2010-11-13 07:36:36

by Zheng, Shaohui

[permalink] [raw]
Subject: [v2,3/8] NUMA Hotplug emulator


From: Haicheng Li <[email protected]>
Subject: hotplug emulator: Userland interface to hotplug-add fake offlined nodes.

Add a sysfs entry "probe" under /sys/devices/system/node/:

- to show all fake offlined nodes:
$ cat /sys/devices/system/node/probe

- to hotadd a fake offlined node, e.g. nodeid is N:
$ echo N > /sys/devices/system/node/probe

Signed-off-by: Haicheng Li <[email protected]>
Signed-off-by: Shaohui Zheng <[email protected]>
---
diff --git a/drivers/base/node.c b/drivers/base/node.c
index 2872e86..d21bea2 100644
--- a/drivers/base/node.c
+++ b/drivers/base/node.c
@@ -538,6 +538,25 @@ void unregister_one_node(int nid)
unregister_node(&node_devices[nid]);
}

+#ifdef CONFIG_NODE_HOTPLUG_EMU
+static ssize_t store_nodes_probe(struct sysdev_class *class,
+ struct sysdev_class_attribute *attr,
+ const char *buf, size_t count)
+{
+ long nid;
+
+ strict_strtol(buf, 0, &nid);
+ if (nid < 0 || nid > nr_node_ids - 1) {
+ printk(KERN_ERR "Invalid NUMA node id: %ld (0 <= nid < %d).\n",
+ nid, nr_node_ids);
+ return -EPERM;
+ }
+ hotadd_hidden_nodes(nid);
+
+ return count;
+}
+#endif
+
/*
* node states attributes
*/
@@ -566,26 +585,35 @@ static ssize_t show_node_state(struct sysdev_class *class,
return print_nodes_state(na->state, buf);
}

-#define _NODE_ATTR(name, state) \
+#define _NODE_ATTR_RO(name, state) \
{ _SYSDEV_CLASS_ATTR(name, 0444, show_node_state, NULL), state }

+#define _NODE_ATTR_RW(name, store_func, state) \
+ { _SYSDEV_CLASS_ATTR(name, 0644, show_node_state, store_func), state }
+
static struct node_attr node_state_attr[] = {
- _NODE_ATTR(possible, N_POSSIBLE),
- _NODE_ATTR(online, N_ONLINE),
- _NODE_ATTR(has_normal_memory, N_NORMAL_MEMORY),
- _NODE_ATTR(has_cpu, N_CPU),
+ [N_POSSIBLE] = _NODE_ATTR_RO(possible, N_POSSIBLE),
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+ [N_HIDDEN] = _NODE_ATTR_RW(probe, store_nodes_probe, N_HIDDEN),
+#endif
+ [N_ONLINE] = _NODE_ATTR_RO(online, N_ONLINE),
+ [N_NORMAL_MEMORY] = _NODE_ATTR_RO(has_normal_memory, N_NORMAL_MEMORY),
#ifdef CONFIG_HIGHMEM
- _NODE_ATTR(has_high_memory, N_HIGH_MEMORY),
+ [N_HIGH_MEMORY] = _NODE_ATTR_RO(has_high_memory, N_HIGH_MEMORY),
#endif
+ [N_CPU] = _NODE_ATTR_RO(has_cpu, N_CPU),
};

static struct sysdev_class_attribute *node_state_attrs[] = {
- &node_state_attr[0].attr,
- &node_state_attr[1].attr,
- &node_state_attr[2].attr,
- &node_state_attr[3].attr,
+ &node_state_attr[N_POSSIBLE].attr,
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+ &node_state_attr[N_HIDDEN].attr,
+#endif
+ &node_state_attr[N_ONLINE].attr,
+ &node_state_attr[N_NORMAL_MEMORY].attr,
+ &node_state_attr[N_CPU].attr,
#ifdef CONFIG_HIGHMEM
- &node_state_attr[4].attr,
+ &node_state_attr[N_HIGH_MEMORY].attr,
#endif
NULL
};
diff --git a/mm/Kconfig b/mm/Kconfig
index f0fb912..a01a679 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -147,6 +147,21 @@ config MEMORY_HOTREMOVE
depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE
depends on MIGRATION

+config NUMA_HOTPLUG_EMU
+ bool "NUMA hotplug emulator"
+ depends on X86_64 && NUMA && MEMORY_HOTPLUG
+
+ ---help---
+
+config NODE_HOTPLUG_EMU
+ bool "Node hotplug emulation"
+ depends on NUMA_HOTPLUG_EMU && MEMORY_HOTPLUG
+ ---help---
+ Enable Node hotplug emulation. The machine will be setup with
+ hidden virtual nodes when booted with "numa=hide=N*size", where
+ N is the number of hidden nodes, size is the memory size per
+ hidden node. This is only useful for debugging.
+
#
# If we have space for more page flags then we can enable additional
# optimizations and functionality.

--
Thanks & Regards,
Shaohui


2010-11-15 06:47:01

by Paul Mundt

[permalink] [raw]
Subject: Re: [v2,3/8] NUMA Hotplug emulator

On Sat, Nov 13, 2010 at 02:15:19PM +0800, Shaohui Zheng wrote:
> diff --git a/mm/Kconfig b/mm/Kconfig
> index f0fb912..a01a679 100644
> --- a/mm/Kconfig
> +++ b/mm/Kconfig
> @@ -147,6 +147,21 @@ config MEMORY_HOTREMOVE
> depends on MEMORY_HOTPLUG && ARCH_ENABLE_MEMORY_HOTREMOVE
> depends on MIGRATION
>
> +config NUMA_HOTPLUG_EMU
> + bool "NUMA hotplug emulator"
> + depends on X86_64 && NUMA && MEMORY_HOTPLUG
> +
How about something like a dummy ARCH_ENABLE_NUMA_EMU instead? The
architecture bits here are quite minimal, and with some further
abstraction it should be trivially supportable by any platform with NUMA
and memblock support (and most (all?) of the NUMA capable platforms
support memblock now already).