From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S265943AbUGZSid (ORCPT ); Mon, 26 Jul 2004 14:38:33 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S265784AbUGZSid (ORCPT ); Mon, 26 Jul 2004 14:38:33 -0400 Received: from mail1.slu.se ([130.238.96.11]:1244 "EHLO mail1.slu.se") by vger.kernel.org with ESMTP id S265943AbUGZQiP (ORCPT ); Mon, 26 Jul 2004 12:38:15 -0400 From: Robert Olsson MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <16645.13126.52445.630789@robur.slu.se> Date: Mon, 26 Jul 2004 18:37:26 +0200 To: Pasi Sjoholm Cc: Francois Romieu , H?ctor Mart?n , Linux-Kernel , , , Subject: Re: ksoftirqd uses 99% CPU triggered by network traffic (maybe RLT-8139 related) In-Reply-To: References: <20040725235927.B30025@electric-eye.fr.zoreil.com> X-Mailer: VM 7.18 under Emacs 21.3.1 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Pasi Sjoholm writes: > Pid: 2, comm: ksoftirqd/0 > EIP: 0060:[] CPU: 0 > EIP is at rtl8139_poll+0xb4/0x100 [8139too] > EFLAGS: 00000247 Not tainted (2.6.7-mm7) > EAX: ffffe000 EBX: 00000040 ECX: df4824f8 EDX: c0441978 > ESI: df482400 EDI: e0868000 EBP: dff85f80 DS: 007b ES: 007b > CR0: 8005003b CR2: b7c5a000 CR3: 1fafd000 CR4: 000006d0 > [] ksoftirqd+0x0/0xc0 > [] net_rx_action+0x6a/0x110 > [] __do_softirq+0xa9/0xb0 > [] do_softirq+0x27/0x30 > [] ksoftirqd+0x68/0xc0 > [] kthread+0xa5/0xb0 > [] kthread+0x0/0xb0 > [] kernel_thread_helper+0x5/0x14 > -- > > I'm not a kernel expert but it would seem that ksoftirqd is in some sort a > loop because I didn't get any "printk("%s wakes ksoftirqd\n", > current->comm);"-lines. Hello! This looks very much like the problem we see when doing route DoS testing with Alexey. In summary: High softirq loads can totally kill userland. The reason is that do_softirq() is run from many places hard interrupts, local_bh_enable etc and bypasses the ksoftirqd protection. It just been discussed at OLS with Andrea and Dipankar and others. Current RCU suffers from this problem as well. I've experimented some code to defer softirq's to ksoftirqd after a time as well as deferring all softirq's to ksoftirqd. Andrea had some ideas as well as Ingo. Cheers. --ro