Skip to contentSkip to conversation

sun neptune mis-detecting ethernet crc faults?

DISCUSSION Chris Friesen 3 participants 3 messages Jun 29, 2009

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

Hi all,

David Miller is busy and suggested someone on the list might be able to help.

We have some boards using the Sun Neptune ethernet adapters. We're seeing behaviour that at this point looks like a hardware glitch in the ethernet CRC validation on the receive path. It appears to be incorrectly detecting a corrupt CRC and dropping the frames. (We've enabled port mirroring on the switch and the frames are received without errors on the eavesdropper board.)

The odd thing is that we're using a TCP connection and once the CRC glitch shows up for a particular chunk of data it continues to drop all the retransmissions for that chunk as having bad CRCs, even though their CRC values are totally different due to different embedded timestamps.

Has anyone heard of anything like this on the Neptune hardware? MTU is set to 2000 if it matters, though we're planning on retesting with it set to 1500.

I'm considering disabling the hardware CRC check as a verification--looking at the niu driver I think I should be able to do this by not including XMAC_CONFIG_RX_CRC_CHK_DIS in the big list of flags being OR'd in niu_init_rx_xmac().

Anyone have any suggestions?

Thanks,

Chris

#1 lore raw
Quoted 11 lines · Chris Friesen

Hi all,

David Miller is busy and suggested someone on the list might be able to help.

We have some boards using the Sun Neptune ethernet adapters. We're seeing behaviour that at this point looks like a hardware glitch in the ethernet CRC validation on the receive path. It appears to be incorrectly detecting a corrupt CRC and dropping the frames. (We've enabled port mirroring on the switch and the frames are received without errors on the eavesdropper board.)

A simplistic question, but are you sure that the eavesdropper board is checking CRCs?

Quoted 4 lines

The odd thing is that we're using a TCP connection and once the CRC glitch shows up for a particular chunk of data it continues to drop all the retransmissions for that chunk as having bad CRCs, even though their CRC values are totally different due to different embedded timestamps.

Do you mean TCP timestamp options?

Quoted 1 line

Has anyone heard of anything like this on the Neptune hardware?

Can't say as I have, but the history of "networking" is littered with data pattern induced bugs in all manner of hardware.

Quoted 2 lines

MTU is set to 2000 if it matters, though we're planning on retesting with it set to 1500.

An MTU of 2000 bytes means a TCP segment with timestamps enabled will be 2032 plus the ethernet header (assuming no vlan tags) of 14 bytes for 2046 and then there is the trailing CRC - which is getting very close to a magic power of two boundary, another place where history is repleat with examples of bugs. One that comes to mind is that the old Alteon AceNICs got very unhappy if one crossed a 4G boundary with a DMA...

rick jones

Quoted 15 lines

I'm considering disabling the hardware CRC check as a verification--looking at the niu driver I think I should be able to do this by not including XMAC_CONFIG_RX_CRC_CHK_DIS in the big list of flags being OR'd in niu_init_rx_xmac().

Anyone have any suggestions?

Thanks,

Chris
--
To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#2 lore raw
Quoted 13 lines · Rick Jones
Quoted 11 lines · Chris Friesen

Hi all,

David Miller is busy and suggested someone on the list might be able to help.

We have some boards using the Sun Neptune ethernet adapters. We're seeing behaviour that at this point looks like a hardware glitch in the ethernet CRC validation on the receive path. It appears to be incorrectly detecting a corrupt CRC and dropping the frames. (We've enabled port mirroring on the switch and the frames are received without errors on the eavesdropper board.)

Quoted 10 lines

A simplistic question, but are you sure that the eavesdropper board is checking CRCs?

Quoted 4 lines

The odd thing is that we're using a TCP connection and once the CRC glitch shows up for a particular chunk of data it continues to drop all the retransmissions for that chunk as having bad CRCs, even though their CRC values are totally different due to different embedded timestamps.

Do you mean TCP timestamp options?

Quoted 1 line

Has anyone heard of anything like this on the Neptune hardware?

At Sun, we haven't seen such RX CRC error before.

Quoted 16 lines

Can't say as I have, but the history of "networking" is littered with data pattern induced bugs in all manner of hardware.

Quoted 2 lines

MTU is set to 2000 if it matters, though we're planning on retesting with it set to 1500.

An MTU of 2000 bytes means a TCP segment with timestamps enabled will be 2032 plus the ethernet header (assuming no vlan tags) of 14 bytes for 2046 and then there is the trailing CRC - which is getting very close to a magic power of two boundary, another place where history is repleat with examples of bugs. One that comes to mind is that the old Alteon AceNICs got very unhappy if one crossed a 4G boundary with a DMA...

rick jones

Quoted 5 lines

I'm considering disabling the hardware CRC check as a verification--looking at the niu driver I think I should be able to do this by not including XMAC_CONFIG_RX_CRC_CHK_DIS in the big list of flags being OR'd in niu_init_rx_xmac().

That is right.

Regards,
Matheos

Quoted 15 lines
Quoted 10 lines

Anyone have any suggestions?

Thanks,

Chris

Signature · 3 lines
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Signature · 3 lines
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html