Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220221122638.7971-1-franciman12@gmail.com> <e32de43c-e00a-5766-e988-fe192d36d719@quicinc.com>
 <CAH4F6utmtAM3CzX2_Fbn94Sb-X8m0patPh8yWwbuBB0t1VYH=g@mail.gmail.com>
 <87o82wvhtk.fsf@kernel.org> <20220224075346.GL3943@kadam> <87fso8vb3w.fsf_-_@kernel.org>
In-Reply-To: <87fso8vb3w.fsf_-_@kernel.org>
From:   Linus Torvalds <torvalds@linuxfoundation.org>
Date:   Thu, 24 Feb 2022 09:45:05 -0800
Message-ID: <CAHk-=whFKYMrF6euVvziW+drw7-yi1pYdf=uccnzJ8k09DoTXA@mail.gmail.com>
Subject: Re: Use of void pointer arithmetic?
To:     Kalle Valo <kvalo@kernel.org>
Cc:     Dan Carpenter <dan.carpenter@oracle.com>,
        Francesco Magliocca <franciman12@gmail.com>,
        Jeff Johnson <quic_jjohnson@quicinc.com>,
        ath10k@lists.infradead.org, rmanohar@qti.qualcomm.com,
        linux-wireless <linux-wireless@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Feb 24, 2022 at 1:59 AM Kalle Valo <kvalo@kernel.org> wrote:
>
> > What does -Wpointer-arith buy us?
>
> A good question. I have always just thought we should avoid void pointer
> arithmetic due to the C standard, but now that you mention it void
> pointers can indeed simplify the code. So I'm not so sure anymore.
>
> Any opinions? Is there a kernel wide recommendation for this?

We consciously use arithmetic on 'void *' in some places, although
it's not exactly _hugely_ common.

It's much more common to turn a pointer into an 'unsigned long' and
then doing basic operations on that.

The advantage of 'void *' is that it does avoid the need to cast the
pointer back.

But at the same time it will never replace the 'cast to an actual
integer type', because the 'void *' arithmetic only does simple
addition and subtraction, and many pointer operations need more (ie
alignment needs to do the bitops).

So I think it's mostly a personal preference. I *do* think that doing
arithmetic on 'void *' is generally superior to doing it the
old-fashioned C way on 'char *' - unless, of course, 'char *' is
simply the native type in question.

So if 'char *' casts (and casting back) is the alternative, then by
all means use 'void *' as a kind of generic and type-independent "byte
pointer". That is how it is meant to be used in the gcc extension.

And no, nobody should ever use -Wpointer-arith on the kernel in
general. Our use of it is not _hugely_ common, but it's does exist,
and it's not really frowned upon.

                    Linus