Received-SPF: softfail (google.com: domain of transitioning linux-kernel-owner@vger.kernel.org does not designate 23.128.96.19 as permitted sender) client-ip=23.128.96.19;
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII;
 format=flowed
Content-Transfer-Encoding: 7bit
Date:   Mon, 21 Mar 2022 22:41:56 +0100
From:   Michael Walle <michael@walle.cc>
To:     Andrew Lunn <andrew@lunn.ch>
Cc:     netdev@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: Clause 45 and Clause 22 PHYs on one MDIO bus
In-Reply-To: <YjjhxbZgKHykJ+35@lunn.ch>
References: <240354b0a54b37e8b5764773711b8aa3@walle.cc>
 <cdb3d3f6ad35d4e26fd8abb23b2e96a3@walle.cc> <YjjhxbZgKHykJ+35@lunn.ch>
User-Agent: Roundcube Webmail/1.4.13
Message-ID: <4d728d267e45fe591c933c86cdfff333@walle.cc>
Precedence: bulk

Am 2022-03-21 21:36, schrieb Andrew Lunn:
>> Actually, it looks like mdiobus_c45_read() is really c45 only and only
>> used for PHYs which just support c45 and not c45-over-c22 (?). I was
>> mistaken by the heavy use of the function in phy_device.c. All the
>> methods in phy-c45.c use phy_*_mmd() functions. Thus it might only be
>> the mxl-gpy doing something fishy in its probe function.
> 
> Yes, there is something odd here. You should search back on the
> mailing list.
> 
> If i remember correctly, it is something like it responds to both c22
> and c45. If it is found via c22, phylib does not set phydev->is_c45,
> and everything ends up going indirect. So the probe additionally tries
> to find it via c45? Or something like that.

Yeah, found it: https://lore.kernel.org/netdev/YLaG9cdn6ewdffjV@lunn.ch/

But that means that if the controller is not c45 capable, it will always
fail to probe, no?

I've added the "if (regnum & MII_ADDR_C45) return -EOPNOTSUPP" to the
mdio driver and the gpy phy will then fail to probe - as expected.

Should it check for -EOPNOTSUPP and just ignore that error and continue
probing? Or make it a no-op if probe_capabilities say it has no c45
access so it would take advantage of a quirk flag (derived from dt)?

>> Nevertheless, I'd still need the opt-out of any c45 access. Otherwise,
>> if someone will ever implement c45 support for the mdio-mscc-mdio
>> driver, I'll run in the erratic behavior.
> 
> Yah, i need to think about that. Are you purely in the DT world, or is
> ACPI also an option?

Just DT world.

> Maybe extend of_mdiobus_register() to look for a DT property to limit
> what values probe_capabilities can take?

I'll have to give it a try. First I was thinking that we wouldn't need
it because a broken PHY driver could just set a quirk 
"broken_c45_access"
or similar. But that would mean it has to be probed before any c45 PHY.
Dunno if that will be true for the future. And it sounds rather fragile.
So yes, a dt property might be a better option.

-michael