dual LAN on SDK7

Hi Eric,

I have asked the factory team for support on your issue. I'll let you know when I have an answer.

Best regards,
Miroslav

Miroslav,

Thanks. Here is another findings.

I added this in device tree.

&phy_sel {
rmii-clock-ext;
};

The port1 does not get package losses. But if I ifup eth1, the packages seems to get flood and my Ethernet switch hangs. 

I have problems with "flooding" too. See the comments to this patch. Miroslav could you assist in this issue or at least confirm such CPSW behavior?

Thanks.

Miroslav,

Did your team have any feedback? 

-Eric

Eric,

what PHY are you using?

I am using SMSC LAN8720 RMII PHY and it works quite fine in SDK6 (v3.2 kernel). I just suspect that the gmii_sel register does not set well in SDK7. 

-Eric

Have you connected both eth0 and eth1 to the same switch?

I knew that eth0 and eth1 have to be in different subnet. However, when testing, I did connected both port in same switch. 

-Eric

So, both 3.12 (SDK 7..0) and the upcoming vanilla 3.15 show the same erroneous behavior, when both eth0 and eth1 (configured for different subnets) are connected to the same switch. Then we are two now having the same problem.

I am using 3.15y from TI's git server indeed.

-Eric

Miroslav,

Do you have any update?

-Eric

Miroslav,

When switch back to v3.12y, the flooding problems are gone. Also the USB works. I will stick at v3.12.10 that from TI SDK at this time.

-Eric

As long as you don't have switches or hosts, that have problem with not removed VLAN IDs you should be fine with 3.12.

What USB problems did you have with 3.15?

"As long as you don't have switches or hosts, that have problem with not removed VLAN IDs you should be fine with 3.12." <-- Could you explain me more?

The USB does not work on 3.15y? At least, my usb mouse is not emulated. I tried BBB's 3.14 kernel and TI's 3.12 with same device tree. Both work except TI's v3.15y.

-Eric

See this thread for more details about dual_emac issue.

Haven't tried the mouse yet.

rfc2217,

Thanks for pointing me the thread. I was just wondering how come i cannot get my ip address by dhcp. The patch in the thread just save me.

-Eric

rfc2217,

Hmm... I think the flooding problem is still there even you use v3.12. My switch hangs again.

-Eric

Miroslav,,

Could you at least confirm this problem?

-Eric

Miroslav,

For now, I have my every driver working in v3.12 except for dual RMII supports. Could you ask someone to support this?

The eth0 work well. If I enable eth1 (ifconfig eth1 up), I got ethernel package "flooding" and I need to reset my switch.

My dts file:

&mac {
pinctrl-names = "default", "sleep";
pinctrl-0 = <&cpsw_default>;
pinctrl-1 = <&cpsw_sleep>;
dual_emac;
};

&davinci_mdio {
pinctrl-names = "default", "sleep";
pinctrl-0 = <&davinci_mdio_default>;
pinctrl-1 = <&davinci_mdio_sleep>;
};

&cpsw_emac0 {
phy_id = <&davinci_mdio>, <0>;
phy-mode = "rmii";
dual_emac_res_vlan = <1>;
};

&cpsw_emac1 {
phy_id = <&davinci_mdio>, <1>;
phy-mode = "rmii";
dual_emac_res_vlan = <2>;
};

&phy_sel {
reg= <0x44e10650 0xf5>;
rmii-clock-ext;
};

 Thanks.

-Eric

Hi Eric,

please test this patch.

Hi rfc2217,

This patch works fine for me. I tested under ti-linux-3.12y and ti-linux-3.15y. The previous "flooding" problem is gone. 

Thanks. How did you find this patch? Is there a place that I can see all the new patches from TI.

-Eric

I'm subscribed to the netdev[at]vger.kernel.org mailing list and I have also contacted the patch author before, so he CC'd me in the e-mail.

linux-omap mailig list is a good place to follow current development.