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;
MIME-Version: 1.0
References: <1656618906-29881-1-git-send-email-radhey.shyam.pandey@amd.com>
 <Yr66xEMB/ORr0Xcp@lunn.ch> <MN0PR12MB59531DFD084FA947084D91B6B7819@MN0PR12MB5953.namprd12.prod.outlook.com>
 <CAGETcx_BUR3EPDLgp9v0Uk9N=8BtYRjFyhpJTQa9kEMHtkgdwQ@mail.gmail.com> <YsSVqknDQxdWqfds@lunn.ch>
In-Reply-To: <YsSVqknDQxdWqfds@lunn.ch>
From:   Saravana Kannan <saravanak@google.com>
Date:   Fri, 15 Jul 2022 12:06:39 -0700
Message-ID: <CAGETcx8dN+STVrG+1OyXbfL5995MUxHtVnzQZp9-xusaHyvTmA@mail.gmail.com>
Subject: Re: [PATCH net-next v2] net: macb: In shared MDIO usecase make MDIO
 producer ethernet node to probe first
To:     Andrew Lunn <andrew@lunn.ch>
Cc:     "Pandey, Radhey Shyam" <radhey.shyam.pandey@amd.com>,
        "nicolas.ferre@microchip.com" <nicolas.ferre@microchip.com>,
        "claudiu.beznea@microchip.com" <claudiu.beznea@microchip.com>,
        "davem@davemloft.net" <davem@davemloft.net>,
        "edumazet@google.com" <edumazet@google.com>,
        "kuba@kernel.org" <kuba@kernel.org>,
        "pabeni@redhat.com" <pabeni@redhat.com>,
        "hkallweit1@gmail.com" <hkallweit1@gmail.com>,
        "linux@armlinux.org.uk" <linux@armlinux.org.uk>,
        "gregkh@linuxfoundation.org" <gregkh@linuxfoundation.org>,
        "rafael@kernel.org" <rafael@kernel.org>,
        "netdev@vger.kernel.org" <netdev@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "git (AMD-Xilinx)" <git@amd.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Jul 5, 2022 at 12:49 PM Andrew Lunn <andrew@lunn.ch> wrote:
>
> > > Thanks for the review.  I want to get your thoughts on the outline of
> > > the generic solution. Is the current approach fine and we can extend it
> > > for all shared MDIO use cases/ or do we see any limitations?
> > >
> > > a) Figure out if the MDIO bus is shared.  (new binding or reuse existing)
> > > b) If the MDIO bus is shared based on DT property then figure out if the
> > > MDIO producer platform device is probed. If not, defer MDIO consumer
> > > MDIO bus registration.
> >
> > Radhey,
> >
> > I think Andrew added me because he's pointing you towards fw_devlink.
> >
> > Andrew,
> >
> > I have intentionally not added phy-handle support to fw_devlink
> > because it would also prevent the generic driver from binding/cause
> > issues with DSA. I have some high level ideas on fixing that but
> > haven't gotten around to it yet.
>
> I took a quick look at macb, and i think it is actually broken in
> other ways. If you where to use NFS root, i suspect it would also
> fail.
>
> This also has nothing to do with shared MDIO busses as such. All it
> requires is some other MDIO bus, not the MACs own MDIO bus.
>
> It is also that we cannot return -EPROBE_DEFER when trying to connect
> the PHY, because it is not performed in the context of the probe, but
> the open.
>
> fw_dewlink might help solve this, bit it is not going to be easy. We
> can also split this into two problems;
>
> 1) probe time
> 2) suspend/resume
>
> macb does seem to probe, for most use cases. So we can probably ignore
> that for now. So we can concentrate on suspend/resume. You say
> suspend/resume is based on probe order. So it must build some sort of
> tree. Can we make phy_attach_direct add an additional link to this
> tree when a MAC device is link to a PHY? Is this what
> device_link_add() is about?

Based on the flags you pass, you can tell device link to only enforce
suspend/resume ordering or also enforce probe ordering.

-Saravana