[PATCH V6 04/11] compiler_types.h: Add __noinstr_section() for noinstr

[Mark Rutland]

On Thu, Oct 20, 2022 at 08:29:28PM +0800, Guo Ren wrote:
> Hi Mark and Lai,
> 
> On Thu, Oct 20, 2022 at 5:15 PM Mark Rutland <mark.rutland at arm.com> wrote:
> >
> > On Sat, Oct 08, 2022 at 09:54:39AM +0800, Guo Ren wrote:
> > > On Mon, Oct 3, 2022 at 7:39 PM Mark Rutland <mark.rutland at arm.com> wrote:
> > > >
> > > > On Sat, Oct 01, 2022 at 09:24:44PM -0400, guoren at kernel.org wrote:
> > > > > From: Lai Jiangshan <laijs at linux.alibaba.com>
> > > > >
> > > > > And it will be extended for C entry code.
> > > > >
> > > > > Cc: Borislav Petkov <bp at alien8.de>
> > > > > Reviewed-by: Miguel Ojeda <ojeda at kernel.org>
> > > > > Reviewed-by: Kees Cook <keescook at chromium.org>
> > > > > Suggested-by: Nick Desaulniers <ndesaulniers at google.com>
> > > > > Suggested-by: Peter Zijlstra <peterz at infradead.org>
> > > > > Signed-off-by: Lai Jiangshan <laijs at linux.alibaba.com>
> > > > > ---
> > > > >  include/linux/compiler_types.h | 8 +++++---
> > > > >  1 file changed, 5 insertions(+), 3 deletions(-)
> > > > >
> > > > > diff --git a/include/linux/compiler_types.h b/include/linux/compiler_types.h
> > > > > index 4f2a819fd60a..e9ce11ea4d8b 100644
> > > > > --- a/include/linux/compiler_types.h
> > > > > +++ b/include/linux/compiler_types.h
> > > > > @@ -227,9 +227,11 @@ struct ftrace_likely_data {
> > > > >  #endif
> > > > >
> > > > >  /* Section for code which can't be instrumented at all */
> > > > > -#define noinstr                                                              \
> > > > > -     noinline notrace __attribute((__section__(".noinstr.text")))    \
> > > > > -     __no_kcsan __no_sanitize_address __no_profile __no_sanitize_coverage
> > > > > +#define __noinstr_section(section)                           \
> > > > > +     noinline notrace __section(section) __no_profile        \
> > > > > +     __no_kcsan __no_sanitize_address __no_sanitize_coverage
> > > > > +
> > > > > +#define noinstr __noinstr_section(".noinstr.text")
> > > >
> > > > One thing proably worth noting here is that while KPROBES will avoid
> > > > instrumenting `.noinstr.text`, that won't happen automatically for other
> > > > __noinstr_section() sections, and that will need to be inhibited through other
> > > > means (e.g. the kprobes blacklist, explicit NOKPROBE_SYMBOL() annotation, or
> > > > otherwise).
> > >
> > > In riscv, "we select HAVE_KPROBES if !XIP_KERNEL", so don't worry
> > > about that. I don't think we could enable kprobe for XIP_KERNEL in the
> > > future.
> >
> > Sure; but someone else might use __noinstr_section() elsewhere where this could
> > matter, and I was suggesting that we could add a comment as above.
> Okay, how about:
> 
> /* Be care KPROBES will avoid instrumenting .noinstr.text, not
> __noinstr_section(). */
> #define __noinstr_section(section)                             \
>        noinline notrace __section(section) __no_profile        \
>        __no_kcsan __no_sanitize_address __no_sanitize_coverage

How about we split this like:

| /*
|  * Prevent the compiler from instrumenting this code in any way
|  * This does not prevent instrumentation via KPROBES, which must be
|  * prevented through other means if necessary.
|  */
| #define __no_compiler_instrument				\
| 	noinline notrace noinline notrace __no_kcsan		\
| 	__no_sanitize_address __no_sanitize_coverage
| 
| /* 
|  * Section for code which can't be instrumented at all.
|  * Any code in this section cannot be instrumented with KPROBES.
|  */
| #define noinstr __no_compiler_instrument section(".noinstr.text")

... then we don't need __noinstr_section(), and IMO the split is
clearer.

Peter?

Thanks,
Mark.