Re: [PATCH v2] random: vDSO: Avoid call to memset() when zeroing reserved in __cvdso_getrandom_data()
From: "Jason A. Donenfeld" <Jason@zx2c4.com>
Date: 2026-09-26 10:02:51
Also in:
linux-arm-kernel, linux-riscv, linux-s390, lkml, llvm, loongarch
On Fri, Sep 25, 2026 at 03:00:46PM -0700, Nick Desaulniers wrote:
On Fri, Sep 25, 2026 at 2:56 PM Nick Desaulniers [off-list ref] wrote:quoted
On Fri, Sep 25, 2026 at 2:53 PM Nick Desaulniers [off-list ref] wrote:quoted
On Fri, Sep 25, 2026 at 2:46 PM Nathan Chancellor [off-list ref] wrote:quoted
After a recent change in LLVM [1], builds with the random vDSO implementation, such as PowerPC and RISC-V, fail when checking the vDSO: arch/powerpc/kernel/vdso/vdso32.so.dbg: dynamic relocations are not supported arch/riscv/kernel/vdso/vdso.so.dbg: dynamic relocations are not supported memset() is now generated when zeroing params->reserved for some builds because LLVM has an optimization (now run in more instances) that can recognize at compile time when it is assigning a static value to a contiguous area of memory and turn that into a call to memset(). Both clang and GCC assume memset() is always available [2]. clang provides a builtin, __builtin_memset_inline [3][4], that can be used to ensure an external function call is not generated when zeroing this memory. Use it when it is available. While GCC has no issues with the current code, it has generated memset() before, as seen in commit b7bad082e113 ("random: vDSO: avoid call to out of line memset()"). GCC 14 provides '-finline-stringops=memset' [5][6] with a similar guarantee to the clang builtin, so use it and __builtin_memset() when available. If no option is available, provide a simple memset_inline() like the one from lib/string.c to avoid adding an ugly ifdef. Link: https://github.com/llvm/llvm-project/commit/90cebef1411617fc3eedd359bdf00cb44b1c2439 [1] Link: https://gcc.gnu.org/onlinedocs/gcc-16.2.0/gcc/Standards.html#index-ffreestanding [2] Link: https://github.com/llvm/llvm-project/commit/38637ee477541370a90b37f149069d8e5c0c2efd [3] Link: https://clang.llvm.org/docs/LanguageExtensions.html#guaranteed-inlined-memset [4] Link: https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=1ff6d9f7428b0668cd8ab0b3e3ab94f1d733124d [5] Link: https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#index-finline-stringops [6] Suggested-by: "Jason A. Donenfeld" <Jason@zx2c4.com> Suggested-by: Nick Desaulniers <ndesaulniers@google.com> Signed-off-by: Nathan Chancellor <nathan@kernel.org> --- Changes in v2: - Switch approach entirely (Jason, Nick) - clang: use __builtin_memset_inline() to avoid external call - GCC 14+: use __builtin_memset + -finline-stringops=memset (hence the massive CC list increase) - GCC < 14: no issues currently but avoid ifdef with simple memset implementation - Link to v1: https://patch.msgid.link/20260916-vdso-getrandom-avoid-memset-llvm-24-v1-1-80a92f2e225a@kernel.org --- arch/arm64/kernel/vdso/Makefile | 2 +- arch/loongarch/vdso/Makefile | 1 + arch/powerpc/kernel/vdso/Makefile | 1 + arch/riscv/kernel/vdso/Makefile | 1 + arch/s390/kernel/vdso/Makefile | 1 + arch/x86/entry/vdso/vdso64/Makefile | 2 +- init/Kconfig | 7 +++++++ lib/vdso/getrandom.c | 19 +++++++++++++++++-- 8 files changed, 30 insertions(+), 4 deletions(-)diff --git a/arch/arm64/kernel/vdso/Makefile b/arch/arm64/kernel/vdso/Makefile index 7dec05dd33b7..f6619e1cb2ce 100644 --- a/arch/arm64/kernel/vdso/Makefile +++ b/arch/arm64/kernel/vdso/Makefile@@ -41,7 +41,7 @@ CC_FLAGS_REMOVE_VDSO := $(CC_FLAGS_FTRACE) -Os $(CC_FLAGS_SCS) \ $(CC_FLAGS_LTO) $(CC_FLAGS_CFI) \ -Wmissing-prototypes -Wmissing-declarations -CC_FLAGS_ADD_VDSO := -O2 -mcmodel=tiny -fasynchronous-unwind-tables +CC_FLAGS_ADD_VDSO := -O2 -mcmodel=tiny -fasynchronous-unwind-tables $(CONFIG_CC_OPT_INLINE_MEMSET) CFLAGS_REMOVE_vgettimeofday.o = $(CC_FLAGS_REMOVE_VDSO) CFLAGS_REMOVE_vgetrandom.o = $(CC_FLAGS_REMOVE_VDSO)diff --git a/arch/loongarch/vdso/Makefile b/arch/loongarch/vdso/Makefile index 9c9181bb4071..0b84892d084b 100644 --- a/arch/loongarch/vdso/Makefile +++ b/arch/loongarch/vdso/Makefile@@ -16,6 +16,7 @@ ccflags-vdso := \ $(filter -m64,$(KBUILD_CFLAGS)) \ $(filter -march=%,$(KBUILD_CFLAGS)) \ $(filter -m%-float,$(KBUILD_CFLAGS)) \ + $(CONFIG_CC_OPT_INLINE_MEMSET) \ $(CLANG_FLAGS) \ -D__VDSO__diff --git a/arch/powerpc/kernel/vdso/Makefile b/arch/powerpc/kernel/vdso/Makefile index 368759f81708..0d1a49529885 100644 --- a/arch/powerpc/kernel/vdso/Makefile +++ b/arch/powerpc/kernel/vdso/Makefile@@ -43,6 +43,7 @@ ccflags-y := -fno-common -fno-builtin -DBUILD_VDSO ccflags-y += $(DISABLE_LATENT_ENTROPY_PLUGIN) ccflags-y += $(call cc-option, -fno-stack-protector) ccflags-y += -DDISABLE_BRANCH_PROFILING +ccflags-y += $(CONFIG_CC_OPT_INLINE_MEMSET) ccflags-y += -ffreestanding -fasynchronous-unwind-tables ccflags-remove-y := $(CC_FLAGS_FTRACE) ldflags-y := -Wl,--hash-style=both -nostdlib -shared -z noexecstack $(CLANG_FLAGS)diff --git a/arch/riscv/kernel/vdso/Makefile b/arch/riscv/kernel/vdso/Makefile index 8dbf2532a573..c023046a3fd7 100644 --- a/arch/riscv/kernel/vdso/Makefile +++ b/arch/riscv/kernel/vdso/Makefile@@ -36,6 +36,7 @@ endif ccflags-y := -fno-stack-protector ccflags-y += -DDISABLE_BRANCH_PROFILING ccflags-y += -fno-builtin +ccflags-y += $(CONFIG_CC_OPT_INLINE_MEMSET) ccflags-y += $(KBUILD_BASE_ISA)$(CFI_MARCH) ccflags-y += $(CFI_FULL) asflags-y += $(KBUILD_BASE_ISA)$(CFI_MARCH)diff --git a/arch/s390/kernel/vdso/Makefile b/arch/s390/kernel/vdso/Makefile index 35c834b895ec..54bcf2984ca1 100644 --- a/arch/s390/kernel/vdso/Makefile +++ b/arch/s390/kernel/vdso/Makefile@@ -29,6 +29,7 @@ KBUILD_CFLAGS_VDSO := $(filter-out -munaligned-symbols,$(KBUILD_CFLAGS_VDSO)) KBUILD_CFLAGS_VDSO := $(filter-out -fno-asynchronous-unwind-tables,$(KBUILD_CFLAGS_VDSO)) KBUILD_CFLAGS_VDSO += -fPIC -fno-common -fno-builtin -fasynchronous-unwind-tables KBUILD_CFLAGS_VDSO += -fno-stack-protector $(DISABLE_KSTACK_ERASE) +KBUILD_CFLAGS_VDSO += $(CONFIG_CC_OPT_INLINE_MEMSET) ldflags-y := -shared -soname=linux-vdso.so.1 \ --hash-style=both --build-id=sha1 \ $(call ld-option, --eh-frame-hdr) -Tdiff --git a/arch/x86/entry/vdso/vdso64/Makefile b/arch/x86/entry/vdso/vdso64/Makefile index 7c0790065b5e..c2353a065279 100644 --- a/arch/x86/entry/vdso/vdso64/Makefile +++ b/arch/x86/entry/vdso/vdso64/Makefile@@ -14,7 +14,7 @@ vobjs-$(CONFIG_X86_SGX) += vsgx.o vobjs-$(CONFIG_FUTEX_ROBUST_UNLOCK) += vfutex.o # Compilation flags -flags-y := -DBUILD_VDSO64 -m64 -mcmodel=small +flags-y := -DBUILD_VDSO64 -m64 -mcmodel=small $(CONFIG_CC_OPT_INLINE_MEMSET) # The location of this include matters! include $(src)/../common/Makefile.includediff --git a/init/Kconfig b/init/Kconfig index 8583d9f06c52..ff3f8475dd2f 100644 --- a/init/Kconfig +++ b/init/Kconfig@@ -173,6 +173,13 @@ config CC_HAS_ALLOC_TOKEN config CC_HAS_MULTIDIMENSIONAL_NONSTRING def_bool $(success,echo 'char tag[][4] __attribute__((__nonstring__)) = { };' | $(CC) $(CLANG_FLAGS) -x c - -c -o /dev/null -Werror) +config CC_HAS_OPT_INLINE_MEMSET + def_bool $(cc-option,-finline-stringops=memset) + +config CC_OPT_INLINE_MEMSET + string + default "-finline-stringops=memset" if CC_HAS_OPT_INLINE_MEMSET + config LD_CAN_USE_KEEP_IN_OVERLAY # ld.lld prior to 21.0.0 did not support KEEP within an overlay description # https://github.com/llvm/llvm-project/pull/130661diff --git a/lib/vdso/getrandom.c b/lib/vdso/getrandom.c index 2851afa9154f..f5cad5641985 100644 --- a/lib/vdso/getrandom.c +++ b/lib/vdso/getrandom.c@@ -29,6 +29,22 @@ } \ } while (0) +#if __has_builtin(__builtin_memset_inline) +#define memset_inline(dst, value, size) __builtin_memset_inline(dst, value, size) +#elif IS_ENABLED(CONFIG_CC_HAS_OPT_INLINE_MEMSET) +#define memset_inline(dst, value, size) __builtin_memset(dst, value, size) +#else +static inline void *memset_inline(void *dst, int value, size_t size) +{ + char *d = dst; + + while (size--) + *d++ = value; + + return d; +} +#endifDoes this work? https://godbolt.org/z/Tx6EaGMfbHmmm...possibly. https://godbolt.org/z/ec9rzd9Yf I don't get it...Just keep it under 2 pages, it will be fine: https://godbolt.org/z/qqoMqfxev
https://godbolt.org/z/9W9f5a7Wx Apparently not on RISCV, which is what prompted this patch in the first place. So I suspect Nathan's v2 here is the way to go. Jason