From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sebastian Andrzej Siewior Subject: Re: [RT PATCH] cputime: remove raw locks introduced by RT patchset Date: Fri, 15 Apr 2016 09:13:34 +0200 Message-ID: <5710949E.1020407@linutronix.de> References: <20160414220943.3d4d7dd2@sluggy.hsv.redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Cc: Thomas Gleixner , RT , LKML To: Clark Williams Return-path: Received: from www.linutronix.de ([62.245.132.108]:51964 "EHLO Galois.linutronix.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751127AbcDOHNh (ORCPT ); Fri, 15 Apr 2016 03:13:37 -0400 In-Reply-To: <20160414220943.3d4d7dd2@sluggy.hsv.redhat.com> Sender: linux-rt-users-owner@vger.kernel.org List-ID: On 04/15/2016 05:09 AM, Clark Williams wrote: > Sebastian, Hi Clark, > This patch removes the raw spinlock operations when updating cputtime > in the vtime_* functions in kernel/sched/cputime.c. > > Based on Frederic's commit b7ce2277f087fd052, there is no need for > the raw spinlocks in vtime_* functions to guard against writer > concurrency and the RT versions of write_seqcount_begin() and > write_seqcount_end() make calls to preempt_disable_rt() and > preempt_enable_rt(), so we'll be in atomic context while updating > cputime. > > I've run this patch on x86_64 4.4.6-rt14 and the RHEL-RT kernel, > with 12h rteval runs on systems with both no tuning and systems > with isolcpus/rcu_nocbs/nohz_full cpus. No ill effects seen. Frederic's patch made it into 4.5-rc1. What I did yesterday was: * sched/cputime: Convert vtime_seqlock to seqcount * sched/cputime: Clarify vtime symbols and document them * Revert "vtime: Split lock and seqcount" which is what you want, correct? > > Signed-off-by: Rik van Riel > Signed-off-by: Clark Williams Sebastian