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;
Date:   Sun, 13 Aug 2023 05:47:02 +0900
From:   Masami Hiramatsu (Google) <mhiramat@kernel.org>
To:     Zheng Yejian <zhengyejian1@huawei.com>
Cc:     Steven Rostedt <rostedt@goodmis.org>, <mhiramat@kernel.org>,
        <laijs@cn.fujitsu.com>, <linux-kernel@vger.kernel.org>,
        <linux-trace-kernel@vger.kernel.org>
Subject: Re: [PATCH] tracing: Fix race when concurrently splice_read
 trace_pipe
Message-Id: <20230813054702.22ce16d9191a1f6b84942a1e@kernel.org>
In-Reply-To: <52369719-dbc7-7cb5-f766-877d24c8400c@huawei.com>
References: <20230810123905.1531061-1-zhengyejian1@huawei.com>
        <20230811152525.2511f8f0@gandalf.local.home>
        <52369719-dbc7-7cb5-f766-877d24c8400c@huawei.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Sat, 12 Aug 2023 09:45:52 +0800
Zheng Yejian <zhengyejian1@huawei.com> wrote:

> On 2023/8/12 03:25, Steven Rostedt wrote:
> > On Thu, 10 Aug 2023 20:39:05 +0800
> > Zheng Yejian <zhengyejian1@huawei.com> wrote:
> > 
> >> When concurrently splice_read file trace_pipe and per_cpu/cpu*/trace_pipe,
> >> there are more data being read out than expected.
> 
> Sorry, I didn't make clear here. It not just read more but also lost
> some data. My case is that, for example:
>    1) Inject 3 events into ring_buffer: event1, event2, event3;
>    2) Concurrently splice_read through trace_pipes;
>    3) Then actually read out: event1, event3, event3. No event2, but 2 
> event3.
> 
> > 
> > Honestly the real fix is to prevent that use case. We should probably have
> > access to trace_pipe lock all the per_cpu trace_pipes too.
> 
> Yes, we could do that, but would it seem not that effective?
> because per_cpu trace_pipe only read its own ring_buffer and not race
> with ring_buffers in other cpus.

I think Steve said that only one of below is usable.

- Read trace_pipe

or

- Read per_cpu/cpu*/trace_pipe concurrently

And I think this makes sence, especially if you use splice (this *moves*
the page from the ring_buffer to other pipe).

Thank you,


> 
> > 
> > -- Steve
> > 
> 


-- 
Masami Hiramatsu (Google) <mhiramat@kernel.org>