Hi,
On Wed, Feb 07, 2024 at 05:14:52PM -0700, Nathan Chancellor wrote:
> Hi all,
>
> This series allows the s390 kernel to be linked with ld.lld (support for
> s390 is under review at [1]). This implicitly depends on [2], which was
> created and sent before it was realized that this series was necessary.
>
> The first chunk of this series enables support for
> CONFIG_LD_ORPHAN_WARN, as it was discovered during testing that the
> kernel fails to build with ld.lld due to differences in orphan section
> handling, which would have been caught with the linker's orphan section
> warnings ahead of the actual build error. There are no warnings when
> building ARCH=s390 defconfig and allmodconfig with GCC 6 through 13 or
> tip of tree Clang using ld.bfd or ld.lld
>
> The final patch resolves a series of errors due to ld.lld having a
> different default for checking for DT_TEXTREL ('-z text') vs ld.bfd,
> which defaults to '-z notext' (but this is configurable at build time).
>
> There is one outstanding issue due to something that ld.lld does not
> support that the kernel relies on:
>
> ld.lld: error: drivers/nvme/host/fc.o:(__bug_table): writable SHF_MERGE section is not supported
>
> This was changed in the kernel in commit e21f8baf8d9a ("s390/bug: add
> entry size to the __bug_table section"). Is this change truly necessary?
> I selectively applied a revert on top of current mainline and I did not
> observe any issues with either Clang or GCC.
>
> diff --git a/arch/s390/include/asm/bug.h b/arch/s390/include/asm/bug.h
> index aebe1e22c7be..c500d45fb465 100644
> --- a/arch/s390/include/asm/bug.h
> +++ b/arch/s390/include/asm/bug.h
> @@ -14,7 +14,7 @@
> ".section .rodata.str,\"aMS\",@progbits,1\n" \
> "1: .asciz \""__FILE__"\"\n" \
> ".previous\n" \
> - ".section __bug_table,\"awM\",@progbits,%2\n" \
> + ".section __bug_table,\"aw\"\n" \
> "2: .long 0b-.\n" \
> " .long 1b-.\n" \
> " .short %0,%1\n" \
> @@ -30,7 +30,7 @@
> #define __EMIT_BUG(x) do { \
> asm_inline volatile( \
> "0: mc 0,0\n" \
> - ".section __bug_table,\"awM\",@progbits,%1\n" \
> + ".section __bug_table,\"aw\"\n" \
> "1: .long 0b-.\n" \
> " .short %0\n" \
> " .org 1b+%1\n" \
>
> If it is necessary, is there any way to work around this error? For
> testing purposes, disabling CONFIG_BUG is easy enough but that is not
> usable in the real world.
>
> To test this series with ld.lld, you'll need to build ld.lld from the
> pull request, which is easy to do following LLVM's instructions [3].
> Here is a TL;DR version I tested that just builds LLD with minimal noise
> during the build.
>
> $ git clone https://github.com/llvm/llvm-project
> $ cd llvm-project
> $ git fetch https://github.com/llvm/llvm-project pull/75643/head
> $ git switch -d FETCH_HEAD
> $ cmake \
> -B build \
> -G Ninja \
> -S llvm \
> --log-level=NOTICE \
> -Wno-dev \
> -DCMAKE_BUILD_TYPE=Release \
> -DLLVM_ENABLE_PROJECTS=lld \
> -DLLVM_ENABLE_WARNINGS=OFF \
> -DLLVM_TARGETS_TO_BUILD=SystemZ
> $ ninja -C build lld
> $ export PATH=$PWD/build/bin:$PATH
>
> Then build the kernel with 'LD=ld.lld' in addition to whatever command
> line you use (I tested both Clang and GCC). I can boot an ld.lld linked
> kernel built with both compilers in QEMU with this series.
Yeah, this all works for me and I am able to boot. I did need to use the
diff present in 0/11 to remove the warnings regarding SHF_MERGE
sections. It should probably be a patch in this series instead of a
inlined diff?
>
> [ 1.386970] Linux version 6.8.0-rc3-00043-g05761ede85d6-dirty (nathan@xxxxxxxxxxxxxxx) (s390-linux-gcc (GCC) 13.2.0, ClangBuiltLinux LLD 19.0.0) #1 SMP Wed Feb 7 16:51:12 MST 2024
>
> [ 0.871923] Linux version 6.8.0-rc3-00043-g05761ede85d6-dirty (nathan@xxxxxxxxxxxxxxx) (ClangBuiltLinux clang version 19.0.0git (https://github.com/llvm/llvm-project 417075e56aeba5a5b20301c7bfeba9c2a800982b), ClangBuiltLinux LLD 19.0.0) #1 SMP Wed Feb 7 17:01:22 MST 2024
>
> [1]: https://github.com/llvm/llvm-project/pull/75643
> [2]: https://lore.kernel.org/r/20240130-s390-vdso-drop-fpic-from-ldflags-v1-1-094ad104fc55@xxxxxxxxxx/
^^^^^^^^^
I needed this too, as I was getting a warnings about -fPIC being an
unknown option.
All in all, works great for me building on clang and booting with qemu.
Tested-by: Justin Stitt <justinstitt@xxxxxxxxxx>
> [3]: https://llvm.org/docs/CMake.html
>
> ---
> Nathan Chancellor (11):
> s390: boot: Add support for CONFIG_LD_ORPHAN_WARN
> s390: vmlinux.lds.S: Handle '.data.rel' sections explicitly
> s390: vmlinux.lds.S: Explicitly handle '.got' and '.plt' sections
> s390: vmlinux.lds.S: Discard unnecessary sections
> s390/boot: vmlinux.lds.S: Handle '.init.text'
> s390/boot: vmlinux.lds.S: Handle '.rela' sections
> s390/boot: vmlinux.lds.S: Handle DWARF debug sections
> s390/boot: vmlinux.lds.S: Handle ELF required sections
> s390/boot: vmlinux.lds.S: Handle commonly discarded sections
> s390: Select CONFIG_ARCH_WANT_LD_ORPHAN_WARN
> s390: Link vmlinux with '-z notext'
>
> arch/s390/Kconfig | 1 +
> arch/s390/Makefile | 2 +-
> arch/s390/boot/Makefile | 5 +++--
> arch/s390/boot/vmlinux.lds.S | 28 ++++++++++++++++++++++++++++
> arch/s390/kernel/vmlinux.lds.S | 28 +++++++++++++++++++++++++++-
> 5 files changed, 60 insertions(+), 4 deletions(-)
> ---
> base-commit: 6613476e225e090cc9aad49be7fa504e290dd33d
> change-id: 20240207-s390-lld-and-orphan-warn-d0ff4ff657b0
>
> Best regards,
> --
> Nathan Chancellor <nathan@xxxxxxxxxx>
>
Thanks
Justin
