Re: Android 11 clang toochain: kernel 5.18 build error with 32 bit target

From: Nick Desaulniers <ndesaulniers@google.com>
To: Mauro Rossi <issor.oruam@gmail.com>
Cc: Arnd Bergmann <arnd@arndb.de>,
	Chih-Wei Huang <cwhuang@android-x86.org>,
	 clang-built-linux <llvm@lists.linux.dev>
Subject: Re: Android 11 clang toochain: kernel 5.18 build error with 32 bit target
Date: Mon, 25 Apr 2022 11:04:35 -0700	[thread overview]
Message-ID: <CAKwvOdmfHS8UwZhEL-tLMJSzCXQEZZZhEMpDXuJHtP5s9yePBw@mail.gmail.com> (raw)
In-Reply-To: <CAK8P3a0Lwtm86wYH+0yFtr=Luc6g3OLu_+vzZaJqxDxNHFmQDQ@mail.gmail.com>

On Mon, Apr 25, 2022 at 8:06 AM Arnd Bergmann <arnd@arndb.de> wrote:
>
> On Mon, Apr 25, 2022 at 3:27 PM Mauro Rossi <issor.oruam@gmail.com> wrote:
> >
> > Hello,
> >
> > while performing my periodic testing of android-x86 build with kernel rc cycle,
> > I found the following building error which seems caused by commit
> > 34737e2698 ("uaccess: add generic __{get,put}_kernel_nofault")
> >
> > Please provide and merge a fix while we're still in kernel 5.18 rc cycle
> >
> > I am available if you need testing of the fix
>
>
> Hi Mauro,
>
> > [x86 32bit target build error with shipped prebuilt toolchain clang 11.0.2]
>
> Thanks for the report! I'm fairly sure that others have tested this path
> with clang-11 on x86-32, so I wonder what is different on your machine.
>
> Is this a clang-11 from your distro (which one?) or from Android?
>
> > /home/utente/r-x86_kernel/kernel/mm/maccess.c:35:3: error: invalid
> > output size for constraint '=a'
> >                 copy_from_kernel_nofault_loop(dst, src, size, u64, Efault);
> >                 ^
> > /home/utente/r-x86_kernel/kernel/mm/maccess.c:17:3: note: expanded
> > from macro 'copy_from_kernel_nofault_loop'
> >                 __get_kernel_nofault(dst, src, type, err_label);
> >          \
> >                 ^
> > /home/utente/r-x86_kernel/kernel/arch/x86/include/asm/uaccess.h:509:18:
> > note: expanded from macro '__get_kernel_nofault'
> >         __get_user_size(*((type *)(dst)), (__force type __user *)(src), \
> >                         ^
> > /home/utente/r-x86_kernel/kernel/mm/maccess.c:35:3: error: invalid
> > output size for constraint '=a'
> > /home/utente/r-x86_kernel/kernel/mm/maccess.c:17:3: note: expanded
> > from macro 'copy_from_kernel_nofault_loop'
> >                 __get_kernel_nofault(dst, src, type, err_label);
> >          \
> >                 ^
> > /home/utente/r-x86_kernel/kernel/arch/x86/include/asm/uaccess.h:509:18:
> > note: expanded from macro '__get_kernel_nofault'
> >         __get_user_size(*((type *)(dst)), (__force type __user *)(src), \
> >                         ^
> > 2 errors generated.

(Plus llvm@ , Christoph and Geert to bcc)

Hi Mauro,
Thanks for testing and for the report. This is vaguely reminiscent of
some i386 assembler issue we fixed back in the kernel in the 5.9
release.
https://github.com/ClangBuiltLinux/linux/issues/194
So definitely surprising to see something similar in mainline; perhaps
config related, or something should be a feature check rather than a
version check.

To start with, do you mind sharing your .config file? If I can repro
with ToT LLVM, then this may be an issue to fix given such a .config.

Otherwise, I'll need to see what's different between upsteam llvm's
release/11.x branch, and that release of AOSP clang.

Orthogonally, I think the latest release of AOSP clang is 14.0.5. Any
reason you're still using 11.0.2?  I expect to drop support for 11
next.

>
> arch/x86/include/asm/uaccess.h:509 indicates that
> CONFIG_CC_HAS_ASM_GOTO_OUTPUT is not set, but I expected
> this to be available with clang-11 or higher. Can you confirm that this is
> disabled in your .config?
>
> Adding Nick to Cc, as he worked on that feature.
>
>        Arnd

-- 
Thanks,
~Nick Desaulniers