FPGA Intellectual Property
PCI Express*, Networking and Connectivity, Memory Interfaces, DSP IP, and Video IP
6400 Discussions

Problem with CXL IP Design Example Type 3 on DK-DEV-AGI027RBES

xalverti
New Contributor I
1,330 Views

Hi,

 

We are using Quartus 24.1 to compile the type 3 ddr memory design example for the board and program the FPGA using the following guide:

 

Agilex™ 7 R-Tile Compute Express Link* (CXL*) FPGA IP Design Example User Guide version 1.12 for quartus 24.1. 

If we program it with the recommended SW4 switch setting (4.1 ON), it does not work and we loose access to the FPGA with a JTAG chain broken error (same as this https://community.intel.com/t5/Programmable-Devices/JTAG-error-after-programming-device/td-p/1594877

The same happens if we leave SW4 to the preset configuration (4.3 ON).

 

If we program the device with all SW4 pins to OFF (as it was for the R1BES version of the board) we can still detect the FPGA but programming fails with the attached configuration errors. The description states:  "External hardware access error. The first i2c command has failed, no response from voltage regulator". 



Could you please let me know what is the correct SW4 setting and how we should proceed?

Best Regards,
Alverti Chloe

Best Regards,

Chloe

 

Labels (1)
0 Kudos
1 Solution
xalverti
New Contributor I
390 Views

Hi again,

 

The problem was on the PCIe slot on the server side. 

We replaced our card and the problem was fixed.

 

For anyone facing the same problems, we had to check the x16 slot with other devices to identify the lane drop issue in BIOS.

CXL IP designs apparently are note recognized in BIOS as PCIe devices even when the design is working -- the corresponding PCIe slot always shows the message "Link did not train" -- as if nothing is attached on the slot.

 

Thank you for your help,

Chloe

View solution in original post

0 Kudos
17 Replies
WZ2
Employee
1,296 Views

Hi Chloe,


I see. The sw4 should be all off. The error u see about I2C is caused the wrong setting in PMBUS. U can check it in Quartus->Assignment->Device->Device and Pin option, the setting of SCL and SDA in Configuration->Configuration pin option; the setting of VR in Power Management.

The right setting can refer from the setting in example design. Or give me u specific opn of your board, let me check it.

Best regards,

WZ


0 Kudos
xalverti
New Contributor I
1,261 Views

Hi,

 

Thank you very much for your reply! 
So this means that the Table 32 in the document is incorrect?

If so could you please then let me know what SW3 should be set to?

 

Setting SW4 to all off did not make the problem go away.

 

With respect to the other settings, shouldn't they be already correct? I am compiling the example design for the device.

Our board opn is DK-DEV-AGI027RBES. And we are compiling the type 3 example design for this board.

 

Best,

Chloe

0 Kudos
xalverti
New Contributor I
1,247 Views

I am attaching some screenshots from our configuration in case that helps.

 

Thank you,

Chloe

0 Kudos
xalverti
New Contributor I
1,103 Views

Hi again,

 

Are there any thoughts on this? Because we are stuck and we cannot really move forward.

 

Thank you very much for your help!

Best Regards,

Chloe

0 Kudos
WZ2
Employee
1,072 Views

Hi there,

I check the setting in quartus is no error, please set all switch to their default setting(just like User guire )

https://www.intel.com/content/www/us/en/docs/programmable/683288/current/default-settings.html

SW4 should be all off, it can enable sdm_pmbus signal and communicate with Voltage regulator. 

WZ2_0-1715910580278.png

Best regards,

WZ

0 Kudos
WZ2
Employee
1,071 Views

Sorry for the miss, for RBES, sw4 should be off off on off

WZ2_2-1715911043700.png

 

WZ2_1-1715911026144.png

 

0 Kudos
xalverti
New Contributor I
1,054 Views

If you see our first post you will read that we have tried this already. 

 

If we go with off/off/on/off we get a jtag chain error. If we go fro off/off/off/off we have the errors you see above.

 

Also please keep in mind that it seems that for CXL IP designs the switches are not the same as the user guide. 

We are following this guide --> Agilex™ 7 R-Tile Compute Express Link* (CXL*) FPGA IP Design Example User Guide version 1.12 for quartus 24.1. 

 

Please give us some advice how to solve this.

 

Chloe

0 Kudos
xalverti
New Contributor I
916 Views

Hi, 

 

I have managed to solve the issue of voltage regulator error and the device is in USER MODE after programming.

However, still I cannot detect 0ddb with lspci and the design is not working.

 

What should I do?

 

Thank you,

Chloe

0 Kudos
xalverti
New Contributor I
862 Views

As some extra information -- in BIOS it seems that PCIe training fails. 

The message "Link did not train" shows up.

 

You can find a figure attached from BIOS.

 

Best Regards,

Chloe

0 Kudos
WZ2
Employee
839 Views

Hi there,

I need to confirm a few pieces of information with you:

  • What was the cause of the previous issue with the PMBUS on the board?
  • Have you tried the PCIe Gen5 example design and did it train successfully (because PCIe is more general)?


Best regards,
WZ

0 Kudos
xalverti
New Contributor I
824 Views

Hi,

 

  • I set all switches to the values you state above that however do not match the "(CXL*) FPGA IP Design Example User Guide", and I set all device configuration values of the cxl design to the ones found in the devkit examples. 
  • Yes I have compiled and run the  PCIe Gen5 example design and it trains successfully and the card is found via "lspci". 

Best,

Chloe

0 Kudos
xalverti
New Contributor I
772 Views

One difference for the example the PCie link width is x8 while for the CXL IP it is x16.

 

Please tell me how I should move forward.

 

Best Regards,

Chloe

0 Kudos
WZ2
Employee
732 Views

Hi there,

Here are some suggestions:


  1. Please confirm that your CPU and PCIe slots support CXL.
  2. For BIOS settings, you can refer to section 6.2 in the Agilex™ 7 R-Tile Compute Express Link* (CXL*) FPGA IP Design Example User Guide.
  3. If the above settings are all correct, I recommend using the debug tool kit. It can help you observe where your training is getting stuck. For using the debug tool kit, you can refer to section 5.1 in the Agilex™ 7 R-Tile Compute Express Link* (CXL*) 1.1/2.0 FPGA IP User Guide.

Best regards,

WZ



0 Kudos
WZ2
Employee
595 Views

Hi there,

Do we have some progress on this issue?

Best regards,

WZ


0 Kudos
xalverti
New Contributor I
559 Views

Hi,

 

Thank you for your reply -- We are trying to figure out if it is a server problem as you pointed out.

Our server is CXL compatible but we may face a PCIe drop lane issue.

 

As a clarification the CXL IP cannot tolerate 8 lanes on the PCIe slot, right?

 

I will keep you posted once we have figured out.

 

Best,

Chloe

0 Kudos
xalverti
New Contributor I
391 Views

Hi again,

 

The problem was on the PCIe slot on the server side. 

We replaced our card and the problem was fixed.

 

For anyone facing the same problems, we had to check the x16 slot with other devices to identify the lane drop issue in BIOS.

CXL IP designs apparently are note recognized in BIOS as PCIe devices even when the design is working -- the corresponding PCIe slot always shows the message "Link did not train" -- as if nothing is attached on the slot.

 

Thank you for your help,

Chloe

0 Kudos
WZ2
Employee
303 Views

Hi there,

So glad to hear that the issue has be solved. For your dilemma about training failed  with *16, to be honest, we never meet the same issue. If we have chance, we will test it based on your description. Thanks for your effort. 

Additionally, we would greatly appreciate it if you could take a moment to fill out our survey. Your feedback is valuable to us and helps us improve our support quality.

Thank you for your time and cooperation.

Best regards,
WZ

0 Kudos
Reply