Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Tue, 14 Jan 2020 14:16:51 +0800
From:   chenqiwu <qiwuchen55@gmail.com>
To:     Christian Brauner <christian.brauner@ubuntu.com>
Cc:     peterz@infradead.org, mingo@kernel.org, tglx@linutronix.de,
        oleg@redhat.com, elena.reshetova@intel.com, jgg@ziepe.ca,
        christian@kellner.me, aarcange@redhat.com, viro@zeniv.linux.org.uk,
        cyphar@cyphar.com, ldv@altlinux.org, linux-kernel@vger.kernel.org,
        chenqiwu <chenqiwu@xiaomi.com>
Subject: Re: [PATCH] kernel/fork: put some fork variables into read-mostly
 section
Message-ID: <20200114061651.GA8818@cqw-OptiPlex-7050>
References: <1578885793-24095-1-git-send-email-qiwuchen55@gmail.com>
 <20200113094342.5ghlgttmhuxfqv2v@wittgenstein>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200113094342.5ghlgttmhuxfqv2v@wittgenstein>
User-Agent: Mutt/1.5.21 (2010-09-15)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Jan 13, 2020 at 10:43:43AM +0100, Christian Brauner wrote:
> On Mon, Jan 13, 2020 at 11:23:13AM +0800, qiwuchen55@gmail.com wrote:
> > From: chenqiwu <chenqiwu@xiaomi.com>
> > 
> > Since total_forks/nr_threads/max_threads global variables are
> > frequently used for process fork, putting these variables into
> > read_mostly section can avoid unnecessary cache line bouncing.
> > 
> > Signed-off-by: chenqiwu <chenqiwu@xiaomi.com>
> > ---
> >  kernel/fork.c | 6 +++---
> >  1 file changed, 3 insertions(+), 3 deletions(-)
> > 
> > diff --git a/kernel/fork.c b/kernel/fork.c
> > index 0808095..163e152 100644
> > --- a/kernel/fork.c
> > +++ b/kernel/fork.c
> > @@ -120,10 +120,10 @@
> >  /*
> >   * Protected counters by write_lock_irq(&tasklist_lock)
> >   */
> > -unsigned long total_forks;	/* Handle normal Linux uptimes. */
> > -int nr_threads;			/* The idle threads do not count.. */
> > +unsigned long total_forks __read_mostly; /* Handle normal Linux uptimes. */
> > +int nr_threads __read_mostly;  /* The idle threads do not count.. */
> 
> total_forks is incremented at every ~CLONE_THREAD and nr_threads at
> CLONE_THREAD I wouldn't exactly say that this qualifies as mostly
> reading.
>

Hi Christian,
I'm holding different views on this matter.
1) total_forks is incremented when any process does sys_fork() reasonablely,
   it counts for every process fork.
2) nr_threads counts for any parent process (except for idle thread) does
   fork successfully, it never exceeds than max_threads.

For an example of arm64 kdump, the system has been running at 119057s,
we can see total_forks is very large, so total_forks is very qualified
as mostly reading because it is referenced frequntly.

crash> p max_threads
max_threads = $2 = 39150
crash> p nr_threads
nr_threads = $1 = 2676
crash> p total_forks
total_forks = $3 = 2413880

nr_threads and max_threads is also qualified as mostly reading because
they are frequntly referenced in the following code of copy_process():
	if (nr_threads >= max_threads)
		goto bad_fork_cleanup_count;
   
> >  
> > -static int max_threads;		/* tunable limit on nr_threads */
> > +static int max_threads __read_mostly; /* tunable limit on nr_threads */
> 
> That make sense.
> 
> Christian

Thanks for your review!
Qiwu