Re: [PATCH 1/4] x86: kmsan: Don't rename memintrinsics in uninstrumented files

From: Marco Elver
Date: Thu Mar 02 2023 - 10:13:53 EST


On Thu, 2 Mar 2023 at 15:28, Alexander Potapenko <glider@xxxxxxxxxx> wrote:
>
> On Thu, Mar 2, 2023 at 12:14 PM Marco Elver <elver@xxxxxxxxxx> wrote:
> >
> > On Wed, 1 Mar 2023 at 15:39, Alexander Potapenko <glider@xxxxxxxxxx> wrote:
> > >
> > > KMSAN should be overriding calls to memset/memcpy/memmove and their
> >
> > You mean that the compiler will override calls?
> > All supported compilers that have fsanitize=kernel-memory replace
> > memintrinsics with __msan_mem*() calls, right?
>
> Right. Changed to:
>
> KMSAN already replaces calls to to memset/memcpy/memmove and their
> __builtin_ versions with __msan_memset/__msan_memcpy/__msan_memmove in
> instrumented files, so there is no need to override them.

But it's not KMSAN - KMSAN is the combined end result of runtime and
compiler - in this case we need to be specific and point out it's the
compiler that's doing it. There is no code in the Linux kernel that
does this replacement.

>
> >
> > > __builtin_ versions in instrumented files, so there is no need to
> > > override them. In non-instrumented versions we are now required to
> > > leave memset() and friends intact, so we cannot replace them with
> > > __msan_XXX() functions.
> > >
> > > Cc: Kees Cook <keescook@xxxxxxxxxxxx>
> > > Suggested-by: Marco Elver <elver@xxxxxxxxxx>
> > > Signed-off-by: Alexander Potapenko <glider@xxxxxxxxxx>
> >
> > Other than that,
> >
> > Reviewed-by: Marco Elver <elver@xxxxxxxxxx>
> Thanks!