Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751322AbWEDFHf (ORCPT ); Thu, 4 May 2006 01:07:35 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751092AbWEDFHf (ORCPT ); Thu, 4 May 2006 01:07:35 -0400 Received: from mga02.intel.com ([134.134.136.20]:16508 "EHLO orsmga101-1.jf.intel.com") by vger.kernel.org with ESMTP id S1751399AbWEDFHe convert rfc822-to-8bit (ORCPT ); Thu, 4 May 2006 01:07:34 -0400 X-IronPort-AV: i="4.05,86,1146466800"; d="scan'208"; a="31261620:sNHT19774076" Content-class: urn:content-classes:message MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 8BIT X-MIMEOLE: Produced By Microsoft Exchange V6.5 Subject: RE: [(repost) git Patch 1/1] avoid IRQ0 ioapic pin collision Date: Thu, 4 May 2006 01:07:29 -0400 Message-ID: X-MS-Has-Attach: X-MS-TNEF-Correlator: Thread-Topic: [(repost) git Patch 1/1] avoid IRQ0 ioapic pin collision Thread-index: AcZtsT6vgePsEi1vT3SuA0EDhIDZnQAAVFawAF4YQWA= From: "Brown, Len" To: "Eric W. Biederman" Cc: "Protasevich, Natalie" , "Andi Kleen" , , "Kimball Murray" , , , , X-OriginalArrivalTime: 04 May 2006 05:07:30.0461 (UTC) FILETIME=[A50D6CD0:01C66F38] Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 1726 Lines: 44 >>> We should never have had a problem with un-connected interrupt lines >>> consuming vectors, as the vectors are handed out at run-time >>> only when interrupts are requested by drivers. >> >>Incorrect. By being requested by drivers I assume you mean by >>request_irq. assign_irq_vector is called long before request_irq >>and in fact there is currently not a call from request_irq to >>assign_irq_vector. Look where assign_irq_vector is called in >io_apic.c Hmmm, on Natalie's ping, I looked at this again. setup_IO_APIC_irqs() actually consumes only 16 vectors. This is because it only sets up the IRQs for pins found with find_irq_entry(), which searches mp_irqs[] which at this point contains only the legacy identify mappings. The PNP code then takes a swing at things, and it registers a handful of gsi's, but they are all duplicates of the legacy mappings already set-up, so no additional vectors are consumed. So on a big system, the large quantity of vectors will not be consumed at IOAPIC initialization time, but later at device probe time. Not via request_irq(), but via pci_enable_device(): pci_enable_device() pci_enable_device_bars() pcibios_enable_device() pcibios_enable_irq() -> acpi_pci_irq_enable() acpi_register_gsi() mp_register_gsi() io_apic_set_pci_routing() entry.vector = assign_irq_vector(irq) So except for the legacy IRQs, we are already allocating the vectors on-demand, and that doesn't need to be fixed. -Len - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/