From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-out.m-online.net (mail-out.m-online.net [212.18.0.10]) by ozlabs.org (Postfix) with ESMTP id 365D2B6F06 for ; Tue, 13 Jul 2010 23:29:00 +1000 (EST) Message-ID: <4C3C6A24.3000906@grandegger.com> Date: Tue, 13 Jul 2010 15:29:08 +0200 From: Wolfgang Grandegger MIME-Version: 1.0 To: Netdev@vger.kernel.org Subject: High process latencies due to MPC5200 FEC hard- soft-irq processing Content-Type: text/plain; charset=ISO-8859-1 Cc: "linuxppc-dev@ozlabs.org" , LKML List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Hello, we realized, that multiple ping floods (ping -f) can cause very large high-priority process latencies (up to a many seconds) on a MPC5200 PowerPC system with FEC NAPI support. The latencies are measured with # cyclictest -p 80 -n The problem is that processing of the ICMP pakets in the Hard-Irq and Soft-IRQ context can last for a long time without returning to the scheduler. Reducing MAX_SOFTIRQ_RESTART from 10 to 2 helps - the latency goes down to 35 ms with 2 "ping -f" - but it's not a configurable parameter, even if it somehow depends on the CPU power. And using the -rt patches seems overkill to me. Any other ideas or comments on how to get rid of such high process latencies? Wolfgang.