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: <20220507024840.42662-1-zhoufeng.zf@bytedance.com>
 <CAEf4BzZD5q2j229_gL_nDFse2v=k2Ea0nfguH+sOA2O1Nm5sQw@mail.gmail.com>
 <CAJD7tkbd8qA-4goUCVW6Tf0xGpj2OSBXncpWhrWFn5y010oBMw@mail.gmail.com> <d20aef2a-273a-3183-0923-bde9657d4418@bytedance.com>
In-Reply-To: <d20aef2a-273a-3183-0923-bde9657d4418@bytedance.com>
From:   Alexei Starovoitov <alexei.starovoitov@gmail.com>
Date:   Mon, 9 May 2022 20:15:16 -0700
Message-ID: <CAADnVQL+Vq5y47J++VCppti1728w3U0maxg9d4SqAtArY+h1yg@mail.gmail.com>
Subject: Re: [External] Re: [PATCH bpf-next] bpf: add bpf_map_lookup_percpu_elem
 for percpu map
To:     Feng Zhou <zhoufeng.zf@bytedance.com>
Cc:     Yosry Ahmed <yosryahmed@google.com>,
        Andrii Nakryiko <andrii.nakryiko@gmail.com>,
        Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Andrii Nakryiko <andrii@kernel.org>, Martin Lau <kafai@fb.com>,
        Song Liu <songliubraving@fb.com>, Yonghong Song <yhs@fb.com>,
        john fastabend <john.fastabend@gmail.com>,
        KP Singh <kpsingh@kernel.org>,
        Steven Rostedt <rostedt@goodmis.org>,
        Ingo Molnar <mingo@redhat.com>, Jiri Olsa <jolsa@kernel.org>,
        Dave Marchevsky <davemarchevsky@fb.com>,
        Joanne Koong <joannekoong@fb.com>,
        Geliang Tang <geliang.tang@suse.com>,
        Networking <netdev@vger.kernel.org>, bpf <bpf@vger.kernel.org>,
        open list <linux-kernel@vger.kernel.org>,
        Xiongchun Duan <duanxiongchun@bytedance.com>,
        Muchun Song <songmuchun@bytedance.com>,
        Dongdong Wang <wangdongdong.6@bytedance.com>,
        Cong Wang <cong.wang@bytedance.com>,
        Chengming Zhou <zhouchengming@bytedance.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Mon, May 9, 2022 at 7:41 PM Feng Zhou <zhoufeng.zf@bytedance.com> wrote:
>
> =E5=9C=A8 2022/5/10 =E4=B8=8A=E5=8D=889:04, Yosry Ahmed =E5=86=99=E9=81=
=93:
> > On Mon, May 9, 2022 at 5:34 PM Andrii Nakryiko
> > <andrii.nakryiko@gmail.com> wrote:
> >> On Fri, May 6, 2022 at 7:49 PM Feng zhou <zhoufeng.zf@bytedance.com> w=
rote:
> >>> From: Feng Zhou <zhoufeng.zf@bytedance.com>
> >>>
> >>> Trace some functions, such as enqueue_task_fair, need to access the
> >>> corresponding cpu, not the current cpu, and bpf_map_lookup_elem percp=
u map
> >>> cannot do it. So add bpf_map_lookup_percpu_elem to accomplish it for
> >>> percpu_array_map, percpu_hash_map, lru_percpu_hash_map.
> >>>
> >>> The implementation method is relatively simple, refer to the implemen=
tation
> >>> method of map_lookup_elem of percpu map, increase the parameters of c=
pu, and
> >>> obtain it according to the specified cpu.
> >>>
> >> I don't think it's safe in general to access per-cpu data from another
> >> CPU. I'd suggest just having either a ARRAY_OF_MAPS or adding CPU ID
> >> as part of the key, if you need such a custom access pattern.
> > I actually just sent an RFC patch series containing a similar patch
> > for the exact same purpose. There are instances in the kernel where
> > per-cpu data is accessed from other cpus (e.g.
> > mem_cgroup_css_rstat_flush()). I believe, like any other variable,
> > percpu data can be safe or not safe to access, based on the access
> > pattern. It is up to the user to coordinate accesses to the variable.
> >
> > For example, in my use case, one of the accessors only reads percpu
> > values of different cpus, so it should be safe. If a user accesses
> > percpu data of another cpu without guaranteeing safety, they corrupt
> > their own data. I understand that the main purpose of percpu data is
> > lockless (and therefore fast) access, but in some use cases the user
> > may be able to safely (and locklessly) access the data concurrently.
> >
>
> Regarding data security, I think users need to consider before using it,
> such
> as hook enqueue_task_fair, the function itself takes the rq lock of the
> corresponding cpu, there is no problem, and the kernel only provides a
> method,
> like bpf_per_cpu_ptr and bpf_this_cpu_ptr, data security needs to be
> guaranteed
> by users in different scenarios, such as using bpf_spin_lock.

Right. The new helper looks useful and is safe.
Please add a selftest and respin.