From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ingo Molnar Subject: Re: 2.6.23-rc2: WARNING: at kernel/irq/resend.c:70 check_irq_resend() Date: Fri, 10 Aug 2007 10:30:50 +0200 Message-ID: <20070810083050.GA4545@elte.hu> References: <18107.11431.838905.331157@stoffel.org> <20070809155445.GA5161@ff.dom.local> <1186733140.12828.45.camel@chaos> <20070810082346.GD1764@ff.dom.local> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Thomas Gleixner , John Stoffel , linux-kernel@vger.kernel.org, shemminger@linux-foundation.org, vignaud@xandmail.fr, marcin.slusarz@gmail.com, torvalds@linux-foundation.org, akpm@linux-foundation.org, alan@lxorguk.ukuu.org.uk, linux-net@vger.kernel.org, netdev@vger.kernel.org To: Jarek Poplawski Return-path: Content-Disposition: inline In-Reply-To: <20070810082346.GD1764@ff.dom.local> Sender: linux-net-owner@vger.kernel.org List-Id: netdev.vger.kernel.org * Jarek Poplawski wrote: > > Hmm. This solution is still just pampering over the real problem. > > The delayed disable just re-sends level interrupts unnecessarily. I > > have a fix (needs some testing) for this, which I send out tomorrow, > > when I'm really back from vacation. > > > > But suppressing the resend is not fixing the driver problem. The > > problem can show up with spurious interrupts and with interrupts on > > a shared PCI interrupt line at any time. It just might take weeks > > instead of minutes. > > Doesn't it look like a little change of mind? [...] what change of mind do you mean exactly? > [...] Well, there are probably (but need more testing) two other > solutions: _SW_RESEND and disabling without delay for levels only... IIRC Marcin tested software-resend and it didnt fix the hang. That strongly points in the direction of a driver bug (or a genirq bug) being made more prominent by the genirq change - not any hardware detail such as the APIC vector-retrigger sequence. While we'd like to see the suspected driver bug (or any higher level genirq bug) fixed, we'll undo the effect of the genirq change (because it is causing a regression). We'll also add a separate, optional irq-debugging feature that generates high-rate interrupts on any shared irq line. (and thus artificially stresses the robustness of the driver and the genirq layer against spurious interrupts.) Ingo