Re: [PATCH] kasan: fix kasan_check_read/write definitions

From: Dmitry Vyukov <dvyukov@google.com>
To: Arnd Bergmann <arnd@arndb.de>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	Andrey Ryabinin <aryabinin@virtuozzo.com>,
	anders.roxell@linaro.org,
	Ard Biesheuvel <ard.biesheuvel@linaro.org>,
	Will Deacon <will.deacon@arm.com>,
	Mark Rutland <mark.rutland@arm.com>,
	Alexander Potapenko <glider@google.com>,
	Andrey Konovalov <andreyknvl@google.com>,
	Stephen Rothwell <sfr@canb.auug.org.au>,
	kasan-dev <kasan-dev@googlegroups.com>,
	LKML <linux-kernel@vger.kernel.org>,
	Linux-MM <linux-mm@kvack.org>
Subject: Re: [PATCH] kasan: fix kasan_check_read/write definitions
Date: Tue, 11 Dec 2018 14:44:50 +0100	[thread overview]
Message-ID: <CACT4Y+bRzY9hO5b=TjHeXTsVVO1z3eBOHz6oLgBhVR4OSm1d1w@mail.gmail.com> (raw)
In-Reply-To: <20181211133453.2835077-1-arnd@arndb.de>

On Tue, Dec 11, 2018 at 2:35 PM Arnd Bergmann <arnd@arndb.de> wrote:
>
> Building little-endian allmodconfig kernels on arm64 started failing
> with the generated atomic.h implementation, since we now try to call
> kasan helpers from the EFI stub:
>
> aarch64-linux-gnu-ld: drivers/firmware/efi/libstub/arm-stub.stub.o: in function `atomic_set':
> include/generated/atomic-instrumented.h:44: undefined reference to `__efistub_kasan_check_write'
>
> I suspect that we get similar problems in other files that explicitly
> disable KASAN for some reason but call atomic_t based helper functions.
>
> We can fix this by checking the predefined __SANITIZE_ADDRESS__ macro
> that the compiler sets instead of checking CONFIG_KASAN, but this in turn
> requires a small hack in mm/kasan/common.c so we do see the extern
> declaration there instead of the inline function.

Alexander, I think you are doing a similar thing for similar reasons
in KMSAN patch (see KMSAN_CHECK_ATOMIC_PARAMS):
https://github.com/google/kmsan/commit/17ebbfe19624c84adf79b0e5a74fd258c49ff12b
Namely, non-KMSAN-instrumented files must not get KMSAN callbacks from
atomics too.

Arnd patch does it the other way around: non-instrumented files need
to opt-in instead of opt-out.
Let's settle on a common way to do this, so that we can use it
consistently across all tools.

> Fixes: b1864b828644 ("locking/atomics: build atomic headers as required")
> Reported-by: Anders Roxell <anders.roxell@linaro.org>
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  include/linux/kasan-checks.h | 2 +-
>  mm/kasan/common.c            | 2 ++
>  2 files changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/include/linux/kasan-checks.h b/include/linux/kasan-checks.h
> index d314150658a4..a61dc075e2ce 100644
> --- a/include/linux/kasan-checks.h
> +++ b/include/linux/kasan-checks.h
> @@ -2,7 +2,7 @@
>  #ifndef _LINUX_KASAN_CHECKS_H
>  #define _LINUX_KASAN_CHECKS_H
>
> -#ifdef CONFIG_KASAN
> +#if defined(__SANITIZE_ADDRESS__) || defined(__KASAN_INTERNAL)
>  void kasan_check_read(const volatile void *p, unsigned int size);
>  void kasan_check_write(const volatile void *p, unsigned int size);
>  #else
> diff --git a/mm/kasan/common.c b/mm/kasan/common.c
> index 03d5d1374ca7..51a7932c33a3 100644
> --- a/mm/kasan/common.c
> +++ b/mm/kasan/common.c
> @@ -14,6 +14,8 @@
>   *
>   */
>
> +#define __KASAN_INTERNAL
> +
>  #include <linux/export.h>
>  #include <linux/interrupt.h>
>  #include <linux/init.h>
> --
> 2.20.0
>