Re: [wireless-next:master 169/199] drivers/net/wireless/brcm80211/brcmfmac/dhd_sdio.c:2469:3: error: implicit declaration of function 'atomic_set_mask'

[Arend van Spriel <arend@broadcom.com>]

On 02/28/14 23:44, Florian Fainelli wrote:
> 2014-02-28 14:33 GMT-08:00 Arend van Spriel<arend@broadcom.com>:
>> On 02/28/2014 11:26 PM, Florian Fainelli wrote:
>>> 2014-02-28 14:22 GMT-08:00 Arend van Spriel<arend@broadcom.com>:
>>>> On 02/28/2014 11:15 PM, kbuild test robot wrote:
>>>>> tree:   git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-next.git master
>>>>> head:   3e3831c4fdc53aabf3a56419ef6d96a841c52435
>>>>> commit: c98db0bec72ac7ef127119c1ed962d6f56802b12 [169/199] brcmfmac: Use atomic functions for intstatus update.
>>>>> config: make ARCH=ia64 allmodconfig
>>>>>
>>>>> All error/warnings:
>>>>>
>>>>>     drivers/net/wireless/brcm80211/brcmfmac/dhd_sdio.c: In function 'brcmf_sdio_intr_rstatus':
>>>>>>> drivers/net/wireless/brcm80211/brcmfmac/dhd_sdio.c:2469:3: error: implicit declaration of function 'atomic_set_mask' [-Werror=implicit-function-declaration]
>>>>>     cc1: some warnings being treated as errors
>>>>
>>>> Hi Fengguang,
>>>>
>>>> So atomic_set_mask() is not a generally available function. I thought
>>>> that if the arch does not provide it, the asm-generic implementation is
>>>> used.
>>>
>>> Though so as well.
>>>
>>>> Also it seems the function prototype is not the same for all
>>>> archs. :-(
>>>
>>> This does not sound like it is expected, and this might need fixing,
>>> do you want to tackle this?
>>
>> I got email m68k and ia64 failing. Some implementations are in assembly.
>> It has been (counting...) over 20 years since I did some baby steps in
>> m68k assembly and do not really feel a strong pull to that dark side.
>
> This is supposed to be a generic function that everyone should be able
> to use without knowing the underlying details. Not quite sure why
> those two architectures are different, maybe we should ask Geert and
> Tony about this?

I was about to when I decided to do some more digging. So taking a 
closer look at include/asm-generic/atomic.h revealed this:

/*
  * Generic C implementation of atomic counter operations. Usable on
  * UP systems only. Do not include in machine independent code.

And further down:

#ifdef CONFIG_SMP
/* Force people to define core atomics */
# if !defined(atomic_add_return) || !defined(atomic_sub_return) || \
      !defined(atomic_clear_mask) || !defined(atomic_set_mask)
#  error "SMP requires a little arch-specific magic"
# endif
#endif

So in these SMP era architecture specific implementations seem required. 
IA64 does not provide it (so CONFIG_SMP not supported?) and m68k has 
different prototype. Apparently, the comment /* Force people to define 
core atomics */ did not fly for all archs (yet) :-p.

Regards,
Arend