Hello,
On Tue, Jan 27, 2026 at 04:05:35PM +0100, Jens Remus wrote:
> This is the implementation of parsing the SFrame V3 stack trace information
> from an .sframe section in an ELF file. It's a continuation of Josh's and
> Steve's work that can be found here:
>
> https://lore.kernel.org/all/cover.1737511963.git.jpoimboe@kernel.org/
> https://lore.kernel.org/all/20250827201548.448472904@kernel.org/
>
> Currently the only way to get a user space stack trace from a stack
> walk (and not just copying large amount of user stack into the kernel
> ring buffer) is to use frame pointers. This has a few issues. The biggest
> one is that compiling frame pointers into every application and library
> has been shown to cause performance overhead.
>
> Another issue is that the format of the frames may not always be consistent
> between different compilers and some architectures (s390) has no defined
> format to do a reliable stack walk. The only way to perform user space
> profiling on these architectures is to copy the user stack into the kernel
> buffer.
>
> SFrame [1] is now supported in binutils (x86-64, ARM64, and s390). There is
> discussions going on about supporting SFrame in LLVM. SFrame acts more like
> ORC, and lives in the ELF executable file as its own section. Like ORC it
> has two tables where the first table is sorted by instruction pointers (IP)
> and using the current IP and finding it's entry in the first table, it will
> take you to the second table which will tell you where the return address
> of the current function is located and then you can use that address to
> look it up in the first table to find the return address of that function,
> and so on. This performs a user space stack walk.
>
> Now because the .sframe section lives in the ELF file it needs to be faulted
> into memory when it is used. This means that walking the user space stack
> requires being in a faultable context. As profilers like perf request a stack
> trace in interrupt or NMI context, it cannot do the walking when it is
> requested. Instead it must be deferred until it is safe to fault in user
> space. One place this is known to be safe is when the task is about to return
> back to user space.
>
> This series makes the deferred unwind user code implement SFrame format V3
> and enables it on x86-64.
>
> [1]: https://sourceware.org/binutils/wiki/sframe
>
>
> This series applies on top of the tip perf/core branch:
>
> git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git perf/core
>
> The to be stack-traced user space programs (and libraries) need to be
> built with the recent SFrame stack trace information format V3, as
> generated by the upcoming binutils 2.46 with assembler option --gsframe.
> It can be built from source from the binutils-2_46-branch branch:
>
> git://sourceware.org/git/binutils-gdb.git binutils-2_46-branch
>
> Namhyung Kim's related perf tools deferred callchain support can be used
> for testing ("perf record --call-graph fp,defer" and "perf report/script").
Is it possible for users to choose the unwinder - frame pointer or
SFrame at runtime? I feel like the option should be
"--call-graph sframe,defer" or just "--call-graph sframe" if it always
uses deferred unwinding.
Thanks,
Namhyung
>
>
> Changes since v12 (see patch notes for details):
> - Rebase on tip perf/core branch (d55c571e4333).
> - Add support for SFrame V3, including its new flexible FDEs. SFrame V2
> is not supported.
>
> Changes since v11 (see patch notes for details):
> - Rebase on tip master branch (f8fdee44bf2f) with Namhyung Kim's
> perf/defer-callchain-v4 branch merged on top.
> - Adjust to Peter's latest undwind user enhancements.
> - Simplify logic by using an internal SFrame FDE representation, whose
> FDE function start address field is an address instead of a PC-relative
> offset (from FDE).
> - Rename struct sframe_fre to sframe_fre_internal to align with
> struct sframe_fde_internal.
> - Remove unused pt_regs from unwind_user_next_common() and its
> callers. (Peter)
> - Simplify unwind_user_next_sframe(). (Peter)
> - Fix a few checkpatch errors and warnings.
> - Minor cleanups (e.g. move includes, fix indentation).
>
> Changes since v10:
> - Support for SFrame V2 PC-relative FDE function start address.
> - Support for SFrame V2 representing RA undefined as indication for
> outermost frames.
>
>
> Patches 1, 4, 11, and 17 have been updated to exclusively support the
> latest SFrame V3 stack trace information format, that is generated by
> the upcoming binutils 2.46 release. Old SFrame V2 sections get rejected
> with dynamic debug message "bad/unsupported sframe header".
>
> Patches 7 and 8 add support to unwind user (sframe) for outermost frames.
>
> Patches 12-15 add support to unwind user (sframe) for the new SFrame V3
> flexible FDEs.
>
> Patch 16 improves the performance of searching the SFrame FRE for an IP.
>
> Regards,
> Jens
>
>
> Jens Remus (7):
> unwind_user: Stop when reaching an outermost frame
> unwind_user/sframe: Add support for outermost frame indication
> unwind_user: Enable archs that pass RA in a register
> unwind_user: Flexible FP/RA recovery rules
> unwind_user: Flexible CFA recovery rules
> unwind_user/sframe: Add support for SFrame V3 flexible FDEs
> unwind_user/sframe: Separate reading of FRE from reading of FRE data
> words
>
> Josh Poimboeuf (11):
> unwind_user/sframe: Add support for reading .sframe headers
> unwind_user/sframe: Store .sframe section data in per-mm maple tree
> x86/uaccess: Add unsafe_copy_from_user() implementation
> unwind_user/sframe: Add support for reading .sframe contents
> unwind_user/sframe: Detect .sframe sections in executables
> unwind_user/sframe: Wire up unwind_user to sframe
> unwind_user/sframe: Remove .sframe section on detected corruption
> unwind_user/sframe: Show file name in debug output
> unwind_user/sframe: Add .sframe validation option
> unwind_user/sframe/x86: Enable sframe unwinding on x86
> unwind_user/sframe: Add prctl() interface for registering .sframe
> sections
>
> MAINTAINERS | 1 +
> arch/Kconfig | 23 +
> arch/x86/Kconfig | 1 +
> arch/x86/include/asm/mmu.h | 2 +-
> arch/x86/include/asm/uaccess.h | 39 +-
> arch/x86/include/asm/unwind_user.h | 69 +-
> arch/x86/include/asm/unwind_user_sframe.h | 12 +
> fs/binfmt_elf.c | 48 +-
> include/linux/mm_types.h | 3 +
> include/linux/sframe.h | 60 ++
> include/linux/unwind_user.h | 18 +
> include/linux/unwind_user_types.h | 46 +-
> include/uapi/linux/elf.h | 1 +
> include/uapi/linux/prctl.h | 6 +-
> kernel/fork.c | 10 +
> kernel/sys.c | 9 +
> kernel/unwind/Makefile | 3 +-
> kernel/unwind/sframe.c | 840 ++++++++++++++++++++++
> kernel/unwind/sframe.h | 87 +++
> kernel/unwind/sframe_debug.h | 68 ++
> kernel/unwind/user.c | 105 ++-
> mm/init-mm.c | 2 +
> 22 files changed, 1414 insertions(+), 39 deletions(-)
> create mode 100644 arch/x86/include/asm/unwind_user_sframe.h
> create mode 100644 include/linux/sframe.h
> create mode 100644 kernel/unwind/sframe.c
> create mode 100644 kernel/unwind/sframe.h
> create mode 100644 kernel/unwind/sframe_debug.h
>
> --
> 2.51.0
>