From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Luck, Tony" Date: Wed, 22 Sep 2004 23:14:53 +0000 Subject: bk pull on ia64 linux tree Message-Id: <200409222314.i8MNErR05567@unix-os.sc.intel.com> List-Id: References: In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: linux-ia64@vger.kernel.org Hi Linus, please do a bk pull http://lia64.bkbits.net/linux-ia64-release-2.6.9 This will update the files shown below. Thanks! -Tony arch/ia64/hp/common/sba_iommu.c | 80 +++++++++++++++++++++++++++++++++++++++- arch/ia64/kernel/acpi.c | 67 +++++++++++++++++++++++++++++++++ arch/ia64/kernel/iosapic.c | 55 ++++++++++++++++++++++++++- arch/ia64/mm/discontig.c | 35 ++++++++++++----- arch/ia64/pci/pci.c | 5 ++ include/asm-ia64/iosapic.h | 3 + 6 files changed, 230 insertions(+), 15 deletions(-) through these ChangeSets: (04/09/09 1.1803.153.3) [IA64] iosapic.h, pci.c, iosapic.c, acpi.c: iosapic NUMA interrupt locality This patch associates IOSAPICs with NUMA nodes such that interrupts gets assigned to a reasonably good default CPU. The patch does not depend on the pxm_to_nid_map fixup, but results will be strange in some configurations without it. This should work on any NUMA box that exposes IOSAPICs with _MAT & _PXM methods, but it's only been tested on an rx8620. There should be no change in behavior for boxes that don't export both of these in ACPI namespace. Signed-off-by: Alex Williamson Signed-off-by: Tony Luck (04/09/09 1.1803.153.2) [IA64] sba_iommu.c: sba_iommu NUMA locality This patch adds NUMA locality info to sba_iommu, allowing coherent DMA mappings to be allocated from the node nearest to the I/O. This patch isn't dependent on the previous pxm_to_nid_map fixup, but may be sub-optimal in some cases without it. Thanks Signed-off-by: Alex Williamson Signed-off-by: Tony Luck (04/09/09 1.1803.153.1) [IA64] discontig.c: fixup pxm_to_nid_map I'd like to be able to use pxm_to_nid_map in several places to discover proximity domain to node id associations. Unfortunately, after reassign_cpu_only_nodes() plays with the node space, the pxm_to_nid_map doesn't necessarily reflect reality. This fixes up the table so it's still valid. Note that nid_to_pxm_map is still potentially broken, but has a one-to-many problem if the above function combines several proximity domains into a single node. Thanks to Bob Picco for the base patch. Signed-off-by: Alex Williamson Signed-off-by: Tony Luck