The Linux Kernel Mailing List
 help / color / mirror / Atom feed
* [PATCH] x86/cfi: Use symmetric SYM_START and SYM_END in __CFI_TYPE()
@ 2026-06-11 15:57 Jens Remus
  2026-06-11 19:28 ` Peter Zijlstra
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Jens Remus @ 2026-06-11 15:57 UTC (permalink / raw)
  To: Sami Tolvanen, Kees Cook, Peter Zijlstra, Thomas Gleixner,
	Ingo Molnar, Borislav Petkov, Dave Hansen, x86, H. Peter Anvin,
	Nathan Chancellor, Nick Desaulniers, Bill Wendling, Justin Stitt,
	linux-kernel, llvm
  Cc: Jens Remus, Heiko Carstens, Sashiko

Commit ccace936eec7 ("x86: Add types to indirectly called assembly
functions") introduced a x86-specific implementation of __CFI_TYPE()
using an asymmetric combination of SYM_START() and SYM_FUNC_END() to
add a symbol to the KCFI type identifier that precedes a function.

This asymmetric combination is an issue if SYM_FUNC_END() ever gets
extended in a way that requires it to be used symmetrically with
SYM_FUNC_START*().  For instance to emit DWARF CFI directives that
denote the start/end of a function. [1]

Use SYM_END() with SYM_T_FUNC instead.  No functional change, as the
generic implementation of SYM_FUNC_END(name) expands into
SYM_END(name, SYM_T_FUNC).

Reported-by: Sashiko <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/patchset/20260522110427.2816637-1-jremus@linux.ibm.com?part=3 [1]
Signed-off-by: Jens Remus <jremus@linux.ibm.com>
---

Notes (jremus):
    This patch applies on top of linus' tree (9716c086c8e8):
    
      git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git  master
    
    I verified as follows that in x86-64 builds of vmlinux using Clang with
    kCFI enabled without and with my patch applied vmlinux.o only differs in
    relocations targeting .rodata.str* and a few symbols in .rodata shifted
    (both likely due to differences in string merging):
    
    $ objdump -d vmlinux.o > vmlinux.o.{old|new}.objdump
    $ readelf -Wa vmlinux.o > vmlinux.o.{old|new}.readelf
    $ diff -u vmlinux.o.old.objdump vmlinux.o.new.objdump
    [no differences]
    $ diff -u0 vmlinux.o.old.readelf vmlinux.o.new.readelf | \
      grep --invert-match -E "\.rodata\.str|@@"
    [see above]

 arch/x86/include/asm/linkage.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/include/asm/linkage.h b/arch/x86/include/asm/linkage.h
index a7294656ad90..c9769a7b6e66 100644
--- a/arch/x86/include/asm/linkage.h
+++ b/arch/x86/include/asm/linkage.h
@@ -103,7 +103,7 @@
 	.byte 0xb8 ASM_NL					\
 	.long __kcfi_typeid_##name ASM_NL			\
 	CFI_POST_PADDING					\
-	SYM_FUNC_END(__cfi_##name)
+	SYM_END(__cfi_##name, SYM_T_FUNC)
 
 /* UML needs to be able to override memcpy() and friends for KASAN. */
 #ifdef CONFIG_UML
-- 
2.53.0


^ permalink raw reply related	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2026-06-28 23:14 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-06-11 15:57 [PATCH] x86/cfi: Use symmetric SYM_START and SYM_END in __CFI_TYPE() Jens Remus
2026-06-11 19:28 ` Peter Zijlstra
2026-06-13 16:29 ` Nathan Chancellor
2026-06-28 23:14 ` [tip: x86/cleanups] " tip-bot2 for Jens Remus

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox