On Wed, May 13, 2020 at 03:58:30PM +0200, Marco Elver wrote:
> On Wed, 13 May 2020 at 15:24, Peter Zijlstra <peterz@infradead.org> wrote:
> >
> > On Wed, May 13, 2020 at 03:15:55PM +0200, Marco Elver wrote:
> > > So far so good, except: both __no_sanitize_or_inline and
> > > __no_kcsan_or_inline *do* avoid KCSAN instrumenting plain accesses, it
> > > just doesn't avoid explicit kcsan_check calls, like those in
> > > READ/WRITE_ONCE if KCSAN is enabled for the compilation unit. That's
> > > just because macros won't be redefined just for __no_sanitize
> > > functions. Similarly, READ_ONCE_NOCHECK does work as expected, and its
> > > access is unchecked.
> > >
> > > This will have the expected result:
> > > __no_sanitize_or_inline void foo(void) { x++; } // no data races reported
> > >
> > > This will not work as expected:
> > > __no_sanitize_or_inline void foo(void) { READ_ONCE(x); }  // data
> > > races are reported
> > >
> > > All this could be fixed if GCC devs would finally take my patch to
> > > make -fsanitize=thread distinguish volatile [1], but then we have to
> > > wait ~years for the new compilers to reach us. So please don't hold
> > > your breath for this one any time soon.
> > > [1] https://gcc.gnu.org/pipermail/gcc-patches/2020-April/544452.html
> >
> > Right, but that does not address the much larger issue of the attribute
> > vs inline tranwreck :/
> 
> Could you check if Clang is equally broken for you? I think GCC and
> Clang have differing behaviour on this. No idea what it takes to fix
> GCC though.

So I have some good and some maybe not so good news.

Given the patch below (on top of tglx's entry-v5-the-rest tag); I did
find that I could actually build alternative.o for gcc-{8,9,10} and
indeed clang-10. Any earlier gcc (I tried, 5,6,7) does not build:

../arch/x86/include/asm/ptrace.h:126:28: error: inlining failed in call to always_inline ‘user_mode’: function attribute mismatch

I dumped the poke_int3_handler output using objdump, find the attached
files.

It looks like clang-10 doesn't want to turn UBSAN off :/ The GCC files
look OK, no funny calls in those.

(the config has KASAN/UBSAN on, it looks like KCSAN and KASAN are
mutually exclusive)

---

diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
index 77c83833d91e..06d8db612efc 100644
--- a/arch/x86/kernel/alternative.c
+++ b/arch/x86/kernel/alternative.c
@@ -990,7 +990,7 @@ static __always_inline int patch_cmp(const void *key, const void *elt)
 	return 0;
 }
 
-int noinstr poke_int3_handler(struct pt_regs *regs)
+int noinstr __no_kcsan __no_sanitize_address __no_sanitize_undefined poke_int3_handler(struct pt_regs *regs)
 {
 	struct bp_patching_desc *desc;
 	struct text_poke_loc *tp;
diff --git a/include/linux/compiler-clang.h b/include/linux/compiler-clang.h
index 2cb42d8bdedc..5e83aada6553 100644
--- a/include/linux/compiler-clang.h
+++ b/include/linux/compiler-clang.h
@@ -15,6 +15,13 @@
 /* all clang versions usable with the kernel support KASAN ABI version 5 */
 #define KASAN_ABI_VERSION 5
 
+#if __has_feature(undefined_sanitizer)
+#define __no_sanitize_undefined \
+               __attribute__((no_sanitize("undefined")))
+#else
+#define __no_sanitize_undefined
+#endif
+
 #if __has_feature(address_sanitizer) || __has_feature(hwaddress_sanitizer)
 /* Emulate GCC's __SANITIZE_ADDRESS__ flag */
 #define __SANITIZE_ADDRESS__
diff --git a/include/linux/compiler-gcc.h b/include/linux/compiler-gcc.h
index 7dd4e0349ef3..8196a121a78e 100644
--- a/include/linux/compiler-gcc.h
+++ b/include/linux/compiler-gcc.h
@@ -138,6 +138,12 @@
 #define KASAN_ABI_VERSION 3
 #endif
 
+#if __has_attribute(__no_sanitize_undefined__)
+#define __no_sanitize_undefined __attribute__((no_sanitize_undefined))
+#else
+#define __no_sanitize_undefined
+#endif
+
 #if __has_attribute(__no_sanitize_address__)
 #define __no_sanitize_address __attribute__((no_sanitize_address))
 #else