Re: [PATCH] kbuild: Set KBUILD_CFLAGS before incl. arch Makefile

[Masahiro Yamada]

2017-11-08 2:37 GMT+09:00 Nick Desaulniers <ndesaulniers@xxxxxxxxxx>:
> On Sat, Nov 4, 2017 at 8:06 PM, Masahiro Yamada
> <yamada.masahiro@xxxxxxxxxxxxx> wrote:
>> ld-option is only used for arch/{arm64,powerpc}/Makefile
>>
>> arch/arm64/Makefile:  ifeq ($(call ld-option, --fix-cortex-a53-843419),)
>> arch/powerpc/Makefile:LDFLAGS_vmlinux += $(call
>> ld-option,--orphan-handling=warn)
>>
>> I think this patch makes sense when it comes along with
>> https://patchwork.kernel.org/patch/10030581/
>
> Good point.
>
>> but, it is now being blocked by 0-day bot
>> due to a x86 problem.
>
> Looks like that is now resolved (unless 0-day bot strikes again).
>
>> The location of CLANG_GCC_TC define
>> only matters after your patch is applied, right?
>
> By "your patch" referring to the 0-day bot thread, yes.
>
>> Did my request for v2 break anything?
>
> Nothing immediately obvious, and no regressions.  It just made this
> patch necessary (along with my previous one) for correctly cross
> compiling with clang for arm64 and powerpc as you point out.
>
>> One more thing: this patch does not apply to kbuild tree.
>
> I absolutely will rebase it on your tree and send a v2.  Just to help
> me understand the contribution model better: none of my other patches
> have yet been requested against any trees other than Linus'.  Is this
> because of where we are in the release cycle, or that a lot of kbuild
> code has changed, or what?


Generally speaking,
a preferred way is to base patches on the subsystem tree.

Kernel developers are supposed to do their development on linux-next,
but, in reality, many people work on Linus' tree since it is more stable and
git history is fast-forward.

In many cases, patches based on Linus' tree can apply to sub-systems as well.

I am happy to fix-up a conflict locally
as long as it is trivial, and there is no other reason for re-spin.

Unfortunately, Kbuild tree changed the top-level Makefile a lot in
this development cycle.

If your patch does not apply cleanly, I do not know which context you
are moving the code to.
Also, I found suspicious description in the commit log.

That's why.


-- 
Best Regards
Masahiro Yamada