linux-kernel - Re: [LKP] Re: [perf/x86] 81ec3f3c4c: will-it-scale.per_process

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200225025748.GB63065@shbuild999.sh.intel.com>
Date:   Tue, 25 Feb 2020 10:57:48 +0800
From:   Feng Tang <feng.tang@...el.com>
To:     Linus Torvalds <torvalds@...ux-foundation.org>
Cc:     Oleg Nesterov <oleg@...hat.com>,
        "Eric W. Biederman" <ebiederm@...ssion.com>,
        Jiri Olsa <jolsa@...hat.com>,
        Peter Zijlstra <peterz@...radead.org>,
        kernel test robot <rong.a.chen@...el.com>,
        Ingo Molnar <mingo@...nel.org>,
        Vince Weaver <vincent.weaver@...ne.edu>,
        Jiri Olsa <jolsa@...nel.org>,
        Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
        Arnaldo Carvalho de Melo <acme@...nel.org>,
        Arnaldo Carvalho de Melo <acme@...hat.com>,
        "Naveen N. Rao" <naveen.n.rao@...ux.vnet.ibm.com>,
        Ravi Bangoria <ravi.bangoria@...ux.ibm.com>,
        Stephane Eranian <eranian@...gle.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        LKML <linux-kernel@...r.kernel.org>, lkp@...ts.01.org,
        andi.kleen@...el.com, "Huang, Ying" <ying.huang@...el.com>
Subject: Re: [LKP] Re: [perf/x86] 81ec3f3c4c: will-it-scale.per_process_ops
 -5.5% regression

On Mon, Feb 24, 2020 at 12:47:14PM -0800, Linus Torvalds wrote:
> [ Adding a few more people that tend to be involved in signal
> handling. Just in case - even if they probably don't care ]
> 
> On Mon, Feb 24, 2020 at 12:09 PM Linus Torvalds
> 
> I've tested it, and the profiles on the silly microbenchmark look much
> nicer. Now it's just the sigpending update shows up, the refcount case
> clearly still occasionally happens, but it's now in the noise.
> 
> I made slight changes to the __sigqueue_alloc() case to generate
> better code: since we now use that atomic_inc_return() anyway, we
> might as well then use the value that is returned for the
> RLIMIT_SIGPENDING check too, instead of reading it again.
> 
> That might avoid another potential cacheline bounce, plus the
> generated code just looks better.
> 
> Updated (and now slightly tested!) patch attached.
> It would be interesting if this is noticeable on your benchmark
> numbers. I didn't actually _time_ anything, I just looked at profiles.
> 
> But my setup clearly isn't going to see the horrible contention case
> anyway, so my timing numbers wouldn't be all that interesting.

Thanks for the optimization patch for signal!

It makes a big difference, that the performance score is tripled!
bump from original 17000 to 54000. Also the gap between 5.0-rc6 and
5.0-rc6+Jiri's patch is reduced to around 2%.

The test I run is inserting your patch right before 5.0-rc6, then
run the test for 5.0-rc6 and 5.0-rc6+Jiri's patch. Sorry it took
quite some time, as the test platform is not local but inside
0day's framework, which takes some time for scheduling, kbuilding
and testing.

Thanks,
Feng

 
>              Linus

>  kernel/signal.c | 23 ++++++++++++++---------
>  1 file changed, 14 insertions(+), 9 deletions(-)
> 
> diff --git a/kernel/signal.c b/kernel/signal.c
> index 9ad8dea93dbb..5b2396350dd1 100644
> --- a/kernel/signal.c
> +++ b/kernel/signal.c
> @@ -413,27 +413,32 @@ __sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimi
>  {
>  	struct sigqueue *q = NULL;
>  	struct user_struct *user;
> +	int sigpending;
>  
>  	/*
>  	 * Protect access to @t credentials. This can go away when all
>  	 * callers hold rcu read lock.
> +	 *
> +	 * NOTE! A pending signal will hold on to the user refcount,
> +	 * and we get/put the refcount only when the sigpending count
> +	 * changes from/to zero.
>  	 */
>  	rcu_read_lock();
> -	user = get_uid(__task_cred(t)->user);
> -	atomic_inc(&user->sigpending);
> +	user = __task_cred(t)->user;
> +	sigpending = atomic_inc_return(&user->sigpending);
> +	if (sigpending == 1)
> +		get_uid(user);
>  	rcu_read_unlock();
>  
> -	if (override_rlimit ||
> -	    atomic_read(&user->sigpending) <=
> -			task_rlimit(t, RLIMIT_SIGPENDING)) {
> +	if (override_rlimit || likely(sigpending <= task_rlimit(t, RLIMIT_SIGPENDING))) {
>  		q = kmem_cache_alloc(sigqueue_cachep, flags);
>  	} else {
>  		print_dropped_signal(sig);
>  	}
>  
>  	if (unlikely(q == NULL)) {
> -		atomic_dec(&user->sigpending);
> -		free_uid(user);
> +		if (atomic_dec_and_test(&user->sigpending))
> +			free_uid(user);
>  	} else {
>  		INIT_LIST_HEAD(&q->list);
>  		q->flags = 0;
> @@ -447,8 +452,8 @@ static void __sigqueue_free(struct sigqueue *q)
>  {
>  	if (q->flags & SIGQUEUE_PREALLOC)
>  		return;
> -	atomic_dec(&q->user->sigpending);
> -	free_uid(q->user);
> +	if (atomic_dec_and_test(&q->user->sigpending))
> +		free_uid(q->user);
>  	kmem_cache_free(sigqueue_cachep, q);
>  }
>