From mboxrd@z Thu Jan 1 00:00:00 1970 From: Pavel Emelyanov Subject: Re: Kernel text size with pid namespace Date: Thu, 20 Sep 2007 15:55:01 +0400 Message-ID: <46F25F95.8000103@openvz.org> References: <20070920001644.GA14880@us.ibm.com> <46F239B2.8080500@openvz.org> <20070920040406.57e40ece.pj@sgi.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20070920040406.57e40ece.pj-sJ/iWh9BUns@public.gmane.org> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: containers-bounces-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org Errors-To: containers-bounces-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org To: Paul Jackson Cc: containers-qjLDD68F18O7TbgM5vRIOg@public.gmane.org, mpm-VDJrAJ4Gl5ZBDgjK7y7TUQ@public.gmane.org List-Id: containers.vger.kernel.org Paul Jackson wrote: >>> functions are used in process creation/termination, we would need to >>> keep them inline, when optimizing for performance. >> I'd keep them inline for performance reasons. > > As Matt Mackall explained more carefully in his reply, it's no longer > clear that inlining is best for performance in as many situations as > it was the past. Cache footprint size tends to dominate performance on > present day processors. True, but AFAIR, when I developed the namespaces I tried to move the task_pid_nr etc calls in kernel/pid.c and the performance on unixbench spawn and nptlperf tests became worse. That's why I said that I'd keep them inline. > See also Matt's comments on the NULL struct pid check. Getting rid > of conditional jumps may be the more important performance issue here. That's a valid argument. I will look over it. Thanks, Pavel