From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751888AbbCYIEL (ORCPT ); Wed, 25 Mar 2015 04:04:11 -0400 Received: from bombadil.infradead.org ([198.137.202.9]:55041 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750803AbbCYIED (ORCPT ); Wed, 25 Mar 2015 04:04:03 -0400 Date: Wed, 25 Mar 2015 09:03:52 +0100 From: Peter Zijlstra To: Chris Metcalf Cc: linux-kernel@vger.kernel.org, Ingo Molnar , "Paul E. McKenney" , Thomas Gleixner , Frederic Weisbecker Subject: Re: [PATCH 1/4] tile: support arch_irq_work_raise Message-ID: <20150325080352.GC27490@worktop.programming.kicks-ass.net> References: <1427224895-30830-1-git-send-email-cmetcalf@ezchip.com> <1427224895-30830-2-git-send-email-cmetcalf@ezchip.com> <20150324211406.GB27490@worktop.programming.kicks-ass.net> <5511DFE2.4010605@ezchip.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <5511DFE2.4010605@ezchip.com> User-Agent: Mutt/1.5.22.1 (2013-10-16) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Tue, Mar 24, 2015 at 06:06:26PM -0400, Chris Metcalf wrote: > On 3/24/2015 5:14 PM, Peter Zijlstra wrote: > >On Tue, Mar 24, 2015 at 03:21:32PM -0400, cmetcalf@ezchip.com wrote: > >>From: Chris Metcalf > >> > >>Tile includes a hypervisor hook to deliver messages to arbitrary > >>tiles, so we can use that to raise an interrupt as soon as > >>possible on our own core. Unfortunately the Tilera hypervisor > >>disabled that support on principle in previous releases, but > >>it will be available in MDE 4.3.4 and later. > >Can you program a timer in the (recent) past which will instantly > >trigger an interrupt? This is what PPC64 does to implement the self-ipi. > > I looked in git history a bit and see commit 105988c015943 from 2009, > which looks like it was the basis for the powerpc support. I'm a little leery > of just randomly changing the in-flight timer decrementer value, though; > does the timer event_handler properly handle being called early, and > then properly handle resetting the decrementer to the right value? > I guess if both of those things are true, it seems plausible to adopt the > approach you suggested. > > On the other hand, the approach I coded up avoids making any of those > slightly scary assumptions about the timer subsystem, and I don't really > have a problem with saying you need a recent Tilera hypervisor binary if > you want to use NOHZ_FULL... Fair enough; and your call obviously. Just wanted to share that there are alternative ways if you really need.