From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ingo Molnar Subject: Re: [PATCH net-next 2/2] net: exit busy loop when another process is runnable Date: Fri, 22 Aug 2014 09:42:24 +0200 Message-ID: <20140822074224.GB7372@gmail.com> References: <1408608310-13579-1-git-send-email-jasowang@redhat.com> <1408608310-13579-2-git-send-email-jasowang@redhat.com> <1408683665.5648.69.camel@marge.simpson.net> <53F6F14B.1030609@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Mike Galbraith , davem@davemloft.net, netdev@vger.kernel.org, linux-kernel@vger.kernel.org, mst@redhat.com, Peter Zijlstra , Ingo Molnar To: Jason Wang Return-path: Content-Disposition: inline In-Reply-To: <53F6F14B.1030609@redhat.com> Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org * Jason Wang wrote: > Polling could be done by either rx busy loop in process > context or NAPI in softirq. [...] Note that this shows another reason why it's a bad idea to query nr_running directly: depending on the softirq processing method, a softirq might run: - directly in process context - in an idle thread's context - or in a ksoftirqd context. 'nr_running' will have different values in these cases, causing assymetries in busy-poll handling! Another class of assymetry is when there are other softirq bits pending, beyond NET_RX (or NET_TX): a nr_running check misses them. The solution I outlined in the previous mail (using a sched_expected_runtime() method) would be able to avoid most of these artifacts. Thanks, Ingo