Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 5 Jan 2023 19:34:45 +0200
From:   Vladimir Oltean <olteanv@gmail.com>
To:     Sean Anderson <sean.anderson@seco.com>
Cc:     Andrew Lunn <andrew@lunn.ch>,
        Heiner Kallweit <hkallweit1@gmail.com>, netdev@vger.kernel.org,
        Russell King <linux@armlinux.org.uk>,
        "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
Message-ID: <20230105173445.72rvdt4etvteageq@skbuf>
References: <20230103220511.3378316-1-sean.anderson@seco.com>
 <20230103220511.3378316-5-sean.anderson@seco.com>
 <20230105140421.bqd2aed6du5mtxn4@skbuf>
 <6ffe6719-648c-36aa-74be-467c8db40531@seco.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <6ffe6719-648c-36aa-74be-467c8db40531@seco.com>
Precedence: bulk

On Thu, Jan 05, 2023 at 11:21:14AM -0500, Sean Anderson wrote:
> > Your code walks through the speed_table[] of media speeds (from 10M up
> > until the max speed of the SERDES) and sees whether the PHY was
> > provisioned, for that speed, to use PAUSE rate adaptation.
> 
> This is because we assume that if a phy supports rate matching for a phy
> interface mode, then it supports rate matching to all slower speeds that
> it otherwise supports. This seemed like a pretty reasonable assumption
> when I wrote the code, but it turns out that some firmwares don't abide
> by this. This is firstly a problem with the firmware (as it should be
> configured so that Linux can use the phy's features), but we have to be
> careful not to end up with an unsupported combination.

When you say "problem with the firmware", you're referring specifically
to my example (10GBASE-R for >1G speeds, SGMII for <=1G speeds)?

Why do you consider this a firmware misconfiguration? Let's say the host
supports both 10GBASE-R and SGMII, but the system designer preferred not
to use PAUSE-based rate adaptation for the speeds where native rate
adaptation was available.

> > If the PHY firmware uses a combination like this: 10GBASE-R/XFI for
> > media speeds of 10G, 5G, 2.5G (rate adapted), and SGMII for 1G, 100M
> > and 10M, a call to your implementation of
> > aqr107_get_rate_matching(PHY_INTERFACE_MODE_10GBASER) would return
> > RATE_MATCH_NONE, right?
> 
> Correct.
> 
> > So only ETHTOOL_LINK_MODE_10000baseT_Full_BIT
> > would be advertised on the media side?
> 
> If the host only supports 10GBASE-R and nothing else. If the host
> supports SGMII as well, we will advertise 10G, 1G, 100M, and 10M. But
> really, this is a problem with the firmware, since if the host supports
> only 10GBASE-R, then the firmware should be set up to rate adapt to all
> speeds.

So we lose the advertisement of 5G and 2.5G, even if the firmware is
provisioned for them via 10GBASE-R rate adaptation, right? Because when
asked "What kind of rate matching is supported for 10GBASE-R?", the
Aquantia driver will respond "None".

> > Shouldn't you take into consideration in your aqr107_rate_adapt_ok()
> > function only the media side link speeds for which the PHY was actually
> > *configured* to use the SERDES protocol @iface?
> 
> No, because we don't know what phy interface modes are actually going to
> be used. The phy doesn't know whether e.g. the host supports both
> 10GBASE-R and SGMII or whether it only supports 10GBASE-R. With the
> current API we cannot say "I support 5G" without also saying "I support
> 1G". If you don't like this, please send a patch for an API returning
> supported speeds for a phy interface mode.