From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752908AbbDIHUq (ORCPT ); Thu, 9 Apr 2015 03:20:46 -0400 Received: from mail-wg0-f49.google.com ([74.125.82.49]:32888 "EHLO mail-wg0-f49.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751592AbbDIHUo (ORCPT ); Thu, 9 Apr 2015 03:20:44 -0400 Date: Thu, 9 Apr 2015 09:20:39 +0200 From: Ingo Molnar To: Peter Zijlstra Cc: Thomas Gleixner , Viresh Kumar , Ingo Molnar , linaro-kernel@lists.linaro.org, linux-kernel@vger.kernel.org, Preeti U Murthy Subject: Re: [PATCH] hrtimer: Replace cpu_base->active_bases with a direct check of the active list Message-ID: <20150409072038.GA30205@gmail.com> References: <20150409062841.GB14259@gmail.com> <20150409065730.GK27490@worktop.programming.kicks-ass.net> <20150409070917.GF14259@gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20150409070917.GF14259@gmail.com> User-Agent: Mutt/1.5.23 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org * Ingo Molnar wrote: > > * Peter Zijlstra wrote: > > > On Thu, Apr 09, 2015 at 08:28:41AM +0200, Ingo Molnar wrote: > > > Btw., does cpu_base->active_bases even make sense? hrtimer bases are > > > fundamentally percpu, and to check whether there are any pending > > > timers is a very simple check: > > > > > > base->active->next != NULL > > > > > > > Yeah, that's 3 pointer dereferences from cpu_base, iow you traded a > > single bit test on an already loaded word for 3 potential cacheline > > misses. > > But the clock bases are not aligned to cachelines, and we have 4 of > them. So in practice when we access one, we'll load the next one > anyway. > > Furthermore the simplification is measurable, and a fair bit of it is > in various fast paths. I'd rather trade a bit of a cacheline footprint > for less overall complexity and faster code. Plus, look at this code in hrtimer_run_queues(): for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) { base = &cpu_base->clock_base[index]; if (!base->active.next) continue; if (gettime) { hrtimer_get_softirq_time(cpu_base); gettime = 0; } if at least one base is active (on my fairly standard system all cpus have at least one active hrtimer base all the time - and many cpus have two bases active), then we run hrtimer_get_softirq_time(), which dirties the cachelines of all 4 clock bases: base->clock_base[HRTIMER_BASE_REALTIME].softirq_time = xtim; base->clock_base[HRTIMER_BASE_MONOTONIC].softirq_time = mono; base->clock_base[HRTIMER_BASE_BOOTTIME].softirq_time = boot; base->clock_base[HRTIMER_BASE_TAI].softirq_time = tai; so in practice we not only touch every cacheline in every timer interrupt, but we _dirty_ them, even the inactive ones. So I'd strongly argue in favor of this patch series of simplification: it makes the code simpler and faster, and won't impact cache footprint in practice. Thanks, Ingo