Re: Regression: cannot compile kernel 4.14 with old gcc

[Sasha Levin <sashal@kernel.org>]

On Fri, Jun 28, 2019 at 09:42:21AM -0700, Nick Desaulniers wrote:
>On Fri, Jun 28, 2019 at 8:53 AM Tony Battersby <tonyb@cybernetics.com> wrote:
>>
>> Old versions of gcc cannot compile 4.14 since 4.14.113:
>>
>> ./include/asm-generic/fixmap.h:37: error: implicit declaration of function ‘__builtin_unreachable’
>>
>> The stable commit that caused the problem is 82017e26e515 ("compiler.h:
>> update definition of unreachable()") (upstream commit fe0640eb30b7).
>> Reverting the commit fixes the problem.
>>
>> Kernel 4.17 dropped support for older versions of gcc in upstream commit
>> cafa0010cd51 ("Raise the minimum required gcc version to 4.6").  This
>> was not backported to 4.14 since that would go against the stable kernel
>> rules.
>>
>> Upstream commit 815f0ddb346c ("include/linux/compiler*.h: make
>> compiler-*.h mutually exclusive") was a fix for cafa0010cd51.  This was
>> not backported to 4.14.
>>
>> Upstream commit fe0640eb30b7 ("compiler.h: update definition of
>> unreachable()") was a fix for 815f0ddb346c.  This is the commit that was
>> backported to 4.14.  But it only fixed a problem introduced in the other
>> commits, and without those commits, it ends up introducing a problem
>> instead of fixing one.  So I recommend reverting that patch in 4.14,
>> which will enable old gcc to compile 4.14 again.  If I understand
>> correctly, I believe that clang will still be able to compile 4.14 with
>> the patch reverted, although I haven't tried to compile with clang.
>>
>> The problematic commit is not present in 4.9.x, 4.4.x, 3.18.x, or 3.16.x.
>
>$ make CC=clang -j71 arch/x86/mm/fault.o
>produces no objtool warning with upstream commit fe0640eb30b7 reverted.
>
>x86 defconfig w/ Clang also have no issue with that reverted on 4.14.y.
>
>Revert away.
>
>Greg, Sasha,
>Do you need a patch file for that, or can you please push a
>$ cd linux-stable
>$ git checkout 4.14.y
>$ git revert 82017e26e51596ee577171a33f357377ec6513b5

You got it.

>-- 
>Thanks,
>~Nick Desaulniers