From: Josh Poimboeuf <jpoimboe@kernel.org>
Enable sframe generation in the VDSO library so kernel and user space
can unwind through it.
[ Jens Remus: Add support for SFrame V3. Prevent GNU_SFRAME program
table entry to empty .sframe section. ]
Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Signed-off-by: Jens Remus <jremus@linux.ibm.com>
---
Notes (jremus):
Changes in v7:
- Rebase on H. Peter Anvin's vDSO changes on tip:x86/entry.
- Simplify adding assembler option -Wa,--gsframe-3. Add for
vdso64 only.
- Align to .eh_frame and mark .sframe as KEEP in vDSO linker script.
Note that GNU linker 2.46 will mark .sframe as KEEP in its default
linker script as well.
arch/Kconfig | 7 +++++++
arch/x86/entry/vdso/common/vdso-layout.lds.S | 11 +++++++++++
arch/x86/entry/vdso/vdso64/Makefile | 1 +
arch/x86/include/asm/dwarf2.h | 5 ++++-
4 files changed, 23 insertions(+), 1 deletion(-)
diff --git a/arch/Kconfig b/arch/Kconfig
index 31220f512b16..8170e492a44c 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -479,6 +479,13 @@ config HAVE_HARDLOCKUP_DETECTOR_ARCH
It uses the same command line parameters, and sysctl interface,
as the generic hardlockup detectors.
+config AS_SFRAME
+ bool
+
+config AS_SFRAME3
+ def_bool $(as-instr,.cfi_startproc\n.cfi_endproc,-Wa$(comma)--gsframe-3)
+ select AS_SFRAME
+
config UNWIND_USER
bool
diff --git a/arch/x86/entry/vdso/common/vdso-layout.lds.S b/arch/x86/entry/vdso/common/vdso-layout.lds.S
index 856b8b9d278c..421eb4ca9131 100644
--- a/arch/x86/entry/vdso/common/vdso-layout.lds.S
+++ b/arch/x86/entry/vdso/common/vdso-layout.lds.S
@@ -60,6 +60,13 @@ SECTIONS
*(.eh_frame.*)
} :text
+#ifdef CONFIG_AS_SFRAME
+ .sframe : {
+ KEEP (*(.sframe))
+ *(.sframe.*)
+ } :text :sframe
+#endif
+
/*
* Text is well-separated from actual data: there's plenty of
* stuff that isn't used at runtime in between.
@@ -89,6 +96,7 @@ SECTIONS
#define PT_GNU_EH_FRAME 0x6474e550
#define PT_GNU_STACK 0x6474e551
#define PT_GNU_PROPERTY 0x6474e553
+#define PT_GNU_SFRAME 0x6474e554
/*
* We must supply the ELF program headers explicitly to get just one
@@ -104,6 +112,9 @@ PHDRS
dynamic PT_DYNAMIC PF_R;
note PT_NOTE PF_R;
eh_frame_hdr PT_GNU_EH_FRAME PF_R;
+#ifdef CONFIG_AS_SFRAME
+ sframe PT_GNU_SFRAME PF_R;
+#endif
gnu_stack PT_GNU_STACK PF_RW;
gnu_property PT_GNU_PROPERTY PF_R;
}
diff --git a/arch/x86/entry/vdso/vdso64/Makefile b/arch/x86/entry/vdso/vdso64/Makefile
index bfffaf1aeecc..459f8026531e 100644
--- a/arch/x86/entry/vdso/vdso64/Makefile
+++ b/arch/x86/entry/vdso/vdso64/Makefile
@@ -14,6 +14,7 @@ vobjs-$(CONFIG_X86_SGX) += vsgx.o
# Compilation flags
flags-y := -DBUILD_VDSO64 -m64 -mcmodel=small
+flags-$(CONFIG_AS_SFRAME3) += -Wa,--gsframe-3
# The location of this include matters!
include $(src)/../common/Makefile.include
diff --git a/arch/x86/include/asm/dwarf2.h b/arch/x86/include/asm/dwarf2.h
index 7cc30500c095..90b42367815e 100644
--- a/arch/x86/include/asm/dwarf2.h
+++ b/arch/x86/include/asm/dwarf2.h
@@ -12,8 +12,11 @@
* For the vDSO, emit both runtime unwind information and debug
* symbols for the .dbg file.
*/
-
+#if defined(__x86_64__) && defined(CONFIG_AS_SFRAME)
+ .cfi_sections .eh_frame, .debug_frame, .sframe
+#else
.cfi_sections .eh_frame, .debug_frame
+#endif
#define CFI_STARTPROC .cfi_startproc
#define CFI_ENDPROC .cfi_endproc
--
2.51.0On 2026-02-03 09:19, Jens Remus wrote: > - > +#if defined(__x86_64__) && defined(CONFIG_AS_SFRAME) > + .cfi_sections .eh_frame, .debug_frame, .sframe > +#else > .cfi_sections .eh_frame, .debug_frame > +#endif > It would be better to: #undef CONFIG_AS_SFRAME /* i386 doesn't support .sframe */ in fake_32bit_build.h. -hpa
On 2/4/2026 1:49 AM, H. Peter Anvin wrote: > On 2026-02-03 09:19, Jens Remus wrote: >> - >> +#if defined(__x86_64__) && defined(CONFIG_AS_SFRAME) My understanding is that you would like above changed to: #ifdef CONFIG_AS_SFRAME >> + .cfi_sections .eh_frame, .debug_frame, .sframe >> +#else >> .cfi_sections .eh_frame, .debug_frame >> +#endif >> > > It would be better to: > > #undef CONFIG_AS_SFRAME /* i386 doesn't support .sframe */ > > in fake_32bit_build.h. Due to (binutils 2.46 pre-release) GNU assembler supporting --gsframe-3, but only for x86-64 not for x86-32 (nor x32) CONFIG_AS_SFRAME is enabled even for my i386 cross build attempt. This then wrongly causes the .cfi_sections with .sframe to be selected (verified by adding an #error for testing). IIUC the .cfi_sections is there to have DWARF in both .eh_frame and .debug_frame. For .sframe it basically has the same effect as specifying the common assembler option --gsframe. Given the more specific option --gsframe-3 is explicitly specified I think it would be better to drop the changes to both dwarf2.h and fake_32bit_build.h and solely rely on --gsframe-3. Josh, Indu, do you agree? Thanks and rRegards, Jens -- Jens Remus Linux on Z Development (D3303) jremus@de.ibm.com / jremus@linux.ibm.com IBM Deutschland Research & Development GmbH; Vorsitzender des Aufsichtsrats: Wolfgang Wendt; Geschäftsführung: David Faller; Sitz der Gesellschaft: Ehningen; Registergericht: Amtsgericht Stuttgart, HRB 243294 IBM Data Privacy Statement: https://www.ibm.com/privacy/
On February 6, 2026 9:48:24 AM PST, Jens Remus <jremus@linux.ibm.com> wrote: >On 2/4/2026 1:49 AM, H. Peter Anvin wrote: >> On 2026-02-03 09:19, Jens Remus wrote: >>> - >>> +#if defined(__x86_64__) && defined(CONFIG_AS_SFRAME) > >My understanding is that you would like above changed to: > >#ifdef CONFIG_AS_SFRAME > >>> + .cfi_sections .eh_frame, .debug_frame, .sframe >>> +#else >>> .cfi_sections .eh_frame, .debug_frame >>> +#endif >>> >> >> It would be better to: >> >> #undef CONFIG_AS_SFRAME /* i386 doesn't support .sframe */ >> >> in fake_32bit_build.h. >Due to (binutils 2.46 pre-release) GNU assembler supporting --gsframe-3, >but only for x86-64 not for x86-32 (nor x32) CONFIG_AS_SFRAME is enabled >even for my i386 cross build attempt. This then wrongly causes the >.cfi_sections with .sframe to be selected (verified by adding an #error >for testing). > >IIUC the .cfi_sections is there to have DWARF in both .eh_frame and >.debug_frame. For .sframe it basically has the same effect as >specifying the common assembler option --gsframe. Given the more >specific option --gsframe-3 is explicitly specified I think it would >be better to drop the changes to both dwarf2.h and fake_32bit_build.h >and solely rely on --gsframe-3. > >Josh, Indu, do you agree? > >Thanks and rRegards, >Jens Seems a lot cleaner to me, too, assuming it actually works.
On Fri, Feb 06, 2026 at 03:56:59PM -0800, H. Peter Anvin wrote: > On February 6, 2026 9:48:24 AM PST, Jens Remus <jremus@linux.ibm.com> wrote: > >On 2/4/2026 1:49 AM, H. Peter Anvin wrote: > >> On 2026-02-03 09:19, Jens Remus wrote: > >>> - > >>> +#if defined(__x86_64__) && defined(CONFIG_AS_SFRAME) > > > >My understanding is that you would like above changed to: > > > >#ifdef CONFIG_AS_SFRAME > > > >>> + .cfi_sections .eh_frame, .debug_frame, .sframe > >>> +#else > >>> .cfi_sections .eh_frame, .debug_frame > >>> +#endif > >>> > >> > >> It would be better to: > >> > >> #undef CONFIG_AS_SFRAME /* i386 doesn't support .sframe */ > >> > >> in fake_32bit_build.h. > >Due to (binutils 2.46 pre-release) GNU assembler supporting --gsframe-3, > >but only for x86-64 not for x86-32 (nor x32) CONFIG_AS_SFRAME is enabled > >even for my i386 cross build attempt. This then wrongly causes the > >.cfi_sections with .sframe to be selected (verified by adding an #error > >for testing). > > > >IIUC the .cfi_sections is there to have DWARF in both .eh_frame and > >.debug_frame. For .sframe it basically has the same effect as > >specifying the common assembler option --gsframe. Given the more > >specific option --gsframe-3 is explicitly specified I think it would > >be better to drop the changes to both dwarf2.h and fake_32bit_build.h > >and solely rely on --gsframe-3. > > > >Josh, Indu, do you agree? > > > >Thanks and rRegards, > >Jens > > Seems a lot cleaner to me, too, assuming it actually works. +1 -- Josh
© 2016 - 2026 Red Hat, Inc.