Re: [net-next PATCH 3/6] net: phylink: Correctly handle PCS probe defer from PCS provider

[Christian Marangi <ansuelsmth@gmail.com>]

On Wed, Mar 19, 2025 at 07:31:04PM +0000, Russell King (Oracle) wrote:
> On Wed, Mar 19, 2025 at 06:35:21PM +0100, Christian Marangi wrote:
> > On Wed, Mar 19, 2025 at 05:02:50PM +0000, Russell King (Oracle) wrote:
> > > My thoughts are that if a PCS goes away after a MAC driver has "got"
> > > it, then:
> > > 
> > > 1. we need to recognise that those PHY interfaces and/or link modes
> > >    are no longer available.
> > > 2. if the PCS was in-use, then the link needs to be taken down at
> > >    minimum and the .pcs_disable() method needs to be called to
> > >    release any resources that .pcs_enable() enabled (e.g. irq masks,
> > >    power enables, etc.)
> > > 3. the MAC driver needs to be notified that the PCS pointer it
> > >    stashed is no longer valid, so it doesn't return it for
> > >    mac_select_pcs().
> > 
> > But why we need all these indirect handling and checks if we can
> > make use of .remove and shutdown the interface. A removal of a PCS
> > should cause the entire link to go down, isn't a dev_close enough to
> > propagate this? If and when the interface will came up checks are done
> > again and it will fail to go UP if PCS can't be found.
> > 
> > I know it's a drastic approach to call dev_close but link is down anyway
> > so lets reinit everything from scratch. It should handle point 2 and 3
> > right?
> 
> Let's look at what dev_close() does. This is how it's documented:
> 
>  * dev_close() - shutdown an interface
>  * @dev: device to shutdown
>  *
>  * This function moves an active device into down state. A
>  * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
>  * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
>  * chain.
> 
> So, this is equivalent to userspace doing:
> 
> # ip li set dev ethX down
> 
> and nothing prevents userspace doing:
> 
> # ip li set dev ethX up
> 
> after that call to dev_close() has returned.
> 
> If this happens, then the netdev driver's .ndo_open will be called,
> which will then call phylink_start(), and that will attempt to bring
> the link back up. That will call .mac_select_pcs(), which _if_ the
> PCS is still "published" means it is _still_ accessible.
> 
> So your call that results in dev_close() with the PCS still being
> published is ineffectual.
> 
> It's *no* different from this crap in the stmmac driver:
> 
>         stmmac_stop_all_dma(priv);
>         stmmac_mac_set(priv, priv->ioaddr, false);
>         unregister_netdev(ndev);
> 
> because *until* that unregister_netdev() call has completed, _userspace_
> still has control over the netdev, and can do whatever it damn well
> pleases.
> 
> Look, this is very very very simple.
> 
> If something is published to another part of the code, it is
> discoverable, and it can be used or manipulated by new users.
> 
> If we wish to take something away, then first, it must be
> unpublished to prevent new users discovering the resource. Then
> existing users need to be dealt with in a safe way. Only at that
> point can we be certain that there are no users, and thus the
> underlying device begin to be torn down.
> 
> It's entirely logical!
>

OK so (I think this was also suggested in the more specific PCS patch)
- 1. unpublish the PCS from the provider
- 2. put down the link...

I feel point 2 is the big effort here to solve. Mainly your problem is
the fact that phylink_major_config should not handle PROBE_DEFER and
should always have all the expected PCS available. (returned from
mac_select_pcs)

So the validation MUST ALWAYS be done before reaching that code path.

That means that when a PCS is removed, the entire phylink should be
refreshed and reevaluated. And at the same time lock userspace from
doing anything fancy (as there might be a possibility for
phylink_major_config)

Daniel at some point in the brainstorm process suggested that we might
need something like phylink_impair() to lock it while it's getting
""refreshed"". Do you think that might be a good path for this?

One of the first implementation of this called phylink_stop (not
dev_stop) so maybe I should reconsider keeping everything phylink
related. But that wouldn't put the interface down from userspace if I'm
not wrong.

It's point 3 (of the old list) "the MAC driver needs to be notified that
the PCS pointer it stashed is no longer valid, so it doesn't return it for
mac_select_pcs()." my problem. I still feel MAC should not track PCS but
only react on the presence (or absence) of them.

And this point is really connected to point 1 so I guess point 1 is the
first to handle, before this. (I also feel it will magically solved once
point 1 is handled)

> > For point 1, additional entry like available_interface? And gets updated
> > once a PCS gets removed??? Or if we don't like the parsing hell we map
> > every interface to a PCS pointer? (not worth the wasted space IMHO)
> 
> At the moment, MAC drivers that I've updated will do things like:
> 
>                 phy_interface_or(priv->phylink_config.supported_interfaces,
>                                  priv->phylink_config.supported_interfaces,
>                                  pcs->supported_interfaces);
> 
> phylink_config.supported_interfaces is the set of interface modes that
> the MAC _and_ PCS subsystem supports. It's not just the MAC, it's both
> together.
> 
> So, if a PCS is going away, then clearing the interface modes that the
> PCS was providing would make sense - but there's a problem here. What
> if the PCS is a bought-in bit of IP where the driver supports many modes
> but the MAC doesn't use it for all those modes. So... which interface
> modes get cleared is up to the MAC driver to decide.
> 

Should we add an OP to handle removal of PCS from a MAC? Like
.mac_release_pcs ? I might be wrong but isn't that giving too much
freedom to the driver?

I need to recheck how the interface validation work and what values are
used but with this removal thing on the table, supported_interfaces OR
with the PCS supported_interface might be problematic and maybe the
original values should be stored somewhere.

> > > There's probably a bunch more that needs to happen, and maybe need
> > > to consider how to deal with "pcs came back".. but I haven't thought
> > > that through yet.
> > 
> > Current approach supports PCS came back as we check the global provider
> > list and the PCS is reachable again there.
> > (we tasted various scenario with unbind/bind while the interface was
> > up/down)
> 
> ... because you look up the PCS in the mac_select_pcs() callback which
> leads to a different race to what we have today, this time inside the
> phylink code which thankfully phylink prints an error which is *NEVER*
> supposed to happen.
>

I want to make sure tho you are ok with the usage of .mac_select_pcs
for re-evaluation task.

Maybe a better approach is to introduce .mac_get_pcs and enforce the
usage only on validation phase? (aka in phylink_validate_mac_and_pcs)

AFAIK in that phase .mac_select_pcs can return errors if the requested
interface is not possible for one reason or another.

What do you think? In short 2 additional OP that with the select one
result in:

- get
- select
- release

-- 
	Ansuel