From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ingo Molnar Subject: Re: [bloat] Measuring header file bloat effects on kernel build performance: a more than 2x slowdown ... Date: Mon, 23 May 2011 19:01:26 +0200 Message-ID: <20110523170126.GA14407@elte.hu> References: <20110520161210.81bbef3a.sfr@canb.auug.org.au> <20110523090918.GA5474@elte.hu> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from mx3.mail.elte.hu ([157.181.1.138]:42955 "EHLO mx3.mail.elte.hu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932123Ab1EWRBp (ORCPT ); Mon, 23 May 2011 13:01:45 -0400 Content-Disposition: inline In-Reply-To: Sender: linux-next-owner@vger.kernel.org List-ID: To: Linus Torvalds Cc: Stephen Rothwell , Thomas Gleixner , "H. Peter Anvin" , David Miller , linux-next@vger.kernel.org, linux-kernel@vger.kernel.org, x86@kernel.org, Peter Zijlstra , Andrew Morton * Linus Torvalds wrote: > On Mon, May 23, 2011 at 2:09 AM, Ingo Molnar wrote: > > > > I've attached a totally hacky patch that removes all the big #include's from > > kernel/pid.c and includes all structure and API definitions explicitly. > > Hmm. > > A less hacky patch might be to split up "sched.h" into multiple > smaller things and at least get *part* of the way. > Yeah, absolutely - Peter already raised that a couple of days ago in a sched.h discussion so it's on the radar. The hacky patch was really just a throw-away attempt to see where we stand - we only had vague impressions about the level of problems we have, now we know some numbers. > A lot of things want "struct task_struct" (and in some cases thread_info, but > that's already split). > > Much fewer care about the signal stuff. > > And many things probably don't even need the task_struct definition, and > might be perfectly happy with just function calls rather than having intimate > knowledge of the structure layout and an inline function. Yeah. Thanks, Ingo