On 2018-03-10 14:41, Kalle Valo wrote:
<yamada.masahiro@xxxxxxxxxxxxx> writes:
>> Kbuild bot found an odd problem with this patch:
>>
>> include/linux/dynamic_debug.h:77:14: error: 'KBUILD_MODNAME'
>> undeclared (first use in this function); did you mean
'KBUILD_BASENAME'?
>>
>> Full report:
>>
>>
http://lists.infradead.org/pipermail/ath10k/2018-February/010907.html
>>
>> Any ideas? Is this is some unrelated issue or what? This patch is not
>> even touching hif.h or ce.c.
>
> I didn't encountered this issue as in my defconfig only
> CONFIG_ATH10K_SNOC was defined. This problem is coming when we define
> CONFIG_ATH10K_SNOC and CONFIG_ATH10K_PCI simultaneously in defconfig
> and this is known issue when multiple modules share objects(in this
> case ce.o). I saw similar reported problem and found
> https://patchwork.kernel.org/patch/10060825/.
>
> After picking the below change issue is not seen.
Let's ask the kbuild maintainer. Masahiro, any chances of getting
this
patch applied anytime soon:
kbuild: define KBUILD_MODNAME even if multiple modules share objects
https://patchwork.kernel.org/patch/10060825/
In ath10k we would need it as otherwise we are not able to link ce.o
both to ath10k_pci.ko and ath10k_snoc.ko. What do you think?
Full discussion and the ath10k patch here:
https://patchwork.kernel.org/patch/10220657/
I plan to submit v2, but even if the undefined KBUILD_MODNAME is
fixed,
I expect another problem from this patch.
If both CONFIG_ATH10K_PCI and CONFIG_ATH10_SNOC are 'y'
two instances of ce.o would be linked into vmliux,
then causes multiple definition error.
Oh, I didn't realise this. Thanks for pointing it out. Govind is
looking
at other ways to fix this.
https://patchwork.kernel.org/patch/10298659/ is raised to address this
problem.
Thanks,
Govind
