DP83867IR: Do not start at 1Gbit

Hi Aniruddha,

I checked with longer cable, kits had connected instantly, same goes to my custom boards. But problem is that I need to connect boards through board-to-board connector, so how can I make them work on short distance? 

I am confused about kits working better with longer cable. Why is that happening? Can longer cable reduce reflection better than short one? How make kits work with short cable?

The shorter cable might be saturating the receiver during the auto-negotiation process.

Can you try setting bit[9] in register 0x1E to see if a longer auto-MDIX helps resolve the issue?

Hi Ross,

Thank you for your response, unfortunately setting bit 9 in register 0x1E didn't resolve the issue. What can I do to confirm receiver saturation?

Register 0x15 have value 0x0, does it mean there is no problem with the receiver?

update: it looks like with bit 9 in register 0x1E 1 Gbit connection starts to establish in 5 minutes, without this bit connection was never established. But still 5 minutes is a long time for link-up.

Hi Sergey,

Could you add 10-15pF caps from signal to GND on each of the MDI lines and let us know if this helps improve the link-up time with short cable?

Hi Ross,

I added 15pF capacitors. Link-up time didn't improve after adding the capacitors, it even got worse. It still works with a long cable, but even with bit 9 in register 0x1E 1 Gbit connection don't establish with a short cable. There is my board and schematic with added capacitors.

Hello Ross,

We are colleagues with Sergey. I am the project manager responsible for a product which relates to this problem. We can't solve this problem since March. I understand it was difficult to find the right decision for this problem when we discussed boards developed by us, but now we are using EVMs developed by TI and Microchip and experiment can be reproduced. Is it possible for you to order both of this EMVs and reproduce experiment? Both of the EVMs available for order. Fortunately, we had an opportunity to debug the software part of the project using a 100Mbit link, but the deadline is coming and we will have to have 1 Gbit link. Please help the problem as soon as possible. It's very important for us currently and maybe the decision will be helpful for someone else in the future.

Regards,

Anton

Hi Anton,

We will try and reproduce this in our lab and get back to you.

Thank you for a quick response. We'll wait for your comments.

Please be aware that this might take me a week to get back to you because we have to order the KSZ board.

Once we have it in our lab we will immediately start the debug.

Hi Ross,

Was the KSZ board delivered to you?

Hi Anton,

I just received the KSZ board.

We will send you an update tomorrow.

Hi Anton,

I see the same issue in lab that you see.

With short cable the link takes about 1min to establish and with longer cable it takes >5min when connected to the EVB-KSZ9897.

I see that link is established faster with the KSZ9031 (on the EVB-KSZ9897). For both devices, auto-negotiation is completing but then link is dropped as soon as the devices switch to sending 1G. If I advertise only 10/100 FD/HD, the link is immediately established on both KSZ9897 and KSZ9031.

I did a google search and found similar threads on Microchip website and Xilinx website about the KSZ9031 having difficulty negotiating to 1Gbps.

In those forums, people said that the issue was with the clock source on the KSZ devices, causing too much jitter on the MDI.

I'll check to see if we can read some additional registers to see where in the process the link is breaking.

I also connected the KSZ9031 (on the EVB-KSZ9897) to the KSZ9897 using the different cable combos and those devices have issues connecting to each other also. Do you see the same on your EVB?

Hi Ross,

Concerning information from your last post:

>With short cable the link takes about 1min to establish and with longer cable it takes >5min when connected to the EVB-KSZ9897.

Can you verify this information because we have opposite results? With short cable (0.1 meters) link between KSZ9897 (on the EVB-KSZ9897) and DP8367IR (on the DP8367ERGZ-R-EVM) is almost never established (it can be established unpredictable after 5 or 10 minutes, but we can’t find any dependencies), and with a long cable (~30 meters) the link is established instantly (less than 1 second).

>I see that link is established faster with the KSZ9031 (on the EVB-KSZ9897).

1 Gbit link between KSZ9031 (on the EVB-KSZ9897) and DP8367IR (on the DP8367ERGZ-R-EVM) is established each time, but it takes about 10 seconds using a short cable and less than 1 second using a long cable. Of course, it can be admitted as a better result, but 10 seconds with a short cable shouldn't be ignored.

>I also connected the KSZ9031 (on the EVB-KSZ9897) to the KSZ9897 using the different cable combos and those devices have issues connecting to each other also. Do you see the same on your EVB?

We don't see any difficulties in the connection between KSZ9031 and KSZ9897 (both on the EVB-KSZ9897). Please, notice that information LEDs work in different modes on KSZ9031 and KSZ9897. LEDs on KSZ9031 work in single-LED mode and LEDs on KSZ9897 works in tri-color dual-LED mode, so they blink differently at 1 Gbit. Please, verify your case using this information.

>If I advertise only 10/100 FD/HD, the link is immediately established on both KSZ9897 and KSZ9031.

Yes, 100 Mbit mode works perfectly in any combination of hardware.

Let me emphasize very important information which we mentioned in previous topics. Please, pay special attention to it.

The main reason why we think that we have a problem with DP8367IR and not with KSZ9897 is that KSZ9897 (it doesn’t matter if it EVB-KSZ9897 or one of our custom board based on this ethernet switch) works fine with DP83867IRPAP (64-pin package) on ZCU102 board (Xilinx EVM which we also use).

There is only one difference that ZCU102 board contains DP83867IRPAP (64-pin package), but DP8367ERGZ-R-EVM and our boards contain DP83867ERGZ (48-pin) andDP83867IRRGZ (also 48-pin) relatively.

We are ready to provide additional information if It’s required ASAP.

Let’s sum up information about hardware which we have for experiments:

• DP8367ERGZ-R-EVM (TI). Contains DP83867ERGZ (48-pin)
• EK-U1-ZCU102-G (Xilinx). Contains DP83867IRPAP (64-pin)
• LS-M07 board (LabSystems). Contains DP83867IRRGZ (48-pin)
• EVB-KSZ9897 (Microchip). Contains KSZ9897RTXC and KSZ9031RNXCC
• LS-M08 board (LabSystems). Contains KSZ9896CTXI
• LS-M10 board (LabSystems). Contains two KSZ9896CTXI

Hi Anton,

1)

<Ross> With short cable the link takes about 1min to establish and with longer cable it takes >5min when connected to the EVB-KSZ9897.

<Anton> Can you verify this information because we have opposite results? With short cable (0.1 meters) link between KSZ9897 (on the EVB-KSZ9897) and DP8367IR (on the DP8367ERGZ-R-EVM) is almost never established (it can be established unpredictable after 5 or 10 minutes, but we can’t find any dependencies), and with a long cable (~30 meters) the link is established instantly (less than 1 second).

<Ross> Yes, I confirm that my original statement holds.

2)

<Ross> I see that link is established faster with the KSZ9031 (on the EVB-KSZ9897).

<Anton> 1 Gbit link between KSZ9031 (on the EVB-KSZ9897) and DP8367IR (on the DP8367ERGZ-R-EVM) is established each time, but it takes about 10 seconds using a short cable and less than 1 second using a long cable. Of course, it can be admitted as a better result, but 10 seconds with a short cable shouldn't be ignored.

3)

<Ross> I also connected the KSZ9031 (on the EVB-KSZ9897) to the KSZ9897 using the different cable combos and those devices have issues connecting to each other also. Do you see the same on your EVB?

<Anton> We don't see any difficulties in the connection between KSZ9031 and KSZ9897 (both on the EVB-KSZ9897). Please, notice that information LEDs work in different modes on KSZ9031 and KSZ9897. LEDs on KSZ9031 work in single-LED mode and LEDs on KSZ9897 works in tri-color dual-LED mode, so they blink differently at 1 Gbit. Please, verify your case using this information.

<Ross> yes, I have observed the link issues with the KSZ devices on the board. I have also seen issues with two KSZ9897 ports, which take a long time to link.

4)

<Ross> If I advertise only 10/100 FD/HD, the link is immediately established on both KSZ9897 and KSZ9031.

<Anton> Yes, 100 Mbit mode works perfectly in any combination of hardware.

5)

<Anton> Let me emphasize very important information which we mentioned in previous topics. Please, pay special attention to it. The main reason why we think that we have a problem with DP8367IR and not with KSZ9897 is that KSZ9897 (it doesn’t matter if it EVB-KSZ9897 or one of our custom board based on this ethernet switch) works fine with DP83867IRPAP (64-pin package) on ZCU102 board (Xilinx EVM which we also use). There is only one difference that ZCU102 board contains DP83867IRPAP (64-pin package), but DP8367ERGZ-R-EVM and our boards contain DP83867ERGZ (48-pin) and DP83867IRRGZ (also 48-pin) relatively.

We are ready to provide additional information if It’s required ASAP.

Let’s sum up information about hardware which we have for experiments:

• DP8367ERGZ-R-EVM (TI). Contains DP83867ERGZ (48-pin)
• EK-U1-ZCU102-G (Xilinx). Contains DP83867IRPAP (64-pin)
• LS-M07 board (LabSystems). Contains DP83867IRRGZ (48-pin)
• EVB-KSZ9897 (Microchip). Contains KSZ9897RTXC and KSZ9031RNXCC
• LS-M08 board (LabSystems). Contains KSZ9896CTXI
• LS-M10 board (LabSystems). Contains two KSZ9896CTXI

<Ross> I will see if there are additional registers you can read/configure to see if it increases link stability with the KSZ devices. I do ask though that you also look into the clock related issues for the KSZ devices in parallel.

Hi Ross,

Are you going to continue finding a decision for the problem? We've done hundreds of experiments and at this point, it looks like some kind of incompatibility only between DP8367 and KSZ989x. Other combinations work stable. If we had time, we would change DP8367 PHY to KSZ9031 (KSZ9031 works well with KSZ989x), but now it's impossible. We'll look forward to any information.

Regards,

Anton

Hi Ross,

We discussed with Aniruddha Khadye that all combinations of devices (which we have) work well using a long cable (30 m). Looks like, there is a saturation during a link establishing process using a short cable. Do you have any idea? 

Hi Anton,

Can you please try the following configurations:

Configuration #1

Configuration #2

Hi Ross,

Thank you for your response. We have tried both configuration. After writing first configuration it seems the board stops any attempts to communicate - all LEDs are off.

After writing second one the board behave exactly like before: instant link-up on the long cable and no link-up on the short one.

We also tried to implement capacitive coupling according to section 2.6 of DP83867ERGZ EVM User's Guide, it seems DP83867ERGZ board works much better in that case: link-up time is about 0 to 10 seconds with short cable.

Hi Sergey,

Thank you for the update. We are still looking to see if there are any other register settings to try and improve the link-up time with the KSZ.