Re: [PATCH net-next v5 4/4] phy: aquantia: Determine rate adaptation support from registers

From: Sean Anderson <sean.anderson@seco.com>
To: "Russell King (Oracle)" <linux@armlinux.org.uk>
Cc: Vladimir Oltean <olteanv@gmail.com>, Andrew Lunn <andrew@lunn.ch>,
	Heiner Kallweit <hkallweit1@gmail.com>,
	netdev@vger.kernel.org, "David S . Miller" <davem@davemloft.net>,
	Paolo Abeni <pabeni@redhat.com>,
	linux-kernel@vger.kernel.org, Jakub Kicinski <kuba@kernel.org>,
	Eric Dumazet <edumazet@google.com>,
	Tim Harvey <tharvey@gateworks.com>
Subject: Re: [PATCH net-next v5 4/4] phy: aquantia: Determine rate adaptation support from registers
Date: Thu, 5 Jan 2023 14:00:23 -0500	[thread overview]
Message-ID: <f777e276-d26e-22f7-f737-c07f49b20bd9@seco.com> (raw)
In-Reply-To: <Y7cd3DrnKxNmHVcp@shell.armlinux.org.uk>

On 1/5/23 13:58, Russell King (Oracle) wrote:
> On Thu, Jan 05, 2023 at 01:03:49PM -0500, Sean Anderson wrote:
>> On 1/5/23 12:55, Vladimir Oltean wrote:
>> > On Thu, Jan 05, 2023 at 07:52:06PM +0200, Vladimir Oltean wrote:
>> >> On Thu, Jan 05, 2023 at 12:43:47PM -0500, Sean Anderson wrote:
>> >> > Again, this is to comply with the existing API assumptions. The current
>> >> > code is buggy. Of course, another way around this is to modify the API.
>> >> > I have chosen this route because I don't have a situation like you
>> >> > described. But if support for that is important to you, I encourage you
>> >> > to refactor things.
>> >> 
>> >> I don't think I'm aware of a practical situation like that either.
>> >> I remember seeing some S32G boards with Aquantia PHYs which use 2500BASE-X
>> >> for 2.5G and SGMII for <=1G, but that's about it in terms of protocol switching.
>> >> As for Layerscape boards, SERDES protocol switching is a very new concept there,
>> >> so they're all going to be provisioned for PAUSE all the way down
>> >> (or USXGMII, where that is available).
>> >> 
>> >> I just pointed this out because it jumped out to me. I don't have
>> >> something against this patch getting accepted as it is.
>> > 
>> > A real-life (albeit niche) scenario where someone might have an Aquantia
>> > firmware provisioned like this would be a 10G capable port that also
>> > wants to support half duplex at 10/100 speeds. Although I'm not quite
>> > sure who cares about half duplex all that much these days.
>> 
>> IMO if we really want to support this, the easier way would be to teach
>> the phy driver how to change the rate adaptation mode. That way we could
>> always advertise rate adaptation, but if someone came along and
>> requested 10HD we could reconfigure the phy to support it. However, this
>> was deemed too risky in the discussion for v1, since we don't really
>> know how the firmware interacts with the registers.
> 
> I'm not sure about "someone came along and requested 10HD". Don't you
> mean "if someone plugged the RJ45 into a 10bT hub which only supports
> 10HD" ? Or are you suggesting that we shouldn't advertise 10HD and
> 100HD along with everything else, and then switch into this special
> mode if someone wants to advertise these and disable all other link
> modes?
> 

The former. "someone" being userspace or the remote end.

--Sean