[PATCH v4] RISC-V: build: Allow LTO to be selected

[PATCH v4] RISC-V: build: Allow LTO to be selected

[Nathan Chancellor]

From: Wende Tan <twd2.me@gmail.com>

Allow LTO to be selected for RISC-V, only when LLD >= 14, since there is
an issue [1] in prior LLD versions that prevents LLD to generate proper
machine code for RISC-V when writing `nop`s.

To avoid boot failures in QEMU [2], '-mattr=+c' and '-mattr=+relax'
need to be passed via '-mllvm' to ld.lld, as there appears to be an
issue with LLVM's target-features and LTO [3], which can result in
incorrect relocations to branch targets [4]. Once this is fixed in LLVM,
it can be made conditional on affected ld.lld versions.

Disable LTO for arch/riscv/kernel/pi, as llvm-objcopy expects an ELF
object file when manipulating the files in that subfolder, rather than
LLVM bitcode.

[1] https://github.com/llvm/llvm-project/issues/50505, resolved by LLVM
    commit e63455d5e0e5 ("[MC] Use local MCSubtargetInfo in writeNops")
[2] https://github.com/ClangBuiltLinux/linux/issues/1942
[3] https://github.com/llvm/llvm-project/issues/59350
[4] https://github.com/llvm/llvm-project/issues/65090

Tested-by: Wende Tan <twd2.me@gmail.com>
Signed-off-by: Wende Tan <twd2.me@gmail.com>
Co-developed-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
---
Testing notes from Nathan:

I tested LLVM 14 through 18 with defconfig + full/thin LTO and
allmodconfig + thin LTO. allmodconfig + thin LTO with LLVM 15 and 16
shows

  ld.lld: error: section size decrease is too large

when linking vmlinux. This appears to be resolved in LLVM 17 with
https://github.com/llvm/llvm-project/commit/9d37ea95df1b84cca9b5e954d8964c976a5e303e
(I did not bisect but the commit message lines up with the issue). I
kept the existing version check because defconfig worked fine but we may
want to bump it to 17.0.0 if randconfigs trip over this.

v4 now boots in QEMU for me with both full and thin LTO.

Testing notes from Wende:

I have tested enabling LTO for `defconfig`. The LLD took ~2m21s and ~3GiB
on our Intel Xeon Gold 6140 server and produced an 18MiB Image. The image
can boot to shell using an archriscv rootfs on QEMU.

I have also tested it for `allyesconfig` without COMPILE_TEST, FTRACE,
KASAN, and GCOV. The LLD took ~7h03m and ~335GiB on the server,
successfully producing a 1.7GiB Image. Unfortunately, we cannot boot this
image because the `create_kernel_page_table()` -> `alloc_pmd_early()` ->
`BUG_ON()` logic limits the image to be < 1GiB. Maybe we can fix it in a
separate patch further.

Changes in v4:
- Pass '-mattr=+c' and '-mattr=+relax' to ld.lld via '-mllvm' flags when
  doing LTO due to LLVM not appearing to honor target-features when
  performing LTO, which can result in incorrect branch targets. This
  fixes boot with CONFIG_LTO_CLANG=y with LLVM 15.0.0+.
- Tidy up commit message and move testing notes to below the fold, as it
  is not too relevant to the git history and it makes the rest of the
  message more of an imperative mood.
- Link to v3: https://lore.kernel.org/r/20231003-riscv-lto-v3-1-8aca61a4ecb4@kernel.org

Changes in v3:
- Disable LTO in arch/riscv/kernel/pi/Makefile, which was added to the
  kernel after the submission of v2. This change matches arm64.
- Link to v2: https://lore.kernel.org/r/20220512205545.992288-1-twd2.me@gmail.com/

Changes in v2:
- Some textual changes suggested by Nick.
- Drop the changes to `arch/riscv/Makefile`, since the LLVM issue is
  filed and resolved.
- Drop the unnecessary changes to `arch/riscv/kernel/vdso/Makefile`.
- Link to v1: https://lore.kernel.org/r/20210719205208.1023221-1-twd2.me@gmail.com/
---
 arch/riscv/Kconfig            | 3 +++
 arch/riscv/Makefile           | 5 +++++
 arch/riscv/kernel/pi/Makefile | 3 +++
 3 files changed, 11 insertions(+)

diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
index d607ab0f7c6d..523640f7441e 100644
--- a/arch/riscv/Kconfig
+++ b/arch/riscv/Kconfig
@@ -46,6 +46,9 @@ config RISCV
 	select ARCH_SUPPORTS_CFI_CLANG
 	select ARCH_SUPPORTS_DEBUG_PAGEALLOC if MMU
 	select ARCH_SUPPORTS_HUGETLBFS if MMU
+	# LLD >= 14: https://github.com/llvm/llvm-project/issues/50505
+	select ARCH_SUPPORTS_LTO_CLANG if LLD_VERSION >= 140000
+	select ARCH_SUPPORTS_LTO_CLANG_THIN if LLD_VERSION >= 140000
 	select ARCH_SUPPORTS_PAGE_TABLE_CHECK if MMU
 	select ARCH_SUPPORTS_PER_VMA_LOCK if MMU
 	select ARCH_USE_MEMTEST
diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile
index 1329e060c548..709400ceac60 100644
--- a/arch/riscv/Makefile
+++ b/arch/riscv/Makefile
@@ -52,6 +52,11 @@ ifndef CONFIG_AS_IS_LLVM
 	KBUILD_CFLAGS += -Wa,-mno-relax
 	KBUILD_AFLAGS += -Wa,-mno-relax
 endif
+# LLVM has an issue with target-features and LTO: https://github.com/llvm/llvm-project/issues/59350
+# Ensure it is aware of linker relaxation with LTO, otherwise relocations may
+# be incorrect: https://github.com/llvm/llvm-project/issues/65090
+else ifeq ($(CONFIG_LTO_CLANG),y)
+	KBUILD_LDFLAGS += -mllvm -mattr=+c -mllvm -mattr=+relax
 endif
 endif
 
diff --git a/arch/riscv/kernel/pi/Makefile b/arch/riscv/kernel/pi/Makefile
index 07915dc9279e..b75f150b923d 100644
--- a/arch/riscv/kernel/pi/Makefile
+++ b/arch/riscv/kernel/pi/Makefile
@@ -9,6 +9,9 @@ KBUILD_CFLAGS	:= $(subst $(CC_FLAGS_FTRACE),,$(KBUILD_CFLAGS)) -fpie \
 		   -fno-asynchronous-unwind-tables -fno-unwind-tables \
 		   $(call cc-option,-fno-addrsig)
 
+# Disable LTO
+KBUILD_CFLAGS	:= $(filter-out $(CC_FLAGS_LTO), $(KBUILD_CFLAGS))
+
 KBUILD_CFLAGS	+= -mcmodel=medany
 
 CFLAGS_cmdline_early.o += -D__NO_FORTIFY

---
base-commit: 8a749fd1a8720d4619c91c8b6e7528c0a355c0aa
change-id: 20231003-riscv-lto-f013beed8587

Best regards,
-- 
Nathan Chancellor <nathan@kernel.org>


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH v4] RISC-V: build: Allow LTO to be selected

[patchwork-bot+linux-riscv]

Hello:

This patch was applied to riscv/linux.git (for-next)
by Palmer Dabbelt <palmer@rivosinc.com>:

On Tue, 17 Oct 2023 15:21:04 -0700 you wrote:
> From: Wende Tan <twd2.me@gmail.com>
> 
> Allow LTO to be selected for RISC-V, only when LLD >= 14, since there is
> an issue [1] in prior LLD versions that prevents LLD to generate proper
> machine code for RISC-V when writing `nop`s.
> 
> To avoid boot failures in QEMU [2], '-mattr=+c' and '-mattr=+relax'
> need to be passed via '-mllvm' to ld.lld, as there appears to be an
> issue with LLVM's target-features and LTO [3], which can result in
> incorrect relocations to branch targets [4]. Once this is fixed in LLVM,
> it can be made conditional on affected ld.lld versions.
> 
> [...]

Here is the summary with links:
  - [v4] RISC-V: build: Allow LTO to be selected
    https://git.kernel.org/riscv/c/021d23428bdb

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html



_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv