When a system has I/O devices (ex. PCI bridges) with their own locality, the following error message shows up.
NUMA: Warning: node ids are out of bound, from=-1 to=-1 distance=10 acpi_numa_slit_init() calls numa_set_distance(), which assumes that all nodes on the system have been parsed with SRAT already. However, SRAT does not list I/O devices. SLIT has the distance table for all localities including I/Os. Hence, the above message shows up when a system has a unique I/O device locality. This patch changes acpi_numa_slit_init() to make sure all the nodes are parsed, so that it can initialize the distance table with all the localities. The following map tables may contain I/O nodes as a result. - numa_distance[], i.e. node distance table - node_states[N_POSSIBLE], aka. node_possible_map - mp_bus_to_node[], i.e. pci bus# to node# map There is no functional change since there is no code that makes use of the I/O nodes. - I/O nodes are set to off-line. - pci_acpi_scan_root() continues to set -1 to pci_sysdata.node of a PCI bridge with a unique locality (per commit b755de8d). Signed-off-by: Toshi Kani <[email protected]> --- arch/x86/mm/srat.c | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-) diff --git a/arch/x86/mm/srat.c b/arch/x86/mm/srat.c index 26f4e12..fb5bd73 100644 --- a/arch/x86/mm/srat.c +++ b/arch/x86/mm/srat.c @@ -45,7 +45,20 @@ static __init inline int srat_disabled(void) /* Callback for SLIT parsing */ void __init acpi_numa_slit_init(struct acpi_table_slit *slit) { - int i, j; + int node, i, j; + + /* SLIT may have I/O nodes, which are not listed in SRAT */ + for (i = 0; i < slit->locality_count; i++) { + if (pxm_to_node(i) != NUMA_NO_NODE) + continue; + + node = setup_node(i); + if (WARN_ONCE(node < 0, "SLIT: Too many proximity domains.\n")) + continue; + + node_set(node, numa_nodes_parsed); + pr_info("SLIT: Node %u PXM %u I/O only\n", node, i); + } for (i = 0; i < slit->locality_count; i++) for (j = 0; j < slit->locality_count; j++) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
