Skip to contentSkip to conversation

(memory hotplug) Fix compile error for i386 with NUMA config

PATCH In review Yasunori Goto 2 participants 5 messages Dec 9, 2006 – Dec 11, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 (memory hotplug) Fix compile error for i386 with NUMA config

Hello.

This patch is to fix compile error when config memory hotplug with numa on i386.

The cause of compile error was missing of arch_add_memory(), remove_memory(), and memory_add_physaddr_to_nid() when NUMA config is on.

This is for 2.6.19, and I tested no compile error of it.

Please apply.

---
 arch/i386/mm/discontig.c |   17 +++++++++++++++++
 arch/i386/mm/init.c      |    4 +---
 2 files changed, 18 insertions(+), 3 deletions(-)
arch/i386/mm/init.c +1 -3
Index: linux-2.6.19/arch/i386/mm/init.c
===================================================================
@@ -681,10 +681,9 @@
  * memory to the highmem for now.
  */
 #ifdef CONFIG_MEMORY_HOTPLUG
-#ifndef CONFIG_NEED_MULTIPLE_NODES
 int arch_add_memory(int nid, u64 start, u64 size)
 {
-	struct pglist_data *pgdata = &contig_page_data;
+	struct pglist_data *pgdata = NODE_DATA(nid);
 	struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 	unsigned long start_pfn = start >> PAGE_SHIFT;
 	unsigned long nr_pages = size >> PAGE_SHIFT;
@@ -697,7 +696,6 @@
 	return -EINVAL;
 }
 #endif
-#endif
 
 kmem_cache_t *pgd_cache;
 kmem_cache_t *pmd_cache;
arch/i386/mm/discontig.c +17 -0
Index: linux-2.6.19/arch/i386/mm/discontig.c
===================================================================
@@ -405,3 +405,20 @@
 	totalram_pages += totalhigh_pages;
 #endif
 }
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+/* This is the case that there is no _PXM on DSDT for added memory */
+int memory_add_physaddr_to_nid(u64 addr)
+{
+	int nid;
+	unsigned long pfn = addr >> PAGE_SHIFT;
+
+	for (nid = 0; nid < MAX_NUMNODES; nid++){
+		if (node_start_pfn[nid] <= pfn &&
+		    pfn < node_end_pfn[nid])
+			return nid;
+	}
+
+	return 0;
+}
+#endif
Signature · 1 line
Yasunori Goto 
#1 lore raw
Quoted 36 lines · Yasunori Goto

Hello.

This patch is to fix compile error when config memory hotplug with numa on i386.

The cause of compile error was missing of arch_add_memory(), remove_memory(), and memory_add_physaddr_to_nid() when NUMA config is on.

This is for 2.6.19, and I tested no compile error of it.

Please apply.

---
 arch/i386/mm/discontig.c |   17 +++++++++++++++++
 arch/i386/mm/init.c      |    4 +---
 2 files changed, 18 insertions(+), 3 deletions(-)

Index: linux-2.6.19/arch/i386/mm/init.c
===================================================================

arch/i386/mm/init.c +1 -2
@@ -681,10 +681,9 @@
  * memory to the highmem for now.
  */
 #ifdef CONFIG_MEMORY_HOTPLUG
-#ifndef CONFIG_NEED_MULTIPLE_NODES
 int arch_add_memory(int nid, u64 start, u64 size)
 {
-	struct pglist_data *pgdata = &contig_page_data;
+	struct pglist_data *pgdata = NODE_DATA(nid);
 	struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 	unsigned long start_pfn = start >> PAGE_SHIFT;
 	unsigned long nr_pages = size >> PAGE_SHIFT;

arch/i386/mm/init.c line 701

Quoted 7 lines
arch/i386/mm/init.c +0 -1
@@ -697,7 +696,6 @@
 	return -EINVAL;
 }
 #endif
-#endif
 
 kmem_cache_t *pgd_cache;
 kmem_cache_t *pmd_cache;

The reason for the #ifndef CONFIG_NEED_MULTIPLE_NODES check seems to solely exist for excluding the NUMA case, so it doesn't appear as though this is the correct fix since your changelog indicates a compile problem with a NUMA build. This hypothesis is supported by the comment which conveniently appears just before arch_add_memory which _explicitly_ states that the following is for non-NUMA cases.

Quoted 16 lines

Index: linux-2.6.19/arch/i386/mm/discontig.c
===================================================================

arch/i386/mm/discontig.c +7 -0
@@ -405,3 +405,20 @@
 	totalram_pages += totalhigh_pages;
 #endif
 }
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+/* This is the case that there is no _PXM on DSDT for added memory */
+int memory_add_physaddr_to_nid(u64 addr)
+{
+	int nid;
+	unsigned long pfn = addr >> PAGE_SHIFT;

arch/i386/mm/discontig.c line 424

Quoted 10 lines
arch/i386/mm/discontig.c +10 -0
+
+	for (nid = 0; nid < MAX_NUMNODES; nid++){
+		if (node_start_pfn[nid] <= pfn &&
+		    pfn < node_end_pfn[nid])
+			return nid;
+	}
+
+	return 0;
+}
+#endif

memory_add_physaddr_to_nid is only declared as extern in include/linux/memory_hotplug.h in the CONFIG_NUMA case so this also doesn't appear as the correct fix but probably worked for your compile since you had CONFIG_MEMORY_HOTPLUG enabled.

David

#2 lore raw

Hi David-san.

Quoted 43 lines · Yasunori Goto
Quoted 42 lines

Hello.

This patch is to fix compile error when config memory hotplug with numa on i386.

The cause of compile error was missing of arch_add_memory(), remove_memory(), and memory_add_physaddr_to_nid() when NUMA config is on.

This is for 2.6.19, and I tested no compile error of it.

Please apply.

---
 arch/i386/mm/discontig.c |   17 +++++++++++++++++
 arch/i386/mm/init.c      |    4 +---
 2 files changed, 18 insertions(+), 3 deletions(-)

Index: linux-2.6.19/arch/i386/mm/init.c
===================================================================

arch/i386/mm/init.c +1 -3
@@ -681,10 +681,9 @@
  * memory to the highmem for now.
  */
 #ifdef CONFIG_MEMORY_HOTPLUG
-#ifndef CONFIG_NEED_MULTIPLE_NODES
 int arch_add_memory(int nid, u64 start, u64 size)
 {
-	struct pglist_data *pgdata = &contig_page_data;
+	struct pglist_data *pgdata = NODE_DATA(nid);
 	struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 	unsigned long start_pfn = start >> PAGE_SHIFT;
 	unsigned long nr_pages = size >> PAGE_SHIFT;
@@ -697,7 +696,6 @@
 	return -EINVAL;
 }
 #endif
-#endif
 
Quoted 9 lines

arch/i386/mm/init.c line 701

Quoted 2 lines
arch/i386/mm/init.c +0 -0
 kmem_cache_t *pgd_cache;
 kmem_cache_t *pmd_cache;

The reason for the #ifndef CONFIG_NEED_MULTIPLE_NODES check seems to solely exist for excluding the NUMA case, so it doesn't appear as though this is the correct fix since your changelog indicates a compile problem with a NUMA build. This hypothesis is supported by the comment which conveniently appears just before arch_add_memory which _explicitly_ states that the following is for non-NUMA cases.

No.
Other arch's arch_add_memory() and remove_memory() have been already used for NUMA case too. But i386 didn't do it because just contig_page_data is used.
Current NODE_DATA() macro is defined both case appropriately. So, this #ifdef is redundant now.

(See: http://marc.theaimsgroup.com/?l=linux-mm&m=116494983531221&w=2)

Quoted 24 lines
Quoted 24 lines

Index: linux-2.6.19/arch/i386/mm/discontig.c
===================================================================

arch/i386/mm/discontig.c +15 -0
@@ -405,3 +405,20 @@
 	totalram_pages += totalhigh_pages;
 #endif
 }
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+/* This is the case that there is no _PXM on DSDT for added memory */
+int memory_add_physaddr_to_nid(u64 addr)
+{
+	int nid;
+	unsigned long pfn = addr >> PAGE_SHIFT;
+
+	for (nid = 0; nid < MAX_NUMNODES; nid++){
+		if (node_start_pfn[nid] <= pfn &&
+		    pfn < node_end_pfn[nid])
+			return nid;
+	}
+
+	return 0;
Quoted 8 lines

arch/i386/mm/discontig.c line 424

Quoted 3 lines
arch/i386/mm/discontig.c +2 -0
+}
+#endif

memory_add_physaddr_to_nid is only declared as extern in include/linux/memory_hotplug.h in the CONFIG_NUMA case so this also doesn't appear as the correct fix but probably worked for your compile since you had CONFIG_MEMORY_HOTPLUG enabled.

memory_add_physaddr_to_nid() is used for memory hotplug to find node id for new memory when ACPI's DSDT doesn't define _PXM for new memory. So, when CONFIG_MEMORY_HOTPLUG is not set, this function is not used.

Bye.

Signature · 1 line
Yasunori Goto 
#3 lore raw
Quoted 7 lines · Yasunori Goto

No.
Other arch's arch_add_memory() and remove_memory() have been already used for NUMA case too. But i386 didn't do it because just contig_page_data is used.
Current NODE_DATA() macro is defined both case appropriately. So, this #ifdef is redundant now.

Then I assume the comment directly above this change is also redundant since it explicitly states that the following code is for the non-NUMA case.

David

#4 lore raw
Quoted 9 lines
Quoted 8 lines

No.
Other arch's arch_add_memory() and remove_memory() have been already used for NUMA case too. But i386 didn't do it because just contig_page_data is used.
Current NODE_DATA() macro is defined both case appropriately. So, this #ifdef is redundant now.

Quoted 3 lines

Then I assume the comment directly above this change is also redundant since it explicitly states that the following code is for the non-NUMA case.

Ahhhhh. Yes indeed.

Here is fixed patch. Thanks for your comment.

Bye.

-------

This patch is to fix compile error when config memory hotplug with numa on i386.

The cause of compile error was missing of arch_add_memory(), remove_memory(), and memory_add_physaddr_to_nid().

This is for 2.6.19, and I tested no compile error of it.

Please apply.

---
 arch/i386/mm/discontig.c |   17 +++++++++++++++++
 arch/i386/mm/init.c      |    9 +--------
 2 files changed, 18 insertions(+), 8 deletions(-)
arch/i386/mm/init.c +1 -8
Index: linux-2.6.19/arch/i386/mm/init.c
===================================================================
@@ -675,16 +675,10 @@
 #endif
 }
 
-/*
- * this is for the non-NUMA, single node SMP system case.
- * Specifically, in the case of x86, we will always add
- * memory to the highmem for now.
- */
 #ifdef CONFIG_MEMORY_HOTPLUG
-#ifndef CONFIG_NEED_MULTIPLE_NODES
 int arch_add_memory(int nid, u64 start, u64 size)
 {
-	struct pglist_data *pgdata = &contig_page_data;
+	struct pglist_data *pgdata = NODE_DATA(nid);
 	struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 	unsigned long start_pfn = start >> PAGE_SHIFT;
 	unsigned long nr_pages = size >> PAGE_SHIFT;
@@ -697,7 +691,6 @@
 	return -EINVAL;
 }
 #endif
-#endif
 
 kmem_cache_t *pgd_cache;
 kmem_cache_t *pmd_cache;
arch/i386/mm/discontig.c +17 -0
Index: linux-2.6.19/arch/i386/mm/discontig.c
===================================================================
@@ -405,3 +405,20 @@
 	totalram_pages += totalhigh_pages;
 #endif
 }
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+/* This is the case that there is no _PXM on DSDT for added memory */
+int memory_add_physaddr_to_nid(u64 addr)
+{
+	int nid;
+	unsigned long pfn = addr >> PAGE_SHIFT;
+
+	for (nid = 0; nid < MAX_NUMNODES; nid++){
+		if (node_start_pfn[nid] <= pfn &&
+		    pfn < node_end_pfn[nid])
+			return nid;
+	}
+
+	return 0;
+}
+#endif
Signature · 1 line
Yasunori Goto