From mboxrd@z Thu Jan 1 00:00:00 1970 From: Steven Rostedt Subject: Re: [RFC][PATCH RT 6/6] vtime: Convert vtime_seqlock into raw_spinlock_t and seqcount combo Date: Thu, 11 Jul 2013 10:30:22 -0400 Message-ID: <1373553022.17876.11.camel@gandalf.local.home> References: <20130626192806.107564905@goodmis.org> <20130626193050.529670718@goodmis.org> <20130710171212.GA13553@windriver.com> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Cc: linux-kernel@vger.kernel.org, linux-rt-users , Thomas Gleixner , Carsten Emde , Sebastian Andrzej Siewior , Clark Williams , "Paul E. McKenney" To: Paul Gortmaker Return-path: In-Reply-To: <20130710171212.GA13553@windriver.com> Sender: linux-kernel-owner@vger.kernel.org List-Id: linux-rt-users.vger.kernel.org On Wed, 2013-07-10 at 13:12 -0400, Paul Gortmaker wrote: > [[RFC][PATCH RT 6/6] vtime: Convert vtime_seqlock into raw_spinlock_t and seqcount combo] On 26/06/2013 (Wed 15:28) Steven Rostedt wrote: > > > The vtime seqlock needs to be taken in true interrupt context on -rt. > > The normal seqlocks are converted to mutexes when PREEMPT_RT_FULL is > > enabled, which will break the vtime code as the calls are done from > > interrupt context. > > > > Convert the vtime seqlock into the raw_spinlock_t and seqcount combo > > that can be done in interrupt context. > > Alternatively, we could revive the raw seqlock patch from Thomas? > > https://lkml.org/lkml/2010/2/17/238 > > Below is a version updating it to 3.8.x-RT. One downside is that > current mainline kernels have raw_seqcount_begin() function, which has > nothing to do with preempt-rt, so the seqcount function namespace can > get confusing unless we rename raw_seqcount_begin to something that > doesn't sound RT-ish. There's a reason that Thomas didn't bring that patch forward, but I don't recall what it was. Thomas? -- Steve