From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ingo Molnar Subject: Re: [RFC][PATCH] sched_clock_cpu() Date: Sun, 4 May 2008 11:00:25 +0200 Message-ID: <20080504090024.GA29318@elte.hu> References: <1209832169.6929.42.camel@lappy> <20080503170130.GB21967@elte.hu> <20080503.212607.157994166.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from mx3.mail.elte.hu ([157.181.1.138]:41250 "EHLO mx3.mail.elte.hu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752262AbYEDJAn (ORCPT ); Sun, 4 May 2008 05:00:43 -0400 Content-Disposition: inline In-Reply-To: <20080503.212607.157994166.davem@davemloft.net> Sender: linux-arch-owner@vger.kernel.org List-ID: To: David Miller Cc: a.p.zijlstra@chello.nl, tglx@linutronix.de, guichaz@gmail.com, andi@firstfloor.org, linux-kernel@vger.kernel.org, linux-arch@vger.kernel.org, efault@gmx.de, dhaval@linux.vnet.ibm.com * David Miller wrote: > > that wont work very well when sched_clock() is called from within > > CONFIG_LOCK_STAT instrumentation. Does the patch below solve the > > boot problems for you? > > Also, no platform can set HAVE_STABLE_CLOCK until we instantiate it in > a Kconfig somewhere. I've choosen to do it in kernel/Kconfig.hz and > here are the sparc/sparc64 bits as well, I've booted this up with > Peter's patch on my 64-cpu niagara2 box and done some basic testing. applied, thanks David. right now this topic looks good in review and in testing but it is stalled on a bug: in overnight testing it triggered an ftrace self-test hang that i bisected down to that patch. While that doesnt affect mainline it's something that shows that the new sched_clock() code is not as widely usable as the old code - have to investigate that some more. > It would be nice if a powerpc person could test the trivial powerpc > Kconfig patch. > > Possibly this should be HAVE_UNSTABLE_CLOCK, then only one platform > needs to set it :-) heh, indeed :) Initially i thought that it's better to first be safe, but this really will only affect x86 in practice, so ... i think we'll switch around the flag, turning this into a no-effort thing for everything but x86. Ingo