Re: [PATCH] lib/string: shrink lib/string.i via IWYU

From: Nick Desaulniers <ndesaulniers@google.com>
To: Andrew Morton <akpm@linux-foundation.org>
Cc: tanzirh@google.com, Kees Cook <keescook@chromium.org>,
	 Andy Shevchenko <andy@kernel.org>,
	linux-hardening@vger.kernel.org,  linux-kernel@vger.kernel.org,
	Nick DeSaulniers <nnn@google.com>,
	llvm@lists.linux.dev
Subject: Re: [PATCH] lib/string: shrink lib/string.i via IWYU
Date: Tue, 5 Dec 2023 13:39:47 -0800	[thread overview]
Message-ID: <CAKwvOdn=og6h5gVdDCjFDANs3MN-_CD4OZ9oRM=o9YAvoTzkzw@mail.gmail.com> (raw)
In-Reply-To: <20231205132452.418722bea8f6878dca88142a@linux-foundation.org>

On Tue, Dec 5, 2023 at 1:24 PM Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Tue, 5 Dec 2023 13:14:16 -0800 Nick Desaulniers <ndesaulniers@google.com> wrote:
>
> > >
> > > The preferred way to import bit-fiddling stuff is to include
> > > <linux/bits.h>.  Under the hood this may include asm/bitsperlong.h.  Or
> > > it may not, depending on Kconfig settings (particularly architecture).
> > >
> >
> > Just triple checking my understanding; it looks like
> > include/linux/bits.h unconditionally includes asm/bitsperlong.h (which
> > is implemented per arch) most of which seem to include
> > asm-generic/bitsperlong.h.
> >
> > include/linux/bits.h also defines a few macros (BIT_MASK, BIT_WORD,
> > BITS_PER_BYTE, GENMASK, etc).  If lib/string.c is not using any of
> > those, why can't we go straight to #including asm/bitsperlong.h?  That
> > should resolve to the arch specific impl which may include
> > asm-generic/bitsperlong.h?
>
> It's just a general rule.  If the higher-level include is present, use
> that.  Because of the above, plus I guess things might change in the
> future.

Hmm...how does one know that linux/bits.h is the higher-level include
of asm/bitsperlong.h?

Do we mention these conventions anywhere under Documentation?

>
> We've been getting better about irregular asm/include files.
>
> But bits.h is a poor example.  A better case to study is spinlock.h.
> If this tool recommended including asm/spinlock.h then that won't work
> on any architecture which doesn't implement SMP (there is no
> arch/nios2/include/asm/spinlock.h).

The tooling Tanzir is working on does wrap IWYU, and does support such
mapping (of 'low level' to 'high level' headers; more so, if it
recommends X you can override to suggest Y instead).

arch/nios/ also doesn't provide a bug.h, which this patch is
suggesting we include directly.  I guess the same goes for
asm/rwonce.h.

-- 
Thanks,
~Nick Desaulniers