Hi Dylan and Mostafa!
On 5/18/2026 7:55 PM, Dylan Hatch wrote:
> On Fri, May 15, 2026 at 4:32 AM Mostafa Saleh <[email protected]> wrote:
>> On Tue, Apr 28, 2026 at 06:36:35PM +0000, Dylan Hatch wrote:
>>> Implement a generic kernel sframe-based [1] unwinder. The main goal is
>>> to improve reliable stacktrace on arm64 by unwinding across exception
>>> boundaries.
>>>
>>> On x86, the ORC unwinder provides reliable stacktrace through similar
>>> methodology, but arm64 lacks the necessary support from objtool to
>>> create ORC unwind tables.
>>>
>>> Currently, there's already a sframe unwinder proposed for userspace: [2].
>>> To maintain common definitions and algorithms for sframe lookup, a
>>> substantial portion of this patch series aims to refactor the sframe
>>> lookup code to support both kernel and userspace sframe sections.
>>>
>>> Currently, only GNU Binutils support sframe. This series relies on the
>>> Sframe V3 format, which is supported in binutils 2.46.
>>>
>>> These patches are based on Steven Rostedt's sframe/core branch [3],
>>> which is and aggregation of existing work done for x86 sframe userspace
>>> unwind, and contains [2]. This branch is, in turn, based on Linux
>>> v7.0-rc3. This full series (applied to the sframe/core branch) is
>>> available on github: [4].
>>>
>>
>> Not sure if related, but after updating my toolchain
>> (aarch64-linux-gnu-gcc (Debian 15.2.0-4) 15.2.0), I hit link errors:
>> ld.lld: error: arch/arm64/kernel/vdso/vgettimeofday.o:(.sframe) is being
>> placed in '.sframe'
>> ld.lld: error: arch/arm64/kernel/vdso/vgetrandom.o:(.sframe) is being placed
>> in '.sframe`
>
> Previously when developing against the SFrame V2 format, I had fixed
> these warnings with the VDSO Makefile change currently in this series:
>
> diff --git a/arch/arm64/kernel/vdso/Makefile b/arch/arm64/kernel/vdso/Makefile
> index 7dec05dd33b7..c60ef921956f 100644
> --- a/arch/arm64/kernel/vdso/Makefile
> +++ b/arch/arm64/kernel/vdso/Makefile
> @@ -38,7 +38,7 @@ ccflags-y += -DDISABLE_BRANCH_PROFILING -DBUILD_VDSO
> CC_FLAGS_REMOVE_VDSO := $(CC_FLAGS_FTRACE) -Os $(CC_FLAGS_SCS) \
> $(RANDSTRUCT_CFLAGS) $(KSTACK_ERASE_CFLAGS) \
> $(GCC_PLUGINS_CFLAGS) \
> - $(CC_FLAGS_LTO) $(CC_FLAGS_CFI) \
> + $(CC_FLAGS_LTO) $(CC_FLAGS_CFI) $(CC_FLAGS_SFRAME) \
> -Wmissing-prototypes -Wmissing-declarations
>
> CC_FLAGS_ADD_VDSO := -O2 -mcmodel=tiny -fasynchronous-unwind-tables
>
> But the warnings seem to have returned after upgrading my toolchain,
> possibly due to SFrame V3 or some confounding change in GCC. The
> --gsframe in the assembler should be set to 'no' by default, so
> perhaps GCC is providing an override --gsframe internally?
Could it be that your build of binutils was configured with
--enable-default-sframe, so that the GNU assembler defaults to generate
.sframe? AFAIK this configure option was meant for distributors and
package maintainers.
You can check as follows whether --gsframe defaults to "no" or "yes":
$ as --help | grep -A1 gsframe
--gsframe[={no|yes}] whether to generate SFrame stack trace information
(default: no)
...
>
>>
>> I applied this series hoping that fix it, but it doesn't, so far I
>> have this hack :
>> diff --git a/arch/arm64/kernel/vdso/vdso.lds.S
>> b/arch/arm64/kernel/vdso/vdso.lds.S
>> index 52314be29191..53bdf757ee44 100644
>> --- a/arch/arm64/kernel/vdso/vdso.lds.S
>> +++ b/arch/arm64/kernel/vdso/vdso.lds.S
>> @@ -77,7 +77,7 @@ SECTIONS
>> /DISCARD/ : {
>> *(.data .data.* .gnu.linkonce.d.* .sdata*)
>> *(.bss .sbss .dynbss .dynsbss)
>> - *(.eh_frame .eh_frame_hdr)
>> + *(.eh_frame .eh_frame_hdr .sframe)
>> }
>> }
>>
>> diff --git a/include/asm-generic/vmlinux.lds.h
>> b/include/asm-generic/vmlinux.lds.h
>> index 60c8c22fd3e4..759903acd6fc 100644
>> --- a/include/asm-generic/vmlinux.lds.h
>> +++ b/include/asm-generic/vmlinux.lds.h
>> @@ -1064,6 +1064,7 @@
>> /* ld.bfd warns about .gnu.version* even when not emitted */ \
>> *(.gnu.version*) \
>> *(__tracepoint_check) \
>> + *(.sframe) \
>>
>> #define DISCARDS \
>> /DISCARD/ : { \
>
> Since this series only handles kernel stacktrace, I believe it's
> better to omit the .sframe section entirely in the case where only
> ARCH_SUPPORTS_UNWIND_KERNEL_SFRAME is enabled. I think this hack may
> work better for this purpose:
>
> diff --git a/arch/arm64/kernel/vdso/Makefile b/arch/arm64/kernel/vdso/Makefile
> index c60ef921956f..29f802bfedb1 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) $(CC_FLAGS_SFRAME) \
> -Wmissing-prototypes -Wmissing-declarations
>
> -CC_FLAGS_ADD_VDSO := -O2 -mcmodel=tiny -fasynchronous-unwind-tables
> +CC_FLAGS_ADD_VDSO := -O2 -mcmodel=tiny -fasynchronous-unwind-tables
> -Wa,--gsframe=no
>
> CFLAGS_REMOVE_vgettimeofday.o = $(CC_FLAGS_REMOVE_VDSO)
> CFLAGS_REMOVE_vgetrandom.o = $(CC_FLAGS_REMOVE_VDSO)
>
> Though, I don't understand why it is necessary to provide --gsframe=no
> explicitly. If this approach seems ok to other folks/maintainers, I
> can fold this into my series.
Maybe build the VDSO separately with V=1 to see what assembler/compiler
options are effectively used (e.g. for vgettimeofday.o and vgetrandom.o
mentioned in the linker error message above)?
$ make mrproper
$ make defconfig
$ ./scripts/config --enable HAVE_UNWIND_KERNEL_SFRAME # enable kernel sframe
unwinder
$ make arch/arm64/kernel/vdso/ V=1
>
> On the topic of SFrame for VDSO, Jens has a patch adding support for
> this as part of a series to support userspace SFrame unwinding for
> arm64:
>
> https://lore.kernel.org/lkml/[email protected]/
Any feedback is very welcome. :-)
Regards,
Jens
--
Jens Remus
Linux on Z Development (D3303)
[email protected] / [email protected]
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/
