NUMA Hotplug emulator

x86: infrastructure of NUMA hotplug emulation

NUMA hotplug emulator introduces a new node state N_HIDDEN to identify the fake offlined node. It firstly hides RAM via E820 table and then emulates fake offlined nodes with the hidden RAM.

After system bootup, user is able to hotplug-add these offlined nodes, which is just similar to a real hardware hotplug behavior.

Using boot option "numa=hide=N*size" to fake offlined nodes:
- N is the number of hidden nodes
- size is the memory size (in MB) per hidden node.

OPEN: Kernel might use part of hidden memory region as RAM buffer,
now emulator directly hide 128M extra space to workaround this issue. Any better way to avoid this conflict?

Signed-off-by Haicheng Li <[email protected]>

Signed-off-by Shaohui Zheng <[email protected]>

---

arch/x86/mm/numa_64.c +96 -3

index 8948f47..3e0d94d 100644
@@ -307,6 +307,87 @@ void __init numa_init_array(void)
 	}
 }
 
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+static char *hp_cmdline __initdata;
+static struct bootnode *hidden_nodes;
+static u64 hp_start, hp_end;
+static long hidden_num, hp_size;
+
+int hotadd_hidden_nodes(int nid)
+{
+	int ret;
+
+	if (!node_hidden(nid))
+		return -EINVAL;
+
+	ret = add_memory(nid, hidden_nodes[nid].start,
+			 hidden_nodes[nid].end - hidden_nodes[nid].start);
+	if (!ret) {
+		node_clear_hidden(nid);
+		return 0;
+	} else {
+		return -EEXIST;
+	}
+}
+
+static void __init numa_hide_nodes(void)
+{
+	char *c;
+	int ret;
+
+	c = strchr(hp_cmdline, '*');
+	if (!c)
+		return;
+	else
+		*c = '\0';
+	ret = strict_strtol(hp_cmdline, 0, &hidden_num);
+	if (ret == -EINVAL)
+		return;
+	ret = strict_strtol(c + 1, 0, &hp_size);
+	if (ret == -EINVAL)
+		return;
+	hp_size <<= 20;
+
+	hp_start = e820_hide_mem(hidden_num * hp_size);
+	if (hp_start <= 0) {
+		printk(KERN_ERR "Hide too much memory, disable node hotplug emualtion.");
+		hidden_num = 0;
+		return;
+	}
+
+	hp_end = hp_start + hidden_num * hp_size;
+
+	/* leave 128M space for possible RAM buffer usage later
+	 * any other better way to avoid this conflict?
+	 */
+	e820_hide_mem(128*1024*1024);
+}
+
+static void __init numa_hotplug_emulation(void)
+{
+	int i, num_nodes = 0;
+
+	for_each_online_node(i)
+		if (i > num_nodes)
+			num_nodes = i;
+
+	i = num_nodes + hidden_num;
+	if (!hidden_nodes) {
+		hidden_nodes = alloc_bootmem(sizeof(struct bootnode) * i);
+		memset(hidden_nodes, 0, sizeof(struct bootnode) * i);
+	}
+
+	if (hidden_num)
+		for (i = 0; i < hidden_num; i++) {
+			int nid = num_nodes + i + 1;
+			node_set(nid, node_possible_map);
+			hidden_nodes[nid].start = hp_start + hp_size * i;
+			hidden_nodes[nid].end = hp_start + hp_size * (i+1);
+			node_set_hidden(nid);
+		}
+}
+#endif /* CONFIG_NODE_HOTPLUG_EMU */
+
 #ifdef CONFIG_NUMA_EMU
 /* Numa emulation */
 static struct bootnode nodes[MAX_NUMNODES] __initdata;

@@ -661,7 +742,7 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
 
 #ifdef CONFIG_NUMA_EMU
 	if (cmdline && !numa_emulation(start_pfn, last_pfn, acpi, k8))
-		return;
+		goto done;
 	nodes_clear(node_possible_map);
 	nodes_clear(node_online_map);
 #endif

@@ -669,14 +750,14 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
 #ifdef CONFIG_ACPI_NUMA
 	if (!numa_off && acpi && !acpi_scan_nodes(start_pfn << PAGE_SHIFT,
 						  last_pfn << PAGE_SHIFT))
-		return;
+		goto done;
 	nodes_clear(node_possible_map);
 	nodes_clear(node_online_map);
 #endif
 
 #ifdef CONFIG_K8_NUMA
 	if (!numa_off && k8 && !k8_scan_nodes())
-		return;
+		goto done;
 	nodes_clear(node_possible_map);
 	nodes_clear(node_online_map);
 #endif

@@ -696,6 +777,12 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
 		numa_set_node(i, 0);
 	e820_register_active_regions(0, start_pfn, last_pfn);
 	setup_node_bootmem(0, start_pfn << PAGE_SHIFT, last_pfn << PAGE_SHIFT);
+done:
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+	if (hidden_num)
+		numa_hotplug_emulation();
+#endif
+	return;
 }
 
 unsigned long __init numa_free_all_bootmem(void)

@@ -723,6 +810,12 @@ static __init int numa_setup(char *opt)
 	if (!strncmp(opt, "fake=", 5))
 		cmdline = opt + 5;
 #endif
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+	if (!strncmp(opt, "hide=", 5)) {
+		hp_cmdline = opt + 5;
+		numa_hide_nodes();
+	}
+#endif
 #ifdef CONFIG_ACPI_NUMA
 	if (!strncmp(opt, "noacpi", 6))
 		acpi_numa = -1;

include/linux/nodemask.h +11 -0

index dba35e4..ba0f82d 100644
@@ -371,6 +371,10 @@ static inline void __nodes_fold(nodemask_t *dstp, const nodemask_t *origp,
  */
 enum node_states {
 	N_POSSIBLE,		/* The node could become online at some point */
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+	N_HIDDEN,		/* The node is hidden at booting time, could be
+				 * onlined in run time */
+#endif
 	N_ONLINE,		/* The node is online */
 	N_NORMAL_MEMORY,	/* The node has regular memory */
 #ifdef CONFIG_HIGHMEM

@@ -470,6 +474,13 @@ static inline int num_node_state(enum node_states state)
 #define node_online(node)	node_state((node), N_ONLINE)
 #define node_possible(node)	node_state((node), N_POSSIBLE)
 
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+#define node_set_hidden(node)	   node_set_state((node), N_HIDDEN)
+#define node_clear_hidden(node)	   node_clear_state((node), N_HIDDEN)
+#define node_hidden(node)	node_state((node), N_HIDDEN)
+extern int hotadd_hidden_nodes(int nid);
+#endif
+
 #define for_each_node(node)	   for_each_node_state(node, N_POSSIBLE)
 #define for_each_online_node(node) for_each_node_state(node, N_ONLINE)
 

Signature · 2 lines

Thanks & Regards,
Shaohui

Quoted 2 lines · On Thu, 13 May 2010 19:45:44 +0800 Shaohui Zheng <[email protected]>…

x86: infrastructure of NUMA hotplug emulation

Hmm. do we have to create this for x86 only ?
Can't we live with lmb ? as

lmb_hide_node() or some.

IIUC, x86-version lmb is now under development.

Quoted 8 lines · KAMEZAWA Hiroyuki

Quoted 2 lines · On Thu, 13 May 2010 19:45:44 +0800 Shaohui Zheng <[email protected]>…

x86: infrastructure of NUMA hotplug emulation

Hmm. do we have to create this for x86 only ?
Can't we live with lmb ? as

Quoted 4 lines

lmb_hide_node() or some.

IIUC, x86-version lmb is now under development.

Indeed. There is very little x86-specific about this patch series at all except for the e820 bits and tying in the CPU topology. Most of what this series is doing wrapping around e820 could be done on top of LMB, which would also make it possible to use on non-x86 architectures.

Quoted 9 lines · KAMEZAWA Hiroyuki

Quoted 2 lines · On Thu, 13 May 2010 19:45:44 +0800 Shaohui Zheng <[email protected]>…

x86: infrastructure of NUMA hotplug emulation

Hmm. do we have to create this for x86 only ?
Can't we live with lmb ? as

Quoted 4 lines

lmb_hide_node() or some.

IIUC, x86-version lmb is now under development.

Indeed. There is very little x86-specific about this patch series at all except for the e820 bits and tying in the CPU topology. Most of what this series is doing wrapping around e820 could be done on top of LMB, which would also make it possible to use on non-x86 architectures.

x86: infrastructure of NUMA hotplug emulation

NUMA hotplug emulator introduces a new node state N_HIDDEN to identify the fake offlined node. It firstly hides RAM via E820 table and then emulates fake offlined nodes with the hidden RAM.

After system bootup, user is able to hotplug-add these offlined nodes, which is just similar to a real hardware hotplug behavior.

Using boot option "numa=hide=N*size" to fake offlined nodes:
- N is the number of hidden nodes
- size is the memory size (in MB) per hidden node.

OPEN: Kernel might use part of hidden memory region as RAM buffer,
now emulator directly hide 128M extra space to workaround this issue. Any better way to avoid this conflict?

Signed-off-by Haicheng Li <[email protected]>

Signed-off-by Shaohui Zheng <[email protected]>

---

arch/x86/mm/numa_64.c +96 -3

index 8948f47..3e0d94d 100644
@@ -307,6 +307,87 @@ void __init numa_init_array(void)
 	}
 }
 
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+static char *hp_cmdline __initdata;
+static struct bootnode *hidden_nodes;
+static u64 hp_start, hp_end;
+static long hidden_num, hp_size;
+
+int hotadd_hidden_nodes(int nid)
+{
+	int ret;
+
+	if (!node_hidden(nid))
+		return -EINVAL;
+
+	ret = add_memory(nid, hidden_nodes[nid].start,
+			 hidden_nodes[nid].end - hidden_nodes[nid].start);
+	if (!ret) {
+		node_clear_hidden(nid);
+		return 0;
+	} else {
+		return -EEXIST;
+	}
+}
+
+static void __init numa_hide_nodes(void)
+{
+	char *c;
+	int ret;
+
+	c = strchr(hp_cmdline, '*');
+	if (!c)
+		return;
+	else
+		*c = '\0';
+	ret = strict_strtol(hp_cmdline, 0, &hidden_num);
+	if (ret == -EINVAL)
+		return;
+	ret = strict_strtol(c + 1, 0, &hp_size);
+	if (ret == -EINVAL)
+		return;
+	hp_size <<= 20;
+
+	hp_start = e820_hide_mem(hidden_num * hp_size);
+	if (hp_start <= 0) {
+		printk(KERN_ERR "Hide too much memory, disable node hotplug emualtion.");
+		hidden_num = 0;
+		return;
+	}
+
+	hp_end = hp_start + hidden_num * hp_size;
+
+	/* leave 128M space for possible RAM buffer usage later
+	 * any other better way to avoid this conflict?
+	 */
+	e820_hide_mem(128*1024*1024);
+}
+
+static void __init numa_hotplug_emulation(void)
+{
+	int i, num_nodes = 0;
+
+	for_each_online_node(i)
+		if (i > num_nodes)
+			num_nodes = i;
+
+	i = num_nodes + hidden_num;
+	if (!hidden_nodes) {
+		hidden_nodes = alloc_bootmem(sizeof(struct bootnode) * i);
+		memset(hidden_nodes, 0, sizeof(struct bootnode) * i);
+	}
+
+	if (hidden_num)
+		for (i = 0; i < hidden_num; i++) {
+			int nid = num_nodes + i + 1;
+			node_set(nid, node_possible_map);
+			hidden_nodes[nid].start = hp_start + hp_size * i;
+			hidden_nodes[nid].end = hp_start + hp_size * (i+1);
+			node_set_hidden(nid);
+		}
+}
+#endif /* CONFIG_NODE_HOTPLUG_EMU */
+
 #ifdef CONFIG_NUMA_EMU
 /* Numa emulation */
 static struct bootnode nodes[MAX_NUMNODES] __initdata;

@@ -661,7 +742,7 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
 
 #ifdef CONFIG_NUMA_EMU
 	if (cmdline && !numa_emulation(start_pfn, last_pfn, acpi, k8))
-		return;
+		goto done;
 	nodes_clear(node_possible_map);
 	nodes_clear(node_online_map);
 #endif

@@ -669,14 +750,14 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
 #ifdef CONFIG_ACPI_NUMA
 	if (!numa_off && acpi && !acpi_scan_nodes(start_pfn << PAGE_SHIFT,
 						  last_pfn << PAGE_SHIFT))
-		return;
+		goto done;
 	nodes_clear(node_possible_map);
 	nodes_clear(node_online_map);
 #endif
 
 #ifdef CONFIG_K8_NUMA
 	if (!numa_off && k8 && !k8_scan_nodes())
-		return;
+		goto done;
 	nodes_clear(node_possible_map);
 	nodes_clear(node_online_map);
 #endif

@@ -696,6 +777,12 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
 		numa_set_node(i, 0);
 	e820_register_active_regions(0, start_pfn, last_pfn);
 	setup_node_bootmem(0, start_pfn << PAGE_SHIFT, last_pfn << PAGE_SHIFT);
+done:
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+	if (hidden_num)
+		numa_hotplug_emulation();
+#endif
+	return;
 }
 
 unsigned long __init numa_free_all_bootmem(void)

@@ -723,6 +810,12 @@ static __init int numa_setup(char *opt)
 	if (!strncmp(opt, "fake=", 5))
 		cmdline = opt + 5;
 #endif
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+	if (!strncmp(opt, "hide=", 5)) {
+		hp_cmdline = opt + 5;
+		numa_hide_nodes();
+	}
+#endif
 #ifdef CONFIG_ACPI_NUMA
 	if (!strncmp(opt, "noacpi", 6))
 		acpi_numa = -1;

include/linux/nodemask.h +4 -0

index dba35e4..ba0f82d 100644
@@ -371,6 +371,10 @@ static inline void __nodes_fold(nodemask_t *dstp, const nodemask_t *origp,
  */
 enum node_states {
 	N_POSSIBLE,		/* The node could become online at some point */
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+	N_HIDDEN,		/* The node is hidden at booting time, could be
+				 * onlined in run time */
+#endif
 	N_ONLINE,		/* The node is online */
 	N_NORMAL_MEMORY,	/* The node has regular memory */
 #ifdef CONFIG_HIGHMEM

@@ -470,6 +474,13 @@ static inline int num_node_state(enum node_states state)
 #define node_online(node)	node_state((node), N_ONLINE)
 #define node_possible(node)	node_state((node), N_POSSIBLE)
 

x86: infrastructure of NUMA hotplug emulation

NUMA hotplug emulator introduces a new node state N_HIDDEN to identify the fake offlined node. It firstly hides RAM via E820 table and then emulates fake offlined nodes with the hidden RAM.

After system bootup, user is able to hotplug-add these offlined nodes, which is just similar to a real hardware hotplug behavior.

Using boot option "numa=hide=N*size" to fake offlined nodes:
       - N is the number of hidden nodes
       - size is the memory size (in MB) per hidden node.

OPEN: Kernel might use part of hidden memory region as RAM buffer, now emulator directly hide 128M extra space to workaround this issue.  Any better way to avoid this conflict?

Signed-off-by Haicheng Li <[email protected]>

Signed-off-by Shaohui Zheng <[email protected]>

---

arch/x86/mm/numa_64.c +61 -0

index 8948f47..3e0d94d 100644
@@ -307,6 +307,87 @@ void __init numa_init_array(void)
        }
  }
 
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+static char *hp_cmdline __initdata;
+static struct bootnode *hidden_nodes;
+static u64 hp_start, hp_end;
+static long hidden_num, hp_size;
+
+int hotadd_hidden_nodes(int nid)
+{
+       int ret;
+
+       if (!node_hidden(nid))
+               return -EINVAL;
+
+       ret = add_memory(nid, hidden_nodes[nid].start,
+                        hidden_nodes[nid].end - hidden_nodes[nid].start);
+       if (!ret) {
+               node_clear_hidden(nid);
+               return 0;
+       } else {
+               return -EEXIST;
+       }
+}
+
+static void __init numa_hide_nodes(void)
+{
+       char *c;
+       int ret;
+
+       c = strchr(hp_cmdline, '*');
+       if (!c)
+               return;
+       else
+               *c = '\0';
+       ret = strict_strtol(hp_cmdline, 0, &hidden_num);
+       if (ret == -EINVAL)
+               return;
+       ret = strict_strtol(c + 1, 0, &hp_size);
+       if (ret == -EINVAL)
+               return;
+       hp_size <<= 20;
+
+       hp_start = e820_hide_mem(hidden_num * hp_size);
+       if (hp_start <= 0) {
+               printk(KERN_ERR "Hide too much memory, disable node hotplug emualtion.");
+               hidden_num = 0;
+               return;
+       }
+
+       hp_end = hp_start + hidden_num * hp_size;
+
+       /* leave 128M space for possible RAM buffer usage later
+        * any other better way to avoid this conflict?
+        */
+       e820_hide_mem(128*1024*1024);
+}
+
+static void __init numa_hotplug_emulation(void)
+{
+       int i, num_nodes = 0;
+
+       for_each_online_node(i)

arch/x86/mm/numa_64.c +25 -3

+               for (i = 0; i < hidden_num; i++) {
+                       int nid = num_nodes + i + 1;
+                       node_set(nid, node_possible_map);
+                       hidden_nodes[nid].start = hp_start + hp_size * i;
+                       hidden_nodes[nid].end = hp_start + hp_size * (i+1);
+                       node_set_hidden(nid);
+               }
+}
+#endif /* CONFIG_NODE_HOTPLUG_EMU */
+
  #ifdef CONFIG_NUMA_EMU
  /* Numa emulation */
  static struct bootnode nodes[MAX_NUMNODES] __initdata;

@@ -661,7 +742,7 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
 
  #ifdef CONFIG_NUMA_EMU
        if (cmdline && !numa_emulation(start_pfn, last_pfn, acpi, k8))
-               return;
+               goto done;
        nodes_clear(node_possible_map);
        nodes_clear(node_online_map);
  #endif

@@ -669,14 +750,14 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
  #ifdef CONFIG_ACPI_NUMA
        if (!numa_off && acpi && !acpi_scan_nodes(start_pfn << PAGE_SHIFT,
                                                  last_pfn << PAGE_SHIFT))
-               return;
+               goto done;
        nodes_clear(node_possible_map);
        nodes_clear(node_online_map);
  #endif
 
  #ifdef CONFIG_K8_NUMA
        if (!numa_off && k8 && !k8_scan_nodes())
-               return;
+               goto done;
        nodes_clear(node_possible_map);
        nodes_clear(node_online_map);
  #endif

@@ -696,6 +777,12 @@ void __init initmem_init(unsigned long start_pfn, unsigned long last_pfn,
                numa_set_node(i, 0);
        e820_register_active_regions(0, start_pfn, last_pfn);
        setup_node_bootmem(0, start_pfn << PAGE_SHIFT, last_pfn << PAGE_SHIFT);
+done:
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+       if (hidden_num)
+               numa_hotplug_emulation();
+#endif
+       return;
  }
 
  unsigned long __init numa_free_all_bootmem(void)

@@ -723,6 +810,12 @@ static __init int numa_setup(char *opt)
        if (!strncmp(opt, "fake=", 5))
                cmdline = opt + 5;
  #endif
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+       if (!strncmp(opt, "hide=", 5)) {
+               hp_cmdline = opt + 5;
+               numa_hide_nodes();
+       }
+#endif
  #ifdef CONFIG_ACPI_NUMA
        if (!strncmp(opt, "noacpi", 6))
                acpi_numa = -1;

include/linux/nodemask.h +0 -0

index dba35e4..ba0f82d 100644
@@ -371,6 +371,10 @@ static inline void __nodes_fold(nodemask_t *dstp, const nodemask_t *origp,
  */
  enum node_states {

       N_POSSIBLE,             /* The node could become online at some point */

include/linux/nodemask.h +4 -0

+#ifdef CONFIG_NODE_HOTPLUG_EMU
+       N_HIDDEN,               /* The node is hidden at booting time, could be
+                                * onlined in run time */
+#endif

       N_ONLINE,               /* The node is online */ N_NORMAL_MEMORY,        /* The node has regular memory */ #ifdef CONFIG_HIGHMEM

include/linux/nodemask.h +7 -1

@@ -470,6 +474,13 @@ static inline int num_node_state(enum node_states state)
  #define node_online(node)      node_state((node), N_ONLINE)
  #define node_possible(node)    node_state((node), N_POSSIBLE)
 
+#ifdef CONFIG_NODE_HOTPLUG_EMU
+#define node_set_hidden(node)     node_set_state((node), N_HIDDEN)
+#define node_clear_hidden(node)           node_clear_state((node), N_HIDDEN)
+#define node_hidden(node)      node_state((node), N_HIDDEN)
+extern int hotadd_hidden_nodes(int nid);
+#endif
+
  #define for_each_node(node)       for_each_node_state(node, N_POSSIBLE)
  #define for_each_online_node(node) for_each_node_state(node, N_ONLINE)
 
--

Thanks & Regards,
Shaohui

Quoted 1 line

+ if (hidden_num)

if (hidden_num) is not required, as next line's for statement is also doing the same thing.

Thanks,
--
Jaswinder Singh.

Quoted 8 lines

arch/x86/mm/numa_64.c +8 -0

+               for (i = 0; i < hidden_num; i++) {
+                       int nid = num_nodes + i + 1;
+                       node_set(nid, node_possible_map);
+                       hidden_nodes[nid].start = hp_start + hp_size * i;
+                       hidden_nodes[nid].end = hp_start + hp_size * (i+1);
+                       node_set_hidden(nid);
+               }
+}