Re: [PATCH net] net: dsa: Do not leave DSA master with NULL netdev_ops

[Florian Fainelli]

On 5/4/2020 1:49 PM, Vladimir Oltean wrote:
> On Mon, 4 May 2020 at 23:40, Florian Fainelli <f.fainelli@xxxxxxxxx> wrote:
>>
>>
>>
>> On 5/4/2020 1:34 PM, Vladimir Oltean wrote:
>>> Hi Florian,
>>>
>>> On Mon, 4 May 2020 at 23:19, Florian Fainelli <f.fainelli@xxxxxxxxx> wrote:
>>>>
>>>> When ndo_get_phys_port_name() for the CPU port was added we introduced
>>>> an early check for when the DSA master network device in
>>>> dsa_master_ndo_setup() already implements ndo_get_phys_port_name(). When
>>>> we perform the teardown operation in dsa_master_ndo_teardown() we would
>>>> not be checking that cpu_dp->orig_ndo_ops was successfully allocated and
>>>> non-NULL initialized.
>>>>
>>>> With network device drivers such as virtio_net, this leads to a NPD as
>>>> soon as the DSA switch hanging off of it gets torn down because we are
>>>> now assigning the virtio_net device's netdev_ops a NULL pointer.
>>>>
>>>> Fixes: da7b9e9b00d4 ("net: dsa: Add ndo_get_phys_port_name() for CPU port")
>>>> Reported-by: Allen Pais <allen.pais@xxxxxxxxxx>
>>>> Signed-off-by: Florian Fainelli <f.fainelli@xxxxxxxxx>
>>>> ---
>>>
>>> The fix makes complete sense.
>>> But on another note, if we don't overlay an ndo_get_phys_port_name if
>>> the master already has one, doesn't that render the entire mechanism
>>> of having a reliable way for user space to determine the CPU port
>>> number pointless?
>>
>> For the CPU port I would consider ndo_get_phys_port_name() to be more
>> best effort than an absolute need unlike the user facing ports, where
>> this is necessary for a variety of actions (e.g.: determining
>> queues/port numbers etc.) which is why there was no overlay being done
>> in that case. There is not a good way to cascade the information other
>> than do something like pX.Y and defining what the X and Y are, what do
>> you think?
>> --
>> Florian
> 
> For the CPU/master port I am not actually sure who is the final
> consumer of the ndo_get_phys_port_name, I thought it is simply
> informational, with the observation that it may be unreliable in
> transmitting that information over.
> Speaking of which, if "informational" is the only purpose, could this
> not be used?

Yes, I had not considered devlink would expose that information,
ndo_phys_port_name() is there now though and since it is exposed through
sysfs so reverting would be an ABI breakage.

> 
> devlink port | grep "flavour cpu"
> pci/0000:00:00.5/4: type notset flavour cpu port 4
> spi/spi2.0/4: type notset flavour cpu port 4
> spi/spi2.1/4: type notset flavour cpu port 4
> 
> Thanks,
> -Vladimir
> 

-- 
Florian