DP83869HM: loosing a link in 1000Base-T and 1000Base-X modes

Hi Slava,

Please help me out by clarifying the following :

1. Without writing to register 0x01EC, does link stay stable?

2. When link is lost after writting register 0x01EC, does link recovery happens by only reseting 869 or only resetting KSZ? 

3. Does this SFP module has inbuilt dc-block caps on the sop, son, sip and sin lines?

4. Did you also try disabling Sgmii auto-neg with register configuraiton of register<0x14>?

--

Regards,

Vikram

I have connected +3.3 V voltage to pin 22 of dp83869 expecting it to switch to 100 media converter mode. However, its behavior has not changed at all. The register values in the responses will relate to the 100base-tx and 100base-fx modes and in 100 mode, I’m able to record waveforms. (After reset, the value of register 0x0000 is 0x3100, 0xc00 is 0x2100, and 0x01df is 0x0045).

1.   The link between ksz9477 and dp83869 remains stable in 100 and 1000 media converter modes

2.   Unfortunately, the hardware reset pins of the both chips are connected to a single controller pin, so I’m not able to reset them but simultaneously. To avoid cutting the board I tried the software reset of DP83869:

a. I tried writing the value of 0xD100 to the register 0x0000. After that in about 2 seconds, I got 100base-tx link but exactly 10 seconds after, it was down again.

b. Then I wrote the value of 0x4000 to register 0x001f. The link was restored in about 2 seconds, but exactly 10 seconds after the software reset, it was down (the same behavior as described above).

c. Then I wrote the value of 0x8000 to the register 0x001f. I got a stable link.

d. I repeated the actions a, b, c described above several times (after the link was stable, I wrote the value of 0x1FFC to the register 0x1EC, and the link went down in 10 seconds).

e. Writing the value of 0xA100 to the 0x0c00 register did not restore the link.

3.   The recommended circuit from the SFP module datasheet:In addition, I tried external capacitors on the SIP/SIN line of dp83869. Also I tried a 100 Ohm resistor near dp83869 between SIP and SIN.

The waveform above is taken from the outputs 12 and 13 of the SFP module (or the inputs 16 and 17 of the dp83869) relative to the ground. The waveform is the same for both inputs 16 and 17. An external 100 Ohm resistor was used between SIP and SIN near dp83869, although it was not necessary. The mode is 100base-fx.

4.    I tried to write different values of 0x2947/2907/2807/2827 to the register 0x14. The register 0x4F has a stable value of 0x0200 (no sync), and 0x0c01 has a stable value of 0x6149.

Hello,

I am checking with team here and hopefully get back to you in two days. I am summarizing my understanding here. Let me know if I got something wrong :

1. Copper side issue : 

a. Link is stable before writting reg<0x01EC> = 0x1FFC. Looses link after 10s of writting this value for both 100BTx and 1000BTx link case.

b. Only writting reg<0x001F>=8000 recovers the link and keep it stable. (this is the hard reset and actually resets the value of reg<0x01EC> to default and hence is linked to observation a)

2. Fiber side issue :

a. No link ever even with different configurations of reg<14>

b. 100 ohms termination should not be placed as 869 has inbuilt termination .

c. SFP already has ac coupling caps -> so extra ac coupling caps are not required.

d. 62.5MHz periodic signal is observed on SFP's outputs. (This does not look like a correct signal type but I am checking again with team).

--

Regards,

Vikram

Hello,

Can you read and share the values of register 0x0002 and 0x0003 of 869 part that you are working on?

--

Regards,

Vikram

Hello,

the value of register 0x0002 is 0x2000,

the value of register 0x0003 is 0xA0F1

Hello,

Thanks for sharing the register values. Here is the summary of my discussion with team related to this issue :

1. Copper side issue :

a. Dont update register 0x01EC

b. Team is looking into possible error in description in datasheet.

2. Fiber side issue :

a. We have not seen this "never linking up" issue on fiber side and we need to look at the way to make it work with selected SFP.

b. Observed periodic signal of 62.5MHz is not expected either in 100BFx or 1000BFx. Both of them should be transmitting random data pattern of 125MHz and 1250MHz rate. May be the observation is limited by oscilloscope capture bandwidth and you will need higher Gbps scope.

b. Can we try toggling tx_disable pin of the SFP module + normal configuration to see if the fiber links up? 

c. Can we try toggling tx_disable pin of the SFP module + fiber forced configuration to see if the fiber links up? 

--

Regards,

Vikram

We recently discovered that putting a stronger pull down ~ 600ohms on tx_disable has made a few SFPs in market work. I am not sure if you also have a pull-down. Can we check?

--

Regards,

Vikram