Bug 30092 - smsc911x.c drops long packets with VLAN tags
Summary: smsc911x.c drops long packets with VLAN tags
Status: RESOLVED CODE_FIX
Alias: None
Product: Drivers
Classification: Unclassified
Component: Network (show other bugs)
Hardware: All Linux
: P1 normal
Assignee: drivers_network@kernel-bugs.osdl.org
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2011-02-28 15:56 UTC by Göran Weinholt
Modified: 2012-08-17 10:57 UTC (History)
1 user (show)

See Also:
Kernel Version: 2.6.36.4
Subsystem:
Regression: No
Bisected commit-id:


Attachments

Description Göran Weinholt 2011-02-28 15:56:39 UTC
The smsc911x.c driver can't receive packets of the maximum MTU if they have a 802.1q tag. Testing with ping -s 1500 shows that the frame with the first IP segment is dropped.

The source of the problem is this line at smsc911x.c:1023 (smsc911x_poll):

  if (unlikely(rxstat & RX_STS_ES_)) {

The datasheet at http://www.smsc.com/media/Downloads_Public/Data_Sheets/9221.pdf describes this bit as a logical OR of bits 11, 7, 6 and 1. The problem is that bit 7 is set if the packet is longer than 1518 bytes, which is true for the 802.1q tagged packets. As a quick and dirty solution, I replaced RX_STS_ES_ with (1<<11|1<<6|1<<1), and now 802.1q tagged packets are received just fine.

This was tested on an IGEPv2, which has a SMSC LAN9221i.
Comment 1 Andrew Morton 2011-03-02 00:16:12 UTC
(switched to email.  Please respond via emailed reply-to-all, not via the
bugzilla web interface).

On Mon, 28 Feb 2011 15:57:14 GMT
bugzilla-daemon@bugzilla.kernel.org wrote:

> https://bugzilla.kernel.org/show_bug.cgi?id=30092
> 
>            Summary: smsc911x.c drops long packets with VLAN tags
>            Product: Drivers
>            Version: 2.5
>     Kernel Version: 2.6.36.4
>           Platform: All
>         OS/Version: Linux
>               Tree: Mainline
>             Status: NEW
>           Severity: normal
>           Priority: P1
>          Component: Network
>         AssignedTo: drivers_network@kernel-bugs.osdl.org
>         ReportedBy: weinholt@csbnet.se
>         Regression: No
> 
> 
> The smsc911x.c driver can't receive packets of the maximum MTU if they have a
> 802.1q tag. Testing with ping -s 1500 shows that the frame with the first IP
> segment is dropped.
> 
> The source of the problem is this line at smsc911x.c:1023 (smsc911x_poll):
> 
>   if (unlikely(rxstat & RX_STS_ES_)) {
> 
> The datasheet at
> http://www.smsc.com/media/Downloads_Public/Data_Sheets/9221.pdf describes
> this
> bit as a logical OR of bits 11, 7, 6 and 1. The problem is that bit 7 is set
> if
> the packet is longer than 1518 bytes, which is true for the 802.1q tagged
> packets. As a quick and dirty solution, I replaced RX_STS_ES_ with
> (1<<11|1<<6|1<<1), and now 802.1q tagged packets are received just fine.
> 
> This was tested on an IGEPv2, which has a SMSC LAN9221i.
> 

Thanks.  Become famous, get more girls: send us a patch as per
Documentation/SubmittingPatches :)
Comment 2 Göran Weinholt 2011-03-02 15:02:04 UTC
Andrew Morton <akpm@linux-foundation.org> writes:

> (switched to email.  Please respond via emailed reply-to-all, not via the
> bugzilla web interface).
>
> On Mon, 28 Feb 2011 15:57:14 GMT
> bugzilla-daemon@bugzilla.kernel.org wrote:
>
>> https://bugzilla.kernel.org/show_bug.cgi?id=30092
>> 
> Thanks.  Become famous, get more girls: send us a patch as per
> Documentation/SubmittingPatches :)

There's a register called VLAN1 that wasn't being configured. Setting it
to 0x8100 fixes the problem I reported. There's also a VLAN2 register,
possibly meant for stacked VLANs, but the description in the datasheet
doesn't make sense to me, and I can't test stacked VLANs, so I'm not
going to change it.

Subject: [PATCH] net/smsc911x.c: Set the VLAN1 register to fix VLAN MTU problem
From: Göran Weinholt <weinholt@csbnet.se>

The smsc911x driver would drop frames longer than 1518 bytes, which is a
problem for networks with VLAN tagging. The VLAN1 tag register is used
to increase the legal frame size to 1522 when a VLAN tag is identified.

Signed-off-by: Göran Weinholt <weinholt@csbnet.se>
---
diff -uprN linux-2.6.36.4-vanilla/drivers/net/smsc911x.c linux-2.6.36.4/drivers/net/smsc911x.c
--- linux-2.6.36.4-vanilla/drivers/net/smsc911x.c	2011-02-18 00:14:38.000000000 +0100
+++ linux-2.6.36.4/drivers/net/smsc911x.c	2011-03-02 14:29:11.884382251 +0100
@@ -1178,6 +1178,11 @@ static int smsc911x_open(struct net_devi
 	smsc911x_reg_write(pdata, HW_CFG, 0x00050000);
 	smsc911x_reg_write(pdata, AFC_CFG, 0x006E3740);
 
+	/* Increase the legal frame size of VLAN tagged frames to 1522 bytes */
+	spin_lock_irq(&pdata->mac_lock);
+	smsc911x_mac_write(pdata, VLAN1, ETH_P_8021Q);
+	spin_unlock_irq(&pdata->mac_lock);
+
 	/* Make sure EEPROM has finished loading before setting GPIO_CFG */
 	timeout = 50;
 	while ((smsc911x_reg_read(pdata, E2P_CMD) & E2P_CMD_EPC_BUSY_) &&
Comment 3 David S. Miller 2011-03-07 20:30:54 UTC
From: weinholt@csbnet.se (Göran Weinholt)
Date: Wed, 02 Mar 2011 15:07:21 +0100

> Subject: [PATCH] net/smsc911x.c: Set the VLAN1 register to fix VLAN MTU
> problem
> From: Göran Weinholt <weinholt@csbnet.se>
> 
> The smsc911x driver would drop frames longer than 1518 bytes, which is a
> problem for networks with VLAN tagging. The VLAN1 tag register is used
> to increase the legal frame size to 1522 when a VLAN tag is identified.
> 
> Signed-off-by: Göran Weinholt <weinholt@csbnet.se>

Applied, thanks.
Comment 4 Anonymous Emailer 2011-03-08 14:29:35 UTC
Reply-To: Steve.Glendinning@smsc.com

> > Subject: [PATCH] net/smsc911x.c: Set the VLAN1 register to fix VLAN 
MTU problem
> > From: Göran Weinholt <weinholt@csbnet.se>
> > 
> > The smsc911x driver would drop frames longer than 1518 bytes, which is 
a
> > problem for networks with VLAN tagging. The VLAN1 tag register is used
> > to increase the legal frame size to 1522 when a VLAN tag is 
identified.
> > 
> > Signed-off-by: Göran Weinholt <weinholt@csbnet.se>
> 
> Applied, thanks.

Thanks David, I was just about to ack this but you beat me to it :-)

As Göran said, the datasheet is a little vague.  I've checked here, and 
there aren't any bad implications of always turning this option on.

Acked-by: Steve Glendinning <steve.glendinning@smsc.com>

Note You need to log in before you can comment on or make changes to this bug.