[PATCH net-next] net: phy: Don't conditionally compile the phy_link_topology creation

[Heiner Kallweit]

On 30.04.2024 13:57, Maxime Chevallier wrote:
> Hello Heiner,
> 
> On Tue, 30 Apr 2024 10:17:31 +0200
> Heiner Kallweit <hkallweit1 at gmail.com> wrote:
> 
>> On 29.04.2024 15:10, Maxime Chevallier wrote:
>>> The core of the phy_link_topology isn't directly tied to phylib, and at
>>> the moment it's initialized, phylib might not be loaded yet. Move the
>>> initialization of the topology to the phy_link_topology_core header,
>>> which contains the bare minimum so that we can initialize it at netdev
>>> creation.
>>>   
>>
>> The change fixes the issue for me, but according to my personal taste
>> the code isn't intuitive and still error-prone. Also there's no good
>> reason to inline a function like phy_link_topo_create() and make it
>> publicly available. Do you expect it to be ever used outside net core?
>> In general it may make sense to add a config symbol for the topology
>> extension, there seem to be very few, specialized use cases for it.
> 
> I think I'm missing the point here then. Do you mean adding a Kconfig
> option to explicitely turn phy_link_topology on ? or build it as a
> dedicated kernel module ?
> 
> Or do you see something such as "if phylib is M or Y, then build the
> topology stuff and make sure it's allocated when a netdev gets created
> ?"
> 
> Thanks,
> 
> Maxime
> 
>>
>>> Signed-off-by: Maxime Chevallier <maxime.chevallier at bootlin.com>
>>> Closes: https://lore.kernel.org/netdev/2e11b89d-100f-49e7-9c9a-834cc0b82f97@gmail.com/
>>> Closes: https://lore.kernel.org/netdev/20240409201553.GA4124869@dev-arch.thelio-3990X/
>>> ---
>>>  drivers/net/phy/phy_link_topology.c    | 23 --------------------
>>>  include/linux/phy_link_topology.h      |  5 -----
>>>  include/linux/phy_link_topology_core.h | 30 +++++++++++++++++---------
>>>  3 files changed, 20 insertions(+), 38 deletions(-)
>>>
>>> diff --git a/drivers/net/phy/phy_link_topology.c b/drivers/net/phy/phy_link_topology.c
>>> index 985941c5c558..960aedd73308 100644
>>> --- a/drivers/net/phy/phy_link_topology.c
>>> +++ b/drivers/net/phy/phy_link_topology.c
>>> @@ -12,29 +12,6 @@
>>>  #include <linux/rtnetlink.h>
>>>  #include <linux/xarray.h>
>>>  
>>> -struct phy_link_topology *phy_link_topo_create(struct net_device *dev)
>>> -{
>>> -	struct phy_link_topology *topo;
>>> -
>>> -	topo = kzalloc(sizeof(*topo), GFP_KERNEL);
>>> -	if (!topo)
>>> -		return ERR_PTR(-ENOMEM);
>>> -
>>> -	xa_init_flags(&topo->phys, XA_FLAGS_ALLOC1);
>>> -	topo->next_phy_index = 1;
>>> -
>>> -	return topo;
>>> -}
>>> -
>>> -void phy_link_topo_destroy(struct phy_link_topology *topo)
>>> -{
>>> -	if (!topo)
>>> -		return;
>>> -
>>> -	xa_destroy(&topo->phys);
>>> -	kfree(topo);
>>> -}
>>> -
>>>  int phy_link_topo_add_phy(struct phy_link_topology *topo,
>>>  			  struct phy_device *phy,
>>>  			  enum phy_upstream upt, void *upstream)
>>> diff --git a/include/linux/phy_link_topology.h b/include/linux/phy_link_topology.h
>>> index 6b79feb607e7..ad72d7881257 100644
>>> --- a/include/linux/phy_link_topology.h
>>> +++ b/include/linux/phy_link_topology.h
>>> @@ -32,11 +32,6 @@ struct phy_device_node {
>>>  	struct phy_device *phy;
>>>  };
>>>  
>>> -struct phy_link_topology {
>>> -	struct xarray phys;
>>> -	u32 next_phy_index;
>>> -};
>>> -
>>>  static inline struct phy_device *
>>>  phy_link_topo_get_phy(struct phy_link_topology *topo, u32 phyindex)
>>>  {
>>> diff --git a/include/linux/phy_link_topology_core.h b/include/linux/phy_link_topology_core.h
>>> index 0a6479055745..0116ec49cd1b 100644
>>> --- a/include/linux/phy_link_topology_core.h
>>> +++ b/include/linux/phy_link_topology_core.h
>>> @@ -2,24 +2,34 @@
>>>  #ifndef __PHY_LINK_TOPOLOGY_CORE_H
>>>  #define __PHY_LINK_TOPOLOGY_CORE_H
>>>  
>>> -struct phy_link_topology;
>>> +#include <linux/xarray.h>
>>>  
>>> -#if IS_REACHABLE(CONFIG_PHYLIB)
>>> -
>>> -struct phy_link_topology *phy_link_topo_create(struct net_device *dev);
>>> -void phy_link_topo_destroy(struct phy_link_topology *topo);
>>> -
>>> -#else
>>> +struct phy_link_topology {
>>> +	struct xarray phys;
>>> +	u32 next_phy_index;
>>> +};
>>>  
>>>  static inline struct phy_link_topology *phy_link_topo_create(struct net_device *dev)
>>>  {
>>> -	return NULL;
>>> +	struct phy_link_topology *topo;
>>> +
>>> +	topo = kzalloc(sizeof(*topo), GFP_KERNEL);
>>> +	if (!topo)
>>> +		return ERR_PTR(-ENOMEM);
>>> +
>>> +	xa_init_flags(&topo->phys, XA_FLAGS_ALLOC1);
>>> +	topo->next_phy_index = 1;
>>> +
>>> +	return topo;
>>>  }
>>>  
>>>  static inline void phy_link_topo_destroy(struct phy_link_topology *topo)
>>>  {
>>> -}
>>> +	if (!topo)
>>> +		return;
>>>  
>>> -#endif
>>> +	xa_destroy(&topo->phys);
>>> +	kfree(topo);
>>> +}
>>>  
>>>  #endif /* __PHY_LINK_TOPOLOGY_CORE_H */  
>>
> 

To go a little bit more into detail:

phy_link_topo_create() and phy_link_topo_destroy() are used in net/core/dev.c
only. Do you expect them to be ever used by other callers?
If not, their functionality could be moved to net/core/dev.c.
Supposedly guarded by IS_ENABLED(CONFIG_PHYLIB), alternatively a new config
symbol for link_topo support.

To get rid of the dependency you could also lazy-inizialize
netdev->link_topo. For this phy_link_topo_add_phy() would have
to take the netdev as first argument, not the topo.
Then the first call to phy_link_topo_add_phy() would initialize
netdev->link_topo.

I think functions like phy_link_topo_get_phy() should also check for
topo != NULL first, maybe combined with a WARN_ON().
They are exported and you have no control over its use.