Re: [RFC] net: phy: read link status twice when phy_check_link_status()

[Heiner Kallweit <hkallweit1@gmail.com>]

On 30.07.2019 06:03, liuyonglong wrote:
> 
> 
> On 2019/7/30 4:57, Heiner Kallweit wrote:
>> On 29.07.2019 05:59, liuyonglong wrote:
>>>
>>>
>>> On 2019/7/27 2:14, Heiner Kallweit wrote:
>>>> On 26.07.2019 11:53, Yonglong Liu wrote:
>>>>> According to the datasheet of Marvell phy and Realtek phy, the
>>>>> copper link status should read twice, or it may get a fake link
>>>>> up status, and cause up->down->up at the first time when link up.
>>>>> This happens more oftem at Realtek phy.
>>>>>
>>>> This is not correct, there is no fake link up status.
>>>> Read the comment in genphy_update_link, only link-down events
>>>> are latched. Means if the first read returns link up, then there
>>>> is no need for a second read. And in polling mode we don't do a
>>>> second read because we want to detect also short link drops.
>>>>
>>>> It would be helpful if you could describe your actual problem
>>>> and whether you use polling or interrupt mode.
>>>>
>>>
>>> [   44.498633] hns3 0000:bd:00.1 eth5: net open
>>> [   44.504273] hns3 0000:bd:00.1: reg=0x1, data=0x79ad -> called from phy_start_aneg
>>> [   44.532348] hns3 0000:bd:00.1: reg=0x1, data=0x798d -> called from phy_state_machine,update link.
>>
>> This should not happen. The PHY indicates link up w/o having aneg finished.
>>
>>>
>>> According to the datasheet:
>>> reg 1.5=0 now, means copper auto-negotiation not complete
>>> reg 1.2=1 now, means link is up
>>>
>>> We can see that, when we read the link up, the auto-negotiation
>>> is not complete yet, so the speed is invalid.
>>>
>>> I don't know why this happen, maybe this state is keep from bios?
>>> Or we may do something else in the phy initialize to fix it?
>>> And also confuse that why read twice can fix it?
>>>
>> I suppose that basically any delay would do.
>>
>>> [   44.554063] hns3 0000:bd:00.1: invalid speed (-1)
>>> [   44.560412] hns3 0000:bd:00.1 eth5: failed to adjust link.
>>> [   45.194870] hns3 0000:bd:00.1 eth5: link up
>>> [   45.574095] hns3 0000:bd:00.1: phyid=3, reg=0x1, data=0x7989
>>> [   46.150051] hns3 0000:bd:00.1 eth5: link down
>>> [   46.598074] hns3 0000:bd:00.1: phyid=3, reg=0x1, data=0x7989
>>> [   47.622075] hns3 0000:bd:00.1: phyid=3, reg=0x1, data=0x79a9
>>> [   48.646077] hns3 0000:bd:00.1: phyid=3, reg=0x1, data=0x79ad
>>> [   48.934050] hns3 0000:bd:00.1 eth5: link up
>>> [   49.702140] hns3 0000:bd:00.1: phyid=3, reg=0x1, data=0x79ad
>>>
>>>>> I add a fake status read, and can solve this problem.
>>>>>
>>>>> I also see that in genphy_update_link(), had delete the fake
>>>>> read in polling mode, so I don't know whether my solution is
>>>>> correct.
>>>>>
>>
>> Can you test whether the following fixes the issue for you?
>> Also it would be interesting which exact PHY models you tested
>> and whether you built the respective PHY drivers or whether you
>> rely on the genphy driver. Best use the second patch to get the
>> needed info. It may make sense anyway to add the call to
>> phy_attached_info() to the hns3 driver.
>>
> 
> [   40.100716] RTL8211F Gigabit Ethernet mii-0000:bd:00.3:07: attached PHY driver [RTL8211F Gigabit Ethernet] (mii_bus:phy_addr=mii-0000:bd:00.3:07, irq=POLL)
> [   40.932133] hns3 0000:bd:00.3 eth7: net open
> [   40.932458] hns3 0000:bd:00.3: invalid speed (-1)
> [   40.932541] 8021q: adding VLAN 0 to HW filter on device eth7
> [   40.937149] hns3 0000:bd:00.3 eth7: failed to adjust link.
> 
> [   40.662050] Generic PHY mii-0000:bd:00.2:05: attached PHY driver [Generic PHY] (mii_bus:phy_addr=mii-0000:bd:00.2:05, irq=POLL)
> [   41.563511] hns3 0000:bd:00.2 eth6: net open
> [   41.563853] hns3 0000:bd:00.2: invalid speed (-1)
> [   41.563943] 8021q: adding VLAN 0 to HW filter on device eth6
> [   41.568578] IPv6: ADDRCONF(NETDEV_CHANGE): eth6: link becomes ready
> [   41.568898] hns3 0000:bd:00.2 eth6: failed to adjust link.
> 
> I am using RTL8211F, but you can see that, both genphy driver and
> RTL8211F driver have the same issue.
> 
>>
>> diff --git a/drivers/net/phy/phy_device.c b/drivers/net/phy/phy_device.c
>> index 6b5cb87f3..fbecfe210 100644
>> --- a/drivers/net/phy/phy_device.c
>> +++ b/drivers/net/phy/phy_device.c
>> @@ -1807,7 +1807,8 @@ int genphy_read_status(struct phy_device *phydev)
>>  
>>  	linkmode_zero(phydev->lp_advertising);
>>  
>> -	if (phydev->autoneg == AUTONEG_ENABLE && phydev->autoneg_complete) {
>> +	if (phydev->autoneg == AUTONEG_ENABLE &&
>> +	    (phydev->autoneg_complete || phydev->link)) {
>>  		if (phydev->is_gigabit_capable) {
>>  			lpagb = phy_read(phydev, MII_STAT1000);
>>  			if (lpagb < 0)
>>
> 
> I have try this patch, have no effect. I suppose that at this time,
> the autoneg actually not complete yet.
> 
> Maybe the wrong phy state is passed from bios? Invalid speed just
> happen at the first time when ethX up, after that, repeat the
> ifconfig down/ifconfig up command can not see that again.
> 
> So I think the bios should power off the phy(writing reg 1.11 to 1)
> before it starts the OS? Or any other way to fix this in the OS?
> 
To get a better idea of whats going on it would be good to see a full
MDIO trace. Can you enable MDIO tracing via the following sysctl file
/sys/kernel/debug/tracing/events/mdio/enable
and provide the generated trace?

Due to polling mode each second entries will be generated, so you
better stop network after the issue occurred.

> 
> 
> 

Heiner