A recent change to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
Also, the LLVM commit notes that these relocation types aren't supported until binutils 2.26. Since we support binutils 2.23+, avoid the relocations regardless of linker.
The proper solution is to build the compressed boot image as position independent. There's a series working its way through code review currently that does that, but it's unlikely to be backported to stable, due to its size. For now, cut a smaller patch that's more likely to be easily picked up into stable, so that we can get our kernels booting again.
Cc: stable@vger.kernel.org # 4.14.y Link: https://github.com/ClangBuiltLinux/linux/issues/1121 Link: https://reviews.llvm.org/rGc41a18cf61790fc898dcda1055c3efbf442c14c0 Link: https://travis-ci.com/github/ClangBuiltLinux/continuous-integration/builds/1... Signed-off-by: Nick Desaulniers ndesaulniers@google.com --- https://lore.kernel.org/lkml/20200731230820.1742553-7-keescook@chromium.org/ is the patch I'm hopeful for building the compressed image as -pie, but I don't think the series will be backported. Regardless, we probably want this for older binutils support.
arch/x86/boot/compressed/Makefile | 7 +++++++ 1 file changed, 7 insertions(+)
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..ab0f7e7dabf9 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -44,6 +44,13 @@ KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS
+# Until we can build arch/x86/boot/compressed/vmlinux as -Wl,-pie, don't emit +# R_X86_64_GOTPCRELX or R_X86_64_REX_GOTPCRELX relocations that LLD will relax +# into absolute addressed operands, and that BFD didn't support until 2.26. +ifdef CONFIG_CC_IS_CLANG +KBUILD_CFLAGS += -Wa,-mrelax-relocations=no +endif + KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n UBSAN_SANITIZE :=n
On Fri, Aug 07, 2020 at 12:41:00PM -0700, Nick Desaulniers wrote:
A recent change to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
It could also cause kernels compiled with gcc and linked with LLD to fail in the same way, no? The gcc/gas combination will generate the relaxed relocations from I think gas-2.26 onward. Although the only troublesome symbol in the case of gcc/gas is trampoline_32bit_src, referenced from pgtable_64.c (gcc doesn't use a GOTPC reloc for _pgtable etc).
I'm a bit surprised you were able to boot with just _pgtable fixed (looking at the CBL issue), there are quite a few more GOTPC relocs with clang -- maybe LLD isn't doing all the optimizations it could yet.
This potential issue was mentioned [0] in one of the earlier threads (see last paragraph).
[0] https://lore.kernel.org/lkml/20200526191411.GA2380966@rani.riverdale.lan/
Also, the LLVM commit notes that these relocation types aren't supported until binutils 2.26. Since we support binutils 2.23+, avoid the relocations regardless of linker.
Note that the GNU assembler won't support the option to disable the relaxations until 2.26, when they were added.
However, it turns out that clang always uses the integrated assembler for the decompressor (and the EFI stub) because the no-integrated-as option gets dropped when building these pieces, due to redefinition of KBUILD_CFLAGS. You might want to mention this in the commit log or a comment to explain why using the option unconditionally is safe. It might need to be made conditional if the CFLAGS ever gets fixed to maintain no-integrated-as.
Thanks.
On Fri, Aug 7, 2020 at 2:29 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Fri, Aug 07, 2020 at 12:41:00PM -0700, Nick Desaulniers wrote:
A recent change to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
It could also cause kernels compiled with gcc and linked with LLD to fail in the same way, no? The gcc/gas combination will generate the relaxed relocations from I think gas-2.26 onward. Although the only troublesome symbol in the case of gcc/gas is trampoline_32bit_src, referenced from pgtable_64.c (gcc doesn't use a GOTPC reloc for _pgtable etc).
Thanks for taking a look, and the feedback. I appreciate it!
$ gcc --version | head -n 1 gcc (Debian 9.3.0-11) 9.3.0 $ make -j71 clean defconfig bzImage $ llvm-readelf -r arch/x86/boot/compressed/*.o | grep -e R_X86_64_GOTPCRELX -e R_X86_64_REX_GOTPCRELX 0000000000000114 000000120000002a R_X86_64_REX_GOTPCRELX 0000000000000000 trampoline_32bit_src - 4 $ llvm-readelf -r arch/x86/boot/compressed/vmlinux | grep -e R_X86_64_GOTPCRELX -e R_X86_64_REX_GOTPCRELX $
So it looks like yes. I guess then we'd need to add a check for CONFIG_LD_IS_LLD and CONFIG_CC_IS_GCC and binutils version is 2.26+? I don't mind adding support for that combination, but I'd like to skip it in this patch for the sake of backporting something small to stable to get our CI green ASAP, since CONFIG_LD_IS_LLD probably doesn't exist for those stable branches, which will complicate the backport of such a patch. So I'd do it in a follow up patch if we're cool with that?
I'm a bit surprised you were able to boot with just _pgtable fixed (looking at the CBL issue), there are quite a few more GOTPC relocs with clang -- maybe LLD isn't doing all the optimizations it could yet.
I am, too. I didn't specify which symbol was problematic or put this flag on just one object file, because it's likely that there's an issue with multiple symbols in multiple object files, though it's just _pgtable that causes observable boot failures.
This potential issue was mentioned [0] in one of the earlier threads (see last paragraph).
[0] https://lore.kernel.org/lkml/20200526191411.GA2380966@rani.riverdale.lan/
Oh, indeed.
Also, the LLVM commit notes that these relocation types aren't supported until binutils 2.26. Since we support binutils 2.23+, avoid the relocations regardless of linker.
Note that the GNU assembler won't support the option to disable the relaxations until 2.26, when they were added.
However, it turns out that clang always uses the integrated assembler for the decompressor (and the EFI stub) because the no-integrated-as option gets dropped when building these pieces, due to redefinition of KBUILD_CFLAGS. You might want to mention this in the commit log or a
That's why I was careful to note in the commit message that it was Clang's integrated assembler (assembler job) vs Clang (compiler job) itself that was producing these. May I add precisely:
``` Note that the GNU assembler won't support the option to disable the relaxations until 2.26, when they were added.
However, it turns out that clang always uses the integrated assembler for the decompressor (and the EFI stub) because the no-integrated-as option gets dropped when building these pieces, due to redefinition of KBUILD_CFLAGS. ``` with your suggested-by tag for a v2?
comment to explain why using the option unconditionally is safe. It might need to be made conditional if the CFLAGS ever gets fixed to maintain no-integrated-as.
Thanks.
On Fri, Aug 07, 2020 at 02:54:39PM -0700, Nick Desaulniers wrote:
On Fri, Aug 7, 2020 at 2:29 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Fri, Aug 07, 2020 at 12:41:00PM -0700, Nick Desaulniers wrote:
A recent change to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
It could also cause kernels compiled with gcc and linked with LLD to fail in the same way, no? The gcc/gas combination will generate the relaxed relocations from I think gas-2.26 onward. Although the only troublesome symbol in the case of gcc/gas is trampoline_32bit_src, referenced from pgtable_64.c (gcc doesn't use a GOTPC reloc for _pgtable etc).
Thanks for taking a look, and the feedback. I appreciate it!
$ gcc --version | head -n 1 gcc (Debian 9.3.0-11) 9.3.0 $ make -j71 clean defconfig bzImage $ llvm-readelf -r arch/x86/boot/compressed/*.o | grep -e R_X86_64_GOTPCRELX -e R_X86_64_REX_GOTPCRELX 0000000000000114 000000120000002a R_X86_64_REX_GOTPCRELX 0000000000000000 trampoline_32bit_src - 4 $ llvm-readelf -r arch/x86/boot/compressed/vmlinux | grep -e R_X86_64_GOTPCRELX -e R_X86_64_REX_GOTPCRELX $
So it looks like yes. I guess then we'd need to add a check for CONFIG_LD_IS_LLD and CONFIG_CC_IS_GCC and binutils version is 2.26+? I don't mind adding support for that combination, but I'd like to skip it in this patch for the sake of backporting something small to stable to get our CI green ASAP, since CONFIG_LD_IS_LLD probably doesn't exist for those stable branches, which will complicate the backport of such a patch. So I'd do it in a follow up patch if we're cool with that?
What if we did it only if we couldn't enable -pie, like the below patch? I think this should cover all the cases without needing LD_IS_LLD checks.
For BFD, the only case that should change is binutils-2.26, which supports relaxations but not -z noreloc-overflow, and will now have relax-relocations disabled. It currently works (with gcc) only because the relaxation of movq foo@GOTPCREL(%rip), %reg to movq $foo, %reg in the non-pie case was only added in 2.27, which is also when -z noreloc-overflow was added, allowing -pie to be enabled. With 2.26, it only gets relaxed to leaq foo(%rip), %reg which is all LLD currently does as well.
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 8abc30b27ba3..d25bb71f195a 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -60,6 +60,13 @@ else KBUILD_LDFLAGS += $(shell $(LD) --help 2>&1 | grep -q "-z noreloc-overflow" \ && echo "-z noreloc-overflow -pie --no-dynamic-linker") endif + +# Disable relocation relaxation if not building as PIE +ifeq ($(filter -pie,$(KBUILD_LDFLAGS)),) +KBUILD_CFLAGS += $(call as-option, -Wa$(comma)-mrelax-relocations=no) +KBUILD_AFLAGS += $(call as-option, -Wa$(comma)-mrelax-relocations=no) +endif + LDFLAGS_vmlinux := -T
hostprogs := mkpiggy
On Fri, Aug 7, 2020 at 6:43 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Fri, Aug 07, 2020 at 02:54:39PM -0700, Nick Desaulniers wrote:
On Fri, Aug 7, 2020 at 2:29 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Fri, Aug 07, 2020 at 12:41:00PM -0700, Nick Desaulniers wrote:
A recent change to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
It could also cause kernels compiled with gcc and linked with LLD to fail in the same way, no? The gcc/gas combination will generate the relaxed relocations from I think gas-2.26 onward. Although the only troublesome symbol in the case of gcc/gas is trampoline_32bit_src, referenced from pgtable_64.c (gcc doesn't use a GOTPC reloc for _pgtable etc).
Thanks for taking a look, and the feedback. I appreciate it!
$ gcc --version | head -n 1 gcc (Debian 9.3.0-11) 9.3.0 $ make -j71 clean defconfig bzImage $ llvm-readelf -r arch/x86/boot/compressed/*.o | grep -e R_X86_64_GOTPCRELX -e R_X86_64_REX_GOTPCRELX 0000000000000114 000000120000002a R_X86_64_REX_GOTPCRELX 0000000000000000 trampoline_32bit_src - 4 $ llvm-readelf -r arch/x86/boot/compressed/vmlinux | grep -e R_X86_64_GOTPCRELX -e R_X86_64_REX_GOTPCRELX $
So it looks like yes. I guess then we'd need to add a check for CONFIG_LD_IS_LLD and CONFIG_CC_IS_GCC and binutils version is 2.26+? I don't mind adding support for that combination, but I'd like to skip it in this patch for the sake of backporting something small to stable to get our CI green ASAP, since CONFIG_LD_IS_LLD probably doesn't exist for those stable branches, which will complicate the backport of such a patch. So I'd do it in a follow up patch if we're cool with that?
What if we did it only if we couldn't enable -pie, like the below patch? I think this should cover all the cases without needing LD_IS_LLD checks.
For BFD, the only case that should change is binutils-2.26, which supports relaxations but not -z noreloc-overflow, and will now have relax-relocations disabled. It currently works (with gcc) only because the relaxation of movq foo@GOTPCREL(%rip), %reg to movq $foo, %reg in the non-pie case was only added in 2.27, which is also when -z noreloc-overflow was added, allowing -pie to be enabled. With 2.26, it only gets relaxed to leaq foo(%rip), %reg which is all LLD currently does as well.
Sure, that will work, too. If you'd like to send it along, please add my: Reviewed-by: Nick Desaulniers ndesaulniers@google.com Tested-by: Nick Desaulniers ndesaulniers@google.com
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 8abc30b27ba3..d25bb71f195a 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -60,6 +60,13 @@ else KBUILD_LDFLAGS += $(shell $(LD) --help 2>&1 | grep -q "-z noreloc-overflow" \ && echo "-z noreloc-overflow -pie --no-dynamic-linker") endif
+# Disable relocation relaxation if not building as PIE +ifeq ($(filter -pie,$(KBUILD_LDFLAGS)),) +KBUILD_CFLAGS += $(call as-option, -Wa$(comma)-mrelax-relocations=no) +KBUILD_AFLAGS += $(call as-option, -Wa$(comma)-mrelax-relocations=no) +endif
LDFLAGS_vmlinux := -T
hostprogs := mkpiggy
The x86-64 psABI [0] specifies special relocation types (R_X86_64_[REX_]GOTPCRELX) for indirection through the Global Offset Table, semantically equivalent to R_X86_64_GOTPCREL, which the linker can take advantage of for optimization (relaxation) at link time. This is supported by LLD and binutils versions 2.26 onwards.
The compressed kernel is position-independent code, however, when using LLD or binutils versions before 2.27, it must be linked without the -pie option. In this case, the linker may optimize certain instructions into a non-position-independent form, by converting foo@GOTPCREL(%rip) to $foo.
This potential issue has been present with LLD and binutils-2.26 for a long time, but it has never manifested itself before now: - LLD and binutils-2.26 only relax movq foo@GOTPCREL(%rip), %reg to leaq foo(%rip), %reg which is still position-independent, rather than mov $foo, %reg which is permitted by the psABI when -pie is not enabled. - gcc happens to only generate GOTPCREL relocations on mov instructions. - clang does generate GOTPCREL relocations on non-mov instructions, but when building the compressed kernel, it uses its integrated assembler (due to the redefinition of KBUILD_CFLAGS dropping -no-integrated-as), which has so far defaulted to not generating the GOTPCRELX relocations.
Nick Desaulniers reports [1,2]: A recent change [3] to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
Patch series [4] is a solution to allow the compressed kernel to be linked with -pie unconditionally, but even if merged is unlikely to be backported. As a simple solution that can be applied to stable as well, prevent the assembler from generating the relaxed relocation types using the -mrelax-relocations=no option.
[0] https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/master/x86-64-ABI/linker-opt... [1] https://lore.kernel.org/lkml/20200807194100.3570838-1-ndesaulniers@google.co... [2] https://github.com/ClangBuiltLinux/linux/issues/1121 [3] https://reviews.llvm.org/rGc41a18cf61790fc898dcda1055c3efbf442c14c0 [4] https://lore.kernel.org/lkml/20200731202738.2577854-1-nivedita@alum.mit.edu/
Signed-off-by: Arvind Sankar nivedita@alum.mit.edu Reported-by: Nick Desaulniers ndesaulniers@google.com Reviewed-by: Nick Desaulniers ndesaulniers@google.com Tested-by: Nick Desaulniers ndesaulniers@google.com Cc: stable@vger.kernel.org # 4.19.x --- arch/x86/boot/compressed/Makefile | 6 ++++++ 1 file changed, 6 insertions(+)
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..c5449bea58ec 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -62,6 +62,12 @@ KBUILD_LDFLAGS += $(shell $(LD) --help 2>&1 | grep -q "-z noreloc-overflow" \ endif LDFLAGS_vmlinux := -T
+# Disable relocation relaxation if not linking as PIE +ifeq ($(filter -pie,$(KBUILD_LDFLAGS)),) +KBUILD_CFLAGS += $(call as-option, -Wa$(comma)-mrelax-relocations=no) +KBUILD_AFLAGS += $(call as-option, -Wa$(comma)-mrelax-relocations=no) +endif + hostprogs := mkpiggy HOST_EXTRACFLAGS += -I$(srctree)/tools/include
On Tue, Aug 11, 2020 at 10:36 AM Arvind Sankar nivedita@alum.mit.edu wrote:
The x86-64 psABI [0] specifies special relocation types (R_X86_64_[REX_]GOTPCRELX) for indirection through the Global Offset Table, semantically equivalent to R_X86_64_GOTPCREL, which the linker can take advantage of for optimization (relaxation) at link time. This is supported by LLD and binutils versions 2.26 onwards.
The compressed kernel is position-independent code, however, when using LLD or binutils versions before 2.27, it must be linked without the -pie option. In this case, the linker may optimize certain instructions into a non-position-independent form, by converting foo@GOTPCREL(%rip) to $foo.
This potential issue has been present with LLD and binutils-2.26 for a long time, but it has never manifested itself before now:
- LLD and binutils-2.26 only relax movq foo@GOTPCREL(%rip), %reg to leaq foo(%rip), %reg which is still position-independent, rather than mov $foo, %reg which is permitted by the psABI when -pie is not enabled.
- gcc happens to only generate GOTPCREL relocations on mov instructions.
- clang does generate GOTPCREL relocations on non-mov instructions, but when building the compressed kernel, it uses its integrated assembler (due to the redefinition of KBUILD_CFLAGS dropping -no-integrated-as), which has so far defaulted to not generating the GOTPCRELX relocations.
Nick Desaulniers reports [1,2]: A recent change [3] to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
Patch series [4] is a solution to allow the compressed kernel to be linked with -pie unconditionally, but even if merged is unlikely to be backported. As a simple solution that can be applied to stable as well, prevent the assembler from generating the relaxed relocation types using the -mrelax-relocations=no option.
[0] https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/master/x86-64-ABI/linker-opt... [1] https://lore.kernel.org/lkml/20200807194100.3570838-1-ndesaulniers@google.co... [2] https://github.com/ClangBuiltLinux/linux/issues/1121 [3] https://reviews.llvm.org/rGc41a18cf61790fc898dcda1055c3efbf442c14c0 [4] https://lore.kernel.org/lkml/20200731202738.2577854-1-nivedita@alum.mit.edu/
Signed-off-by: Arvind Sankar nivedita@alum.mit.edu Reported-by: Nick Desaulniers ndesaulniers@google.com Reviewed-by: Nick Desaulniers ndesaulniers@google.com Tested-by: Nick Desaulniers ndesaulniers@google.com Cc: stable@vger.kernel.org # 4.19.x
Thanks Arvind, good write up. Just curious about this stable tag, how come you picked 4.19? I can see boot failures in our CI for x86+LLD back to 4.9. Can we amend that tag to use `# 4.9`? I'd be happy to help submit backports should they fail to apply cleanly. https://travis-ci.com/github/ClangBuiltLinux/continuous-integration/builds/1...
arch/x86/boot/compressed/Makefile | 6 ++++++ 1 file changed, 6 insertions(+)
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..c5449bea58ec 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -62,6 +62,12 @@ KBUILD_LDFLAGS += $(shell $(LD) --help 2>&1 | grep -q "-z noreloc-overflow" \ endif LDFLAGS_vmlinux := -T
+# Disable relocation relaxation if not linking as PIE +ifeq ($(filter -pie,$(KBUILD_LDFLAGS)),) +KBUILD_CFLAGS += $(call as-option, -Wa$(comma)-mrelax-relocations=no) +KBUILD_AFLAGS += $(call as-option, -Wa$(comma)-mrelax-relocations=no) +endif
hostprogs := mkpiggy HOST_EXTRACFLAGS += -I$(srctree)/tools/include
-- 2.26.2
On Tue, Aug 11, 2020 at 10:58:40AM -0700, Nick Desaulniers wrote:
Cc: stable@vger.kernel.org # 4.19.x
Thanks Arvind, good write up. Just curious about this stable tag, how come you picked 4.19? I can see boot failures in our CI for x86+LLD back to 4.9. Can we amend that tag to use `# 4.9`? I'd be happy to help submit backports should they fail to apply cleanly. https://travis-ci.com/github/ClangBuiltLinux/continuous-integration/builds/1...
4.19 renamed LDFLAGS to KBUILD_LDFLAGS. For 4.4, 4.9 and 4.14 the patch needs to be modified, KBUILD_LDFLAGS -> LDFLAGS, so I figured we should submit backports separately. For 4.19 onwards, it should apply without changes I think.
Thanks.
On Tue, Aug 11, 2020 at 3:44 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 11, 2020 at 10:58:40AM -0700, Nick Desaulniers wrote:
Cc: stable@vger.kernel.org # 4.19.x
Thanks Arvind, good write up. Just curious about this stable tag, how come you picked 4.19? I can see boot failures in our CI for x86+LLD back to 4.9. Can we amend that tag to use `# 4.9`? I'd be happy to help submit backports should they fail to apply cleanly. https://travis-ci.com/github/ClangBuiltLinux/continuous-integration/builds/1...
4.19 renamed LDFLAGS to KBUILD_LDFLAGS. For 4.4, 4.9 and 4.14 the patch needs to be modified, KBUILD_LDFLAGS -> LDFLAGS, so I figured we should submit backports separately. For 4.19 onwards, it should apply without changes I think.
Cool, sounds good. I'll keep an eye out for when stable goes to pick this up.
tglx, Ingo, BP, can we pretty please get this in tip/urgent for inclusion into 5.9?
On Tue, Aug 11, 2020 at 04:04:40PM -0700, Nick Desaulniers wrote:
On Tue, Aug 11, 2020 at 3:44 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 11, 2020 at 10:58:40AM -0700, Nick Desaulniers wrote:
Cc: stable@vger.kernel.org # 4.19.x
Thanks Arvind, good write up. Just curious about this stable tag, how come you picked 4.19? I can see boot failures in our CI for x86+LLD back to 4.9. Can we amend that tag to use `# 4.9`? I'd be happy to help submit backports should they fail to apply cleanly. https://travis-ci.com/github/ClangBuiltLinux/continuous-integration/builds/1...
4.19 renamed LDFLAGS to KBUILD_LDFLAGS. For 4.4, 4.9 and 4.14 the patch needs to be modified, KBUILD_LDFLAGS -> LDFLAGS, so I figured we should submit backports separately. For 4.19 onwards, it should apply without changes I think.
Cool, sounds good. I'll keep an eye out for when stable goes to pick this up.
tglx, Ingo, BP, can we pretty please get this in tip/urgent for inclusion into 5.9? -- Thanks, ~Nick Desaulniers
Another alternative is to just do this unconditionally instead of even checking for the -pie flag. None of the GOTPCRELs are in the decompressor, so they shouldn't be performance-sensitive at all.
It still wouldn't apply cleanly to all the stable versions, but backporting would be even simpler.
What do you think?
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..10c2ba59d192 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -43,6 +43,7 @@ KBUILD_CFLAGS += -Wno-pointer-sign KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS +KBUILD_CFLAGS += $(call as-option,-Wa$(comma)-mrelax-relocations=no)
KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n
On Tue, Aug 11, 2020 at 4:43 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 11, 2020 at 04:04:40PM -0700, Nick Desaulniers wrote:
On Tue, Aug 11, 2020 at 3:44 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 11, 2020 at 10:58:40AM -0700, Nick Desaulniers wrote:
Cc: stable@vger.kernel.org # 4.19.x
Thanks Arvind, good write up. Just curious about this stable tag, how come you picked 4.19? I can see boot failures in our CI for x86+LLD back to 4.9. Can we amend that tag to use `# 4.9`? I'd be happy to help submit backports should they fail to apply cleanly. https://travis-ci.com/github/ClangBuiltLinux/continuous-integration/builds/1...
4.19 renamed LDFLAGS to KBUILD_LDFLAGS. For 4.4, 4.9 and 4.14 the patch needs to be modified, KBUILD_LDFLAGS -> LDFLAGS, so I figured we should submit backports separately. For 4.19 onwards, it should apply without changes I think.
Cool, sounds good. I'll keep an eye out for when stable goes to pick this up.
tglx, Ingo, BP, can we pretty please get this in tip/urgent for inclusion into 5.9? -- Thanks, ~Nick Desaulniers
Another alternative is to just do this unconditionally instead of even checking for the -pie flag. None of the GOTPCRELs are in the decompressor, so they shouldn't be performance-sensitive at all.
It still wouldn't apply cleanly to all the stable versions, but backporting would be even simpler.
What do you think?
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..10c2ba59d192 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -43,6 +43,7 @@ KBUILD_CFLAGS += -Wno-pointer-sign KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS +KBUILD_CFLAGS += $(call as-option,-Wa$(comma)-mrelax-relocations=no)
We'd still want it for KBUILD_AFLAGS, too, just to be safe. Maybe a one line comment to the effect of `# remove me once we can link as -pie` would help us rip off this band-aid in the future? It's more obvious that the added hunk can be reverted once -pie linkage is achieved with the current patch; either are fine by me. Thanks!
KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n
On Tue, Aug 11, 2020 at 04:51:23PM -0700, Nick Desaulniers wrote:
On Tue, Aug 11, 2020 at 4:43 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 11, 2020 at 04:04:40PM -0700, Nick Desaulniers wrote:
On Tue, Aug 11, 2020 at 3:44 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 11, 2020 at 10:58:40AM -0700, Nick Desaulniers wrote:
Cc: stable@vger.kernel.org # 4.19.x
Thanks Arvind, good write up. Just curious about this stable tag, how come you picked 4.19? I can see boot failures in our CI for x86+LLD back to 4.9. Can we amend that tag to use `# 4.9`? I'd be happy to help submit backports should they fail to apply cleanly. https://travis-ci.com/github/ClangBuiltLinux/continuous-integration/builds/1...
4.19 renamed LDFLAGS to KBUILD_LDFLAGS. For 4.4, 4.9 and 4.14 the patch needs to be modified, KBUILD_LDFLAGS -> LDFLAGS, so I figured we should submit backports separately. For 4.19 onwards, it should apply without changes I think.
Cool, sounds good. I'll keep an eye out for when stable goes to pick this up.
tglx, Ingo, BP, can we pretty please get this in tip/urgent for inclusion into 5.9? -- Thanks, ~Nick Desaulniers
Another alternative is to just do this unconditionally instead of even checking for the -pie flag. None of the GOTPCRELs are in the decompressor, so they shouldn't be performance-sensitive at all.
It still wouldn't apply cleanly to all the stable versions, but backporting would be even simpler.
What do you think?
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..10c2ba59d192 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -43,6 +43,7 @@ KBUILD_CFLAGS += -Wno-pointer-sign KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS +KBUILD_CFLAGS += $(call as-option,-Wa$(comma)-mrelax-relocations=no)
We'd still want it for KBUILD_AFLAGS, too, just to be safe. Maybe a
KBUILD_CFLAGS gets included into KBUILD_AFLAGS, so this already does that.
one line comment to the effect of `# remove me once we can link as -pie` would help us rip off this band-aid in the future? It's more obvious that the added hunk can be reverted once -pie linkage is achieved with the current patch; either are fine by me. Thanks!
KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n
-- Thanks, ~Nick Desaulniers
The x86-64 psABI [0] specifies special relocation types (R_X86_64_[REX_]GOTPCRELX) for indirection through the Global Offset Table, semantically equivalent to R_X86_64_GOTPCREL, which the linker can take advantage of for optimization (relaxation) at link time. This is supported by LLD and binutils versions 2.26 onwards.
The compressed kernel is position-independent code, however, when using LLD or binutils versions before 2.27, it must be linked without the -pie option. In this case, the linker may optimize certain instructions into a non-position-independent form, by converting foo@GOTPCREL(%rip) to $foo.
This potential issue has been present with LLD and binutils-2.26 for a long time, but it has never manifested itself before now: - LLD and binutils-2.26 only relax movq foo@GOTPCREL(%rip), %reg to leaq foo(%rip), %reg which is still position-independent, rather than mov $foo, %reg which is permitted by the psABI when -pie is not enabled. - gcc happens to only generate GOTPCREL relocations on mov instructions. - clang does generate GOTPCREL relocations on non-mov instructions, but when building the compressed kernel, it uses its integrated assembler (due to the redefinition of KBUILD_CFLAGS dropping -no-integrated-as), which has so far defaulted to not generating the GOTPCRELX relocations.
Nick Desaulniers reports [1,2]: A recent change [3] to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
Patch series [4] is a solution to allow the compressed kernel to be linked with -pie unconditionally, but even if merged is unlikely to be backported. As a simple solution that can be applied to stable as well, prevent the assembler from generating the relaxed relocation types using the -mrelax-relocations=no option. For ease of backporting, do this unconditionally.
[0] https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/master/x86-64-ABI/linker-opt... [1] https://lore.kernel.org/lkml/20200807194100.3570838-1-ndesaulniers@google.co... [2] https://github.com/ClangBuiltLinux/linux/issues/1121 [3] https://reviews.llvm.org/rGc41a18cf61790fc898dcda1055c3efbf442c14c0 [4] https://lore.kernel.org/lkml/20200731202738.2577854-1-nivedita@alum.mit.edu/
Signed-off-by: Arvind Sankar nivedita@alum.mit.edu Reported-by: Nick Desaulniers ndesaulniers@google.com Reviewed-by: Nick Desaulniers ndesaulniers@google.com Tested-by: Nick Desaulniers ndesaulniers@google.com Cc: stable@vger.kernel.org Signed-off-by: Arvind Sankar nivedita@alum.mit.edu --- arch/x86/boot/compressed/Makefile | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..ff7894f39e0e 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -43,6 +43,8 @@ KBUILD_CFLAGS += -Wno-pointer-sign KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS +# Disable relocation relaxation in case the link is not PIE. +KBUILD_CFLAGS += $(call as-option,-Wa$(comma)-mrelax-relocations=no)
KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n
On Tue, Aug 11, 2020 at 5:43 PM Arvind Sankar nivedita@alum.mit.edu wrote:
The x86-64 psABI [0] specifies special relocation types (R_X86_64_[REX_]GOTPCRELX) for indirection through the Global Offset Table, semantically equivalent to R_X86_64_GOTPCREL, which the linker can take advantage of for optimization (relaxation) at link time. This is supported by LLD and binutils versions 2.26 onwards.
The compressed kernel is position-independent code, however, when using LLD or binutils versions before 2.27, it must be linked without the -pie option. In this case, the linker may optimize certain instructions into a non-position-independent form, by converting foo@GOTPCREL(%rip) to $foo.
This potential issue has been present with LLD and binutils-2.26 for a long time, but it has never manifested itself before now:
- LLD and binutils-2.26 only relax movq foo@GOTPCREL(%rip), %reg to leaq foo(%rip), %reg which is still position-independent, rather than mov $foo, %reg which is permitted by the psABI when -pie is not enabled.
- gcc happens to only generate GOTPCREL relocations on mov instructions.
- clang does generate GOTPCREL relocations on non-mov instructions, but when building the compressed kernel, it uses its integrated assembler (due to the redefinition of KBUILD_CFLAGS dropping -no-integrated-as), which has so far defaulted to not generating the GOTPCRELX relocations.
Nick Desaulniers reports [1,2]: A recent change [3] to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
Patch series [4] is a solution to allow the compressed kernel to be linked with -pie unconditionally, but even if merged is unlikely to be backported. As a simple solution that can be applied to stable as well, prevent the assembler from generating the relaxed relocation types using the -mrelax-relocations=no option. For ease of backporting, do this unconditionally.
[0] https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/master/x86-64-ABI/linker-opt... [1] https://lore.kernel.org/lkml/20200807194100.3570838-1-ndesaulniers@google.co... [2] https://github.com/ClangBuiltLinux/linux/issues/1121 [3] https://reviews.llvm.org/rGc41a18cf61790fc898dcda1055c3efbf442c14c0 [4] https://lore.kernel.org/lkml/20200731202738.2577854-1-nivedita@alum.mit.edu/
Signed-off-by: Arvind Sankar nivedita@alum.mit.edu Reported-by: Nick Desaulniers ndesaulniers@google.com Reviewed-by: Nick Desaulniers ndesaulniers@google.com Tested-by: Nick Desaulniers ndesaulniers@google.com Cc: stable@vger.kernel.org Signed-off-by: Arvind Sankar nivedita@alum.mit.edu
LGTM
arch/x86/boot/compressed/Makefile | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..ff7894f39e0e 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -43,6 +43,8 @@ KBUILD_CFLAGS += -Wno-pointer-sign KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS +# Disable relocation relaxation in case the link is not PIE. +KBUILD_CFLAGS += $(call as-option,-Wa$(comma)-mrelax-relocations=no)
KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n -- 2.26.2
On Wed, Aug 12, 2020 at 2:43 AM Arvind Sankar nivedita@alum.mit.edu wrote:
The x86-64 psABI [0] specifies special relocation types (R_X86_64_[REX_]GOTPCRELX) for indirection through the Global Offset Table, semantically equivalent to R_X86_64_GOTPCREL, which the linker can take advantage of for optimization (relaxation) at link time. This is supported by LLD and binutils versions 2.26 onwards.
The compressed kernel is position-independent code, however, when using LLD or binutils versions before 2.27, it must be linked without the -pie option. In this case, the linker may optimize certain instructions into a non-position-independent form, by converting foo@GOTPCREL(%rip) to $foo.
This potential issue has been present with LLD and binutils-2.26 for a long time, but it has never manifested itself before now:
- LLD and binutils-2.26 only relax movq foo@GOTPCREL(%rip), %reg to leaq foo(%rip), %reg which is still position-independent, rather than mov $foo, %reg which is permitted by the psABI when -pie is not enabled.
- gcc happens to only generate GOTPCREL relocations on mov instructions.
- clang does generate GOTPCREL relocations on non-mov instructions, but when building the compressed kernel, it uses its integrated assembler (due to the redefinition of KBUILD_CFLAGS dropping -no-integrated-as), which has so far defaulted to not generating the GOTPCRELX relocations.
Nick Desaulniers reports [1,2]: A recent change [3] to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
Patch series [4] is a solution to allow the compressed kernel to be linked with -pie unconditionally, but even if merged is unlikely to be backported. As a simple solution that can be applied to stable as well, prevent the assembler from generating the relaxed relocation types using the -mrelax-relocations=no option. For ease of backporting, do this unconditionally.
[0] https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/master/x86-64-ABI/linker-opt... [1] https://lore.kernel.org/lkml/20200807194100.3570838-1-ndesaulniers@google.co... [2] https://github.com/ClangBuiltLinux/linux/issues/1121 [3] https://reviews.llvm.org/rGc41a18cf61790fc898dcda1055c3efbf442c14c0 [4] https://lore.kernel.org/lkml/20200731202738.2577854-1-nivedita@alum.mit.edu/
Signed-off-by: Arvind Sankar nivedita@alum.mit.edu Reported-by: Nick Desaulniers ndesaulniers@google.com Reviewed-by: Nick Desaulniers ndesaulniers@google.com Tested-by: Nick Desaulniers ndesaulniers@google.com Cc: stable@vger.kernel.org Signed-off-by: Arvind Sankar nivedita@alum.mit.edu
Thanks for the patch.
Tested-by: Sedat Dilek sedat.dilek@gmail.com
- Sedat -
[1] https://github.com/ClangBuiltLinux/linux/issues/1120#issuecomment-674409705
arch/x86/boot/compressed/Makefile | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..ff7894f39e0e 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -43,6 +43,8 @@ KBUILD_CFLAGS += -Wno-pointer-sign KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS +# Disable relocation relaxation in case the link is not PIE. +KBUILD_CFLAGS += $(call as-option,-Wa$(comma)-mrelax-relocations=no)
KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n -- 2.26.2
-- You received this message because you are subscribed to the Google Groups "Clang Built Linux" group. To unsubscribe from this group and stop receiving emails from it, send an email to clang-built-linux+unsubscribe@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/clang-built-linux/20200812004308.1448603-1....
Hi Ingo, I saw you picked up Arvind's other series into x86/boot. Would you mind please including this, as well? Our CI is quite red for x86...
EOM
On Sat, Aug 15, 2020 at 8:49 AM Sedat Dilek sedat.dilek@gmail.com wrote:
On Wed, Aug 12, 2020 at 2:43 AM Arvind Sankar nivedita@alum.mit.edu wrote:
The x86-64 psABI [0] specifies special relocation types (R_X86_64_[REX_]GOTPCRELX) for indirection through the Global Offset Table, semantically equivalent to R_X86_64_GOTPCREL, which the linker can take advantage of for optimization (relaxation) at link time. This is supported by LLD and binutils versions 2.26 onwards.
The compressed kernel is position-independent code, however, when using LLD or binutils versions before 2.27, it must be linked without the -pie option. In this case, the linker may optimize certain instructions into a non-position-independent form, by converting foo@GOTPCREL(%rip) to $foo.
This potential issue has been present with LLD and binutils-2.26 for a long time, but it has never manifested itself before now:
- LLD and binutils-2.26 only relax movq foo@GOTPCREL(%rip), %reg to leaq foo(%rip), %reg which is still position-independent, rather than mov $foo, %reg which is permitted by the psABI when -pie is not enabled.
- gcc happens to only generate GOTPCREL relocations on mov instructions.
- clang does generate GOTPCREL relocations on non-mov instructions, but when building the compressed kernel, it uses its integrated assembler (due to the redefinition of KBUILD_CFLAGS dropping -no-integrated-as), which has so far defaulted to not generating the GOTPCRELX relocations.
Nick Desaulniers reports [1,2]: A recent change [3] to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
Patch series [4] is a solution to allow the compressed kernel to be linked with -pie unconditionally, but even if merged is unlikely to be backported. As a simple solution that can be applied to stable as well, prevent the assembler from generating the relaxed relocation types using the -mrelax-relocations=no option. For ease of backporting, do this unconditionally.
[0] https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/master/x86-64-ABI/linker-opt... [1] https://lore.kernel.org/lkml/20200807194100.3570838-1-ndesaulniers@google.co... [2] https://github.com/ClangBuiltLinux/linux/issues/1121 [3] https://reviews.llvm.org/rGc41a18cf61790fc898dcda1055c3efbf442c14c0 [4] https://lore.kernel.org/lkml/20200731202738.2577854-1-nivedita@alum.mit.edu/
Signed-off-by: Arvind Sankar nivedita@alum.mit.edu Reported-by: Nick Desaulniers ndesaulniers@google.com Reviewed-by: Nick Desaulniers ndesaulniers@google.com Tested-by: Nick Desaulniers ndesaulniers@google.com Cc: stable@vger.kernel.org Signed-off-by: Arvind Sankar nivedita@alum.mit.edu
Thanks for the patch.
Tested-by: Sedat Dilek sedat.dilek@gmail.com
- Sedat -
[1] https://github.com/ClangBuiltLinux/linux/issues/1120#issuecomment-674409705
arch/x86/boot/compressed/Makefile | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..ff7894f39e0e 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -43,6 +43,8 @@ KBUILD_CFLAGS += -Wno-pointer-sign KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS +# Disable relocation relaxation in case the link is not PIE. +KBUILD_CFLAGS += $(call as-option,-Wa$(comma)-mrelax-relocations=no)
KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n -- 2.26.2
-- You received this message because you are subscribed to the Google Groups "Clang Built Linux" group. To unsubscribe from this group and stop receiving emails from it, send an email to clang-built-linux+unsubscribe@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/clang-built-linux/20200812004308.1448603-1....
On Sat, Aug 15, 2020 at 10:57 PM Nick Desaulniers ndesaulniers@google.com wrote:
Hi Ingo, I saw you picked up Arvind's other series into x86/boot. Would you mind please including this, as well? Our CI is quite red for x86...
EOM
+1
- Sedat -
On Sat, Aug 15, 2020 at 8:49 AM Sedat Dilek sedat.dilek@gmail.com wrote:
On Wed, Aug 12, 2020 at 2:43 AM Arvind Sankar nivedita@alum.mit.edu wrote:
The x86-64 psABI [0] specifies special relocation types (R_X86_64_[REX_]GOTPCRELX) for indirection through the Global Offset Table, semantically equivalent to R_X86_64_GOTPCREL, which the linker can take advantage of for optimization (relaxation) at link time. This is supported by LLD and binutils versions 2.26 onwards.
The compressed kernel is position-independent code, however, when using LLD or binutils versions before 2.27, it must be linked without the -pie option. In this case, the linker may optimize certain instructions into a non-position-independent form, by converting foo@GOTPCREL(%rip) to $foo.
This potential issue has been present with LLD and binutils-2.26 for a long time, but it has never manifested itself before now:
- LLD and binutils-2.26 only relax movq foo@GOTPCREL(%rip), %reg to leaq foo(%rip), %reg which is still position-independent, rather than mov $foo, %reg which is permitted by the psABI when -pie is not enabled.
- gcc happens to only generate GOTPCREL relocations on mov instructions.
- clang does generate GOTPCREL relocations on non-mov instructions, but when building the compressed kernel, it uses its integrated assembler (due to the redefinition of KBUILD_CFLAGS dropping -no-integrated-as), which has so far defaulted to not generating the GOTPCRELX relocations.
Nick Desaulniers reports [1,2]: A recent change [3] to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot.
Patch series [4] is a solution to allow the compressed kernel to be linked with -pie unconditionally, but even if merged is unlikely to be backported. As a simple solution that can be applied to stable as well, prevent the assembler from generating the relaxed relocation types using the -mrelax-relocations=no option. For ease of backporting, do this unconditionally.
[0] https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/master/x86-64-ABI/linker-opt... [1] https://lore.kernel.org/lkml/20200807194100.3570838-1-ndesaulniers@google.co... [2] https://github.com/ClangBuiltLinux/linux/issues/1121 [3] https://reviews.llvm.org/rGc41a18cf61790fc898dcda1055c3efbf442c14c0 [4] https://lore.kernel.org/lkml/20200731202738.2577854-1-nivedita@alum.mit.edu/
Signed-off-by: Arvind Sankar nivedita@alum.mit.edu Reported-by: Nick Desaulniers ndesaulniers@google.com Reviewed-by: Nick Desaulniers ndesaulniers@google.com Tested-by: Nick Desaulniers ndesaulniers@google.com Cc: stable@vger.kernel.org Signed-off-by: Arvind Sankar nivedita@alum.mit.edu
Thanks for the patch.
Tested-by: Sedat Dilek sedat.dilek@gmail.com
- Sedat -
[1] https://github.com/ClangBuiltLinux/linux/issues/1120#issuecomment-674409705
arch/x86/boot/compressed/Makefile | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..ff7894f39e0e 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -43,6 +43,8 @@ KBUILD_CFLAGS += -Wno-pointer-sign KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS +# Disable relocation relaxation in case the link is not PIE. +KBUILD_CFLAGS += $(call as-option,-Wa$(comma)-mrelax-relocations=no)
KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n -- 2.26.2
-- You received this message because you are subscribed to the Google Groups "Clang Built Linux" group. To unsubscribe from this group and stop receiving emails from it, send an email to clang-built-linux+unsubscribe@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/clang-built-linux/20200812004308.1448603-1....
-- Thanks, ~Nick Desaulniers
On Sat, Aug 15, 2020 at 01:56:49PM -0700, Nick Desaulniers wrote:
Hi Ingo, I saw you picked up Arvind's other series into x86/boot. Would you mind please including this, as well? Our CI is quite red for x86...
EOM
Hi Ingo, while this patch is unnecessary after the series in tip/x86/boot, it is still needed for 5.9 and older. Would you be able to send it in for the next -rc? It shouldn't hurt the tip/x86/boot series, and we can add a revert on top of that later.
Thanks.
On Tue, Aug 25, 2020 at 10:56:52AM -0400, Arvind Sankar wrote:
On Sat, Aug 15, 2020 at 01:56:49PM -0700, Nick Desaulniers wrote:
Hi Ingo, I saw you picked up Arvind's other series into x86/boot. Would you mind please including this, as well? Our CI is quite red for x86...
EOM
Hi Ingo, while this patch is unnecessary after the series in tip/x86/boot, it is still needed for 5.9 and older. Would you be able to send it in for the next -rc? It shouldn't hurt the tip/x86/boot series, and we can add a revert on top of that later.
Thanks.
Ping.
Thanks.
On Tue, Aug 25, 2020 at 10:56:52AM -0400, Arvind Sankar wrote:
On Sat, Aug 15, 2020 at 01:56:49PM -0700, Nick Desaulniers wrote:
Hi Ingo, I saw you picked up Arvind's other series into x86/boot. Would you mind please including this, as well? Our CI is quite red for x86...
EOM
Hi Ingo, while this patch is unnecessary after the series in tip/x86/boot, it is still needed for 5.9 and older. Would you be able to send it in for the next -rc? It shouldn't hurt the tip/x86/boot series, and we can add a revert on top of that later.
Thanks.
Ping.
https://lore.kernel.org/lkml/20200812004308.1448603-1-nivedita@alum.mit.edu/
On Mon, 14 Sep 2020 at 01:34, Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 25, 2020 at 10:56:52AM -0400, Arvind Sankar wrote:
On Sat, Aug 15, 2020 at 01:56:49PM -0700, Nick Desaulniers wrote:
Hi Ingo, I saw you picked up Arvind's other series into x86/boot. Would you mind please including this, as well? Our CI is quite red for x86...
EOM
Hi Ingo, while this patch is unnecessary after the series in tip/x86/boot, it is still needed for 5.9 and older. Would you be able to send it in for the next -rc? It shouldn't hurt the tip/x86/boot series, and we can add a revert on top of that later.
Thanks.
Ping.
https://lore.kernel.org/lkml/20200812004308.1448603-1-nivedita@alum.mit.edu/
Acked-by: Ard Biesheuvel ardb@kernel.org
* Ard Biesheuvel ardb@kernel.org wrote:
On Mon, 14 Sep 2020 at 01:34, Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 25, 2020 at 10:56:52AM -0400, Arvind Sankar wrote:
On Sat, Aug 15, 2020 at 01:56:49PM -0700, Nick Desaulniers wrote:
Hi Ingo, I saw you picked up Arvind's other series into x86/boot. Would you mind please including this, as well? Our CI is quite red for x86...
EOM
Hi Ingo, while this patch is unnecessary after the series in tip/x86/boot, it is still needed for 5.9 and older. Would you be able to send it in for the next -rc? It shouldn't hurt the tip/x86/boot series, and we can add a revert on top of that later.
Thanks.
Ping.
https://lore.kernel.org/lkml/20200812004308.1448603-1-nivedita@alum.mit.edu/
Acked-by: Ard Biesheuvel ardb@kernel.org
Thanks guys - queued up in tip:x86/urgent.
Ingo
On Mon, Sep 14, 2020 at 11:16 AM Ingo Molnar mingo@kernel.org wrote:
- Ard Biesheuvel ardb@kernel.org wrote:
On Mon, 14 Sep 2020 at 01:34, Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 25, 2020 at 10:56:52AM -0400, Arvind Sankar wrote:
On Sat, Aug 15, 2020 at 01:56:49PM -0700, Nick Desaulniers wrote:
Hi Ingo, I saw you picked up Arvind's other series into x86/boot. Would you mind please including this, as well? Our CI is quite red for x86...
EOM
Hi Ingo, while this patch is unnecessary after the series in tip/x86/boot, it is still needed for 5.9 and older. Would you be able to send it in for the next -rc? It shouldn't hurt the tip/x86/boot series, and we can add a revert on top of that later.
Thanks.
Ping.
https://lore.kernel.org/lkml/20200812004308.1448603-1-nivedita@alum.mit.edu/
Acked-by: Ard Biesheuvel ardb@kernel.org
Thanks guys - queued up in tip:x86/urgent.
Thanks. Did you push it?
Git-Web on <git.kernel.org> seems to be slow this Morning - checked Linus Git an hour ago. Does Anyone know what's going on?
- Sedat
The following commit has been merged into the x86/urgent branch of tip:
Commit-ID: 09e43968db40c33a73e9ddbfd937f46d5c334924 Gitweb: https://git.kernel.org/tip/09e43968db40c33a73e9ddbfd937f46d5c334924 Author: Arvind Sankar nivedita@alum.mit.edu AuthorDate: Tue, 11 Aug 2020 20:43:08 -04:00 Committer: Ingo Molnar mingo@kernel.org CommitterDate: Mon, 14 Sep 2020 11:14:45 +02:00
x86/boot/compressed: Disable relocation relaxation
The x86-64 psABI [0] specifies special relocation types (R_X86_64_[REX_]GOTPCRELX) for indirection through the Global Offset Table, semantically equivalent to R_X86_64_GOTPCREL, which the linker can take advantage of for optimization (relaxation) at link time. This is supported by LLD and binutils versions 2.26 onwards.
The compressed kernel is position-independent code, however, when using LLD or binutils versions before 2.27, it must be linked without the -pie option. In this case, the linker may optimize certain instructions into a non-position-independent form, by converting foo@GOTPCREL(%rip) to $foo.
This potential issue has been present with LLD and binutils-2.26 for a long time, but it has never manifested itself before now:
- LLD and binutils-2.26 only relax movq foo@GOTPCREL(%rip), %reg to leaq foo(%rip), %reg which is still position-independent, rather than mov $foo, %reg which is permitted by the psABI when -pie is not enabled.
- GCC happens to only generate GOTPCREL relocations on mov instructions.
- CLang does generate GOTPCREL relocations on non-mov instructions, but when building the compressed kernel, it uses its integrated assembler (due to the redefinition of KBUILD_CFLAGS dropping -no-integrated-as), which has so far defaulted to not generating the GOTPCRELX relocations.
Nick Desaulniers reports [1,2]:
"A recent change [3] to a default value of configuration variable (ENABLE_X86_RELAX_RELOCATIONS OFF -> ON) in LLVM now causes Clang's integrated assembler to emit R_X86_64_GOTPCRELX/R_X86_64_REX_GOTPCRELX relocations. LLD will relax instructions with these relocations based on whether the image is being linked as position independent or not. When not, then LLD will relax these instructions to use absolute addressing mode (R_RELAX_GOT_PC_NOPIC). This causes kernels built with Clang and linked with LLD to fail to boot."
Patch series [4] is a solution to allow the compressed kernel to be linked with -pie unconditionally, but even if merged is unlikely to be backported. As a simple solution that can be applied to stable as well, prevent the assembler from generating the relaxed relocation types using the -mrelax-relocations=no option. For ease of backporting, do this unconditionally.
[0] https://gitlab.com/x86-psABIs/x86-64-ABI/-/blob/master/x86-64-ABI/linker-opt... [1] https://lore.kernel.org/lkml/20200807194100.3570838-1-ndesaulniers@google.co... [2] https://github.com/ClangBuiltLinux/linux/issues/1121 [3] https://reviews.llvm.org/rGc41a18cf61790fc898dcda1055c3efbf442c14c0 [4] https://lore.kernel.org/lkml/20200731202738.2577854-1-nivedita@alum.mit.edu/
Reported-by: Nick Desaulniers ndesaulniers@google.com Signed-off-by: Arvind Sankar nivedita@alum.mit.edu Signed-off-by: Ingo Molnar mingo@kernel.org Tested-by: Nick Desaulniers ndesaulniers@google.com Tested-by: Sedat Dilek sedat.dilek@gmail.com Acked-by: Ard Biesheuvel ardb@kernel.org Reviewed-by: Nick Desaulniers ndesaulniers@google.com Cc: stable@vger.kernel.org Link: https://lore.kernel.org/r/20200812004308.1448603-1-nivedita@alum.mit.edu --- arch/x86/boot/compressed/Makefile | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f59..ff7894f 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -43,6 +43,8 @@ KBUILD_CFLAGS += -Wno-pointer-sign KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS +# Disable relocation relaxation in case the link is not PIE. +KBUILD_CFLAGS += $(call as-option,-Wa$(comma)-mrelax-relocations=no)
KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n
On Tue, Aug 11, 2020 at 5:42 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 11, 2020 at 04:51:23PM -0700, Nick Desaulniers wrote:
On Tue, Aug 11, 2020 at 4:43 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 11, 2020 at 04:04:40PM -0700, Nick Desaulniers wrote:
On Tue, Aug 11, 2020 at 3:44 PM Arvind Sankar nivedita@alum.mit.edu wrote:
On Tue, Aug 11, 2020 at 10:58:40AM -0700, Nick Desaulniers wrote:
> Cc: stable@vger.kernel.org # 4.19.x
Thanks Arvind, good write up. Just curious about this stable tag, how come you picked 4.19? I can see boot failures in our CI for x86+LLD back to 4.9. Can we amend that tag to use `# 4.9`? I'd be happy to help submit backports should they fail to apply cleanly. https://travis-ci.com/github/ClangBuiltLinux/continuous-integration/builds/1...
4.19 renamed LDFLAGS to KBUILD_LDFLAGS. For 4.4, 4.9 and 4.14 the patch needs to be modified, KBUILD_LDFLAGS -> LDFLAGS, so I figured we should submit backports separately. For 4.19 onwards, it should apply without changes I think.
Cool, sounds good. I'll keep an eye out for when stable goes to pick this up.
tglx, Ingo, BP, can we pretty please get this in tip/urgent for inclusion into 5.9? -- Thanks, ~Nick Desaulniers
Another alternative is to just do this unconditionally instead of even checking for the -pie flag. None of the GOTPCRELs are in the decompressor, so they shouldn't be performance-sensitive at all.
It still wouldn't apply cleanly to all the stable versions, but backporting would be even simpler.
What do you think?
diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile index 3962f592633d..10c2ba59d192 100644 --- a/arch/x86/boot/compressed/Makefile +++ b/arch/x86/boot/compressed/Makefile @@ -43,6 +43,7 @@ KBUILD_CFLAGS += -Wno-pointer-sign KBUILD_CFLAGS += $(call cc-option,-fmacro-prefix-map=$(srctree)/=) KBUILD_CFLAGS += -fno-asynchronous-unwind-tables KBUILD_CFLAGS += -D__DISABLE_EXPORTS +KBUILD_CFLAGS += $(call as-option,-Wa$(comma)-mrelax-relocations=no)
We'd still want it for KBUILD_AFLAGS, too, just to be safe. Maybe a
KBUILD_CFLAGS gets included into KBUILD_AFLAGS, so this already does that.
Ah, right, just below it in the diff.
one line comment to the effect of `# remove me once we can link as -pie` would help us rip off this band-aid in the future? It's more obvious that the added hunk can be reverted once -pie linkage is achieved with the current patch; either are fine by me. Thanks!
KBUILD_AFLAGS := $(KBUILD_CFLAGS) -D__ASSEMBLY__ GCOV_PROFILE := n
-- Thanks, ~Nick Desaulniers
linux-stable-mirror@lists.linaro.org
-
Ard Biesheuvel -
Arvind Sankar -
Ingo Molnar -
Nick Desaulniers -
Sedat Dilek -
tip-bot2 for Arvind Sankar