Re: Link down reasons

[Andrew Lunn <andrew@lunn.ch>]

On Thu, May 28, 2020 at 06:54:24PM +0200, Petr Machata wrote:
> 
> Andrew Lunn <andrew@lunn.ch> writes:
> 
> >> Andrew, pardon my ignorance in these matters, can a PHY driver in
> >> general determine that the issue is with the cable, even without running
> >> the fairly expensive cable test?
> >
> > No. To diagnose a problem, you need the link to be idle. If the link
> > peer is sending frames, they interfere with TDR. So all the cable
> > testing i've seen first manipulates the auto-negotiation to make the
> > link peer go quiet. That takes 1 1/2 seconds. There are some
> > optimizations possible, e.g. if the cable is so broken it never
> > establishes link, you can skip this. But Ethernet tends to be robust,
> > it drops back to 100Mbps only using two pairs if one of the four pairs
> > is broken, for example.
> 
> OK, thanks. I suspect our FW is doing this behind the scenes, because it
> can report a shorted cable.
> 
> In another e-mail you suggested this:
> 
>     Link detected: no (cable issue)
> 
> But if the link just silently falls back to 100Mbps, there would never
> be an opportunity for phy to actually report a down reason. So there
> probably is no way for the phy layer to make use of this particular
> down reason.

It is called downshift. And we have support for it in the phylib core,
if the PHY has the needed vendor register.

https://elixir.bootlin.com/linux/v5.7-rc7/source/drivers/net/phy/phy-core.c#L341
https://elixir.bootlin.com/linux/v5.7-rc7/source/drivers/net/phy/phy.c#L95

There are also standard phylib/ethtool ways to configure it, how many
times the PHY should try to establish a 1G link before downshifting to
100M.

So in theory we could report:

Link detected: yes (downshifted)

Assuming your proposed API support a reason why it is up, not just a
reason why it is down?

	Andrew