Re: [PATCH net-next 1/3] net: netdevsim: Add PHY support in netdevsim

[Maxime Chevallier <maxime.chevallier@bootlin.com>]

Hi Simon,

On Fri, 4 Jul 2025 13:43:36 +0100
Simon Horman <horms@kernel.org> wrote:

> On Wed, Jul 02, 2025 at 10:28:03AM +0200, Maxime Chevallier wrote:
> > With the introduction of phy_link_topology, we have the ability to keep
> > track of PHY devices that sit behind a net_device. While we still can
> > only attach one single PHY to a netdev, we can look at all these PHYs
> > through netlink, with the ETHTOOL_MSG_PHY_GET command.
> > 
> > Moreover, netlink commands that are targeting PHY devices also now
> > allow specifying which PHY we want to address in a given netlink
> > command.
> > 
> > That whole process comes with its own complexity, and a few bugs were
> > dicovered over the months following the introduction of  
> 
> Hi Maxime,
> 
> As it seems like there will be a v2 anyway: discovered

Thanks :)

> > phy_link_topology.  
> 
> ...
> 
> > +static struct phy_driver nsim_virtual_phy_drv[] = {
> > +	{
> > +		.name			= "Netdevsim virtual PHY driver",
> > +		.get_features		= nsim_get_features,
> > +		.match_phy_device	= nsim_match_phy_device,
> > +		.config_aneg		= nsim_config_aneg,
> > +		.read_status		= nsim_read_status,
> > +	},
> > +};
> > +
> > +module_phy_driver(nsim_virtual_phy_drv);  
> 
> I see that this has been flagged by Kernel Test Robot,
> but as I had already written most of this it seems worth sending anyway.
> 
> I am somewhat guessing at the why here, but
> I see build failures with this patch applied:
> 
> ld: drivers/net/netdevsim/phy.o: in function `phy_module_init':
> phy.c:(.init.text+0x0): multiple definition of `init_module'; drivers/net/netdevsim/netdev.o:netdev.c:(.init.text+0x0): first defined here
> ld: drivers/net/netdevsim/phy.o: in function `phy_module_exit':
> phy.c:(.exit.text+0x0): multiple definition of `cleanup_module'; drivers/net/netdevsim/netdev.o:netdev.c:(.exit.text+0x0): first defined here
> 
> I am guessing that this is because above module_phy_driver() will define
> init_module and phy_module_exit functions.  But the following lines near
> the end of drivers/net/netdevsim/netdev.c also define functions with those
> names.
> 
> module_init(nsim_module_init);
> module_exit(nsim_module_exit);
> 
> ...

I just received the kernel test robot report indeed :( Thanks for the
investigation ! I'll rework that part, sorry about that :/

Maxime