Messages in this thread Patch in this message | | | Date | Fri, 22 Sep 2006 10:12:18 +0900 | From | KAMEZAWA Hiroyuki <> | Subject | [BUGFIX][PATCH] node hotplug : symblic link between node and cpu fixup |
| |
This patch is tesed in my environment, but I want a review form other people, who does node-hotplug (maybe x86_64...?). = When adding cpu to the NUMA system, node-id for it should be determined. But at boot time, possible cpu may not belongs to any node if cpu is physically offlined. (found in ia64)
This patch does: 1. avoid creating symbolic link from a node to the cpu if node_to_cpu_mask is not set. (cpu_to_node_map[] is initialized by '0', not useful here) 2. register cpu_hotplug_notifier to bind a cpu to a node. Because cpu is up here, cpu-to-node relationship must be fixed before here. 3. at cpu offlining, remove symbolic link to cpu from node.
Tested on ia64/NUMA system, which supports physical node-hot-plug.
Signed-Off-By: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
drivers/base/node.c | 37 +++++++++++++++++++++++++++++++++++-- 1 files changed, 35 insertions(+), 2 deletions(-)
Index: linux-2.6.18/drivers/base/node.c =================================================================== --- linux-2.6.18.orig/drivers/base/node.c 2006-09-22 09:30:25.000000000 +0900 +++ linux-2.6.18/drivers/base/node.c 2006-09-22 09:42:22.000000000 +0900 @@ -169,19 +169,26 @@ } struct node node_devices[MAX_NUMNODES]; +cpumask_t registered_cpus; /* * register cpu under node */ int register_cpu_under_node(unsigned int cpu, unsigned int nid) { - if (node_online(nid)) { + int err; + if (node_online(nid) && + cpu_isset(cpu, node_to_cpumask(nid)) && + !cpu_isset(cpu, registered_cpus)) { struct sys_device *obj = get_cpu_sysdev(cpu); if (!obj) return 0; - return sysfs_create_link(&node_devices[nid].sysdev.kobj, + err = sysfs_create_link(&node_devices[nid].sysdev.kobj, &obj->kobj, kobject_name(&obj->kobj)); + if (!err) + cpu_set(cpu, registered_cpus); + return err; } return 0; @@ -191,9 +198,11 @@ { if (node_online(nid)) { struct sys_device *obj = get_cpu_sysdev(cpu); + WARN_ON(!cpu_isset(cpu, registered_cpus)); if (obj) sysfs_remove_link(&node_devices[nid].sysdev.kobj, kobject_name(&obj->kobj)); + cpu_clear(cpu, registered_cpus); } return 0; } @@ -223,6 +232,28 @@ } +static int +cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) +{ + long cpu = (long)hcpu; + int nid = cpu_to_node(cpu); + + switch(action) { + case CPU_ONLINE: + register_cpu_under_node(cpu, nid); + break; + case CPU_DEAD: + unregister_cpu_under_node(cpu, nid); + break; + default: + break; + } + return NOTIFY_OK; +} + +static struct notifier_block node_cpu_notifier = { &cpu_callback, NULL, 0}; + + void unregister_one_node(int nid) { unregister_node(&node_devices[nid]); @@ -230,6 +261,8 @@ static int __init register_node_type(void) { + register_cpu_notifier(&node_cpu_notifier); return sysdev_class_register(&node_class); } + postcore_initcall(register_node_type); - 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/
| |