[PATCHv2 1/3] net: phy: prevent linking breakage

[Alexandre Belloni]

On 04/06/2013 18:09, Florian Fainelli wrote:
> 2013/6/4 Arnd Bergmann <arnd at arndb.de>:
>> On Tuesday 04 June 2013 16:36:50 Florian Fainelli wrote:
>>> It seems to me that what David proposes is to have say an
>>> arch/arm/mach-foo/phy-fixups.c file which is only enabled when
>>> CONFIG_PHYLIB is set (obj-$(CONFIG_PHYLIB) += phy-fixup.o), such that
>>> it does not need to have any conditionnals when calling
>>> phy_register_fixup. This sounds a little unusual, but why not.
>> I don't think it would actually help us, because then we still need
>> to declare a local function that gets called from the board init
>> code. Instead of doing
>>
>>         if (IS_ENABLED(CONFIG_PHYLIB))
>>                  phy_register_fixup_for_uid(phy_id, foo_phy_fixup);
>>
>> we would then do
>>
>>         if (IS_ENABLED(CONFIG_PHYLIB))
>>                 foo_phy_fixup_register();
>>
>> which is not much different at all.
> You would just need to define a stub for your arch_foo_phy_fixup()
> which has a different definition depending on whether CONFIG_PHYLIB is
> defined or not.
>
> This would be just one function, instead of the whole bunch of stubs
> needed for phylib. Right now its probably 1 vs 3, so it does not make
> that much of a difference but who knows, if we had more phylib stubs
> and forget to update the stubs? (which tends to happen pretty often).

The fact is that for now it is called on 6 different platforms. Of those:
 - two (imx and mxs) are now doing it right *after* discovering the issue
 - two always selecting PHYLIB and can't be compiled without NET
 - two will break at compile time.

I'm really concerned about getting more platforms getting it wrong. I
think it boils down to who wants to take the maintenance burden. I would
believe that doing it in phylib is more future proof. But I'm fine
with whatever the consensus will be.

However, it is becoming urgent for Atmel to get the fix in 3.10 so that
they never get a kernel compilation that breaks.

Regards,

-- 
Alexandre Belloni, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com