Re: [PATCH net-next] net: phy: micrel: Add support for lan8842

[Oleksij Rempel <o.rempel@pengutronix.de>]

On Wed, Jul 23, 2025 at 11:01:45AM +0200, Horatiu Vultur wrote:
> The 07/23/2025 07:34, Oleksij Rempel wrote:
> > 
> > Hi Horatiu,
> 
> Hi Olekij,
> 
> > 
> > On Mon, Jul 21, 2025 at 09:14:05AM +0200, Horatiu Vultur wrote:
> > 
> > > +static int lan8842_config_init(struct phy_device *phydev)
> > > +{
> > > +     int val;
> > > +     int ret;
> > > +
> > > +     /* Reset the PHY */
> > > +     val = lanphy_read_page_reg(phydev, 4, LAN8814_QSGMII_SOFT_RESET);
> > 
> > It would be good to use defines for MMD pages.
> 
> Those are extended pages and not MMD pages. Currently in the entire
> source code I can see we used hardcoded values, also in the register
> description it looks like all these extended pages do not have really
> meaningfull names: Extended Page 0, Extended Page 4, Extended Page 5...

I'll be happy with xxxx_EXT_PAGE_0, etc 

> > > +     val = lanphy_read_page_reg(phydev, 0, LAN8842_FLF);
> > > +     if (val < 0)
> > > +             return val;
> > > +     val |= LAN8842_FLF_ENA | LAN8842_FLF_ENA_LINK_DOWN;
> > 
> > If I see it correctly, FLF support will make link fail after ~1ms, while
> > IEEE 802.3 recommends 750ms. Since a link recovery of a PHY with autoneg
> > support usually takes multiple seconds, I see the benefit for FLF
> > support only mostly for SyncE environment at same time it seems to be
> > a disadvantage for other environments.
> 
> Why would be a disadvantage?

The disadvantage in a standard network without a backup link (like one
using RSTP) comes from how the system handles recoverable, temporary
errors like a short burst of noise.

# Standard PHY Behavior (Grace Period)
When a standard 1000BASE-T link becomes unstable, the IEEE 802.3
standard requires the PHY to attempt to retrain and recover the
connection on its own. It has a timeout window of up to 750 ms to do
this, which acts as a grace period.

If the link issue was temporary and the PHY recovers within this window,
the operating system never sees a "link down" event. Applications only
experience a brief moment of packet loss, which is often handled
gracefully by protocols like TCP.

# FLF Behavior (Immediate Failure)
An FLF-enabled PHY is designed to report link instability almost
immediately (~1 ms). Instead of trying to recover silently, it
immediately reports a hard link failure to the operating system.

# The Disadvantage in a Single-Link System

For a system with only one link, this "fail-fast" approach can be a
disadvantage. Consider a short, recoverable noise burst:

- Without FLF: The PHY uses its 750 ms grace period to recover. The
link stays up, and the service interruption is limited to brief packet
loss.

- With FLF: The PHY reports "link down" after ~1 ms. The operating
system tears down the network interface. Even if the hardware recovers
quickly, the OS has to bring the interface back up, re-run DHCP, and
re-establish all application connections. This system-level recovery
often takes much longer than the original glitch.

In short, FLF can turn a minor, recoverable physical-layer glitch into a
more disruptive, longer-lasting outage at the application level when
there is no backup link to switch to.

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |