From mboxrd@z Thu Jan 1 00:00:00 1970 From: Thomas Gleixner Subject: Re: [PATCH v2] irq: Add node_affinity CPU masks for smarter irqbalance hints Date: Tue, 24 Nov 2009 23:23:44 +0100 (CET) Message-ID: References: <20091124093518.3909.16435.stgit@ppwaskie-hc2.jf.intel.com> <20091124.095703.107687163.davem@davemloft.net> <1259100343.2631.78.camel@ppwaskie-mobl2> Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Cc: David Miller , "linux-kernel@vger.kernel.org" , "arjan@linux.jf.intel.com" , "mingo@elte.hu" , "yong.zhang0@gmail.com" , "netdev@vger.kernel.org" To: Peter P Waskiewicz Jr Return-path: In-Reply-To: <1259100343.2631.78.camel@ppwaskie-mobl2> Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org On Tue, 24 Nov 2009, Peter P Waskiewicz Jr wrote: > On Tue, 2009-11-24 at 13:56 -0800, Thomas Gleixner wrote: > > On Tue, 24 Nov 2009, David Miller wrote: > > > > > From: Thomas Gleixner > > > Date: Tue, 24 Nov 2009 12:07:35 +0100 (CET) > > > > > > > And what does the kernel do with this information and why are we not > > > > using the existing device/numa_node information ? > > > > > > It's a different problem space Thomas. > > > > > > If the device lives on NUMA node X, we still end up wanting to > > > allocate memory resources (RX ring buffers) on other NUMA nodes on a > > > per-queue basis. > > > > > > Otherwise a network card's forwarding performance is limited by the > > > memory bandwidth of a single NUMA node, and on a multiqueue cards we > > > therefore fare much better by allocating each device RX queue's memory > > > resources on a different NUMA node. > > > > > > It is this NUMA usage that PJ is trying to export somehow to userspace > > > so that irqbalanced and friends can choose the IRQ cpu masks more > > > intelligently. > > > > So you need a preferred irq mask information on a per IRQ basis and > > that mask is not restricted to the CPUs of a single NUMA node, right ? > > > Just to clarify, I need a preferred CPU mask on a per IRQ basis. And > yes, that mask may not be restricted to the CPUs of a single NUMA node. > But in the normal case, the mask will be restricted to CPUs of a single > node. Right, but the normal case does not help much if we need to consider the special case of multiple nodes affected which requires another cpumask in irq_desc. That's what I really want to avoid. I at least understand the exact problem you guys want to solve. Will think more about it. Thanks, tglx