Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
From:   Andy Lutomirski <luto@amacapital.net>
Mime-Version: 1.0 (1.0)
Subject: Re: [PATCH v3] seccomp: Improve performace by optimizing rmb()
Date:   Wed, 24 Feb 2021 08:18:36 -0800
Message-Id: <638D44BA-0ACA-4041-8213-217233656A70@amacapital.net>
References: <1614156585-18842-1-git-send-email-wanghongzhe@huawei.com>
Cc:     keescook@chromium.org, andrii@kernel.org, ast@kernel.org,
        bpf@vger.kernel.org, daniel@iogearbox.net,
        john.fastabend@gmail.com, kafai@fb.com, kpsingh@kernel.org,
        linux-kernel@vger.kernel.org, netdev@vger.kernel.org,
        songliubraving@fb.com, wad@chromium.org, yhs@fb.com,
        tongxiaomeng@huawei.com
In-Reply-To: <1614156585-18842-1-git-send-email-wanghongzhe@huawei.com>
To:     wanghongzhe <wanghongzhe@huawei.com>
Precedence: bulk


> On Feb 24, 2021, at 12:03 AM, wanghongzhe <wanghongzhe@huawei.com> wrote:
>=20
> =EF=BB=BFAs Kees haved accepted the v2 patch at a381b70a1 which just
> replaced rmb() with smp_rmb(), this patch will base on that and just adjus=
t
> the smp_rmb() to the correct position.
>=20
> As the original comment shown (and indeed it should be):
>   /*
>    * Make sure that any changes to mode from another thread have
>    * been seen after SYSCALL_WORK_SECCOMP was seen.
>    */
> the smp_rmb() should be put between reading SYSCALL_WORK_SECCOMP and readi=
ng
> seccomp.mode to make sure that any changes to mode from another thread hav=
e
> been seen after SYSCALL_WORK_SECCOMP was seen, for TSYNC situation. Howeve=
r,
> it is misplaced between reading seccomp.mode and seccomp->filter. This iss=
ue
> seems to be misintroduced at 13aa72f0fd0a9f98a41cefb662487269e2f1ad65 whic=
h
> aims to refactor the filter callback and the API. So let's just adjust the=

> smp_rmb() to the correct position.
>=20
> A next optimization patch will be provided if this ajustment is appropriat=
e.

Would it be better to make the syscall work read be smp_load_acquire()?

>=20
> v2 -> v3:
> - move the smp_rmb() to the correct position
>=20
> v1 -> v2:
> - only replace rmb() with smp_rmb()
> - provide the performance test number
>=20
> RFC -> v1:
> - replace rmb() with smp_rmb()
> - move the smp_rmb() logic to the middle between TIF_SECCOMP and mode
>=20
> Signed-off-by: wanghongzhe <wanghongzhe@huawei.com>
> ---
> kernel/seccomp.c | 15 +++++++--------
> 1 file changed, 7 insertions(+), 8 deletions(-)
>=20
> diff --git a/kernel/seccomp.c b/kernel/seccomp.c
> index 1d60fc2c9987..64b236cb8a7f 100644
> --- a/kernel/seccomp.c
> +++ b/kernel/seccomp.c
> @@ -1160,12 +1160,6 @@ static int __seccomp_filter(int this_syscall, const=
 struct seccomp_data *sd,
>    int data;
>    struct seccomp_data sd_local;
>=20
> -    /*
> -     * Make sure that any changes to mode from another thread have
> -     * been seen after SYSCALL_WORK_SECCOMP was seen.
> -     */
> -    smp_rmb();
> -
>    if (!sd) {
>        populate_seccomp_data(&sd_local);
>        sd =3D &sd_local;
> @@ -1291,7 +1285,6 @@ static int __seccomp_filter(int this_syscall, const s=
truct seccomp_data *sd,
>=20
> int __secure_computing(const struct seccomp_data *sd)
> {
> -    int mode =3D current->seccomp.mode;
>    int this_syscall;
>=20
>    if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
> @@ -1301,7 +1294,13 @@ int __secure_computing(const struct seccomp_data *s=
d)
>    this_syscall =3D sd ? sd->nr :
>        syscall_get_nr(current, current_pt_regs());
>=20
> -    switch (mode) {
> +    /*=20
> +     * Make sure that any changes to mode from another thread have
> +     * been seen after SYSCALL_WORK_SECCOMP was seen.
> +     */
> +    smp_rmb();
> +
> +    switch (current->seccomp.mode) {
>    case SECCOMP_MODE_STRICT:
>        __secure_computing_strict(this_syscall);  /* may call do_exit */
>        return 0;
> --=20
> 2.19.1
>=20