From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755574Ab0IGH3D (ORCPT ); Tue, 7 Sep 2010 03:29:03 -0400 Received: from mail-qw0-f46.google.com ([209.85.216.46]:42753 "EHLO mail-qw0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751494Ab0IGH25 convert rfc822-to-8bit (ORCPT ); Tue, 7 Sep 2010 03:28:57 -0400 MIME-Version: 1.0 X-Originating-IP: [81.243.35.243] In-Reply-To: <201009061400.05637.trenn@suse.de> References: <201009061243.00479.trenn@suse.de> <201009061400.05637.trenn@suse.de> Date: Tue, 7 Sep 2010 09:28:56 +0200 Message-ID: Subject: Re: [PATCH] tracing, perf: add more power related events From: Jean Pihet To: Thomas Renninger Cc: discuss@lesswatts.org, linux-pm@lists.linux-foundation.org, Ingo Molnar , Arjan van de Ven , Peter Zijlstra , Len Brown , arjan@infradead.org, Kevin Hilman , linux-kernel@vger.kernel.org Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 8BIT Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Thomas, On Mon, Sep 6, 2010 at 2:00 PM, Thomas Renninger wrote: > On Monday 06 September 2010 13:27:13 Jean Pihet wrote: >> On Mon, Sep 6, 2010 at 12:42 PM, Thomas Renninger wrote: > ... >> In order to reach an efficient low power mode you >> need to ensure that all clocks in a clock domain are idled or cut off >> (automatically by the HW or with the intervention of SW drivers). >> >> The HW complexity requires the tracing of all those clocks/power >> states, that is why new events are proposed. > Hm, but if the HW can switch clocks behind your back, how do you > realize that to throw a perf clock power event then? Such clocks are automatically shut off when there is no activity on the module, this allows the system to reach a lower power state. All that is needed is to ensure the modules are correctly configured to shut off unneeded clocks automatically. > Theoretically you need to poll to keep track of HW switching the > clocks? No. I would not poll hundreds of clocks. Furthermore there is no event to be thrown away, in the case the reached power mode is not 'deep' enough you need a trace point showing what blocked the transition. There is still work to do to achieve that, it is coming soon. > >    Thomas > Note: a re-spin of the patch has been sent on a new thread. Jean