Re: [PATCH RFC net-next] net: pcs: pcs-mtk-lynxi fix mtk_pcs_lynxi_get_state() for 2500base-x

[Eric Woudstra]

With some extra info:

echo "file drivers/net/phy/* +p" > /sys/kernel/debug/dynamic_debug/control

The log looks like this when sfp inserted:

With the path (on original net-next, no further modifications), Traffic OK:

[   71.212634] sfp sfp-1: mod-def0 0 -> 1
[   71.216403] sfp sfp-1: tx-fault 1 -> 0
[   71.220140] sfp sfp-1: SM: enter empty:up:down event insert
[   71.225716] sfp sfp-1: SM: exit probe:up:down
[   71.230059] sfp sfp-1: SM: enter probe:up:down event tx_clear
[   71.235803] sfp sfp-1: SM: exit probe:up:down
[   71.240195] sfp sfp-1: tx-fault 0 -> 1
[   71.243939] sfp sfp-1: SM: enter probe:up:down event tx_fault
[   71.249688] sfp sfp-1: SM: exit probe:up:down
[   71.254052] sfp sfp-1: tx-fault 1 -> 0
[   71.257810] sfp sfp-1: SM: enter probe:up:down event tx_clear
[   71.263542] sfp sfp-1: SM: exit probe:up:down
[   71.534808] sfp sfp-1: SM: enter probe:up:down event timeout
[   71.570662] sfp sfp-1: module OEM              SFP-2.5G-T       rev 1.0  sn SK2301110007     dc 230110  
[   71.580153] mtk_soc_eth 15100000.ethernet eth1: optical SFP: interfaces=[mac=2-4,22-23, sfp=23]
[   71.588848] mtk_soc_eth 15100000.ethernet eth1:  interface 23 (2500base-x) rate match none supports 10,13-14,47
[   71.598941] mtk_soc_eth 15100000.ethernet eth1: optical SFP: chosen 2500base-x interface
[   71.607605] mtk_soc_eth 15100000.ethernet eth1: requesting link mode inband/2500base-x with support 00,00000000,00008000,00006400
[   71.619321] sfp sfp-1: tx disable 1 -> 0
[   71.623287] sfp sfp-1: SM: exit present:up:wait
[   71.636489] hwmon hwmon0: temp1_input not attached to any thermal zone
[   71.684749] sfp sfp-1: SM: enter present:up:wait event timeout
[   71.690587] sfp sfp-1: SM: exit present:up:wait_los
[   74.704872] sfp sfp-1: los 1 -> 0
[   74.708199] sfp sfp-1: SM: enter present:up:wait_los event los_low
[   74.714389] sfp sfp-1: SM: exit present:up:link_up
[   74.714422] mtk_soc_eth 15100000.ethernet eth1: Link is Up - 2.5Gbps/Full - flow control off

Without the patch, No traffic possible:

[  261.515414] sfp sfp-1: los 1 -> 0
[  261.518740] sfp sfp-1: SM: enter empty:up:down event los_low
[  261.524418] sfp sfp-1: SM: exit empty:up:down
[  261.528799] sfp sfp-1: mod-def0 0 -> 1
[  261.532541] sfp sfp-1: los 0 -> 1
[  261.535843] sfp sfp-1: SM: enter empty:up:down event insert
[  261.541406] sfp sfp-1: SM: exit probe:up:down
[  261.545748] sfp sfp-1: SM: enter probe:up:down event los_high
[  261.551481] sfp sfp-1: SM: exit probe:up:down
[  261.555859] sfp sfp-1: tx-fault 1 -> 0
[  261.559595] sfp sfp-1: SM: enter probe:up:down event tx_clear
[  261.565330] sfp sfp-1: SM: exit probe:up:down
[  261.859940] sfp sfp-1: SM: enter probe:up:down event timeout
[  261.895690] sfp sfp-1: module OEM              SFP-2.5G-T       rev 1.0  sn SK2301110007     dc 230110  
[  261.905218] mtk_soc_eth 15100000.ethernet eth1: optical SFP: interfaces=[mac=2-4,22-23, sfp=23]
[  261.913929] mtk_soc_eth 15100000.ethernet eth1:  interface 23 (2500base-x) rate match none supports 10,13-14,47
[  261.924104] mtk_soc_eth 15100000.ethernet eth1: optical SFP: chosen 2500base-x interface
[  261.932199] mtk_soc_eth 15100000.ethernet eth1: requesting link mode inband/2500base-x with support 00,00000000,00008000,00006400
[  261.945449] sfp sfp-1: tx disable 1 -> 0
[  261.950886] sfp sfp-1: SM: exit present:up:wait
[  261.973346] hwmon hwmon0: temp1_input not attached to any thermal zone
[  262.009896] sfp sfp-1: SM: enter present:up:wait event timeout
[  262.015771] sfp sfp-1: SM: exit present:up:wait_los
[  264.842218] sfp sfp-1: los 1 -> 0
[  264.845544] sfp sfp-1: SM: enter present:up:wait_los event los_low
[  264.851770] sfp sfp-1: SM: exit present:up:link_up
[  264.851801] mtk_soc_eth 15100000.ethernet eth1: Link is Up - Unknown/Unknown - flow control off


So if phylink_mii_c22_pcs_decode_state() should not set the speed, then it is not correctly set somewhere else.

On 1/2/24 13:55, Eric Woudstra wrote:
>> Please describe your setup more fully. What is the link partner on this
>> 2500base-X link?
> 
> I use a BananaPi R3, with the oem-sfp2.5g-t module. It has the SFP quirk that disables autoneg. I was trying Marek's rtl8221b patchset, but found that even with unmodified code,  original net-next unmodified, I could get link up, but no traffic is going through.
> 
> On the other side is a.rock5b with rtl8125b.
> 
> Only after applying this patch, it works and eth1 reports link up with 2.5Gbps instead of unknown speed.
> 
> If you need more debugging info, I can supply it at a later time.
> 
> 
> On January 2, 2024 1:10:01 PM GMT+01:00, "Russell King (Oracle)" <linux@xxxxxxxxxxxxxxx> wrote:
>> On Tue, Jan 02, 2024 at 08:44:08AM +0100, Eric Woudstra wrote:
>>> From: Daniel Golle <daniel@xxxxxxxxxxxxxx>
>>>
>>> Need to fix mtk_pcs_lynxi_get_state() in order for the pcs to function
>>> correctly when the interface is set to 2500base-x, even when
>>> PHYLINK_PCS_NEG_INBAND_DISABLED is set.
>>
>> Please describe your setup more fully. What is the link partner on this
>> 2500base-X link?
>>
>> In PHYLINK_PCS_NEG_INBAND_DISABLED mode, this means that phylink is
>> operating in inband mode, but Autoneg is clear in the advertisement
>> mask, meaning Autoneg is disabled and we are using a "fixed" setting.
>> state->speed and state->duplex should already be initialised.
>>
>>> When the pcs is set to 2500base-x, the register values are not compatible
>>> with phylink_mii_c22_pcs_decode_state(). It results in parameters such as
>>> speed unknown and such. Then the mac/pcs are setup incorrectly and do not
>>> function.
>>
>> Since Autoneg is clear, phylink_mii_c22_pcs_decode_state() won't
>> change state->speed and state->duplex, which should already be
>> correctly set.
>>