Return-path: Received: from mail-qw0-f46.google.com ([209.85.216.46]:50912 "EHLO mail-qw0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757382Ab0JLRPd (ORCPT ); Tue, 12 Oct 2010 13:15:33 -0400 Received: by qwa26 with SMTP id 26so527311qwa.19 for ; Tue, 12 Oct 2010 10:15:32 -0700 (PDT) MIME-Version: 1.0 In-Reply-To: <201010121933.59209.br1@einfach.org> References: <201010121933.59209.br1@einfach.org> Date: Tue, 12 Oct 2010 13:15:29 -0400 Message-ID: Subject: Re: [ath5k-devel] How to profile this strange effect under high load? From: Bob Copeland To: Bruno Randolf Cc: linux-wireless , ath5k-devel@lists.ath5k.org Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-wireless-owner@vger.kernel.org List-ID: On Tue, Oct 12, 2010 at 6:33 AM, Bruno Randolf wrote: > So my question is: How can I find out, what [phy1] is so busy doing? Any ideas > how to profile this? /proc/interrupts shows lots of interrupts on phy1 line? Is it shared irq? What options did you try with oprofile? You can try perf -- as long as you limit the functions you profile it should have about the same overhead as oprofile. But then you should get about the same results with either; I just find perf to be a little better these days. One thing I know of is that all the linear interpolation we do for setting the power curves is pretty heavy cpu-wise and we do it on every reset. There's some low-hanging fruit there like precomputing the reciprocal of some scaling factors instead of some divisions... -- Bob Copeland %% www.bobcopeland.com