Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 19 Nov 2018 13:33:36 +0100
From:   Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
To:     Masahiro Yamada <yamada.masahiro@socionext.com>
Cc:     Andrew Morton <akpm@linux-foundation.org>,
        Nick Desaulniers <ndesaulniers@google.com>,
        Kees Cook <keescook@chromium.org>,
        Josh Triplett <josh@joshtriplett.org>,
        Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>,
        Arnd Bergmann <arnd@arndb.de>, linux-kernel@vger.kernel.org,
        Paul Burton <paul.burton@mips.com>
Subject: Re: [PATCH v3 1/3] compiler_types.h: make
 __builtin_types_compatible_p() noop for Sparse
Message-ID: <20181119123334.6xr23ok4uy5cflzy@ltop.local>
References: <1542623503-3755-1-git-send-email-yamada.masahiro@socionext.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1542623503-3755-1-git-send-email-yamada.masahiro@socionext.com>
User-Agent: NeoMutt/20180622
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Nov 19, 2018 at 07:31:41PM +0900, Masahiro Yamada wrote:
> When I tried to delete BUILD_BUG_ON stubs for sparse, the kbuild test
> robot reported lots of Sparse warnings from container_of(), which
> seem false positive.
> 
> The following checker in container_of() seems to be causing something
> strange for Sparse.
> 
>   BUILD_BUG_ON_MSG(!__same_type(*(ptr), ((type *)0)->member) &&   \
>                    !__same_type(*(ptr), void),                    \
>                    "pointer type mismatch in container_of()");    \
> 
> I narrowed down the problem into the following test code:
> 
>   --------------------(test_code.c begin)--------------------
>   struct foo {
>           int (*callback)(void);
>   };
> 
>   void assert(int);
> 
>   static inline struct foo *get_foo(void)
>   {
>           assert(__builtin_types_compatible_p(void, void));
> 
>           return (struct foo *)0;
>   }
> 
>   int test(void);
>   int test(void)
>   {
>           return get_foo()->callback();
>   }
>   ---------------------(test_code.c end)---------------------
> 
> Of course, GCC (and Clang as well) can compile it:
> 
>   $ gcc -Wall -c -o test_code.o test_code.c
> 
> However, Sparse complains about this obviously correct code:
> 
>   $ sparse test_code.c
>   test_code.c:9:45: warning: unknown expression (4 0)
>   test_code.c:9:51: warning: unknown expression (4 0)
> 
> Interstingly, just removing the 'inline' keyword in the test code
> makes Sparse happy.
> 
> I concluded that Sparse cannot handle __builtin_types_compatible_p()
> correctly.

I think it's only caused by comparing 'void' (which is never
an l-value).
I'll investigate. Thanks for the small test-case.

> Make it no-op.

...

> diff --git a/include/linux/compiler_types.h b/include/linux/compiler_types.h
> index 4a3f9c0..9e7da0b 100644
> --- a/include/linux/compiler_types.h
> +++ b/include/linux/compiler_types.h
> @@ -23,6 +23,7 @@
>  extern void __chk_user_ptr(const volatile void __user *);
>  extern void __chk_io_ptr(const volatile void __iomem *);
>  # define ACCESS_PRIVATE(p, member) (*((typeof((p)->member) __force *) &(p)->member))
> +# define __builtin_types_compatible_p(t1, t2)	(1)

Now, BUILD_BUG_ON() becomes a no-op for sparse but all the other usages
of __builtin_types_compatible_p() become potentially wrong and can now
create their onw false warnings.

Regards,
-- Luc