e1000 packet corruption problem

e1000 packet corruption problem

[Michal Vanco]

[-- Attachment #1: Type: text/plain, Size: 1680 bytes --]

Hi,

I saw this issue mentioned several times in various lists, but no
satisfactory answer was given.

I have 4-port Intel e1000 card in my dual Amd Opteron machine:
0000:02:04.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
Controller (rev 02)
0000:02:05.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
Controller (rev 02)
0000:02:06.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
Controller (rev 02)
0000:02:07.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
Controller (rev 02)

with this driver:
Intel(R) PRO/1000 Network Driver - version 5.7.6
Copyright (c) 1999-2004 Intel Corporation.
ACPI: PCI interrupt 0000:02:04.0[A] -> GSI 28 (level, low) -> IRQ 201
e1000: eth3: e1000_probe: Intel(R) PRO/1000 Network Connection
ACPI: PCI interrupt 0000:02:05.0[A] -> GSI 29 (level, low) -> IRQ 209
e1000: eth4: e1000_probe: Intel(R) PRO/1000 Network Connection
ACPI: PCI interrupt 0000:02:06.0[A] -> GSI 30 (level, low) -> IRQ 217
e1000: eth5: e1000_probe: Intel(R) PRO/1000 Network Connection
ACPI: PCI interrupt 0000:02:07.0[A] -> GSI 31 (level, low) -> IRQ 225
e1000: eth6: e1000_probe: Intel(R) PRO/1000 Network Connection

eth3 is autonegotiated at 100 Mbps FDX.

Trying to forward packets through eth3 causes corruption of packets.
I've got 'Corrupted MAC on input' trying to download (or copy) something
using scp. Using ftp doesn't emit any error, but all files downloaded
are apparently corrupted.

After that I've tried to disable all off-loading. In this case no error
is visible but all download stalls forever.

Is this problem related only to e1000 driver or any GigE cards? Is there
any fix available?

regards,
michal

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 256 bytes --]

Re: e1000 packet corruption problem

[Michal Vanco]

[-- Attachment #1: Type: text/plain, Size: 3355 bytes --]

Brandeburg, Jesse wrote:
>>I have 4-port Intel e1000 card in my dual Amd Opteron machine:
>>0000:02:04.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
>>Controller (rev 02)
>>0000:02:05.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
>>Controller (rev 02)
>>0000:02:06.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
>>Controller (rev 02)
>>0000:02:07.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
>>Controller (rev 02)
>>
>>with this driver:
>>Intel(R) PRO/1000 Network Driver - version 5.7.6
>>Copyright (c) 1999-2004 Intel Corporation.
>
>
> Standard questions: what kernel version, what exact machine (and bios),
> lspci -n will help here.

Ooops. Sorry. It's 2.6.12-rc1 but the same behaviour on 2.6.10 and 2.6.11.

# cat /proc/cpuinfo | egrep 'proc|model name'
processor       : 0
model name      : AMD Opteron(tm) Processor 246
processor       : 1
model name      : AMD Opteron(tm) Processor 246

# lspci -n
0000:00:06.0 0604: 1022:7460 (rev 07)
0000:00:07.0 0601: 1022:7468 (rev 05)
0000:00:07.1 0101: 1022:7469 (rev 03)
0000:00:07.2 0c05: 1022:746a (rev 02)
0000:00:07.3 0680: 1022:746b (rev 05)
0000:00:0a.0 0604: 1022:7450 (rev 12)
0000:00:0a.1 0800: 1022:7451 (rev 01)
0000:00:0b.0 0604: 1022:7450 (rev 12)
0000:00:0b.1 0800: 1022:7451 (rev 01)
0000:00:18.0 0600: 1022:1100
0000:00:18.1 0600: 1022:1101
0000:00:18.2 0600: 1022:1102
0000:00:18.3 0600: 1022:1103
0000:00:19.0 0600: 1022:1100
0000:00:19.1 0600: 1022:1101
0000:00:19.2 0600: 1022:1102
0000:00:19.3 0600: 1022:1103
0000:01:03.0 0604: 12d8:8154 (rev 01)
0000:02:04.0 0200: 8086:100e (rev 02)
0000:02:05.0 0200: 8086:100e (rev 02)
0000:02:06.0 0200: 8086:100e (rev 02)
0000:02:07.0 0200: 8086:100e (rev 02)
0000:03:06.0 0100: 9005:801d (rev 10)
0000:03:06.1 0100: 9005:801d (rev 10)
0000:03:09.0 0200: 14e4:1648 (rev 03)
0000:03:09.1 0200: 14e4:1648 (rev 03)
0000:04:00.0 0c03: 1022:7464 (rev 0b)
0000:04:00.1 0c03: 1022:7464 (rev 0b)
0000:04:05.0 0180: 1095:3114 (rev 02)
0000:04:06.0 0300: 1002:4752 (rev 27)
0000:04:08.0 0200: 8086:1229 (rev 10)

>
> You might be having interrupt routing problems, have you tried
> pci=noapic as a boot parameter?
>
> The data corruption is puzzling.
>

I didn't but I'll tomorrow.

>
>>eth3 is autonegotiated at 100 Mbps FDX.
>>
>>Trying to forward packets through eth3 causes corruption of packets.
>>I've got 'Corrupted MAC on input' trying to download (or copy) something
>>using scp. Using ftp doesn't emit any error, but all files downloaded
>>are apparently corrupted.
>>
>>After that I've tried to disable all off-loading. In this case no error
>>is visible but all download stalls forever.
>
>
> How did you disable all offloading? With ethtool? Did you disable TSO
> and TX/RX checksumming?  You left scatter gather on, right?
>

Well. Actually I did this:

# ethtool -K eth3 rx off tx off sg off tso off

>
>>Is this problem related only to e1000 driver or any GigE cards? Is
>
> there
>
>>any fix available?
>
>
> Can you try the card in another (non opteron) machine?  Does your
> opteron machine pci bus support PCI-X/133?
>

I did testing on Wi****s 2K3 and it worked perfectly on the same machine.

> If we can reproduce this we can likely get a fix, but without a
> reproduction its unlikely because we haven't seen problems like this
> around here.
>
> Jesse
>

regards,
michal

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 256 bytes --]

RE: e1000 packet corruption problem

[Brandeburg, Jesse]

>I have 4-port Intel e1000 card in my dual Amd Opteron machine:
>0000:02:04.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
>Controller (rev 02)
>0000:02:05.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
>Controller (rev 02)
>0000:02:06.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
>Controller (rev 02)
>0000:02:07.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet
>Controller (rev 02)
>
>with this driver:
>Intel(R) PRO/1000 Network Driver - version 5.7.6
>Copyright (c) 1999-2004 Intel Corporation.

Standard questions: what kernel version, what exact machine (and bios),
lspci -n will help here.

You might be having interrupt routing problems, have you tried
pci=noapic as a boot parameter?

The data corruption is puzzling.

>eth3 is autonegotiated at 100 Mbps FDX.
>
>Trying to forward packets through eth3 causes corruption of packets.
>I've got 'Corrupted MAC on input' trying to download (or copy)
something
>using scp. Using ftp doesn't emit any error, but all files downloaded
>are apparently corrupted.
>
>After that I've tried to disable all off-loading. In this case no error
>is visible but all download stalls forever.

How did you disable all offloading? With ethtool? Did you disable TSO
and TX/RX checksumming?  You left scatter gather on, right?

>Is this problem related only to e1000 driver or any GigE cards? Is
there
>any fix available?

Can you try the card in another (non opteron) machine?  Does your
opteron machine pci bus support PCI-X/133?

If we can reproduce this we can likely get a fix, but without a
reproduction its unlikely because we haven't seen problems like this
around here.

Jesse

Re: e1000 packet corruption problem

[Michal Vanco]

[-- Attachment #1: Type: text/plain, Size: 295 bytes --]

Brandeburg, Jesse wrote:
> You might be having interrupt routing problems, have you tried
> pci=noapic as a boot parameter?
>
hi

pci=noapic doesn't work (PCI: Unknown option `noapic' in dmesg).

tried also noapic and pci=noacpi params with the same result (packet
corruption).

regards,
michal

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 256 bytes --]

Re: e1000 packet corruption problem

[Michal Vanco]

[-- Attachment #1: Type: text/plain, Size: 392 bytes --]

Hello,

I've made some more testing. This one was pretty surprising:

>From 192.168.0.2:

# sudo ping 192.168.0.1 -p fefffefffefffeff -i 0.01 -c 100
100 packets transmitted, 100 received, 0% packet loss, time 901ms

# sudo ping 192.168.0.1 -p fffefffefffefffe -i 0.01 -c 100
100 packets transmitted, 18 received, 82% packet loss, time 1068ms

Any ideas of how to solve this?

regards,
michal

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 256 bytes --]