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;
MIME-Version: 1.0
References: <20181209032715.3466040-1-liuxiaozhou@bytedance.com>
In-Reply-To: <20181209032715.3466040-1-liuxiaozhou@bytedance.com>
From:   Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>
Date:   Thu, 13 Dec 2018 22:59:10 +0100
Message-ID: <CANiq72nqFuJm5f69G-Pc_9QHz+anwaon-WfVsPrkZWah1OP7Fg@mail.gmail.com>
Subject: Re: [PATCH v2] Compiler Attributes: don't pollute userspace with
 macro definitions
To:     liuxiaozhou@bytedance.com, Greg KH <gregkh@linuxfoundation.org>
Cc:     linux-kernel <linux-kernel@vger.kernel.org>,
        Linus Torvalds <torvalds@linux-foundation.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Xiaozhou,

Couple of comments.

On Sun, Dec 9, 2018 at 4:27 AM Xiaozhou Liu <liuxiaozhou@bytedance.com> wrote:
>
> v2: update commit message.

This line should go below the "---", since it shouldn't be part of the
commit message.

> +#ifdef __KERNEL__
> +
>  #ifdef CONFIG_ENABLE_MUST_CHECK
>  #define __must_check           __attribute__((__warn_unused_result__))
>  #else
> @@ -215,4 +217,6 @@ struct ftrace_likely_data {
>   */
>  #define noinline_for_stack noinline
>
> +#endif /* __KERNEL */

I wonder if we can/should simply move them into the __KERNEL__ &&
!__ASSEMBLY__ block that is above, instead. It would be simpler to
read, and there aren't apparently dependencies to those outside it
that go after the block.

I took a look at where the macros were at each "step", and, on one
hand, compiler-gcc.h was (and is) included entirely inside it, which
is from where most of the macros come originally from. On the other
hand, not all do: __must_check (the generic version, not the one in
-gcc.h) and noinline_for_stack were defined in __KERNEL__ (only)
before commit 815f0ddb346c ("include/linux/compiler*.h: make
compiler-*.h mutually exclusive"). But anyway using those two in
assembly does not make sense, right?

What do you think?

Greg/Linus, are you going to pick this (or a v3) directly? If not, I
can pick it up in compiler-attributes tree linux-next.

PS: In addition (related to this but not for this patch), we should
review whether other macros that are currently outside should be there
or simply pushed back into __KERNEL__ (and possibly __ASSEMBLY__). For
instance, __latent_entropy (the generic one) is outside, but it is
only defined in compiler-gcc.h, so the generic one should be inside
the __KERNEL__ && !__ASSEMBLY__ block, no?

Cheers,
Miguel