper layers should pass the aligned buffer pointers
Or
this patch should go in net/ethic, so that other drivers can benefit

Well, I've looked around a bit and other drivers also handle it this way - see greth.c for example. I agree that it would be nice to handle it at an upper layer, but that would require a good code review to find the potential places where this problem could occur. I feel I'm a bit too U-boot green to do that as of now :-)

I'll submit a new patch with this malloc'ed though.

// Simon

#18 lore raw

PATCH 4/4 : arm: Kirkwood: See to it that sent data is 8-byte aligned

Quoted 12 lines · Simon Kagstrom
Quoted 8 lines · On Wed, 8 Jul 2009 05:17:08 -0700 Prafulla Wadaskar <[email protected]>…
Quoted 4 lines

U-boot might use non-8-byte-aligned addresses for sending data, which the kwgbe_send doesn't accept (bootp does this for me). This patch copies the data to be sent to a temporary buffer if it is non-aligned.

Kirkwood egiga driver may not be the only client for this kind of requirements. Any Ethernet controller using DMA will require aligned data pointer. I would suggest either upper layers should pass the aligned buffer pointers
Or
this patch should go in net/ethic, so that other drivers can benefit

Quoted 7 lines

Well, I've looked around a bit and other drivers also handle it this way - see greth.c for example. I agree that it would be nice to handle it at an upper layer, but that would require a good code review to find the potential places where this problem could occur. I feel I'm a bit too U-boot green to do that as of now :-)

I'll submit a new patch with this malloc'ed though.

I agree with Prafulla
Ben what do you think?

Best Regards,
J.

#19 lore raw

PATCH 4/4 : arm: Kirkwood: See to it that sent data is 8-byte aligned

In reply to Simon Kagstrom
Quoted 35 lines · On Wednesday 08 July 2009 13:05:42 Simon Kagstrom wrote:

U-boot might use non-8-byte-aligned addresses for sending data, which the kwgbe_send doesn't accept (bootp does this for me). This patch copies the data to be sent to a temporary buffer if it is non-aligned.

---
 drivers/net/kirkwood_egiga.c |   26 ++++++++++++++++++++------
 1 files changed, 20 insertions(+), 6 deletions(-)
drivers/net/kirkwood_egiga.c +1 -6
index 537343f..24269c1 100644
@@ -481,7 +481,7 @@ static int kwgbe_halt(struct eth_device *dev)
 	return 0;
 }
 
-static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
+static int kwgbe_send_aligned(struct eth_device *dev, volatile void
 *dataptr, int datasize)
 {
 	struct kwgbe_device *dkwgbe = to_dkwgbe(dev);
@@ -489,11 +489,6 @@ static int kwgbe_send(struct eth_device *dev, volatile
 void *dataptr, struct kwgbe_txdesc *p_txdesc = dkwgbe->p_txdesc;
 	u32 cmd_sts;
 
-	if ((u32) dataptr & 0x07) {
-		printf("Err..(%s) xmit dataptr not 64bit aligned\n",
-			__FUNCTION__);
-		return -1;
-	}
 	p_txdesc->cmd_sts = KWGBE_ZERO_PADDING | KWGBE_GEN_CRC;
 	p_txdesc->cmd_sts |= KWGBE_TX_FIRST_DESC | KWGBE_TX_LAST_DESC;
 	p_txdesc->cmd_sts |= KWGBE_BUFFER_OWNED_BY_DMA;

drivers/net/kirkwood_egiga.c static int kwgbe_send(struct eth_device *dev, volatile · line 523

Quoted 7 lines
drivers/net/kirkwood_egiga.c +4 -0
@@ -522,6 +517,25 @@ static int kwgbe_send(struct eth_device *dev, volatile
 void *dataptr, return 0;
 }
 
+static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
+		      int datasize)
+{
+	static u8 __attribute__((aligned(8))) aligned_buf[9000];

I would prefer to malloc such big area's.

And I second Prafulla's comment, that this should be handled by the upper network layer.

Thanks.

Best regards,
Stefan

=====================================================================
DENX Software Engineering GmbH, MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-0 Fax: +49-8142-66989-80 Email: office at denx.de
=====================================================================

#20 lore raw

PATCH 4/4 : arm: Kirkwood: See to it that sent data is 8-byte aligned

Quoted 4 lines

I would prefer to malloc such big area's.

And I second Prafulla's comment, that this should be handled by the upper network layer.

Here is a second version which uses malloc instead. It's still handled in the driver though.

// Simon
--

#21 lore raw

PATCH 4/4 : arm: Kirkwood: See to it that sent data is 8-byte aligned

In reply to Simon Kagstrom

Simon,

Quoted 35 lines · Simon Kagstrom

U-boot might use non-8-byte-aligned addresses for sending data, which the kwgbe_send doesn't accept (bootp does this for me). This patch copies the data to be sent to a temporary buffer if it is non-aligned.

---
 drivers/net/kirkwood_egiga.c |   26 ++++++++++++++++++++------
 1 files changed, 20 insertions(+), 6 deletions(-)
drivers/net/kirkwood_egiga.c +1 -6
index 537343f..24269c1 100644
@@ -481,7 +481,7 @@ static int kwgbe_halt(struct eth_device *dev)
 	return 0;
 }
 
-static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
+static int kwgbe_send_aligned(struct eth_device *dev, volatile void *dataptr,
 		      int datasize)
 {
 	struct kwgbe_device *dkwgbe = to_dkwgbe(dev);
@@ -489,11 +489,6 @@ static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
 	struct kwgbe_txdesc *p_txdesc = dkwgbe->p_txdesc;
 	u32 cmd_sts;
 
-	if ((u32) dataptr & 0x07) {
-		printf("Err..(%s) xmit dataptr not 64bit aligned\n",
-			__FUNCTION__);
-		return -1;
-	}
 	p_txdesc->cmd_sts = KWGBE_ZERO_PADDING | KWGBE_GEN_CRC;
 	p_txdesc->cmd_sts |= KWGBE_TX_FIRST_DESC | KWGBE_TX_LAST_DESC;
 	p_txdesc->cmd_sts |= KWGBE_BUFFER_OWNED_BY_DMA;

drivers/net/kirkwood_egiga.c static int kwgbe_send(struct eth_device *dev, volatile void *dataptr, · line 524

Quoted 8 lines
drivers/net/kirkwood_egiga.c +4 -0
@@ -522,6 +517,25 @@ static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
 	return 0;
 }
 
+static int kwgbe_send(struct eth_device *dev, volatile void *dataptr,
+		      int datasize)
+{
+	static u8 __attribute__((aligned(8))) aligned_buf[9000];
 

Why do you need to send a jumbo frame? U-boot is hard-coded in some ways to only handle frames with an MTU of 1518 bytes on Rx.

regards,
Ben