Re: [PATCH 3/4] mfd: intel_quark_i2c_gpio: Utilize i2c-designware.h

[Florian Fainelli <florian.fainelli@broadcom.com>]

[-- Attachment #1: Type: text/plain, Size: 2277 bytes --]

On 4/24/24 05:37, Andy Shevchenko wrote:
> On Wed, Apr 24, 2024 at 4:28 AM Florian Fainelli
> <florian.fainelli@broadcom.com> wrote:
>> On 4/23/2024 5:01 PM, Andy Shevchenko wrote:
>>> Tue, Apr 23, 2024 at 04:36:21PM -0700, Florian Fainelli kirjoitti:
>>>> Rather than open code the i2c_designware string, utilize the newly
>>>> defined constant in i2c-designware.h.
> 
> ...
> 
>>>> -#define INTEL_QUARK_I2C_CONTROLLER_CLK "i2c_designware.0"
>>>> +#define INTEL_QUARK_I2C_CONTROLLER_CLK I2C_DESIGNWARE_NAME ".0"
>>>
>>> So, if you build a module separately for older version of the kernel (assuming
>>> it allows you to modprobe), this won't work anymore.
>>
>> Sorry not following, was that comment supposed to be for patch #1 where
>> I changed the i2c-designware-pci to i2c_designware-pci? modprobe
>> recognizes both - and _ as interchangeable BTW.
> 
> I'm talking about something different. Let's assume you have a running
> kernel (w.o. signature or version requirement for the modules), then
> you have a new patch on top of it and then for an unknown reason you
> changed. e.g., designware to DW in that definition. The newly built
> module may not be loaded on the running kernel. Also note, here is the
> instance name and not an ID in use. The replacement is wrong
> semantically.
> 

See my response in the cover letter, the instance base name is not 
independent from the i2c-designware-platdrv::driver::name because 
otherwise the clock lookup done by devm_clk_get_optional() will fail. So 
that change in this patch is entirely intentional and actually ensures 
correctness if someone were to change the i2c_designware platform driver 
name in the future for whatever reasons.

As far as catering to the specific example you gave, is not this just 
fraught with peril regardless of what is being changed in the kernel? 
Any constant that is serves as a contract between independent parts 
getting out of sync will result in some misbehavior. The only solution 
that I can think of which is edging towards over engineering is to 
export a string symbol which contains I2C_DESIGNWARE_NAME and then make 
other modules dependent upon that symbol to enforce some sort of runtime 
resolution, though I think your example could still be made to fail.
-- 
Florian


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4221 bytes --]