From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
        id S1754157AbcIAJtO (ORCPT <rfc822;w@1wt.eu>);
        Thu, 1 Sep 2016 05:49:14 -0400
Received: from bombadil.infradead.org ([198.137.202.9]:43299 "EHLO
        bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
        with ESMTP id S1751243AbcIAJtN (ORCPT
        <rfc822;linux-kernel@vger.kernel.org>);
        Thu, 1 Sep 2016 05:49:13 -0400
Date: Thu, 1 Sep 2016 11:49:06 +0200
From: Peter Zijlstra <peterz@infradead.org>
To: Stanislaw Gruszka <sgruszka@redhat.com>
Cc: linux-kernel@vger.kernel.org,
        Giovanni Gherdovich <ggherdovich@suse.cz>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Mel Gorman <mgorman@techsingularity.net>,
        Mike Galbraith <mgalbraith@suse.de>,
        Paolo Bonzini <pbonzini@redhat.com>, Rik van Riel <riel@redhat.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Wanpeng Li <wanpeng.li@hotmail.com>, Ingo Molnar <mingo@kernel.org>
Subject: Re: [PATCH 1/3] sched/cputime: Improve scalability of
 times()/clock_gettime() on 32 bit cpus
Message-ID: <20160901094906.GP10153@twins.programming.kicks-ass.net>
References: <1472722064-7151-1-git-send-email-sgruszka@redhat.com>
 <1472722064-7151-2-git-send-email-sgruszka@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1472722064-7151-2-git-send-email-sgruszka@redhat.com>
User-Agent: Mutt/1.5.23.1 (2014-03-12)
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Thu, Sep 01, 2016 at 11:27:42AM +0200, Stanislaw Gruszka wrote:
> My previous commit:
> 
>   a1eb1411b4e4 ("sched/cputime: Improve scalability by not accounting thread group tasks pending runtime")
> 
> helped to achieve good performance of SYS_times() and
> SYS_clock_gettimes(CLOCK_PROCESS_CPUTIME_ID) on 64 bit architectures.
> However taking task_rq_lock() when reading t->se.sum_exec_runtime on
> 32 bit architectures still make those syscalls slow.
> 
> The reason why we take the lock is to make 64bit sum_exec_runtime
> variable consistent. While a inconsistency scenario is very very unlike,
> I assume it still may happen at least on some 32 bit architectures.
> 
> To protect the variable I introduced new seqcount lock. Performance
> improvements on machine with 32 cores (32-bit cpus) measured by
> benchmarks described in commit:

No,.. running 32bit kernels on a machine with 32 cores is insane, full
stop.

You're now making rather hot paths slower to benefit a rather slow path,
that too is backwards.

[ also, seqcount is not a lock ].

Really, people should not expect process wide numbers to be fast.