Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
To:     Marek Vasut <marex@denx.de>,
        Oleksij Rempel <o.rempel@pengutronix.de>,
        Andrew Lunn <andrew@lunn.ch>
Cc:     David Jander <david@protonic.nl>, kernel@pengutronix.de,
        linux-kernel@vger.kernel.org, netdev@vger.kernel.org,
        Russell King <linux@armlinux.org.uk>, mkl@pengutronix.de
References: <20201006080424.GA6988@pengutronix.de>
 <2cc5ea02-707e-dbb5-c081-4c5202bd5815@gmail.com>
 <42d4c4b2-d3ea-9130-ef7f-3d1955116fdc@denx.de>
From:   Florian Fainelli <f.fainelli@gmail.com>
Subject: Re: PHY reset question
Message-ID: <0687984c-5768-7c71-5796-8e16169f5192@gmail.com>
Date:   Tue, 6 Oct 2020 14:11:48 -0700
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:78.0) Gecko/20100101
 Firefox/78.0 Thunderbird/78.3.1
MIME-Version: 1.0
In-Reply-To: <42d4c4b2-d3ea-9130-ef7f-3d1955116fdc@denx.de>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 10/6/2020 1:24 PM, Marek Vasut wrote:
> On 10/6/20 9:36 PM, Florian Fainelli wrote:
> [...]
>>> - Use compatible ("compatible = "ethernet-phy-id0022.1560") in the
>>> devicetree,
>>>     so that reading the PHYID is not needed
>>>     - easy to solve.
>>>     Disadvantage:
>>>     - losing PHY auto-detection capability
>>>     - need a new devicetree if different PHY is used (for example in
>>> different
>>>       board revision)
>>
>> Or you can punt that to the boot loader to be able to tell the
>> difference and populate different compatible, or even manage the PHY
>> reset to be able to read the actual PHY OUI. To me that is still the
>> best solution around.
> 
> Wasn't there some requirement for Linux to be bootloader-independent ?

What kind of dependency does this create here? The fact that Linux is 
capable of parsing a compatible string of the form 
"ethernet-phyAAAA.BBBB" is not something that is exclusively applicable 
to Linux. Linux just so happens to support that, but so could FreeBSD or 
any OS for that matter.

This is exactly the way firmware should be going, that is to describe 
accurately the hardware, while making the life of the OS much easier 
when it can. If we supported ACPI that is exactly what would have to 
happen IMHO.

> Some systems cannot replace their bootloaders, e.g. if the bootloader is
> in ROM, so this might not be a solution.

It is always possible to chain load a field updateable boot loader, and 
even when that is not desirable you could devise a solution that allows 
to utilize say a slightly different DTB that you could append to the 
kernel. Again, if you want to use strictly the same DTB, then you have 
to do what I just suggested and have the boot loader absorb some of this 
complexit

> 
>>> - modify PHY framework to deassert reset before identifying the PHY.
>>>     Disadvantages?
> 
> If this happens on MX6 with FEC, can you please try these two patches?
> 
> https://patchwork.ozlabs.org/project/netdev/patch/20201006135253.97395-1-marex@denx.de/
> 
> https://patchwork.ozlabs.org/project/netdev/patch/20201006202029.254212-1-marex@denx.de/

Your patches are not scaling across multiple Ethernet MAC drivers 
unfortunately, so I am not sure this should be even remotely considered 
a viable solution.
-- 
Florian