Home Forum Software discussion Kernel vanilla

This topic contains 28 replies, has 8 voices, and was last updated by  cpwrunner 1 week ago.

Viewing 14 posts - 16 through 29 (of 29 total)
  • Author
    Posts
  • #7097

    alpha_one_x86
    Participant

    The problem above is more important.
    When I try ethtool -m eth3
    Cannot get module EEPROM information: Operation not supported
    I need this too, to debug a fiber problem.

    #7098

    alpha_one_x86
    Participant

    You config change nothing. I have the single shot, and I use:
    arch/arm64/boot/dts/marvell/armada-8040-mcbin.dtb

    When I try ethtool -m eth3
    Cannot get module EEPROM information: Operation not supported
    I need this too, to debug a fiber problem.
    I try debug too https://www.fs.com/products/48929.html

    #7103

    alpha_one_x86
    Participant

    Fixed with this patch:

    diff --git a/arch/arm64/boot/dts/marvell/armada-8040-mcbin.dts b/arch/arm64/boot/dts/marvell/armada-8040-mcbin.dts
    index 56fa44860909..3da9b4cd78a4 100644
    --- a/arch/arm64/boot/dts/marvell/armada-8040-mcbin.dts
    +++ b/arch/arm64/boot/dts/marvell/armada-8040-mcbin.dts
    @@ -227,22 +227,6 @@
     	};
     };
     
    -&cp0_xmdio {
    -	status = "okay";
    -
    -	phy0: ethernet-phy@0 {
    -		compatible = "ethernet-phy-ieee802.3-c45";
    -		reg = <0>;
    -		sfp = <&sfp_eth0>;
    -	};
    -
    -	phy8: ethernet-phy@8 {
    -		compatible = "ethernet-phy-ieee802.3-c45";
    -		reg = <8>;
    -		sfp = <&sfp_eth1>;
    -	};
    -};
    -
     &cp0_ethernet {
     	status = "okay";
     };
    @@ -250,10 +234,11 @@
     &cp0_eth0 {
     	status = "okay";
     	/* Network PHY */
    -	phy = <&phy0>;
     	phy-mode = "10gbase-kr";
     	/* Generic PHY, providing serdes lanes */
     	phys = <&cp0_comphy4 0>;
    +	managed = "in-band-status";
    +	sfp = <&sfp_eth0>;
     };
     
     &cp0_sata0 {
    @@ -288,10 +273,11 @@
     &cp1_eth0 {
     	status = "okay";
     	/* Network PHY */
    -	phy = <&phy8>;
     	phy-mode = "10gbase-kr";
     	/* Generic PHY, providing serdes lanes */
     	phys = <&cp1_comphy4 0>;
    +	managed = "in-band-status";
    +	sfp = <&sfp_eth1>;
     };
     
     &cp1_eth1 {
    #7104

    alpha_one_x86
    Participant

    When I try ethtool -m eth3
    Cannot get module EEPROM information: Operation not supported
    I need this too, to debug a fiber problem.
    I try debug too https://www.fs.com/products/48929.html
    SPF and SFP+ speed are dynamic?

    #7123

    cpwrunner
    Participant

    alpha_one_x86,

    I have applied your patch to a vanilla kernel 4.18 version and does a fantastic job of bringing up the 2 10G interfaces on my single-shot. I can not thank you engough for working so diligently to get this to work with the single shot. I still have an issue though. I can connect both interfaces to another sfp interface on a netgear switch and I believe I am getting a 10G connection. However, when I connect to the mcbin to an intel 520 adapter the intel gets a carrier signal and seems ready to connect but the marvell sfp adapters on the mcbin (eth0,eth1) do not see any link. This connection worked fine on using the marvell kernel. Why does it not work with the mainline and your patch?

    Thanks

    #7124

    alpha_one_x86
    Participant

    Hi,
    This patch is not mine.
    I have some problem with vanilla too, 1G module on SFP not work, SFP seam be 2.5G not 1G, it’s why I ask before.

    #7307

    amos083
    Participant

    I have the same situation here with a single-shot board, fiber connections.

    With the 4.4.52 kernel form Marvell’s site, I got it to work, but eth0 & eth1 only work at 1G, and even then only after setting ethtool -s eth0 speed 1000.

    With a vanilla 4.9.19 kernel (essentially the same config suggested here) and the patch above, also tried a patch from https://patchwork.kernel.org/patch/10668803/ (which for the single-shot case is essentially the same as the patch above). The kernel recognizes all devices, but it still cannot sense link up/down on the 10G lines; ethtool shows the correct parameters, but claims it cannot set speed.

    Is there a solution yet?

    Thanks

    #7309

    alpha_one_x86
    Participant

    ethtool -s eth0 speed 1000 limit the speed at 1000Mbps, then 1Gbps
    ethtool -s eth0 speed 100 limit the speed at 100Mbps

    I’m trying see with ths tech support to do move solidrun to fix vanilla kernel and have correct SFP support, same here SFP not work with vanilla

    #7310

    amos083
    Participant

    The trouble is, ethtool shows only MII as supported port on the 10G ports (but shows FIBRE on eth2 & eth3); when trying to set it says “Cannot advertise speed” on any given speed and port.

    #7315

    nova_robotics
    Participant

    I’m having the exact same problem. The patch posted gets me further than I have been ever been able to get with the SFP+ ports on this board, and I am extremely grateful. But I think there’s a piece of the puzzle still missing.

    I have tried numerous 10G SFPs and they all kinda do the same thing. Bringing the connection up turns the laser on, and ethtool -m will show output power and received power as being good, and what you would expect to see. But Linux still seems to think there is no carrier, even though the laser is on and there is output and receive power according to ethtool.

    Is there any way to force carrier detection? I’m not above soldering to trick hardware signals.

    #7316

    alpha_one_x86
    Participant

    From tkos.co.il (tech support for solidrun), I have an proforma of 5940€ to fix this.
    1) it should be to solidrun to paid this?
    2) I can’t paid this alone
    This hardware is useless without right and good SFP+ support.
    Cheers,

    #7334

    amos083
    Participant

    Digging into drivers code, I found a suspicious code which may be a bug in the mvpp2 driver, which prevents phylink from setting interface 10GBase-KR if 1000BaseX or 2500BaseX are advertised. I’ll try to consult some experts.

    #7335

    amos083
    Participant

    Apparently there is a bug in the mvpp2 driver; got some patches from Solidrun support, seem to work — we can get a 10G connection, though ethtool still doesn’t list “advertised” and “supported” bits correctly.

    These may be integrated into the next 4.20 release.

    #7344

    cpwrunner
    Participant

    Are there any updates on the status of getting the single shot sfp ports working on a vanilla kernel. It’s great that solidrun has provided patches. Have any of them resolved the link advertisement issues. Will they be put into the vanilla kernel?

    Thanks

Viewing 14 posts - 16 through 29 (of 29 total)

You must be logged in to reply to this topic.

Technical specification tables can not be displayed on mobile. Please view on desktop