Re: [PATCH v5 7/7] fbdev: Make registered_fb[] private to fbmem.c

[Javier Martinez Canillas <javierm@redhat.com>]

Hello Guenter,

On 5/11/22 19:17, Guenter Roeck wrote:
> On 5/11/22 10:00, Sam Ravnborg wrote:

[snip]

>>>   struct fb_info *registered_fb[FB_MAX] __read_mostly;
>>> -EXPORT_SYMBOL(registered_fb);
>>> -
>>>   int num_registered_fb __read_mostly;
>>> +#if IS_ENABLED(CONFIG_FB_OLPC_DCON)
>>> +EXPORT_SYMBOL(registered_fb);
>>>   EXPORT_SYMBOL(num_registered_fb);
>>> +#endif
>>
>> It is stuff like this I refer to as "ugly" in the comment above.
>>
> 
> My "solution" for that kind of thing is to use a namespace,
> such as
> 
> EXPORT_SYMBOL_NS(registered_fb, FB_OLPC_DCON);
> EXPORT_SYMBOL_NS(num_registered_fb, FB_OLPC_DCON);
>

Using a namespace in this case is indeed a great idea I think.

I've used in the past to limit the export of a symbol for within a driver
that could be scattered across different compilations units, but it never
occurred to me using it to limit symbols exported by core code.
 
> and import it from the offending code. That avoids ifdefs
> while at the same time limiting the use of the symbols
> to the expected scope. Of course that could be abused but
> that abuse would be obvious.
>

Agreed. For the next revision, besides using an namespaced export symbol
as you suggested, I'll include a comment to make clear that it shouldn't
by any other driver and FB_OLPC_DCON fixed instead.


-- 
Best regards,

Javier Martinez Canillas
Linux Engineering
Red Hat