Date:   Wed, 28 Jul 2021 11:59:59 -0700
From:   Nick Desaulniers <ndesaulniers@...gle.com>
To:     Masahiro Yamada <masahiroy@...nel.org>
Cc:     Nathan Chancellor <nathan@...nel.org>,
        Miguel Ojeda <ojeda@...nel.org>,
        Fangrui Song <maskray@...gle.com>,
        Michal Marek <michal.lkml@...kovi.net>,
        Arnd Bergmann <arnd@...nel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Linux Kbuild mailing list <linux-kbuild@...r.kernel.org>,
        clang-built-linux <clang-built-linux@...glegroups.com>,
        Geert Uytterhoeven <geert@...ux-m68k.org>,
        Christoph Hellwig <hch@...radead.org>,
        Linus Torvalds <torvalds@...ux-foundation.org>
Subject: Re: [PATCH v2 2/2] Makefile: infer CROSS_COMPILE from SRCARCH for
 LLVM=1 LLVM_IAS=1

On Tue, Jul 20, 2021 at 8:50 PM Masahiro Yamada <masahiroy@...nel.org> wrote:
>
> On Wed, Jul 21, 2021 at 2:30 AM Nathan Chancellor <nathan@...nel.org> wrote:
> >
> > On 7/20/2021 1:04 AM, Masahiro Yamada wrote:
> > > On Fri, Jul 9, 2021 at 8:25 AM 'Nick Desaulniers' via Clang Built
> > > Linux <clang-built-linux@...glegroups.com> wrote:
> > >>
> > >> diff --git a/scripts/Makefile.clang b/scripts/Makefile.clang
> > >> index 297932e973d4..956603f56724 100644
> > >> --- a/scripts/Makefile.clang
> > >> +++ b/scripts/Makefile.clang
> > >> @@ -1,6 +1,36 @@
> > >> -ifneq ($(CROSS_COMPILE),)
> > >> +# Individual arch/{arch}/Makfiles should use -EL/-EB to set intended endianness
> > >> +# and -m32/-m64 to set word size based on Kconfigs instead of relying on the
> > >> +# target triple.
> > >> +ifeq ($(CROSS_COMPILE),)
> > >> +ifneq ($(LLVM),)
> > >
> > >
> > > Do you need to check $(LLVM) ?
> > >
> > >
> > > LLVM=1 is a convenient switch to change all the
> > > defaults, but yet you can flip each tool individually.
> > >
> > > Instead of LLVM=1, you still should be able to
> > > get the equivalent setups by:
> > >
> > >
> > >    make CC=clang LD=ld.lld AR=llvm-ar OBJCOPY=llvm-objcopy ...
> > >
> > >
> > > The --target option is passed to only
> > > KBUILD_CFLAGS and KBUILD_AFLAGS.
> > >
> > > So, when we talk about --target=,
> > > we only care about whether $(CC) is Clang.
> > > Not caring about $(AR), $(LD), or $(OBJCOPY).
> > >
> > >
> > > scripts/Makefile.clang is already guarded by:
> > >
> > > ifneq ($(findstring clang,$(CC_VERSION_TEXT)),
> >
> > $ make ARCH=arm64 CC=clang LLVM_IAS=1
> >
> > will use the right compiler and assembler but none of the other binary
> > tools because '--prefix=' will not be set so CROSS_COMPILE needs to be
> > specified still, which defeats the purpose of this whole change. This
> > patch is designed to work for the "normal" case of saying "I want to use
> > all of the LLVM tools", not "I want to use clang by itself".
>
>
> I disagree.
>
> LLVM=1 is a shorthand.
>
>
>
> make LLVM=1 LLVM_IAS=1
>
>   should be equivalent to:
>
> make CC=clang LD=ld.lld AR=llvm-ar NM=llvm-nm STRIP=llvm-strip \
>   OBJCOPY=llvm-objcopy OBJDUMP=llvm-objdump READELF=llvm-readelf \
>   HOSTCC=clang HOSTCXX=clang++ HOSTAR=llvm-ar HOSTLD=ld.lld \
>   LLVM_IAS=1
>
>
>
> We do not care about the origin of CC=clang,
> whether it came from LLVM=1 or every tool was explicitly,
> individually specified.
>
>
>
> ifneq ($(LLVM),) is a garbage code
> that checks a pointless thing.

Masahiro,
Nathan is correct.  Test for yourself; if you apply these two patches,
then apply:

diff --git a/scripts/Makefile.clang b/scripts/Makefile.clang
index 956603f56724..a1b46811bdc6 100644
--- a/scripts/Makefile.clang
+++ b/scripts/Makefile.clang
@@ -2,7 +2,6 @@
 # and -m32/-m64 to set word size based on Kconfigs instead of relying on the
 # target triple.
 ifeq ($(CROSS_COMPILE),)
-ifneq ($(LLVM),)
 ifeq ($(LLVM_IAS),1)
 ifeq ($(SRCARCH),arm)
 CLANG_FLAGS    += --target=arm-linux-gnueabi
@@ -26,7 +25,6 @@ else
 $(error Specify CROSS_COMPILE or add '--target=' option to
scripts/Makefile.clang)
 endif # SRCARCH
 endif # LLVM_IAS
-endif # LLVM
 else
 CLANG_FLAGS    += --target=$(notdir $(CROSS_COMPILE:%-=%))
 endif # CROSS_COMPILE

Then build as Nathan specified:
$ ARCH=arm64 make CC=clang LLVM_IAS=1 -j72 defconfig all
...
arch/arm64/Makefile:25: ld does not support --fix-cortex-a53-843419;
kernel may be susceptible to erratum
...
  LD      arch/arm64/kernel/vdso/vdso.so.dbg
ld: unrecognised emulation mode: aarch64linux
Supported emulations: elf_x86_64 elf32_x86_64 elf_i386 elf_iamcu
elf_l1om elf_k1om i386pep i386pe
make[1]: *** [arch/arm64/kernel/vdso/Makefile:56:
arch/arm64/kernel/vdso/vdso.so.dbg] Error 1
make: *** [arch/arm64/Makefile:193: vdso_prepare] Error 2

Nathan referred to --prefix, but in this failure, because
CROSS_COMPILE was never set, the top level Makefile set LD to:
 452 LD    = $(CROSS_COMPILE)ld
in this case `ld` in my path was my host x86 linker, which is not
correct for a cross compilation of arm64 target.

Perhaps we can somehow support "implicit CROSS_COMPILE" with just
CC=clang, and not LLVM=1, but I think it would be inflexible to
hardcode such target triple prefixes.  What if someone has
arm-linux-gnueabi-as but not arm-linux-gnueabihf-as installed?  That's
the point of CROSS_COMPILE in my opinion to provide such flexibility
at the cost of additional command line verbosity.

For the common case of LLVM=1 though, this series is a simplification.
If users want to specify CC=clang, then they MUST use CROSS_COMPILE
when cross compiling.

Please review the current approach and see if there's more I can
improve in a v3; otherwise I still think this series is good to go.
-- 
Thanks,
~Nick Desaulniers

Powered by blists - more mailing lists