Re: [PATCH] LoongArch: vDSO: Emit GNU_EH_FRAME correctly
From: Jinyang He
Date: Wed Feb 25 2026 - 20:31:01 EST
On 2026-02-25 18:45, Xi Ruoyao wrote:
With -fno-asynchronous-unwind-tables and --no-eh-frame-hdr (the defaultCould we use `.cfi_return_column 0` here? Since I recommended
of the linker), the GNU_EH_FRAME segment (specified by vdso.lds.S) is
empty. This is not valid, as the current DWARF specification mandates
the first byte of the EH frame to be the version number 1. It causes
some unwinders to complain, for example the ClickHouse query profiler
spams the log with messages:
clickhouse-server[365854]: libunwind: unsupported .eh_frame_hdr
version: 127 at 7ffffffb0000
Here "127" is just the byte located at the p_vaddr (0, i.e. the
beginning of the vDSO) of the empty GNU_EH_FRAME segment.
Cross-checking with /proc/365854/maps has also proven 7ffffffb0000 is
the start of vDSO in the process VM image.
In LoongArch the -fno-asynchronous-unwind-tables option seems just a
MIPS legacy, and MIPS only uses this option to satisfy the MIPS-specific
"genvdso" program, per the commit cfd75c2db17e ("MIPS: VDSO: Explicitly
use -fno-asynchronous-unwind-tables"). IIUC it indicates some inherent
limitation of the MIPS ELF ABI and has nothing to do with LoongArch. So
we can simply flip it over to -fasynchronous-unwind-tables and pass
--eh-frame-hdr for linking the vDSO, allowing the profilers to unwind the
stack for statistics even if the sample point is taken when the PC is in
the vDSO.
However simply adjusting the options above would exploit an issue: when
the libgcc unwinder saw the invalid GNU_EH_FRAME segment, it silently
falled back to a machine-specific routine to match the code pattern of
rt_sigreturn and extract the registers saved in the sigframe if the code
pattern is matched. As unwinding from signal handlers is vital for
libgcc to support pthread cancellation etc., the fall-back routine had
been silently keeping the LoongArch Linux systems functioning since
Linux 5.19. But when we start to emit GNU_EH_FRAME with the correct
format, fall-back routine will no longer be used and libgcc will fail
to unwind the sigframe, and unwinding from signal handlers will no
longer work, causing dozens of glibc test failures. To make it possible
to unwind from signal handlers again, it's necessary to code the unwind
info in __vdso_rt_sigreturn via .cfi_* directives.
The offsets in the .cfi_* directives depend on the layout of struct
sigframe, notably the offset of sigcontect in the sigframe. To use the
offset in the assembly file, factor out struct sigframe into a header to
allow asm-offsets.c to output the offset for assembly.
To work around a long-term issue in the libgcc unwinder (the pc is
unconditionally substracted by 1: doing so is technically incorrect for
a signal frame), a nop instruction is included with the two real
instructions in __vdso_rt_sigreturn in the same FDE PC range. The same
hack has been used on x86 for a long time.
Fixes: c6b99bed6b8f ("LoongArch: Add VDSO and VSYSCALL support")
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Xi Ruoyao <xry111@xxxxxxxxxxx>
---
arch/loongarch/include/asm/sigframe.h | 22 ++++++++++++++
arch/loongarch/kernel/asm-offsets.c | 2 ++
arch/loongarch/kernel/signal.c | 6 +---
arch/loongarch/vdso/Makefile | 4 +--
arch/loongarch/vdso/sigreturn.S | 44 ++++++++++++++++++++++++---
5 files changed, 67 insertions(+), 11 deletions(-)
create mode 100644 arch/loongarch/include/asm/sigframe.h
diff --git a/arch/loongarch/include/asm/sigframe.h b/arch/loongarch/include/asm/sigframe.h
new file mode 100644
index 000000000000..6889bcf5dc88
--- /dev/null
+++ b/arch/loongarch/include/asm/sigframe.h
@@ -0,0 +1,22 @@
+/* SPDX-License-Identifier: GPL-2.0+ */
+/*
+ * Separated from arch/loongarch/kernel/signal.c:
+ *
+ * Author: Hanlu Li <lihanlu@xxxxxxxxxxx>
+ * Huacai Chen <chenhuacai@xxxxxxxxxxx>
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ *
+ * Derived from MIPS:
+ * Copyright (C) 1991, 1992 Linus Torvalds
+ * Copyright (C) 1994 - 2000 Ralf Baechle
+ * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
+ * Copyright (C) 2014, Imagination Technologies Ltd.
+ */
+
+#include <uapi/asm/ucontext.h>
+#include <asm/siginfo.h>
+
+struct rt_sigframe {
+ struct siginfo rs_info;
+ struct ucontext rs_uctx;
+};
diff --git a/arch/loongarch/kernel/asm-offsets.c b/arch/loongarch/kernel/asm-offsets.c
index 3017c7157600..2cc953f113ac 100644
--- a/arch/loongarch/kernel/asm-offsets.c
+++ b/arch/loongarch/kernel/asm-offsets.c
@@ -16,6 +16,7 @@
#include <asm/ptrace.h>
#include <asm/processor.h>
#include <asm/ftrace.h>
+#include <asm/sigframe.h>
#include <vdso/datapage.h>
static void __used output_ptreg_defines(void)
@@ -220,6 +221,7 @@ static void __used output_sc_defines(void)
COMMENT("Linux sigcontext offsets.");
OFFSET(SC_REGS, sigcontext, sc_regs);
OFFSET(SC_PC, sigcontext, sc_pc);
+ OFFSET(RT_SIGFRAME_SC, rt_sigframe, rs_uctx.uc_mcontext);
BLANK();
}
diff --git a/arch/loongarch/kernel/signal.c b/arch/loongarch/kernel/signal.c
index c9f7ca778364..e297d54ea638 100644
--- a/arch/loongarch/kernel/signal.c
+++ b/arch/loongarch/kernel/signal.c
@@ -37,6 +37,7 @@
#include <asm/lbt.h>
#include <asm/ucontext.h>
#include <asm/vdso.h>
+#include <asm/sigframe.h>
#ifdef DEBUG_SIG
# define DEBUGP(fmt, args...) printk("%s: " fmt, __func__, ##args)
@@ -51,11 +52,6 @@
#define lock_lbt_owner() ({ preempt_disable(); pagefault_disable(); })
#define unlock_lbt_owner() ({ pagefault_enable(); preempt_enable(); })
-struct rt_sigframe {
- struct siginfo rs_info;
- struct ucontext rs_uctx;
-};
-
struct _ctx_layout {
struct sctx_info *addr;
unsigned int size;
diff --git a/arch/loongarch/vdso/Makefile b/arch/loongarch/vdso/Makefile
index 520f1513f07d..294c16b9517f 100644
--- a/arch/loongarch/vdso/Makefile
+++ b/arch/loongarch/vdso/Makefile
@@ -26,7 +26,7 @@ cflags-vdso := $(ccflags-vdso) \
$(filter -W%,$(filter-out -Wa$(comma)%,$(KBUILD_CFLAGS))) \
-std=gnu11 -fms-extensions -O2 -g -fno-strict-aliasing -fno-common -fno-builtin \
-fno-stack-protector -fno-jump-tables -DDISABLE_BRANCH_PROFILING \
- $(call cc-option, -fno-asynchronous-unwind-tables) \
+ $(call cc-option, -fasynchronous-unwind-tables) \
$(call cc-option, -fno-stack-protector)
aflags-vdso := $(ccflags-vdso) \
-D__ASSEMBLY__ -Wa,-gdwarf-2
@@ -41,7 +41,7 @@ endif
# VDSO linker flags.
ldflags-y := -Bsymbolic --no-undefined -soname=linux-vdso.so.1 \
- $(filter -E%,$(KBUILD_CFLAGS)) -shared --build-id -T
+ $(filter -E%,$(KBUILD_CFLAGS)) -shared --build-id --eh-frame-hdr -T
#
# Shared build commands.
diff --git a/arch/loongarch/vdso/sigreturn.S b/arch/loongarch/vdso/sigreturn.S
index 9cb3c58fad03..e46c1deacb9e 100644
--- a/arch/loongarch/vdso/sigreturn.S
+++ b/arch/loongarch/vdso/sigreturn.S
@@ -12,13 +12,49 @@
#include <asm/regdef.h>
#include <asm/asm.h>
+#include <asm/asm-offsets.h>
.section .text
- .cfi_sections .debug_frame
-SYM_FUNC_START(__vdso_rt_sigreturn)
+ .cfi_startproc
+ .cfi_signal_frame
+ /*
+ * There is a struct rt_sigframe at $sp, set CFA to the address of
+ * the struct sigcontext in the rt_sigframe to simplify the
+ * offsets below.
+ */
+ .cfi_def_cfa 3, RT_SIGFRAME_SC
+
+ /*
+ * 72 is DWARF 2 CFA column for the return address from a signal
+ * handler context on LoongArch, i.e. the PC stored in the
+ * sigcontext.
+ */
llvm::libunwind that restore register to zero not destroy it
and we can use it as a hack, [1]. But I don't know whether the
libgcc works ok or not in this case.
Otherwise I think we should update [2] first.
Others look good to me.
[1] https://reviews.llvm.org/D137010#3907282
[2] https://github.com/loongson/la-abi-specs/blob/release/ladwarf.adoc
Thanks,
Jinyang
+ .cfi_return_column 72
+ .cfi_offset 72, SC_PC
+
+ /* The GPRs of the "caller" are also stored in the sigcontext. */
+.irp num, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, \
+ 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
+ .cfi_offset \num, SC_REGS + \num * SZREG
+.endr
+
+ /*
+ * HACK: The dwarf2 unwind routine will subtract 1 from the return
+ * address to get an address in the middle of the persumed call
+ * instruction. While in libgcc there exists a logic to avoid
+ * subtracting 1 for the signal frame (a frame with the 'S'
+ * augmentation that we've already added via .cfi_signal_frame),
+ * unfortunately it doesn't really work: the check of signal frame
+ * is at libgcc/unwind-dw2:1008 in GCC 15.2.0, but the flag it
+ * checks will only get updated by the extract_cie_info call at line
+ * 1025. So include a nop before the real start to make up for it.
+ * This is also the reason we don't use SYM_FUNC_START.
+ */
+ nop
+SYM_START(__vdso_rt_sigreturn, SYM_L_GLOBAL, SYM_A_ALIGN)
li.w a7, __NR_rt_sigreturn
syscall 0
-
-SYM_FUNC_END(__vdso_rt_sigreturn)
+ .cfi_endproc
+SYM_END(__vdso_rt_sigreturn, SYM_T_FUNC)