Bad UDP checksum with 82540EM

Bad UDP checksum with 82540EM

[Andreas Schwab]

I'm getting bad UPD checksums in outgoing packets with the e1000 driver
(version 5.2.16) in 2.6.1 when using HW checksumming on a HP branded
82540EM on ia64 (TCP works fine).  The same driver works fine with an
Intel branded chip.  The two devices are identified as follows:

  Model: "Hewlett-Packard Company 82540EM Gigabit Ethernet Controller"
  Vendor: pci 0x8086 "Intel Corporation"
  Device: pci 0x100e "82540EM Gigabit Ethernet Controller"
  SubVendor: pci 0x103c "Hewlett-Packard Company"
  SubDevice: pci 0x1274 

  Model: "Intel 82540EM Gigabit Ethernet Controller"
  Vendor: pci 0x8086 "Intel Corporation"
  Device: pci 0x100e "82540EM Gigabit Ethernet Controller"
  SubVendor: pci 0x8086 "Intel Corporation"
  SubDevice: pci 0x3402 

After disabling tx-checksumming UDP is working again.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Bad UDP checksum with 82540EM

[Jon D Mason]

This probably isn't helpful, but it sounds like a hardware error.  Is the 
error occurring on multiple/all HP adapters, or only one? 

Also, your driver code is old, the latest is 5.2.30.1 (goto 
http://sourceforge.net/projects/e1000/).  I don't think that has anything 
to do with it though.

Jon Mason                jonmason@us.ibm.com
Software Engineer        Phone:(512)838.4162
Linux eServer I/O        Fax:  (512)838.3509

Re: Bad UDP checksum with 82540EM

[Andreas Schwab]

Jon D Mason <jonmason@us.ibm.com> writes:

> This probably isn't helpful, but it sounds like a hardware error.  Is the 
> error occurring on multiple/all HP adapters, or only one? 

I've also seen it with a HP branded BCM5701 on IA64, but also with a
Broadcom BCM5702 on AMD64 (both using the tg3 driver).  It seems like
broken UDP checksumming is rather common. :-(

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Bad UDP checksum with 82540EM

[Andi Kleen]

On Sat, 07 Feb 2004 23:42:21 +0100
Andreas Schwab <schwab@suse.de> wrote:

> Jon D Mason <jonmason@us.ibm.com> writes:
> 
> > This probably isn't helpful, but it sounds like a hardware error.  Is the 
> > error occurring on multiple/all HP adapters, or only one? 
> 
> I've also seen it with a HP branded BCM5701 on IA64, but also with a
> Broadcom BCM5702 on AMD64 (both using the tg3 driver).  It seems like
> broken UDP checksumming is rather common. :-(

It could be still a software bug.  When hardware checksumming is available
the UDP packets use a slightly different path through the stack.

Could you perhaps test if the problem occurs in a 32bit box with the 
same NIC ? Maybe it is some 64bit problem somewhere in software.

-Andi

Re: Bad UDP checksum with 82540EM

[Andreas Schwab]

Andi Kleen <ak@suse.de> writes:

> On Sat, 07 Feb 2004 23:42:21 +0100
> Andreas Schwab <schwab@suse.de> wrote:
>
>> Jon D Mason <jonmason@us.ibm.com> writes:
>> 
>> > This probably isn't helpful, but it sounds like a hardware error.  Is the 
>> > error occurring on multiple/all HP adapters, or only one? 
>> 
>> I've also seen it with a HP branded BCM5701 on IA64, but also with a
>> Broadcom BCM5702 on AMD64 (both using the tg3 driver).  It seems like
>> broken UDP checksumming is rather common. :-(
>
> It could be still a software bug.  When hardware checksumming is available
> the UDP packets use a slightly different path through the stack.
>
> Could you perhaps test if the problem occurs in a 32bit box with the 
> same NIC ? Maybe it is some 64bit problem somewhere in software.

Just tested with a "3Com 3C996B-T 1000Base-T" (BCM5701 based, with tg3
driver) on an Athlon.  It shows the same bug, and disabling HW
checksumming on tx fixes it.  So it doesn't look like a 64bit issue.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Bad UDP checksum with 82540EM

[Andreas Schwab]

"cramerj" <cramerj@intel.com> writes:

> Just for clarification...how are you determining a bad checksum?  Are
> you using tcpdump/ethereal on those test machines (that you're
> transmitting from), or are you capturing packets on the wire (from some
> other system receiving the packets).

Actually both.  I have first noticed the issue because my DHCP clients
didn't accept the answers from the server.

> If the former, then please see the following thread.
>
> http://marc.theaimsgroup.com/?t=107422099800006&r=1&w=2
>
> Is this perhaps the issue you're seeing?

I just double checked, the packets are really going out with a bad
checksum.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Maxfeldstraße 5, 90409 Nürnberg, Germany
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."