TMDS243EVM: CONNECTION WITH UBUNTU THROUGH PCIe

Marta de Madariaga

Part Number: TMDS243EVM
Other Parts Discussed in Thread: UNIFLASH

Tool/software:

Hi all,

So I have a sitara board AM243, that i want to connect to a Dell Precision 5820 tower with ubuntu by PCIe using this cable: https://www.digikey.ch/en/products/detail/3m/8KJ2-0743-0250/5128345, I have tried to execute the example shown here: AM243x MCU+ SDK: PCIE EP Enumeration, being the board the EP and the ubuntu server the RC.

I have modified the resistors and jumpers on the board:

remove Resistors R661, R662, R667 & R668
populate Resistors R665, R666, R679 & R680 (all 0 ohm)
remove j34 jumper

and i made sure that the ubuntu that im using (ubuntu 24.04 with the linux kernel 6.8.0-38-generic), had these configurations:

CONFIG_IOMMU, CONFIG_INTEL_IOMMU
CONFIG_VFIO, CONFIG_VFIO_PCI
GRUB_CMDLINE_LINUX="intel_iommu=on"

I flashed the pcie_enumerate_ep to the OSPI of the board using the default_sbl_ospi.cfg configuration file and the uart_uniflash.py script:

"python uart_uniflash.py -p COM13 --cfg=sbl_prebuilt/am243x-evm/default_sbl_ospi.cfg"

and connected it through a pci cable to the pci slot1 of the dell tower.

When i turn the tower on, and powers on the board this is all i get:

And the ubuntu does not recognize the board on the lspci, also in the BIOS system information the slot appears as empty.

Things we have tried:

rebooting the ubuntu so that the board is already booted when it starts, is still doesnt recognize it.
try if the server has hot plug for pci, it doesn't.
connect an ethernet controller board to the pci slot, it detect it.
use the default_sbl_null.cfg configuration file to see igf with other bootloader boots faster and gets detected, it doesn't.
turn the ubuntu server after having the board turned on and waiting for link (powered by the power cable instead of directly from PCI cable, didn't work either.
set in BIOS PCIe linkspeed as GEN2, didnt work.

If anyone has an idea of what im doing wrong or what i shoud try, id love to hear :)

thank you,

Marta.

over 1 year ago

0 Ashwani Goel over 1 year ago

TI__Mastermind 27440 points

Hi Marta,

Thanks for your query.

Hopefully, below documentation will be a help.

AM64x and AM243x: MCU+ SDK-Based PCIe End Point (ti.com)

Regards

Ashwani

0 Dominic Rath over 1 year ago

Mastermind 7560 points

Hello Marta,

Marta de Madariaga said:
using this cable: https://www.digikey.ch/en/products/detail/3m/8KJ2-0743-0250/5128345

this isn't the recommended cable. Could you verify how your cable connects the signals on both ends? Especially RefClk, Lane 0 transmit data and Lane 0 receive data:

(https://en.wikipedia.org/wiki/PCI_Express)

From the 3M description "THIS IS A FEED THROUGH ASSEMBLY. THE PIN ASSIGNMENTS REMAIN THE SAME ON BOTH ENDS" I assume that it doesn't cross-over the transmitter/receiver lanes, which I believe is necessary.

The recommended cable is a PE-FLEX1-G2.MMCX-12-TI1. I got one at the office, and can check later if my assumption is correct.

One more thing to try it J35. This controls how PRNST is connected on the board. With the recommended cable this shouldn't matter, as the cable internally connects PRSTN1# and PRSTN2#, but if your cable doesn't, you should put jumper J35 in the 2+3 position.

Regards,

Dominic

0 Dominic Rath over 1 year ago in reply to Dominic Rath

Mastermind 7560 points

Hello Marta,

I verified that the PE-FLEX1-G2.MMCX-12-TI1 connects transmit lane on one end with the receive lane on the other end and vice versa. In the pinout above B14/15 connect to A16/17.

The RefClk pins (A13/A14) are connected 1:1. This might require soldering 0-Ohm resistors on the Adex Electronics cable (we have three such cables, and all three came without RefClk connection).

I also verified that the Adex cable connects PRSTN1# and PRSTN2#. If the 3M cable doesn't, you need J35 in the 2+3 position.

It would be great if you could measure the 3M cable and reply here with your findings.

Regards,

Dominic

0 Marta de Madariaga over 1 year ago in reply to Dominic Rath

Prodigy 10 points

Good morning Dominic,

We checked the Adex cable as you suggested, and it was crossing the reception and transmission lines. We tried to replicate this manually with our cable, as a temporary solution until we get the recommended cable, but it still didn't work. Jumper 35 is placed as you mentioned, and the rest of the transistors changed as mentioned in the first message:

I am uploading some images of the clock signals we are measuring since some of them are concerning us a little bit.

This is the clock on J27 pins

Clock on R680, R679 (left side)

Clock on R680, R679 (right side)

Thanks a lot for your feedback, and please let us know if you see anything in the images that we should change or check.

0 Ashwani Goel over 1 year ago in reply to Marta de Madariaga

TI__Mastermind 27440 points

Hi Marta de Madariaga,

Thanks for info.

We will check internally and get back to you.

Regards

Ashwani

0 Ashwani Goel over 1 year ago in reply to Ashwani Goel

TI__Mastermind 27440 points

Hi Marta de Madariaga

Thanks for having patience.

It took long time as I was out for a week.

The board (EVM) and cable modification looks to be okay.

Did you get chance to run more test on this setup?

Any follow-up question?

Regards

Ashwani

0 Marta de Madariaga over 1 year ago in reply to Ashwani Goel

Prodigy 10 points

Hello!

we made the same modifications in another board and we could detect that, seems like there was some resistor poorly soldered in the original one. But we could fix that one too.

Thanks for the support

Marta.

0 Ashwani Goel over 1 year ago in reply to Marta de Madariaga

TI__Mastermind 27440 points

Thanks Marta for confirmation and choosing TI

Regards

Ashwani

Arm-based microcontrollers

Arm-based microcontrollers forum

TMDS243EVM: CONNECTION WITH UBUNTU THROUGH PCIe