From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S270038AbUJSWsp (ORCPT ); Tue, 19 Oct 2004 18:48:45 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S270039AbUJSWnI (ORCPT ); Tue, 19 Oct 2004 18:43:08 -0400 Received: from adsl-63-197-226-105.dsl.snfc21.pacbell.net ([63.197.226.105]:35733 "EHLO cheetah.davemloft.net") by vger.kernel.org with ESMTP id S269880AbUJSWig (ORCPT ); Tue, 19 Oct 2004 18:38:36 -0400 Date: Tue, 19 Oct 2004 15:33:08 -0700 From: "David S. Miller" To: Lee Revell Cc: herbert@gondor.apana.org.au, vda@port.imtp.ilyichevsk.odessa.ua, netdev@oss.sgi.com, linux-kernel@vger.kernel.org, maxk@qualcomm.com, irda-users@lists.sourceforge.net Subject: Re: tun.c patch to fix "smp_processor_id() in preemptible code" Message-Id: <20041019153308.488d34c1.davem@davemloft.net> In-Reply-To: <1098223857.23367.35.camel@krustophenia.net> References: <1098222676.23367.18.camel@krustophenia.net> <20041019215401.GA16427@gondor.apana.org.au> <1098223857.23367.35.camel@krustophenia.net> X-Mailer: Sylpheed version 0.9.12 (GTK+ 1.2.10; sparc-unknown-linux-gnu) X-Face: "_;p5u5aPsO,_Vsx"^v-pEq09'CU4&Dc1$fQExov$62l60cgCc%FnIwD=.UF^a>?5'9Kn[;433QFVV9M..2eN.@4ZWPGbdi<=?[:T>y?SD(R*-3It"Vj:)"dP Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Tue, 19 Oct 2004 18:10:58 -0400 Lee Revell wrote: > /* > * Since receiving is always initiated from a tasklet (in iucv.c), > * we must use netif_rx_ni() instead of netif_rx() > */ > > This implies that the author thought it was a matter of correctness to > use netif_rx_ni vs. netif_rx. But it looks like the only difference is > that the former sacrifices preempt-safety for performance. You can't really delete netif_rx_ni(), so if there is a preemptability issue, just add the necessary preemption protection around the softirq checks.