From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754497AbZHPOZS (ORCPT ); Sun, 16 Aug 2009 10:25:18 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1754323AbZHPOZR (ORCPT ); Sun, 16 Aug 2009 10:25:17 -0400 Received: from www.tglx.de ([62.245.132.106]:57261 "EHLO www.tglx.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752298AbZHPOZQ (ORCPT ); Sun, 16 Aug 2009 10:25:16 -0400 Date: Sun, 16 Aug 2009 16:25:13 +0200 (CEST) From: Thomas Gleixner To: Michael Buesch cc: linux-kernel@vger.kernel.org Subject: Re: Threaded interrupt handlers broken? In-Reply-To: <200908161546.46328.mb@bu3sch.de> Message-ID: References: <200908161153.14081.mb@bu3sch.de> <200908161445.53147.mb@bu3sch.de> <200908161546.46328.mb@bu3sch.de> User-Agent: Alpine 2.00 (LFD 1167 2008-08-23) MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sun, 16 Aug 2009, Michael Buesch wrote: > On Sunday 16 August 2009 15:22:29 Thomas Gleixner wrote: > > > > + if (0&&unlikely(desc->status & IRQ_DISABLED)) { > > > > So the interrupt is marked disabled. How do you setup the handler > > ? And what does the primary handler do ? Can you post your driver > > code please? > > This patch converts the b43 driver to threaded interrupts: > http://bu3sch.de/patches/wireless-testing/20090816-1535/patches/002-b43-threaded-irq-handler.patch On the first glance this looks not too bad. the unlocked access to the irq status registers looks a bit scary, but that is not relevant for the problem at hand. > It kind of works with this hack applied to kernel/irq/manage.c Hmm. Is the interrupt of the device shared ? If yes, what's the other device on that interrupt line ? what puzzles me is the fact that the IRQ_DISABLED flag is set. Is there anything unusual in dmesg ? > > So you wake when the thread counter is != 0 after the decrement. > > > > #define atomic_dec_and_test(v) (atomic_sub_return(1, (v)) == 0) > > Yeah, isn't that what we want to do? I read the test as "wake other threads, > if there are other threads" or something like that. No, it's for synchronize_irq(). When there are threaded handlers in progress, then sychronize_irq() waits on the waitqueue until they are finished. So we wake the queue when the last threaded handler of this irq line returns from thread_fn. Thanks, tglx