Re: [RFC PATCH net] net: phy: Fix "link partner" information disappear issue

[Heiner Kallweit]

On 11.10.2019 07:55, Yonglong Liu wrote:
> 
> 
> On 2019/10/11 3:17, Heiner Kallweit wrote:
>> On 10.10.2019 11:30, Yonglong Liu wrote:
>>> Some drivers just call phy_ethtool_ksettings_set() to set the
>>> links, for those phy drivers that use genphy_read_status(), if
>>> autoneg is on, and the link is up, than execute "ethtool -s
>>> ethx autoneg on" will cause "link partner" information disappear.
>>>
>>> The call trace is phy_ethtool_ksettings_set()->phy_start_aneg()
>>> ->linkmode_zero(phydev->lp_advertising)->genphy_read_status(),
>>> the link didn't change, so genphy_read_status() just return, and
>>> phydev->lp_advertising is zero now.
>>>
>> I think that clearing link partner advertising info in
>> phy_start_aneg() is questionable. If advertising doesn't change
>> then phy_config_aneg() basically is a no-op. Instead we may have
>> to clear the link partner advertising info in genphy_read_lpa()
>> if aneg is disabled or aneg isn't completed (basically the same
>> as in genphy_c45_read_lpa()). Something like:
>>
>> if (!phydev->autoneg_complete) { /* also covers case that aneg is disabled */
>> 	linkmode_zero(phydev->lp_advertising);
>> } else if (phydev->autoneg == AUTONEG_ENABLE) {
>> 	...
>> }
>>
> 
> If clear the link partner advertising info in genphy_read_lpa() and
> genphy_c45_read_lpa(), for the drivers that use genphy_read_status()
> is ok, but for those drivers that use there own read_status() may
> have problem, like aqr_read_status(), it will update lp_advertising
> first, and than call genphy_c45_read_status(), so will cause
> lp_advertising lost.
> 
Right, in genphy_read_lpa() we shouldn't clear all lpa bits but only
those ones the generic functions care about. Basically the same as
in the c45 version. Then a vendor-specific part isn't affected.

aqr_read_status() is a good example. It deals with 1Gbps mode that
isn't covered by the generic c45 functions. Therefore the 1Gbps-related
bits won't be overwritten by the generic functions.

> Another question, please see genphy_c45_read_status(), if clear the
> link partner advertising info in genphy_c45_read_lpa(), if autoneg is
> off, phydev->lp_advertising will not clear.
> 

If autoneg is off, lp_advertising should never be set, so there's
nothing to clear. However we may have to look at the case that user
switches to fixed speed mode via ethtool.

>>> This patch call genphy_read_lpa() before the link state judgement
>>> to fix this problem.
>>>
>>> Fixes: 88d6272acaaa ("net: phy: avoid unneeded MDIO reads in genphy_read_status")
>>> Signed-off-by: Yonglong Liu <liuyonglong@xxxxxxxxxx>
>>> ---
>>>  drivers/net/phy/phy_device.c | 8 ++++----
>>>  1 file changed, 4 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/drivers/net/phy/phy_device.c b/drivers/net/phy/phy_device.c
>>> index 9d2bbb1..ef3073c 100644
>>> --- a/drivers/net/phy/phy_device.c
>>> +++ b/drivers/net/phy/phy_device.c
>>> @@ -1839,6 +1839,10 @@ int genphy_read_status(struct phy_device *phydev)
>>>  	if (err)
>>>  		return err;
>>>  
>>> +	err = genphy_read_lpa(phydev);
>>> +	if (err < 0)
>>> +		return err;
>>> +
>>>  	/* why bother the PHY if nothing can have changed */
>>>  	if (phydev->autoneg == AUTONEG_ENABLE && old_link && phydev->link)
>>>  		return 0;
>>> @@ -1848,10 +1852,6 @@ int genphy_read_status(struct phy_device *phydev)
>>>  	phydev->pause = 0;
>>>  	phydev->asym_pause = 0;
>>>  
>>> -	err = genphy_read_lpa(phydev);
>>> -	if (err < 0)
>>> -		return err;
>>> -
>>>  	if (phydev->autoneg == AUTONEG_ENABLE && phydev->autoneg_complete) {
>>>  		phy_resolve_aneg_linkmode(phydev);
>>>  	} else if (phydev->autoneg == AUTONEG_DISABLE) {
>>>
>>
>>
>> .
>>
> 
>