LX2162A clearfog SFP+ optical transceivers not working

NXP LX2160
1

Hey!
I’m using LX2162A SOM (rev 1.1) + LX2162A Clearfog board (rev.2). So these are the most recent HW revisions for this time.
I connected and configured DAC cable to dpmac.5 and dpmac.6 (due to article https://solidrun.atlassian.net/wiki/spaces/developer/pages/199131187/ClearFog+LX2162A+Quick+Start+Guide) - and everything works ok. Both ethernet interfaces are up and running.
Then I did the same with SFP+ optical cable - linux was not able to read EEPROM inside SFP+ modules and the link was down:

[  504.572693] sfp sfp-at: please wait, module slow to respond
[  560.864692] sfp sfp-at: failed to read EEPROM: -6

I’ve found 2 HW bugs with LX2162A clearfog board:

  1. Low 3.3 voltage on the board. I got only 2.6V instead of 3.3V.
    Due to schematics 3.3V is passed through BAT54C diode D48 which has about 800mV forward voltage drop at 100mA:
    image
    So i have 3.3V at point 1 and 2.6V at point 3. This is extremely low for EEPROM inside SFP+ transceivers - thats why EEPROM is not functional. Also this 3.3V (2.6V in fact) is feeding usb hub + marvell phy chip (haven’t tested them yet).
    I’ve shorted points 1 and 3 on diode D48 and got true 3.3V on the board. After that linux was able to read EEPROM inside SFP+ module.
    But link was still down.

  2. TX_DISABLE (AT_TX_DIS) pin on SFP+ connector is left floating (not connected to anywhere).

    2
    21107×891 84.7 KB

This means that SFP+ transceiver is turned off. I tied this pin to ground and after that i got fully functional SFP+ optical connection.

@jnettlet - please address these bugs.

2

Your post has been forwarded to the hardware team.

3

Hello everyone.

I got into what it seems the same issue but with SFP28 DACs (10Gtek 25G SFP28 DAC 3-m). I looked into the board and SOM and found no printed hardware versions. I couldn’t find that information on u-boot either. No labels were found on the shipping box which have that information.

Once the OS is running, the voltage on the D48 pin 3 is around 2.6 V and I get the same error messages on dmesg and no connection (voltage drops to around 2.58 V). If I add a power supply on the ATX 3.3 V pin to raise the 3.3 V rail voltage to 2.7 (before attaching the SFP28), the error messages are gone (voltage drops to around 2.62 V) but I still get no comms. I also tested an optical transceiver, just to get the voltage reading and it reported 0.0 V. The same transceiver, on a PC, it reported 3.3 V. (ethtool -m ethX)

On the board I cannot find the path of TX_DISABLE to ground it and test. However, on u-boot, I am able to perform DHCP requests and those are received on a PC. On that state, the 3.3 V rail voltage is around 2.7 V (without power supply on ATX 3.3V hack) and I see two options for TX_DISABLE:

  • Current draw is lower than during OS runtime and the noise collected by the “open” TX_DISABLE path is low enough to make it to be evaluated as 0 (grounded);
  • The TX_DISABLE path is connected actually to some controller (GPIO/…) and it is incorrectly set by the OS.

I have a few questions:

  • Were this two issues solved on newer hardware releases?
  • It is actually shortening the pins 1 and 3 of D48 the correct way to move forward?
  • Solution for the TX_DISABLE path to get the SFP28 working correctly?
  • Where to get the hardware versions from?

Thanks.

4

Here is an errata from SolidRun with instructions and pictures

5

Actually, I will use this opportunity to suggest https://dev.solid-run.com/. We have worked hard to make it a much nicer documentation experience than the old atlassian site.

6

I performed the hardware changes as per errata and the SFP28 ports at 25 Gbit/s are not yet working at that speed.

What I got working was 4x 10 Gbit/s, by using the firmware which sets SERDES to 3 9 0. For this I used 1x 10G DAC and 3x 10G RJ45 SFPs from different vendors. The firmware image was built from 4e90f55, both SERDES options.

By using the firmware which sets SERDES to 18 9 0, I only got 1x 10 Gbit/s working. For this I used 2x 10G RJ45 SFPs from different vendors and 2x 25G DACs (10Gtek CAB-ZSP/ZSP-P3M).

I also connected directly both SFP28 cages by using the cables CAB-ZSP/ZSP-P3M and CAB-ZSP/ZSP-P0.5M without success.

According to the readme on git (kernel 6.6.52), 25 Gbit/s speeds are not supported. The same disclaimer is not present on the kernel 5.15.71, but I prefer to use the latest possible kernel. The same readme has related information about the retimer, which it seems that the DACs (passive) I am using does not have. On the other hand, the page “SFP Modules: Tested on LX216x” has at least one (active) DAC at 25 Gbit/s. Is this information up-to-date?

Thanks.