From mboxrd@z Thu Jan 1 00:00:00 1970 From: Stephane Eranian Date: Thu, 16 Nov 2006 16:58:25 +0000 Subject: Re: [PATCH] ia64 add idle loop entry/exit notifier Message-Id: <20061116165825.GH18162@frankl.hpl.hp.com> List-Id: References: <20061115222334.GE17238@frankl.hpl.hp.com> In-Reply-To: <20061115222334.GE17238@frankl.hpl.hp.com> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: linux-ia64@vger.kernel.org Ken, On Wed, Nov 15, 2006 at 06:07:19PM -0800, Chen, Kenneth W wrote: > Stephane Eranian wrote on Wednesday, November 15, 2006 2:24 PM > > Here is a patch which adds an idle notifier to IA-64. It clones > > the one in X86-64. You can register a callback via the notifier > > and you get called when: > > - entering the lowest level of the idle loop > > - exiting the lowest level of the idle loop, either normally > > or to process an interrupt > > > > Basically, you can monitor useful vs. useless work accomplished > > by the idle thread on each processor. > > Is idle notifier a light weight callback thingy? I happen to work Well, it uses the notifier interface. I have not changed this part. > on an application environment that cycles CPUs into gazillion nano > second sleep due to tons of sleep / wake up activity from thousands > of processes. I'm just checking and hope this notifier chain isn't > going to add significant overhead when cycling through busy/idle. > I would suggest you measure the overhead when no callback is registered. The patch is such that there will systematically be a callback registered for SGI machines. But if you are cycling busy/idle, I would think that you biggest cost if coming out of PAL_HALT. -- -Stephane