BUG report: sun4i-emac and mdio-sun4i

[qianfan]

在 2023/5/11 1:27, Florian Fainelli 写道:
> On 5/10/23 08:22, Maxime Ripard wrote:
>> Hi,
>>
>> On Wed, May 10, 2023 at 05:40:25PM +0800, qianfan wrote:
>>> Hi:
>>>
>>> Now the linux driver for allwinner sun4i emac has split two seprate 
>>> drivers,
>>> emac and mdio. But there has minor problem.
>>>
>>> Most of resources such as clk are prepared when emac driver loaded, 
>>> and the mdio
>>> driver can't work if the clk is not enabled.
>>>
>>> So if we reigster mdio driver before emac driver, the top level will 
>>> read phy's
>>> id right now when mdio driver registed. But the emac resoure have 
>>> not prepared
>>> that will cause sun4i_mdio_read always return zero.
>>>
>>> Listen mdio_access trace event, and then 'insmod mdio-sun4i.ko' you 
>>> can find
>>> the read result: (all of the register are read as zero)
>>>
>>> insmod-3744 ..: mdio_access: 1c0b080.mdio-mii read  phy:0x00 
>>> reg:0x02 val:0x0000
>>> insmod-3744 ..: mdio_access: 1c0b080.mdio-mii read  phy:0x00 
>>> reg:0x03 val:0x0000
>>>
>>> That will cause phy_id always zero, the phy's special driver can not 
>>> loaded:
>>> $ cat /sys/bus/mdio_bus/devices/1c0b080.mdio-mii\:00/phy_id
>>> 0x00000000
>>>
>>> I had tried capture the mdc/mdio bus signal, noting is sent.
>>>
>>> If both mdio-sun4i.c and sun4i-emac.c drivers are built with Y, the 
>>> kernel
>>> will always load mdio-sun4i.c first.
>>>
>>> So can we find a easy was to make sure mdio driver are loaded later, 
>>> at least
>>> after emac driver? Or merge those two drivers?
>>
>> Why not enable the clock from the MDIO driver as well?
>
> Another option is to use a a module soft dependency, e.g.:
>
> MODULE_SOFTDEP("pre: mdio-bcm-unimac");
But MODULE_SOFTDEP can not work if all of the drivers are build with Y 
option.