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:   Mon, 16 Jan 2023 21:10:37 +0000
From:   Al Viro <viro@zeniv.linux.org.uk>
To:     Matthew Wilcox <willy@infradead.org>
Cc:     maobibo <maobibo@loongson.cn>,
        Hongchen Zhang <zhanghongchen@loongson.cn>,
        Andrew Morton <akpm@linux-foundation.org>,
        David Howells <dhowells@redhat.com>,
        Sedat Dilek <sedat.dilek@gmail.com>,
        "Christian Brauner (Microsoft)" <brauner@kernel.org>,
        Luis Chamberlain <mcgrof@kernel.org>,
        Mauro Carvalho Chehab <mchehab@kernel.org>,
        Eric Dumazet <edumazet@google.com>,
        "Fabio M. De Francesco" <fmdefrancesco@gmail.com>,
        Christophe JAILLET <christophe.jaillet@wanadoo.fr>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v3] pipe: use __pipe_{lock,unlock} instead of spinlock
Message-ID: <Y8W9TR5ifZmRADLB@ZenIV>
References: <20230107012324.30698-1-zhanghongchen@loongson.cn>
 <9fcb3f80-cb55-9a72-0e74-03ace2408d21@loongson.cn>
 <4b140bd0-9b7f-50b5-9e3b-16d8afe52a50@loongson.cn>
 <Y8TUqcSO5VrbYfcM@casper.infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Y8TUqcSO5VrbYfcM@casper.infradead.org>
Sender: Al Viro <viro@ftp.linux.org.uk>
Precedence: bulk

On Mon, Jan 16, 2023 at 04:38:01AM +0000, Matthew Wilcox wrote:
> On Mon, Jan 16, 2023 at 11:16:13AM +0800, maobibo wrote:
> > Hongchen,
> > 
> > I have a glance with this patch, it simply replaces with
> > spinlock_irqsave with mutex lock. There may be performance
> > improvement with two processes competing with pipe, however
> > for N processes, there will be complex context switches
> > and ipi interruptts.
> > 
> > Can you find some cases with more than 2 processes competing
> > pipe, rather than only unixbench?
> 
> What real applications have pipes with more than 1 writer & 1 reader?
> I'm OK with slowing down the weird cases if the common cases go faster.

From commit 0ddad21d3e99c743a3aa473121dc5561679e26bb:
    While this isn't a common occurrence in the traditional "use a pipe as a
    data transport" case, where you typically only have a single reader and
    a single writer process, there is one common special case: using a pipe
    as a source of "locking tokens" rather than for data communication.
    
    In particular, the GNU make jobserver code ends up using a pipe as a way
    to limit parallelism, where each job consumes a token by reading a byte
    from the jobserver pipe, and releases the token by writing a byte back
    to the pipe.