From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Rafael J. Wysocki" Date: Wed, 21 May 2014 19:09:08 +0000 Subject: Re: [regression][bisected] zx1 AGP GART no more detected since merge of 'acpi-scan' branch Message-Id: <537CF9D4.2070504@intel.com> List-Id: References: In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="windows-1252" Content-Transfer-Encoding: quoted-printable To: linux-ia64@vger.kernel.org Hi, On 5/21/2014 5:31 PM, =C3=89meric MASCHINO wrote: > Again, > > I'm now quite confident that the regression comes from commit > 66345d5f79fcfa0214f5d98763643d4ee8e6965d as the description talks > about an ordering issue on hp rx2600 between the SBA I/O MMU driver > and the container scan handler. > > First, I'm running on hp zx6000 that's a rx2600 with an AGP slot. > Then, defining DEBUG_SBA_INIT in sba_iommu.c clearly shows in dmesg > logs that, for working kernel 3.10.40, ioc_init (and thus > ioc_iova_init) is called once the ACPI/PCI resources (is this the > right terminology?) have been parsed, whereas for non-working kernel > 3.11-rc1, ioc_init is called at the very beginning. Quite evidently, acpi_sba_ioc_add() has to be executed after=20 acpi_pci_root_add() and we need to enforce that ordering somehow. Please create a bug entry for that at bugzilla.kernel.org (against=20 ACPI), mark it as a regression and assign it to rjw@rjwysocki.net. Thanks, Rafael