Re: [PATCH v3 10/11] RFC: usb-storage: export symbols in USB_STORAGE namespace

[Matthias Maennich <maennich@google.com>]

On Wed, Aug 21, 2019 at 04:13:29PM -0700, Christoph Hellwig wrote:
>On Wed, Aug 21, 2019 at 12:49:25PM +0100, Matthias Maennich wrote:
>> Modules using these symbols are required to explicitly import the
>> namespace. This patch was generated with the following steps and serves
>> as a reference to use the symbol namespace feature:
>>
>>  1) Define DEFAULT_SYMBOL_NAMESPACE in the corresponding Makefile
>>  2) make  (see warnings during modpost about missing imports)
>>  3) make nsdeps
>>
>> Instead of a DEFAULT_SYMBOL_NAMESPACE definition, the EXPORT_SYMBOL_NS
>> variants can be used to explicitly specify the namespace. The advantage
>> of the method used here is that newly added symbols are automatically
>> exported and existing ones are exported without touching their
>> respective EXPORT_SYMBOL macro expansion.
>
>So what is USB_STORAGE here?  It isn't a C string, so where does it
>come from?  To me using a C string would seem like the nicer interface
>vs a random cpp symbol that gets injected somewhere.

To be honest, I would also prefer an interface that uses C strings or
literals for the new EXPORT_SYMBOLS* macros:

  EXPORT_SYMBOL_NS(mysym, "USB_STORAGE");

  or

  const char USB_STORAGE_NS[] = "USB_STORAGE";
  EXPORT_SYMBOL_NS(mysym, USB_STORAGE_NS);

The DEFAULT_SYMBOL_NAMESPACE define within Makefiles would get a bit
more verbose in that case to express the literal:
  ccflags-y += -DDEFAULT_SYMBOL_NAMESPACE="\"USB_STORAGE\""


The main reason against that, is, that in the expansion of
EXPORT_SYMBOL_NS, we define the ksymtab entry, which name is
constructed partly by the name of the namespace:

   static const struct kernel_symbol __ksymtab_##sym##__##ns  ...
                                                        ^^^^

For that we depend on a cpp symbol to construct the name. I am not sure
there is a reasonable way of getting rid of that without ending up
constructing the ksymtab entries completely in asm as it is already done
in case of PREL32_RELOCATIONS. But I am happy to be corrected.

For reference that is done in patch 03/11 of this series:
https://lore.kernel.org/lkml/20190821114955.12788-4-maennich@google.com/

Cheers,
Matthias