Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.35 as permitted sender) client-ip=23.128.96.35;
MIME-Version: 1.0
References: <20231010164234.140750-1-ubizjak@gmail.com> <CAHk-=whYWhZN52SJN-Th9x2L2V-vHtAXUgiy_nSJ3+vQU6ak4Q@mail.gmail.com>
 <CAFULd4ZqH3FeG8_mjDvUAU9QiGB36wDu3MzUtadgAgoVuQ9QRg@mail.gmail.com> <108536FC-445C-41F9-8A87-B7D31A45AC4B@vmware.com>
In-Reply-To: <108536FC-445C-41F9-8A87-B7D31A45AC4B@vmware.com>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Tue, 10 Oct 2023 11:42:49 -0700
Message-ID: <CAHk-=wj9dGfmA_zPof+CZdcZq6ZA41e6ycQus6P5CShLPiMoVw@mail.gmail.com>
Subject: Re: [PATCH v2 -tip] x86/percpu: Use C for arch_raw_cpu_ptr()
To:     Nadav Amit <namit@vmware.com>
Cc:     Uros Bizjak <ubizjak@gmail.com>,
        "the arch/x86 maintainers" <x86@kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Andy Lutomirski <luto@kernel.org>,
        Brian Gerst <brgerst@gmail.com>,
        Denys Vlasenko <dvlasenk@redhat.com>,
        "H . Peter Anvin" <hpa@zytor.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Josh Poimboeuf <jpoimboe@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Tue, 10 Oct 2023 at 11:25, Nadav Amit <namit@vmware.com> wrote:
>
> As a minor note the proposed assembly version seems to be missing
> __FORCE_ORDER as an input argument to prevent reordering past preempt_ena=
ble
> and preempt_disable. But that=E2=80=99s really not the main point.

Hmm. No, it's probably *is* the main point - see my reply to Uros that
the CSE on the inline asm itself gets rid of duplication.

And yes, we currently rely on that asm CSE for doing 'current' and not
reloading the value all the time.

So yes, we'd like to have a barrier for not moving it across the
preemption barriers, and __FORCE_ORDER would seem to be a good way to
do that.

I really suspect that 'rdgsbase' is better than a memory load in
practice, but I have no numbers to back that up, apart from a "it's
not a slow instruction, and core CPU is generally better than memory".

          Linus