Re: [PATCH net-next 00/15] Realtek RTL822x PHY rework to c45 and SerDes interface switching

[Heiner Kallweit <hkallweit1@gmail.com>]

On 25.12.2023 11:28, Marek Behún wrote:
> On Sat, 23 Dec 2023 20:09:33 +0100
> Heiner Kallweit <hkallweit1@gmail.com> wrote:
> 
>> On 20.12.2023 17:25, Marek Behún wrote:
>>> On Wed, 20 Dec 2023 17:20:07 +0100
>>> Heiner Kallweit <hkallweit1@gmail.com> wrote:
>>>   
>>>> On 20.12.2023 16:55, Marek Behún wrote:  
>>>>> Hi,
>>>>>
>>>>> this series reworks the realtek PHY driver's support for rtl822x
>>>>> 2.5G transceivers:
>>>>>
>>>>> - First I change the driver so that the high level driver methods
>>>>>   only use clause 45 register accesses (the only clause 22 accesses
>>>>>   are left when accessing c45 registers indirectly, if the MDIO bus
>>>>>   does not support clause 45 accesses).
>>>>>   The driver starts using the genphy_c45_* methods.
>>>>>
>>>>>   At this point the driver is ready to be used on a MDIO bus capable
>>>>>   of only clause 45 accesses, but will still work on clause 22 only
>>>>>   MDIO bus.
>>>>>
>>>>> - I then add support for SerDes mode switching between 2500base-x
>>>>>   and sgmii, based on autonegotiated copper speed.
>>>>>
>>>>> All this is done so that we can support another 2.5G copper SFP
>>>>> module, which is enabled by the last patch.
>>>>>     
>>>>
>>>> Has been verified that the RTL8125-integrated PHY's still work
>>>> properly with this patch set?
>>>>  
>>>
>>> Hi Heiner,
>>>
>>> no, I wanted to send you an email to test this. I do not have the
>>> controllers with integrates PHYs.
>>>
>>> Can you test this?
>>>
>>> Also do you have a controller where the rtlgen driver is used but it
>>> only supports 1gbps ? I.e. where the PHY ID is RTL_GENERIC_PHYID
>>> (0x001cc800).
>>>
>>> I am asking because I am told that it also is clause 45, so the drivers
>>> can potentially be merged completely (the rtl822x_ functions can be
>>> merged with rtlgen_ functions and everything rewritten to clause 45,
>>> and gentphy_c45_ functions can be used).
>>>   
>> At least on RTL8168h indirect MMD reads return 0 always.
>> IIRC this was the reason why the rtlgen functions use the vendor-specific
>> registers.
> 
> Looking at the code in r8169_phy_config.c, I see function
>   rtl8168h_config_eee_phy()
> with three paged writes to vendor registers, but the writes do not access
> the same registers as the .read_mmd() methods for the PCS_EEE / AN_EEE registers
> in realtek.c PHY driver.
> 
That's some other undocumented EEE-related magic copied from the vendor driver.

> It seems for now it would be best to keep the methods for paged
> accesses.
> 
> Could you test the patchset without the patch that removes the paged
> access methods?
> 
> The rewrite of the read_mmd / write_mmd methods should not cause
> problems. I am told by the realtek contact you gave me that:
> 

I tested on RTL8125A and RTL8125B, and on both indirect MMD reads return 0.
I tested with reading MDIO_MMD_PCS / MDIO_PCS_EEE_ABLE. Reading this register
should return 6. So it seems indirect MMD access was somehow and for whatever
reason disabled by Realtek for the RTL8125-internal PHY's (provided that
indirect C22 MMD access is supported by the standalone versions).

Note:
Internal PHY of RTL8125A has PHY ID 0x001cc800 (RTL8226 2.5Gbps PHY)
Internal PHY of RTL8125B has PHY ID 0x001cc840 (RTL8226B_RTL8221B 2.5Gbps PHY)

Consequence is that we can't replace reading the vendor-specific registers
with standard MMD reads.

>   If FE PHY supports EEE, then it will support MMD register and it will
>   also support use internal registers to access theses MMD registers.
> 
> Marek

Heiner