Re: [PATCH net-next v2 2/5] net: rnpgbe: Report link settings through ethtool
From: Yibo Dong
Date: Tue Nov 17 2026 - 20:18:29 EST
On Tue, Nov 17, 2026 at 02:21:46PM +0100, Andrew Lunn wrote:
> On Tue, Nov 17, 2026 at 10:55:26AM +0800, Yibo Dong wrote:
> > On Tue, Nov 17, 2026 at 02:54:45AM +0100, Andrew Lunn wrote:
> > > On Tue, Nov 17, 2026 at 09:49:02AM +0800, Yibo Dong wrote:
> > > > On Mon, Nov 16, 2026 at 02:18:28PM +0100, Andrew Lunn wrote:
> > > > > > link, speed, duplex and mdix are updated asynchronously by
> > > > > > the mailbox event handler, outside RTNL. link_lock protects these updates
> > > > > > and provides a consistent snapshot in the getter.
> > > > >
> > > > > O.K, thanks for the explanation.
> > > > >
> > > > > I removed too much context, so i cannot see here. What about
> > > > > lp_advertising? That also comes from the firmware, reported on link up
> > > > > when autoneg completes.
> > > > >
> > > > The firmware manages autonegotiation and reports the negotiated speed
> > > > and duplex through mailbox link events. However, the current mailbox
> > > > interface does not expose the link partner's advertisement bitmap.
> > > >
> > > > Therefore, lp_advertising is currently left zero to indicate that this
> > > > information is unknown to the driver.
> > > >
> > > > If reporting lp_advertising is must, we would need to add a new
> > > > firmware mailbox command to retrieve the link partner's advertisement,
> > > > or add lp_advertising to link-event mbx alongwith link, speed and
> > > > duplex.
> > >
> > > It is not a must, but it is useful.
> > >
> > > Or you can just throw away your firmware, expose the MDIO bus, and let
> > > phylink drive the hardware. Less code, less bugs, more functionality
> > > for free.
> > The GMAC provides MDIO register access, so exposing an MDIO bus is
> > technically possible. However, the firmware currently manages the PHY,
> > and its interface does not provide a PHY ownership handover mechanism.
> > Introducing host-side PHY management would require coordinated firmware
> > and driver changes to avoid independently configuring the same PHY.
> >
> > In addition, on this SoC, the firmware configures MDI/MDI-X through
> > AXI-mapped control registers driving the PHY's pi_force_mdix_en and
> > pi_force_mdix_mode inputs. This control path is outside the PHY's MDIO
> > register interface.
> >
> > The firmware also supports link management before the host driver is
> > loaded and while the host is shut down, including wake-up handling.
> > On OCP platforms, it supports preserving the PHY link across driver
> > reloads and host shutdown. These capabilities are already implemented
> > in firmware, although driver support for them is not included in this
> > series. Any future ownership handover would need to preserve them.
> >
> > Would it be acceptable for this series to retain the firmware-managed
> > PHY design without registering an MDIO bus, and leave lp_advertising
> > zero for now? If a mailbox command is added to expose the link partner's
> > advertisement, we can populate lp_advertising in a follow-up patch.
>
> You have not implemented anything special, other systems offer the
> same features, using phylink. However, it is fine to implement it all
> your own, rather than reuse existing software. SoC devices tend to
> make use of Linux to do all this, and PCIe devices tend to reinvent
> the wheel and do it in firmware. We are happy either way.
>
Thanks for the clarification.
> Andrew
>