Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
MIME-Version: 1.0
References: <YouKQw72H7y9EJQK@alley> <CAHk-=wgC47n_7E6UtFx_agkJtLmWOXGsjdFjybBFYNA1AheQLQ@mail.gmail.com>
 <87pmk1tgmx.fsf@jogness.linutronix.de> <Yo9t9Vg/DpvMqCs7@alley>
In-Reply-To: <Yo9t9Vg/DpvMqCs7@alley>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Thu, 26 May 2022 10:13:29 -0700
Message-ID: <CAHk-=wjny_OCPeBfr0uLzcMsAK2aa1TX3UPkp2V2scLD4g=7mw@mail.gmail.com>
Subject: Re: [GIT PULL] printk for 5.19
To:     Petr Mladek <pmladek@suse.com>
Cc:     John Ogness <john.ogness@linutronix.de>,
        Sergey Senozhatsky <senozhatsky@chromium.org>,
        Steven Rostedt <rostedt@goodmis.org>,
        Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
        Rasmus Villemoes <linux@rasmusvillemoes.dk>,
        Jan Kara <jack@suse.cz>, Peter Zijlstra <peterz@infradead.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, May 26, 2022 at 5:09 AM Petr Mladek <pmladek@suse.com> wrote:
>
> Yeah, it is a shame. I have met code with exclusive waiters only once,
> 8 years ago and completely forgot about it.

Yeah, the exclusive waiters thing is quite unusual, and not a normal pattern.

It's basically only used for a very few performance-critical cases
where the "multiple readers, but only one writer should wake things
up" can happen.

And it's rare. It's things like "accept()" on a socket - where it's
common to have a lot of threads waiting for a new connection, but you
don't want to wake them all just because one new connection comes in,
because only one of them will successfully take it.

There are not very many patterns like that where you have that "many
people waiting for the same thing, only one will get it". It happens
for socket readers (kind of like accept, but for the "UDP packet with
command comes in"), or with things like pipes (which people use for
similar things and distribute work to potentially lots of threads
through a pipe).

So the common pattern tends to be "some event that gets a threaded
server using one thread to complete", but there just aren't that many
of those cases.

It _could_ also be used for things like locking (ie think "lock
release wakes up waiter"), but I think all our lock primitives end up
doing their wakeup routines at a lower level (ie using
"wake_up_process()" directly) so those don't actually use it.

End result: the exclusive waiters are an important case, but they are
rare - and thus the "all()" version of wakeup is also quite rare.

                        Linus